HGHB

Progress in Materials Science Vol. 35, pp.
205-418, 1991
Printed in Great Britain. All rights reserved.
0079-6425/91 $0.00 + .50 1991 Pergamon Press plc
SEMICONDUCTORS
FOR
SOLAR
CELL
APPLICATIONS
Hans Joachim M6ller Department of Materials Science and Engineering, Case Western Reserve University, Cleveland, OH 44106, U.S.A.
CONTENTS
1. INTRODUCTION
1.1. Historical Background of the Solar Cell Development 1.2. Technical Concepts for Photovoltaic Solar Energy Conversion 1.3. Material Aspects of Solar Cell Semiconductors
2. PHYSICALPRINCIPLES OF PHOTOVOLTAIC ENERGY CONVERSION
2.1. Solar Spectrum in Space and at the Earth's Surface 2.2. Absorption of Light in Semiconductors 2.3. Principle of the p-n Junction Solar Cell 2.4. Fundamental Material Parameters 2.4.1. Mobility of charge carriers 2.4.2. Bulk recombination processes 2.4.3. Surface recombination 2.4.4. Recombination at extended bulk lattice defects 2.5. Efficiency Requirements for the Solar Cell Material 2.5.1. Short-circuit current 2.5.2. Open-circuit voltage 2.5.3. Temperature effect
3. TECHNOLOGY OF SOLAR CELL DEVICES
3.1. 3.2. 3.3. 3.4. 3.5. 3.6.
Single-Junction Devices Heterojunction Devices Schottky Barrier and MIS Solar Cells Thin Film Solar Cells Optical Concentration Multi-Junction Structures
4. STRUCTURAL AND PHYSICALPROPERTIES OF LATTICE DEFECTS IN SEMICONDUCTORS
4.1. Diffusion and Electrical Behavior of Point Defects 4.1.1. Intrinsic point defects 4.1.2. Carbon and oxygen 4.1.3. Transition metal impurities 4.1.4. Hydrogen passivation of deep and shallow level defects 4.2. Radiation Induced Defects 4,3. Dislocations 4.3.1. Dislocation velocity 4.3,2. Dislocation multiplication 4.4. Grain Boundaries 4.4.1. Atomic structure and electronic states 4.4.2. Electrical activity of grain boundaries 4.5. Interaction of Impurities with Grain Boundaries 4.5.1. Segregation and precipitation 4.5.2. Interface diffusion 4.6. Electrical Transport Properties of Polycrystalline Semiconductors
5. SINGLE CRYSTAL AND POLYCRYSTALLINESILICON
5.1. Cz and Fz Grown Silicon 5.2. Polycrystalline Silicon 5.2.1. Ingot technique 5.2.2. Ribbon and sheet growth techniques
JPMS 35-3/4--A
206 206 207 211 213 213 214 218 224 225 228 233 237 245 247 248 248 249 250 257 259 261 264 265 267 268 268 277 284 291 296 302 303 306 308 309 318 329 330 338 340 347 347 349 350 361
205
206
PROGRESS
IN MATERIALS
SCIENCE
6. SINGLE CRYSTAL AND EPITAXlAL GALLIUM ARSENIDE
6.1. Single Crystal Growth Techniques 6.2. Epitaxial Thin Films of GaAs and Related Ternary Compounds
7. POLYCRYSTALLINETHIN FILMS 7.1. I I - V I Compounds
7. I. 1. Basic properties 7.1.2. Thin film preparation 7.1.3. Electrical properties 7.2. Multinary Compounds
7.3.
Copper Ternary Compounds
7.3.1. Film and crystal growth techniques 7.3.2. Band structure and opticalproperties 7.3.3. Lattice defects and electrical properties 7.3.4. Interface properties and oxidation behavior 8. AMORPHOUSTHIN FILMS
8.1. Structure and Electro-Optical Properties of Amorphous Silicon 8.1.1. Band gap states 8.1.2. Doping in a-Si :H 8.1.3. Electrical transport 8.1.4. Optical absorption 8.1.5. Amorphous silicon alloys 8.2. Stability and Light Induced Degradation
REFERENCES
366 367 370 373 373 375 375 376 377 379 381 382 385 393 395 396 396 399 402 406 4O8 408 412
1. INTRODUCTION
1.1. Historical Background of the Solar Cell Development

The conversion of sunlight directly into electricity using the electronic properties of suitable materials is the most elegant energy conversion process. Being a laboratory curiosity for more than a hundred years the solar cell technology has seen an enormous development during the last three decades initially for providing electrical power for spacecraft and more recently for terrestrial systems. The driving force for the technological development is the realization that the traditional fossil energy resources coal, oil and gas are not only limited but contribute also to unpredictable and possibly irreversible climate changes in the near future through the emission of carbon dioxide. The increasing concern for environmental pollution problems in industrialized countries has also discredited nuclear power as a long term alternative energy concept. From this perspective the use of sunlight offers a conceivable alternative to the world wide energy problems. Solar cells utilize the photovoltaic effect for their operation which was initially discovered in 1839 by Becquerel who studied the behavior of solids in electrolyte solutions. He observed that when metal plates immersed in a suitable electrolyte were exposed to sun light, a small voltage and current were produced. The first solid state materials that showed a significant light dependent voltage between two contacts were selenium in 1876 and later cuprous oxide 0-5) which indicated already that semiconductors would become the most promising class of materials for photovoltaic energy conversion The technological development began with the development of a diffused silicon p-n junction in 1954, ~6)a forerunner of the present silicon solar cell, which converted light into electricity with a reasonable efficiency. Almost coincident with the beginning of the silicon solar cell technology was the first development of cuprous sulphide/cadmium sulphide heterojunctions which served as the basis for the intense research on thin film solar cell devices. ~7) The initial applications were of a small scale and the first real impact of solar cells was only realized with the advent of space exploration. The demand for a reliable long lasting power source was the major reason for the application of solar cells and by 1958 the first silicon solar cells were used in spacecraft.
SEMICONDUCTORS
FOR SOLAR CELL APPLICATIONS
207
The interest in solar cells as an alternative energy source for terrestrial applications awoke in the mid 1970s after the political crisis in the Middle East and the oil embargo, and the realization of limitations in fossil fuel sources. The increased research efforts resulted not only in the further improvement of the efficiencies of silicon solar cells and a considerable reduction of the energy costs, but also in the development of new photovoltaic materials and devices. The increase in production volume and new cost efficient solar cell technologies rendered this achievement possible. The cost target for electricity from a photovoltaic plant operating for 30 years has been established in 1986 to be equal to about 0.06 SUS/kWhr. It was estimated that this requires module efficiencies in the range of 15-20% for a fiat plate panel system and 25-30% for a system which operates under concentrated sun light (for a reviewCS)). With the beginning of the 80s new solar cell materials and innovative device concepts have been evaluated at the pilot production stage and currently a number of promising options for a future development are available. However, the expected breakthrough of the solar technology as a competitive energy source could not be achieved, mainly because the required efficiencies could only be obtained for laboratory solar cells so far. Part of the barrier to the more widespread use of the photovoltaic technology is also the still rather expensive fossil and nuclear power energy. A comprehensive comparison of an alternative energy system in general is not trivial and has to take into account the full costs of the energy system to society such as environmental and climatic effects, hidden subsidies or production of intermediate goods. These external costs are only beginning to be considered, but have already some impact on the energy policies of several European countries. Though the solar cell technology is still not mature and expectations are that it will contribute not more than 5-7% of the total energy consumptions over the next decade, it offers a realistic and promising energy technology for the future. A photovoltaic system consists of several components beginning from the solar cell to a module, the array of modules, and the entire plant. If one examines the individual components the major hurdle to technological feasibility remains with the basic component, the solar cell itself. Therefore research into photovoltaic alternatives is imperative to make the technology competitive. This includes the development of low cost techniques, higher efficiency cells utilizing new materials and cell concepts, and thin films which require less material. Inherent to any option is successful tailoring of the semiconductor material and the control of the electro-optical properties during each processing step. A number of the various cell concepts will be discussed in the following, with a particular emphasis on the material properties which eventually determine the solar cell performance.
1.2. Technical Concepts for Photovoltaic Solar Energy Conversion

Semiconductors have emerged as the most promising group of materials which can convert sunlight directly into electrical energy. They utilize the fundamental physical process that a photon that penetrates into the semiconductor and is absorbed can generate electron-hole pairs. Because of their opposite charges they can be separated by an internal electrical field and collected at two contacts thus giving rise to a voltage and photocurrent if the two contacts are connected externally. In semiconductors internal electric fields occur in connection with space charges at junctions and a variety of technological concepts are used to produce a built-in voltage. The most widely used device principle is the operation of a solar cell as a diode or p - n junction. Alternative concepts are heterojunction devices where the materials on either side of the junction are different semiconductors. Less common structures for a solar cell utilize metal-semiconductor junctions (Schottky contacts) or metal-insulator-
208
P R O G R E S S IN M A T E R I A L S SCIENCE
semiconductor (MIS) structures, and recently the latter principle has been successfully used for a silicon based solar cell with efficiencies of about 17%. c9~ These are conventional device structures which are also used for other microelectronic devices and thus benefit from the rapid development of the semiconductor technology. It shall be mentioned, however, that other solar cell concepts are also pursued. For instance, in photoelectrochemical cells the internal electric field can be created at the surface of a semiconductor which is brought into contact with an electrolyte. Though it requires only minimal processing to produce such a device and energy-conversion efficiencies of about 12% have been reported ~ there are still major technological problems that have to be solved like for instance the photoenhanced corrosion of the semiconductor. Nonetheless this technology may become important in connection with the hydrogen technology, the photoelectrolytic decomposition of water. The electrochemical cell technology and its problems, however, are beyond the scope of this paper and will not be considered in the following. In general, in semiconductor junction solar cells the maximum voltage (or open circuit voltage which will be introduced in Chap. 2.3) that can be obtained is limited by the band gap of the material whereas the generated photocurrent is determined by optical excitation and recombination process. The generation of electron-hole pairs occurs if the energy of the incident photon is sufficient to excite electrons into the conduction band of the semiconductor. Photons with lower energies do not contribute to the photocurrent whereas photons with energies larger than the band gap generate electrons which transform part of their energy into heat until they occupy energy states at the conduction band edge. Therefore materials with small band gaps not only waste the energy of part of the photons but have also a small open circuit voltage which in addition reduces the energy-conversion efficiency of the solar cell as will be discussed in Chap. 2.5. Consequently for a single-junction cell photons with energies around the band gap energy are most efficiently utilized. Theoretical calculations show c~ that ~ the optimum band gap energy is approximately 1.5 eV with respect to the spectral energy distribution of the terrestrial sunlight (Chap. 2.1-2.2). The research has consequently focused on semiconductors with band gap energies close to the optimum value. Historically silicon was the first commercially used solar cell material and is also the most extensively studied one although its band gap energy of 1.12 eV (for the crystalline material) is somewhat too small for reaching the theoretical maximum conversion efficiency. The advantage is the mature processing technology because of its use for microelectronic devices, the large abundance in the crust of the earth, and from the perspective of environmental concerns the non-toxicity. Monocrystalline silicon as an indirect semiconductor has a rather low absorption coefficient which requires a relatively large volume to absorb the light ( ~ 1000 #m thickness) compared to the width of the space charge region where the charge carriers are separated. Therefore recombination mechanisms which limit the lifetime of electrons or holes before they can diffuse to the junction region are of great importance. It will be discussed in Chap. 2.3 that high purity silicon is necessary for maximum efficiency cells and after 30 years of solar cell research the optimization of material properties and of the cell design has resulted in efficiencies of 23% for silicon under normal sunlight. 02~ Such high efficiencies, however, are only obtained in laboratory experiments on small cells with areas of a few cm 2 and a large gap still exists between the lower efficiencies of about 10-13% for commercial cells in modules. Another drawback of high efficiency concepts is that the production costs for the solar cells because of the sophisticated processing technology are still too high so that despite the gain in efficiency the energy costs of photovoltaic power probably cannot become competitive in the near future.
SEMICONDUCTORS
209
The cost problem for monocrystalline silicon has led to the development of low cost technologies at the expense of the efficiency of the solar cell. Inexpensive processes so far only allow the production of low purity, polycrystalline ("terrestrial solar grade") silicon. Nonetheless this technology has been established commercially in particular areas of the energy market and has reduced the energy price to about $2/KWhr mainly due to scaling-up the production. Most commonly polycrystalline silicon is produced by casting and solidification of the melt in a crucible which produces blocks (ingots) of about 20 kg (see Chap. 5.2). This technique requires the subsequent sawing of the ingots into wafers of 10 x 10 cm 2 and a thickness of about 400 #m and results in the waste of aboul 40% of the silicon. Efficiencies of about 17% have been obtained with cast polycrystalline silicon wafers (~3) by using high efficiency concepts for the solar cell design (Chap. 3). Several strategies have been pursued in the past to further improve the poly-silicon solar cell efficiency and as far as they are related to material issues will be addressed in the next sections. One technological way is to reduce the cell thickness and to employ light trapping concepts (Chap. 3). (~4) Then the recombination probability of light-generated electron-hole pairs is lower and the cell performance becomes less dependent on the minority carrier lifetime. A further reduction of the costs is expected from technologies that avoid the slicing step by growing the silicon directly in the final shape in form of sheets or ribbons. A number of continuous ribbon growth techniques which will be discussed in Chap. 5.2, are currently developed (144s) and will allow the mass production of solar cells in the future. However, still tremendous material problems have to be solved before solar cell efficiencies sufficient for a competitive commercial product can be achieved. Nonetheless, the technologies which have been developed for mono- and polycrystaUine silicon are the most advanced so far and continue to be improved. Parallel to the development of the solar cell technology based on crystalline silicon other concepts are pursued that try to avoid some of the fundamental disadvantages of silicon. A wide variety of compound semiconductors like GaAs, CdTe, InP, CdS has a band gap that is more suitable to the sunlight spectrum and therefore has the potential for higher efficiency. In fact, recent developments of GaAs solar ceils have demonstrated efficiencies of about 25% .~9-24)Despite this success most of the semiconductor compounds cannot yet be produced inexpensively enough or with a sufficient quality to be competitive in the near future. Most of the compounds are direct semiconductors with high absorption coefficients. Therefore only thin layers of about a few/~m thickness are necessary to absorb the light. The electron-hole pairs can be generated then close to or in the junction and the loss of charge carriers due to recombination is greatly reduced. Because of the technological developments in the microelectronics industry a number of thin film deposition techniques like chemical vapor deposition (CVD, OMCVD etc.) or molecular beam epitaxy (MBE) are available, which can also be used for the growth of high quality thin film solar ceils. Some of the particular problems connected with homo- and heteroepitaxial thin film growth will be discussed in Chap. 6.2 for gallium arsenide. Since the growth of high quality semiconductor films is a rather time consuming and expensive technology its application for a large scale solar cell production is still limited. In addition some of the compounds contain toxic elements which may prevent their widespread use because of environmental concerns. Nonetheless thin film solar cells have the advantage that they can be deposited (at least in principle) on different substrates (for instance polymers, steel or glass) and only a fraction of material is necessary compared to bulk solar cells. Since many of the processing techniques are also used in the microelectronic device production the integration of other electronic components is also
210
PROGRESS
IN MATERIALS
SCIENCE
feasible. Therefore thin film solar cells may have considerable advantages in the future if the technology and the material properties can be further improved. One strategy to cut down the production costs is to develop or apply less expensive techniques and obtain thin films of lower quality. One of the most advanced technologies in this direction is the deposition of amorphous silicon. Compared to monocrystalline silicon it has a different band structure and a high absorption so that it can be used as a thin film material. The photovoltaic properties have been explored for more than a decade now and amorphous silicon solar cells are already commercially used in small electronic devices like watches and calculators. Despite this success fundamental problems with regard to the improvement of the efficiency and the long term stability of the material have not been solved yet and hamper the utilization on a larger scale (see Chap. 8.2). A wide range of new semiconductor materials for bulk and thin film devices has been opened up with the exploration of ternary and quaternary compounds. The most promising candidate is the compound CulnSez which has the highest reported absorption coefficient and shows good photovoltaic properties as a polycrystalline thin film. (25) This offers a number of less expensive deposition techniques like vacuum evaporation, spraying methods or electrodeposition. Efficiencies of about 12% have already been reported (26~for solar cells made by electrodeposition and the theoretical limit is estimated to be about 20%. The technology and properties of ternary and quaternary compounds will be discussed in Chap. 7. None of the photovoltaic semiconductors has a band gap that can utilize the entire spectral distribution of the sunlight. The selection of the optimum band gap for the solar cell material can be considered a compromise between choosing a band gap wide enough so that not too many electrons are wasted, and yet narrow enough so that enough photons can create electron-hole pairs. The strategy to overcome the problem is to use semiconductors with different band gaps which are optimized for different wavelength ranges of the spectrum. Two concepts for directing the sunlight with different wavelengths have been developed known as spectrum splitting and the tandem cell approach. In the first case the sunlight is directed on the appropriate cell by spectrally sensitive mirrors, whereas in the second case two or more different semiconductors are stacked upon each other. For instance a wider band gap material on top that absorbs the blue part of the sun spectrum whereas the red part with longer wavelengths passes through and is absorbed by the lower cell with a smaller band gap. This tandem concept has proven to be very efficient in some cases and has recently yielded the record efficiency of 30% for a GaAs-Si tandem cell. This strategy also offers a large number of cell concepts for multinary compound semiconductors where band gaps can be modified by the composition. Another possible approach to reducing the cost of photovoltaic power is to concentrate the sunlight on the active area of the solar cell. This requires in addition to the solar cell concentration elements like lenses or mirrors and possibly a sun tracking system. The higher the concentration ratio, the smaller is the range of angles of light rays that the system can accept. For ratios greater than 10 only direct sunlight can be used and the sun must be tracked in its path across the sky. Maximum possible concentration ratios of about 45,000 are theoretically possible, but concentrating the sunlight will also increase the operating temperature, which decreases the efficiency or requires an additional cooling system. The different technical concepts and aspects that have been mentioned here will be described briefly in Chap. 3. An introduction to the operation principles of solar cells and the different concepts can be found in the books of Hovel(27)and Green. (28'29~Recent developments are given in the latest proceedings of the regular IEEE Photovoltaic Specialists Conferences, the European Photovoltaic Solar Energy Conferences, or other Solar Cell Conferences. Many
SEMICONDUCTORS
211
of the advances and improvements of solar cells during the last decade have become possible because of innovative cell designs and the utilization of new process technologies. Pulfrey and Johnston have reviewed the photovoltaic power generation~3'31)and the material aspects of solar cells prior to 1979 have been discussed by Bachmann.~32)Particular topics have also been reviewed in several papers and books and will be cited in following chapters. The research in the last decade has greatly enhanced the knowledge about the structural, electronic and technological aspects of semiconductors, mainly because of their utilization for microelectronic applications. Further improvements of photovoltaic semiconductors can certainly be achieved if these advances are utilized and a better understanding of the electronic properties of the solar cell materials is obtained. The main emphasis in this paper will therefore be to review the current knowledge about electronic properties that are relevant for photovoltaic materials. The material issues that will be addressed are summarized in the following chapter. 1.3. Material Aspects of Solar Cell Semiconductors The electronic properties of semiconductors are determined to a large extent by point defects (impurities, intrinsic defects) and extended lattice defects like grain boundaries, dislocations and precipitates. Contrary to high purity electronic grade material solar cell semiconductors usually contain higher concentrations of impurities and lattice defects. This is a consequence of the application of low cost techniques and the utilization of novel materials. The further improvement and successful development of solar cell materials therefore requires a thorough understanding of the defect structure and the ramifications for the photovoltaic properties. A main emphasis in this paper will therefore be to review the current knowledge about the physical properties of crystal defects in semiconductors and their process related generation. In Chap. 4 general aspects of point and extended lattice defects will be discussed as far as they are relevant for photovoltaic properties. Since much of the basic knowledge about lattice defects has been acquired for silicon the results will be discussed preferentially for this material, but they are also important from a more general perspective. Some of the general criteria which will determine the choice of a particular semiconductor for photovoltaic applications in the future are for instance the environmental condition (terrestrial or space applications, duration of sunshine etc.), efficiency and cost considerations, and the availability and toxicity of the raw materials. The choice of a solar cell device structure for a particular semiconductor and for optimal performance is primarily determined by the available processing techniques and the photovoltaic properties which can be achieved for this material. The defect structure and chemistry of the semiconductors which mainly determines the performance of the material is linked to the fabrication process and the particular device structure. Therefore the material issues vary with the different solar cell devices and the emphasis will be on different types of lattice defects. A particular problem arises for solar cells which are used in spacecraft. Low weight and high efficiencies are required for solar powered systems. Therefore for many years the exploration of space has been the driving force for the development of high efficiency solar cells. Empirically it has been observed that the photovoltaic properties of silicon and to a lower degree gallium arsenide solar cells degrade under the intense radiation in space. The investigation of the radiation induced defects and the search for radiation resistant semiconductors such as indium phosphide is therefore an important issue for space applications. In single crystalline homo- and heterojunction devices the control and the electrical behavior of the dominant intrinsic point defects (vacancies, antisites, etc.), impurities, and
212
PROGRESS IN MATERIALS SCIENCE
precipitates, and the recombination properties of the surfaces and interfaces between different materials are the most important material issues. Silicon and to a certain extent now gallium arsenide are the only semiconductors where the defect structure in single crystals is quite well understood and can be controlled, but the efficiencies of solar cells fabricated from compound semiconductors such as InP or CdTe are still limited by the poorer quality of these materials and the difficulty to control the electrical properties. Therefore the development of solar cells from these and other semiconductors still requires the need to identify the dominant defects which limit the efficiency. The fabrication and defect structure of III-V and II-VI compound semiconductors will be discussed in Chaps 6 and 7. It has been pointed out that some of the most promising solar cell developments use polycrystalline semiconductors such as silicon in sheet form or thin film copper based ternary semiconductors such as CulnSe2. Grain boundaries are therefore part of the microstructure and can have a strong impact on the material properties. The significance of these defects has stimulated the interest in their investigation in the past decade and many fundamental properties could be solved. Particularly polycrystalline thin films typically with grain sizes in the micron range show electronic properties which can differ considerably from those of monocrystalline semiconductors. The reason for the different behavior is that the electronic properties of individual grain boundaries dominate the material behavior for small grain sizes. In a similar way the electronic properties of a material containing a high density of dislocations is also modified. Since dislocations can be introduced inadvertently during epitaxial (Chap. 6.2) or ribbon growth (Chap. 5.2) they are also a major concern with respect to the material quality. In Chap. 4.6 the basic concepts about the electronic behavior of a polycrystaUine material (and similarly of a material containing high dislocation densities) will be discussed as far as they are relevant for the photovoltaic properties. The most promising materials for thin film solar cells are the copper ternary semiconductors and their properties will be discussed in Chap. 7 together with other multinary compounds. In general the electronic properties of multinary semiconductors change with the stoichiometry and composition which offers the possibility to tailor and optimize certain properties like the band gap energy, the electron affinity or the lattice constant325) The systematic investigation of entire groups of compounds and understanding their common features is therefore an essential part of the current investigation of solar cell semiconductors. Thin film solar cells can also utilize amorphous semiconductors as the active photovoltaic material, however, so far only silicon and germanium have been used for commercial devices. Since the discovery in 1958 amorphous silicon has been thoroughly investigated primarily at first because of the scientific interest in semiconductors with disordered structures. A particular problem is the degradation of the efficiency during the first hours of operation which is undesirable for photovoltaic applications333-37) The defect analysis in these structures is therefore a particular challenge and of great importance for the understanding of the stability of the material under illumination. Though still many fundamental problems have not been solved great progress has been made in recent years by the invention of new device structures and the utilization and incorporation of new amorphous materials such as SiC. The particular problems of amorphous silicon will be reviewed in Chap. 8. In most cases grain boundaries, dislocations and other lattice defects have a detrimental effect on the lifetime of minority carriers and the efficiency of a solar cell. Since in low cost materials these defects have to be tolerated to some extent the passivation of the defects may be a necessary technological step to improve the performance of the cells. A frequently employed process is the hydrogen passivation of recombination centres. The properties of
SEMICONDUCTORS
213
hydrogen especially in silicon have thus attracted considerable interest in the past. Although the interaction with lattice defects is not completely understood yet great progress has been made in understanding the general behavior of hydrogen in semiconductors and the main results will be discussed as a separate issue in Chaps 4 and 8. 2. PHYSICALPRINCIPLESOF PHOTOVOLTAICENERGYCONVERSION In this chapter the fundamental physical principles of a typical p-n junction solar cell will be described and the electrical parameters which characterize the performance are introduced. The physical processes are the light adsorption, the minority carrier transport and the recombination mechanisms and they are determined by the electro-optical properties of the material. The corresponding material parameters and the technical design of the solar cell finally determine and limit the efficiency of the energy conversion process. The relation between material properties and the important electrical parameters will be discussed. To a large extent the performance of a solar cell can also be improved by the technical cell design. Though this aspect has not the main emphasis here a brief summary of different cell concepts will be given. For a more comprehensive discussion the reader is referred to Refs. (27, 28, 38). Solar cells are devices which are optimized with respect to the sunlight and therefore the solar spectrum will be presented first. 2.1. Solar Spectrum in Space and at the Earth's Surface The spectral distribution of the radiation emitted from the sun is determined by the temperature of the surface (photosphere) of the sun which is about 6000 K. The wavelength distribution of the sunlight (power per unit area and per unit wavelength) follows approximately the radiation distribution of a black body at this temperature as can be seen in Fig. 1. The deviations at certain wavelengths are due to adsorption effects in the sun's atmosphere. For the comparison of the conversion efficiency of different solar cells the knowledge of the exact spectral distribution is important since they can respond differently to different wavelengths. The total energy per unit area integrated over the entire spectrum and measured outside the atmosphere perpendicular to the direction of the sun is essentially constant. This radiation power is referred to as the solar constant or air mass zero (AM0) radiation. Measurements taken at high altitudes have yielded the currently accepted average value of 1.353 kW/m2. 09'4) The spectral distribution is changed considerably when the sunlight penetrates through the earth's atmosphere. Even for a clear sky the light intensity is attenuated by at least 30% because of scattering at molecules, aerosols and dust particles, and adsorption by its constituent gases like water vapor, ozone or carbon monoxide. The attenuation mechanisms are wavelength dependent which explains the strong absorption bands in the spectral distribution measured at the earth's surface and which are also given in Fig. 1. (41-43) The measurement of the radiant power (per unit area) is further complicated by the fact that the degree of attenuation is highly variable because of the constantly changing position of the sun and the corresponding change of the light path through the atmosphere. One has defined the ratio of the actual path length to its minimum value when the sun is directly overhead as optical air mass and the corresponding radiation as air mass one (AM1) radiation. For any given angle 0 with respect to the overhead position the air mass takes the value AMm, in which the air mass number, m, is represented as m = 1/cos 0, and thus measures the atmospheric path length relative to the maximum path length when the sun is directly
214
2.5
1
PROGRESS i ~
IN MATERIALS I ~
SCIENCE r I I i
2.0 --
"%_
1 6 0 0 0 K black body
1.6 -
AM0 radiation
,.o-
}
uJ 0.5/
o
0
/2
0.2
"[
,
0.4
,
,
0.6
,
0.8
\
1.0
v,
1.2 1.4 1.6 1.8 2.0
Wavelength {#m)
FIG. l. Spectral distribution of sunlight. S h o w n are the radiation outside the earth's a t m o s p h e r e (AM0) and at the surface (AM1.5). D a s h e d line indicates the radiation distribution expected from the sun if it were a black body at a temperature of 6000 K . (41-43)
overhead. It is evident that the terrestrial sunlight varies greatly both in intensity and spectral distribution. Therefore it became necessary to define a terrestrial standard which allows a meaningful comparison of different solar cells tested at different locations. In the U.S. photovoltaic program the spectral distribution for a AM1.5 radiation with a total power density of 844 kW/m 2 (for 0 = 45 ) has been adopted as a standard. Tabulated values of the spectral distribution are given in Ref. (43) and refer to sunlight in a cloudless sky. The AM 1 spectrum is for an angle of 0 = 0 and has an incident power of about 925 kW/m 2. The spectrum has contributions from direct and diffuse (or scattered) sunlight. The diffusive component can account for about 20% of the total radiation in a clear sky, but can increase considerably with the cloud coverage and the position of the sun. Since the spectral distribution for diffuse radiation is different from direct sunlight, there exists a great variability both in the intensity and the spectral composition at different locations on the earth. This is especially important for photovoltaic systems using concentrator solar cells (see Chap. 3.5) which can only accept radiation from a limited range of angles and require a tracking system to utilize the direct component of the sunlight. Since the diffusive component is wasted in this case the design of a photovoltaic system requires detailed records of the solar insolation at the site of installation. Reasonable estimates of the global radiation received annually are available now for most regions of the world, though there are still uncertainties involved that can be caused by local geographical conditions. It is evident from these considerations that an optimized photovoltaic system has to utilize solar cells which are adjusted to the average spectral distribution of the local sunlight. 2.2. Absorption of Light in Semiconductors The light-generated current, IL, is determined by the absorption behavior of the semiconductor. The fraction of incident light D = 1 - R that actually penetrates the absorbing
SEMICONDUCTORS
215 ~c where ~ is the
material can be calculated from the complex refraction index n~ = n - i extinction coefficient, and R is given by
(n 1) 2 + ~2
R =
(n + 1) 2 Jr- t 2
(1)
n(2) and ~(2) are functions of the wavelength 2 of the incident light. Substituting the appropriate values shows that for many semiconductors a considerable fraction of the light is reflected. It is therefore desirable from the perspective of making an efficient solar cell to decrease R. This can be achieved by an anti-reflection coating or by selectively etching the cell surface, which can produce a textured structure on an initially flat surface if an etchant is used that preferentially attacks particular crystallographic planes (see for instance Fig. 22d). The light which passes through the material is absorbed and the number of generated electron-hole pairs depends on the number of incident photons So (v) (per unit area, unit time and unit energy) which can be calculated from the spectral distribution of the sunlight in Fig. 1. The frequency or the photon energy hv are related to the wavelength 2 by the relation [#m] = c/v = 1.24/hv [eV]. Inside the crystal the photon flux S(x, v) decreases exponentially according to
S(x, v) = S0(v)exp{-~tx)
with ct(v) -
4n~:v
(2)
where v is the frequency and c the speed of the light. The absorption coefficient ~(v) is determined by the absorption process in the semiconductor. The important process for the photovoltaic conversion is the excitation of electrons from the valence band into empty states of the conduction band which can occur if the energy of the incident photons is larger than the band gap energy. In direct band gap semiconductors like GaAs and many other compound semiconductors the minimum of the conduction band and the maximum of the valence band occur for the same wave vector k in the Brillouin zone. Since both the energy and the momentum in such a b a n d - b a n d transition has to be conserved and photons carry a large energy (by) and only a small momentum (hk), the most likely transitions are therefore between states close to the minimum and maximum of the conduction and valence band, respectively. A theoretical calculation of the probability for these direct (allowed) transitions gives the following result for the absorption coefficient ~d as a function of the frequency y.(44,45)
(hv
~d (v) = ct0
Eg) 1/2
hv
(3)
A parabolic dependence of the condition and valence band states as a function of the wave vector k near the edges is assumed here. Eg is the band gap energy and e0 is a constant which is obtained from the calculation in Ref. (45), but is usually fitted to experimental data. The approximation is only valid for photon energies near the band gap edges and unpolarized light. The frequency dependence deviates also from eq. (3) for photon energies very close to the band gap energy because of the Coulomb interaction of electrons and holes, which has not been taken into account here. A different frequency dependence is also obtained for direct forbidden transitions between electronic states with the same parity, but shall not be considered here. In direct semiconductors like silicon and germanium where the minimum and maximum of the conduction and valence band, respectively, occur for different wave vectors, optical transitions between the states close to the band edges require a change in momentum of the
216
PROGRESS
IN MATERIALS
SCIENCE
electrons and are therefore only possible with the participation of phonons. Unlike photons the phonons have a large momentum hq and a small energy hVph, therefore electron transitions between the valence and conduction can already occur when the photon energy is a little smaller than the band gap energy. The most likely photon absorption processes are one-phonon transitions, either by the absorption or the emission of a phonon, between the band edges. The energy and momentum conservation laws yield in these cases Ev (k) + hv + hVph ( q ) = Ec (k'); Ev(k) + hv - hVph(q) = Ec (k'); hk + hq = hk' hk - hq = hk'. (4a) (4b)
Analogous to the direct band-band transitions allowed and forbidden indirect transitions have to be distinguished. In the case of the allowed transition the calculations of the transition probabilities yield for the absorption coefficient
(hv - Eg + hvphI )2 (hv - Eg - hVph2)2 Ctin(V) = ~l exp(hvphJKT) - 1 + ~2 1 - exp(hvph2/KT )
(5)
Contrary to the direct transitions the absorption coefficient here is also temperature dependent. Empirical expressions for the absorption coefficient which require a higher degree of accuracy over a larger photon energy range use linear combinations of eqs (3 and 5) with free parameters t~ and E which are fitted to the experimental data. i The fundamental absorption processes for direct and indirect semiconductors which have been described so far occur for photons with energies close to the band gap energy Eg. For larger photon energies direct and phonon-assisted transitions become possible between other states of the bandstructure within the Brillouin zone. This may lead to additional features in the absorption spectra if the probability for a transition in a certain region of wave vectors k of the band structure is high. Optical transitions within a band can also occur, but their probability is much lower, as might be expected, because the concentrations of electrons (in the conduction band) or unoccupied states (in the valence band) are low. Finally, two optical processes shall be mentioned which change the absorption behavior of a heavily doped semiconductor. Since solar cells can have high dopant concentrations in the emitter region of the p-n junction absorption processes in this part of the cell are affected. This can be especially important in thin film solar cells where the junction region is a major part of the entire absorbing volume. The electrical field in the space charge region and the corresponding distortion of the valence and conduction band allows band-band transitions where the electrons can tunnel through the forbidden gap. Optical transitions below the band gap energy become possible then and depend on the strength of the electrical field E, which is known as the Franz-Keldysch effect. For the simple case of a direct allowed transition without the participation of phonons the absorption coefficient for energies hv < Eg and E g - hv >> E r is given by(44'45)
AE~/2 f ~ 4 (Eg-hv~3/2 ~ tFK(V)=v(Eg-hv)exp(-3\ er / J

with the critical parameter
(6)
= l-y
/ 2rceh E \ 2/3
The constant A and the mass m* are given in(**'45). The expression for the absorption coefficient for energies hv > Eg as derived in eq. (5) is also changed for E # 0 but will not be given here.
S E M I C O N D U C T O R S FOR SOLAR CELL A P P L I C A T I O N S
217
High (dopant) impurity concentrations also effect the shape and the band edges of a semiconductor. The density of states distributions of valence and conduction band are "smeared out" then (band tails) and the width of the band gap is effectively reduced. A calculation of the band gap narrowing AEg as a function of the doping concentration, N, has been given by Lanyon and T u f t (46)
AEg-
3e 2 (e2N)~/2 3e 2 16zte ~ ~'~e~J = 16rCELD
(7)
and is in good agreement with experimental data in silicon for N >/l017 c m -3. (E is the permittivity of the semiconductor and L o the screening or Debye length.) If the semiconductor becomes degenerate optical transitions in states below the Fermi energy EF (for n-type doping) are almost impossible. Therefore the optical absorption edge is shifted by the energy A(hv) = (EF -- E) for n-type or A(hv) = (Ev - EF) for p-type crystals, respectively (BurnsteinMoss shift). Therefore in degenerate crystals the optical band gap appears larger (Fig. 2). In compensated and highly doped semiconductors with a Fermi level within the forbidden gap electron transitions between the band tail states become possible. Again absorption for photon energies below the gap energy hv < Eg can occur and be described by the expression (Urbach law)(44,45) ~u(v)= C(v)exp{
Eg-hv~
(8)
C(v) is a slowly varying function of the frequency, and Eo(T) a characteristic energy which is proportional to the temperature for sufficiently high temperatures (e.g. in GaAs for T > 100 K). Like in heavily doped semiconductors the band structure of amorphous semiconductors (e.g. silicon) is also characterized by band tails in the density of states distribution. Experimental data in fact show that the same law can be used to describe the absorption behavior of some amorphous semiconductors below Eg at least in the temperature range that is interesting for photovoltaic applications (above room temperature). Since the band gap energy Eg is not clearly defined in these materials it has to be considered a free parameter which can be adjusted to fit experimental results. As will be discussed in Chap. 8 the "optical band gap" Eg is in general different from the single crystal value. The absorption behavior of semiconductors is not only important for the photovoltaic response of the material but can also be investigated to obtain information about the band structure of a crystal. Since the development of novel solar cell materials is an essential part of the photovoltaic research optical measurements play a role in the material characterization. Experimentally one usually measures the reflectivity, R, and the index of refraction, n, rather than the light that is transmitted, and calculates the absorption coefficient from eqs (1 and 2). It should also be mentioned that the energy required to generate electron-hole pairs can also be supplied by electrons with energies in the same range. Therefore electron reflectance measurements and related techniques are also applied to investigate the band structure. The previous discussion has shown that in many cases the fundamental absorption edge is not a sharp transition. Absorption can also occur below the band gap energy and there are several effects that can contribute to electron-hole pair generation at lower energies. In the cases where a new material is investigated it can therefore be difficult to determine the band gap energy and other details of the band structure precisely, and it may be required, in addition, to characterize carefully the microstructure and the chemistry of a material.
218
P R O G R E S S IN M A T E R I A L S S C I E N C E
Recombination at front surface
%
Xp ........
N (E)
N (E) (b)
Wp
(a)
Recombination at back surface
FIG. 2. Schematic representation of optical transitions between valenceand conduction band. (a) Compensated semiconductor with band tail states showing possible transitions below the conduction and valence band edge. (b) Highlydoped, degeneratesemiconductorshowing transitions into empty states above the Fermi level and apparent increase of the band gap width (Burnstein-Moss shift).
FIG. 3. Schematicdiagram of a p-n junction solar cell, defining basic parameters.
The absorption coefficient can be used now to calculate the generation rate G(x, v) of electron-hole pairs (per unit time, volume and energy) at a distance x from the semiconductor surface. The fraction of photons which penetrate into the crystal is So(v) (1 - R ) , therefore using eq. (2) the generation rate can be written as
G(x, v) = o~(v)So(v ) (1 - R )exp(-o~(v )x ).
(9)
Here it is assumed that each absorbed photon generates one electron-hole pair. For photons with energies higher than the band gap energy the electrons and holes carry excess (kinetic) energy which will be dissipated to the lattice until they occupy states near the band edges. If the kinetic energy is large enough more electrons may be generated through impact ionization, but this process is not important in the energy range where the visible light is absorbed. Therefore the kinetic energy does not contribute to the photocurrent and is wasted from the perspective of the energy conversion.
2.3. Principle of the p-n Junction Solar Cell The p - n junction solar cells are still the most widely used devices for solar energy conversion, therefore the operation of an ideal device and the parameters that are important for the characterization of a solar cell will be discussed briefly. For a more detailed description the reader is referred to the numerous references about the subject. A comprehensive treatment on the basic solar cell characteristics has been given for instance by Hovel/2v) A typical schematic representative of a p - n junction cell is given in Fig. 3. It consists of a shallow p - n junction formed near the front surface, a front ohmic contact in form of stripes and fingers, and a back ohmic contact that covers the entire back surface. When light is incident on the front surface and penetrates into the crystal the number of electrons and holes which are generated at a distance x from the surface is given by the generation rate G(x, v) determined in the previous chapter [eq. (9)]. Ifpo and no are the carrier concentrations in thermodynamical equilibrium, which are determined by the doping
SEMICONDUCTORS
219
concentrations of donors and acceptors on either side of the junction, the changes in the electron (6n) and hole concentrations (6p) can be expressed by
6p = p --po;
~n = n - n o .
(10)
The carrier concentrations for a non-degenerate semiconductor are given by
Arc exp
k--T-
where the potential is related to the local electric field by E = - g r a d ~p. Nc and Nv are the density of states, EF. and EFp are the quasi-Fermi energies for electrons and holes, n, p their concentrations in the conduction and valence band, respectively, and e the electronic charge. The basic equations which describe the flux of electrons and holes in a semiconductor are the current-density equations Jp = e (p/ip E - Dp grad p); and the continuity equations ~p_ ~t ldivJp+Gp e ~P" Tp ' ~n_ldivJ.+G.___.~n ~t e z, (13) Jn = e (n/i. E + D. grad n ) (12)
The diffusion coefficients Dn, D o are related to the mobility of the carriers /i., /iv by the Einstein relationship D = (kT/e)/i. For light-generated carriers the generation rates for electrons and holes are equal Gn = Gp = G. Additionally, if one assumes fin = 6p one can combine the eqs (13) and obtains for instance for the holes
Op
0t
/iEdivp +DAp +G---
ap
(14)
where the ambipolar diffusion coefficient D and mobility/i are given by
D
=
(n/i.Dp +p/ipDn)
(n/l n + P/ip ) ;
/i --
/i./ip(n - p )
(/,ill n "-[- p/ip
(15)
)
A corresponding equation is obtained for electrons ifp is replaced by n. The relaxation times for electrons and holes are equal in this case l"n = l'p = "C. These equations are required in the intrinsic case whereas for n- or p-type semiconductors the expressions can be greatly simplified. The determination of the electric field E requires in addition Poisson's equation div r = ~ (No - Na - n + p )
CEo
(16)
where Nd(r) and Na(r) are in general a function of the position and usually equal to the concentrations of completely ionized acceptors and donors on each side of the junction, c is the dielectric constant of the material, and co the permittivity of the vacuum. In general one has also to include the electrons and holes trapped in energy levels in the forbidden gap. Such a case will be treated in Chap. 4.4 when the potential barrier of a grain boundary is calculated. In most cases the p-n junction solar cell is treated as a one-dimensional device, though for a realistic calculation of the current-voltage charactersitics the geometry of the particular device has to be taken into account. The usual approximation, which is made here for small
220
currents and an abrupt p-n junction with uniform doping concentrations on each side of the junction, is that n = p within a "depletion width" W. The electrical field outside the depletion region is small and can be neglected for the minority currents. Under steady state conditions the time derivatives o f p and n in the continuity equations are zero and eq. (13) outside the depletion region on the n- and p-side, respectively, simplifies to
02p = c32n 6n Dp-~xZ -G+rP;zp D. ~ x 2 = -G+--.Zn
(17)
Considering the schematic solar cell given in Fig. 3 one can use the expression (9) for the generation rate G and obtains for instance for the minority holes in the n-type front region the following differential equation
Dp
O2p
+ ~So (1 - R ) e x p ( - ~ x ) - -~ = 0.
6p
(18)
The corresponding equation for electrons in the p-base region is easily obtained from the second eq. (17). The light generated current can be calculated at the edges of the depletion region where the electrical field is zero and the total current mainly determined by the minority current at this point. The minority currents from eq. (12) are given then by
Jp = --eDp 0p. ~,
On Jn = eD. d-~.
(19)
At the depletion edges the excess carrier concentration is small due to the electric field in the depletion region, which accelerates the minority carriers to the other side of the junction. Therefore one has approximately the boundary conditions 6p ,~ 0 at x = x. and 6n ,~ 0 at x = Xp. At the front and back surfaces the excess concentration 6p and fin, respectively, are determined by the surface recombination velocities Sp, S,.
@ Dp ~x = Sp 6p D.~x=Snfin
0n
at
x=O,
at
x=Wp+xp.
(20)
Wp is the width of the p-base neutral region. Using these conditions one can solve eq. (18) for 6p(x) and determine the minority current density at x = Xn from eq (19). (32) ~Lp Jp = eSo(1 -- R) (~Lp)2 _ 1 {Fp, -- Fp2 e x p ( - ~Wn)}. W. is the width of the n-type front region, the coefficients Fpt and _ ] / L - ~ p sinh -~-p + cosh ~-~-p , (21)
Fp2 are given by

(22a)
E Dp
Wn + sinh Lp A/L Dp sinh -Lp + cosh Lp j -
(22b)
The diffusion lengths Lp,. = (Zp,n Dp,.) 1/2are related to the lifetime z. and % of the minority carriers. This photocurrent would be collected from the front side of an n-on-p junction solar cell at a given wavelength, assuming this n-type region to be uniform in doping level, lifetime and mobility. For the p-type base region a similar set of equations can be derived. With the
SEMICONDUCTORS
221
same boundary conditions given above for the back side surface at x -- Wp + Xp, and for the depletion edge on the p-type side at x = Xp, the photocurrent which is collected from the base of the cell at x = Xp is given by
J. = eSo (1 - R) (~Ln)2 __ 1 e x p ( - ~Xp) {F~l -- F~2 exp(-- ~ Wp)}

with the coefficients given by
~Ln
(23)
Fn'=~Ln LDn
-
FS.L. coshW,+sinhT. j / l-b-~-. sinh -w, + cosh wa. wa/r o, , L. Ln L. d

sinhT. cosh +W"
(24a)
=F -.s"LoL
(24b)
Some photocurrent generation also takes place in the depletion region. If the electrical field in this region is high, the generated electron-hole pairs are accelerated out of the region before they can recombine. The photocurrent density in this case is equal to the number of photons absorbed
Jdp = eSo (1 -- R)exp(--~x,) (1 -- exp(--~W))
(25)
where W = xp - x, is the width of the depletion region. Implicitly it has been assumed here that the recombination in the depleted region can be neglected. If the doping concentrations, lifetimes and mobilities are a function of the distance, analytical expressions cannot be derived anymore for the minority carrier currents and numerical methods must be used. The total photocurrent IL can be calculated then from the photocurrent density JL = Jp + J, + Jdp and is given by IL = A
;/
(Jp + J~ + Jdp) dv
(26)
where hvm is the smallest photon energy corresponding to the absorption edge of the semiconductor and A the active area of the solar cell. The depletion region and the usually much larger volume within the diffusion length are therefore the active collection regions of a p-n junction cell. The photocurrent IL = A eS o (1 - R) const, is proportional to the light intensity that enters the crystal, therefore the optical performance of a solar cell is frequently characterized by the normalized photocurrent JL (hv) as a function of the photon energy, hv, or wavelength, 2. The external spectral response (or quantum efficiency) of the cell is the total photocurrent JL divided by eSo and the internal spectral response SR(hv) of the cell by eSo ( 1 - R), respectively,
SR(hv) = Sp (hv) + J. (by) + Ja, (by) eSo(hv) (1 - R(hv))
(27)
The individual contributions from each of the three regions are shown in Fig. 4 where the internal spectral response of the cell SR(hv) has been calculated for a silicon n-on-p solar cell.(25) In the ideal case this is a step function which equals zero for energies below the band gap energy and unity for larger energies. It can be seen that for small photon energies most of the current is generated in the base region, whereas for energies above 2.5 eV the front region takes over. The recombination velocity at the front surface has a profound effect on the front photocurrent and this is the reason why the total current can depart substantially
JPMS 35~3/4--B
222
1.0
0.8 i0.6
.,I nr
0.4
U,l
0.2 2.0 5,0 hu(ev) 4.0 50
'to
FIG. 4. Total spectral response of a p-n junction solar cell as a function of the photon energy together with the contributions of the front, base and depletion region. Dashed line indicates ideal behavior, t27)
from the idealized step function for high photon energies. In general, to increase the total photocurrent over the useful energy range, one should increase both Lp and Ln, and decrease both Sp and Sn. The maximum photocurrent that can be generated in a solar cell is given by IL. A solar cell in an electrical circuit will produce a lower current which is determined by the external load resistance and the corresponding operation point on the current-voltage characteristics of the device. If a current flows the minority carrier concentrations at the depletion edges are raised and the previous boundary conditions have to be replaced by 6,o= ' exp~-~-I
atx=Xn atx=xp
(28a) (28b)
6n=NAA~exp
n~/ ~eU- - 1 ) /
where U is the voltage drop across the junction. Solving again for the total current, I, one obtains the following well known current-voltage characteristics for an ideal p-n junction solar cell(38) I = Is exp ~-~ - 1 - IL where the saturation current is given by (29)
f DnFp DpFn ~ / Es Is = eANcNv_~L--~ + L - ~ ; exp ~--~-~).
(30)
The correction factors Fn and F v again account for the finite recombination velocity of electrons and holes at the front and back surfaces (Sn, 8v).(47) Fp and Fa are given by Sn cosh( Wv/Ln) + (Dn/Ln) sinh ( Wp/Ln) Fv = Sn sinh(Wp/Ln) + (DdLn)Cosh(Wp/Ln) Sp cosh( Wn/Lp) + (Dv/Lp) sinh (Wn/Lp) Fn = Sp sinh(W~/L v) + (Dv/Lp)cosh(Wn/Lv) (31a) (31b)
SEMICONDUCTORS
223
-- V
v,~li
IL
I
I
im
~ ne u iad llm t
1,
FIG. 5. Current-voltage characteristics of a p - n junction diode in the dark and when illuminated defining the basic parameters: light generated current (le), short circuit current (I~), and open circuit voltage (Vow). Fill factor (FF) defined by maximum rectangle given by V.p and Imp.
with Wn = x. the width of the n-type front region. These equations show how the surface recombination and the dimensions of the solar cell affect the saturation current and the efficiency, which will be defined next. The typical current-voltage characteristics of a p-n junction diode in the dark and when illuminated is given in Fig. 5. Three parameters are usually used to characterize the solar cell performance. The first one is the short circuit current, I~, which is ideally equal to the light-generated current, I~ = 1L. The second parameter is the open-circuit voltage, Vow,which is obtained from eq. (29) by setting I = 0, Vo~= K___T In e + 1 . (32)
Vo~is determined by the ratio IL/Is and thus by the absorption and light-generation process and the efficiency with which the charge carriers reach the depletion region. The power output for a given operating point in the fourth quadrant of the current-voltage curve, P = I V, is equal to the area of the rectangle that is defined by the corresponding values (Vp, Ip) and given by P=IpK----Te In l + - ~ s
(33)
The maximum possible area (V,p, Imp) defines the fill factor FF and the energy-conversion efficiency, t/, of the solar cell.
VmpImp VmpImp FFV~I~ FF= Vo~I~ r/ = Pin Pin
(34)
Pi, is the total power of the incident light. Evidently FF is larger the more "square-like" the current-voltage curve is. Typically it has a value of 0.7 to 0.9 for cells with a reasonable efficiency. The maximum area and efficiency can be calculated from the current-voltage characteristics. The transcendental equation, which is obtained after taking the derivative
224
with respect to V in eq. (33), yields a relationship for F F which is only dependent on the open-circuit voltage and in the ideal case (with Rs = 0) can be approximated by
F F = vo - In(roe + 0.72) vow+ 1
(35)
with a normalized open-circuit voltage Vo= eVo~/kT. In general F F depends on the actual I-V characteristics which usually deviates from the ideal behavior of a diode as will be discussed in Chap. 2.4. The efficiency ~/, in general is determined by all three parameters Vo, I~, and FF. It can be seen from these equations that the material parameters, lifetime and mobility of the minority charge carriers, determine both the open-circuit voltage and the short-circuit current and hence the efficiency of the solar cell. These parameters are not independent from each other and are controlled by physical processes which will be discussed in the following chapters. That will allow the estimation of the efficiency of solar cells from the perspective of the material properties. Since the technical design of a solar device is closely linked to the material parameters practical solar cells will have to compromise between different aspects and therefore have efficiencies lower than the ideal values. A discussion of the major limiting factors and basic requirements for a solar cell design will be given in Chap. 2.5. 2.4. Fundamental Material Parameters The band gap width, the lifetime of minority carriers and their mobility have been identified as the main material parameters which control the performance of solar cells. The perfect crystal lattice of a semiconductor determines the band structure and the ideal values for the lifetime and the mobility. Real crystals, however, contain imperfections in the crystal structure such as the surface or bulk lattice defects, which change these parameters and are primarily responsible for the degradation of the efficiency of a solar cell. Generally, the mobility/t and the lifetime z of electrons and holes are determined by a variety of processes. If several scattering or recombination mechanisms operate at the same time the contribution of the individual values ~i and zi to the total mobility and lifetime is usually given by the summation rules
1 1 -p = E ~;
1 1 -z = )-" -Ti
(36)
Different relationships have to be used however if the mechanisms are not independent from each other or can be attributed only to a fraction of the total carriers. This can be the case for instance for amorphous semiconductors where electron (or hole) conduction can occur in extended and localized states (Chap. 8). If the occupancy of the different states is Ni one has to use the following relationship for the mobility instead(as)
1 ai --; ~ =E/./i
ai-
Ni/~i2
ENi/.f 2"
(37)
Quantities which are frequently used to describe the electrial transport in a semiconductor under homogeneous illumination are the photoconductivity and diffusion length. If both electrons and holes (e.g. in near intrinsic material) contribute to the current one has to use the general relations for D and z from eq. (15) for ambipolar carrier transport and obtains
aph=tro+eG(#nzn+#pzp); L = ~ 2 K T ~ s I ~ - - z - P ( n + P ) ~ '/2 ( e n#n Zn + p#pzp )
(38)
SEMICONDUCTORS FOR SOLAR CELL APPLICATIONS
225
(tro the conductivity without illumination). Photoconductivity and diffusion length are thus essentially determined by the products #n z, and #p %. If one carrier type dominates these equation reduce to trph = ao + e G#,zn and L = (Dp Zp)~/2(for n-type conductivity for instance). The physical mechanisms which determine the magnitudes of the mobility and lifetime will be discussed in the following sections. 2.4.1. M o b i l i t y o f charge carriers The drift mobility of charge carriers is determined by scattering mechanisms both with lattice vibrations (phonons) and lattice imperfections. In single crystal semiconductors doping impurities are usually the crystal defects with the highest concentrations, whereas in polycrystalline thin films the grain boundaries play the dominant role as scattering centers. It should be noted that in non-stoichiometric compound semiconductors doping can also occur through intrinsic point defects such as vacancies or antisite defects. These point defects scatter electrons and holes in a similar way as foreign atom impurities and contribute to a decrease of the mobility. Theoretical calculations for various scattering mechanisms in single crystals give the following general relationship for the temperature dependence of the mobility
#i = C T s
(39)
where C is a constant or only weakly (logarithmically) dependent on the temperature. A fundamental scattering mechanism in any crystal is the interaction with phonons. Calculations for non-polar crystals show that scattering with acoustic and non-polar optical phonons is characterized by the exponents s = - 3 / 2 and the constants C given by(38'45)
eMh4c 2 CAp = 3rc7/2(2m *)5/2K3/2f~o E 2 eMh4092o
(40a)
Cor~p 37~7/2(2m*)5/2K3/2~ oE2o
(40b)
with the effective mass of the scattered charge carrier, m *, the volume of the unit cell of the crystal, f2o, the atomic mass, M, Planck's constant, h, the electronic charge, e, and the sound velocity, cs. El and Eo ( per unit length) are the deformation potentials for acoustic and optical phonons. In polar crystals with two different kinds of atoms of mass M~ and M2 (e.g. in compound semiconductors) the interaction with polar optical phonons is an essential contribution and characterized by the exponent s = - 1 / 2 and
MiM2h2f~otO 2
Cop = 3(M1 + M2)ZEe37r7/E(2m*)3/2K 1/2
(41)
where coo is the characteristic frequency of optical phonons. The phonon scattering is different for electrons and holes because of their different effective masses, and is characterized by a decrease of their mobility with increasing temperature. The theory for scattering at ionized (usually doping) point defects yields s = 3/2 and hence the slope of the temperature dependence is positive, whereas for neutral impurities, which dominate at low temperatures, one obtains s = 0, and the mobility becomes temperature independent. (44'45)In the first case C depends on the concentration, NT, of the defects and its charge, Z e , and the mobility is given by the Brooks-Herring equation345'49)
(_~)3/2
#=
4e2
Z2e3m,l/2N v
{ 241rm,E,o(KT)2"~-I
In e2-~v j
T3/2
,-~(Nv) r. (42)
226
If one neglects the weak dependence on NT in the logarithm then r = 1. In addition to the scattering mechanism discussed above, other mechanisms also affect the mobility. Therefore in real crystals the exponents and prefactors differ from the theoretical values and are better determined empirically. For instance for pure silicon measurements near room temperature yield for the mobilities of electrons and holes the exponents s = 2.42 and s = 2.20, respectively, and for GaAs s = 1.0 and s = 2.1, respectively. In crystals containing charged point defects one measures a combined mobility according to Matthiessen's rule (eq. (36)). Because of the positive and negative temperature coefficients for the point defect and phonon scattering the mobility has a maximum in a certain temperature range which depends on the impurity concentration. Measurements on n-type silicon for instance show a maximum # ~ 5000 cm2/Vs at T = 50 K for a doping concentration N ~ 1017cm -3. At room temperature the mobility is mainly determined by phonon scattering but the additional impurity contribution decreases the mobility with increasing impurity concentrations N i> 1015cm -3 and exponents of r = 0.38 for holes and r = 0.42 for electrons have been determined, which is smaller compared to the theoretical estimation r ~ 1.'51) Such high doping concentrations occur in various types of silicon solar cells in the emitter region (see Chap. 3), hence the doping dependence of the mobility has to be taken into account in a realistic calculation of a solar cell device352'53) Dislocations and grain boundaries also scatter charge carriers and reduce the mobility in deformed and polycrystalline semiconductors, respectively. A major factor which contributes to the scattering mechanism is the Coulomb interaction with the electrical charge that accumulates at the lattice defects and the corresponding screening charge in the vicinity. The origin of the defect charge is a consequence of the electronic levels of these defects, which are introduced in the forbidden band gap, as will be discussed in detail in the next section for surfaces, grain boundaries and dislocations. Experimental data on the mobility in the presence of grain boundaries and dislocations can be obtained from conductivity and Hall effect measurements in polycrystalline or deformed semiconductors, respectively. For dislocations a comprehensive study of the mobility #D in deformed low doped silicon and germanium has shown that at low temperatures (T < 150 K) scattering at dislocations dominates and the mobility decreases with decreasing temperature354) A theoretical calculationc55) of the mobility yields two terms which are the contributions of the scattering of charge carriers in the long-range strain field, and at the line and screening charge
1 #D
A No + B 2No Q2 T T 3/2
(43)
where Qo is the line charge (per unit length) and No the dislocation density. The line and the screening charge produce an electric field and a potential barrier, VD, at the dislocation, which can be calculated from Poisson's equation (Fig. 6). The results for Vo which will be discussed in more detail in Chap. 4.3 show that the decay of the potential with the distance from the dislocation, r, is characterized by a screening length, 2. Several authors have derived approximations for 2 which are valid for different doping and temperature regimes. If the temperatures are not too low 2 is given by a Debye screening length, and for a p-type crystal the following expressions for the dislocation charge Qo and J[ are obtained. :EeKT'~ l/z
e--
J ;
= e--YU
Na --P
(44)
where Na is the acceptor and p the average hole concentration. Because of the trapping of charge carriers at the dislocations the average free carrier concentration p in the crystal differs
SEMICONDUCTORS V(r)
227
FIG. 6. Schematic representation of the potential barrier V(r) at a charged dislocation as a function of the distance, r, from the dislocation line. Er is the occupation limit of the energy states of neutral dislocation, and ro the radius of the dislocation core.
from the acceptor concentration, particularly in cases where the dislocation density is high. In general, the line charge QD and p are complex functions of the temperature and doping level. For low doping levels and high dislocation densities one can approximate QD ~ e Na/ND, whereas for high doping levels all dislocation traps are filled, so that QD remains constant and p ~ N a. The constants A and B in eq. (43) are parameters which are calculated in the model but can also be determined empirically. A good agreement with mobility measurements as a function of the temperature has been obtained with adjusted parameters in silicon and germanium. The results show that dislocation scattering becomes important for dislocation densities No > 10 7 c m -2. In solar cell semiconductors the dislocation densities usually can be kept below this value so that their contribution to the mobility can be ignored, but they can still be effective recombination centers at lower densities, as will be discussed in Chap. 4.3. Mobility measurements on polycrystalline thin film semiconductors yield information about the influence of grain boundaries. Because of the technological importance of these materials not only for solar cell applications, the conductivity and the mobility of charge carriers has been studied extensively in a number of semiconductors and will therefore be discussed separately in Chap. 4.6. The main result is that in the case of small grain sizes, L, the grain boundary potentials, VB, overlap and the crystal is almost completely depleted from carriers. For this situation the mobility /~B of electrons in an n-type poly-crystal can be calculated from eq. (199)
where A (T) is a prefactor which depends on the model used to calculate the conductivity. In all models A (T) is weakly dependent on the temperature. The grain boundary potential as will be discussed in Chap. 4.4 is a complex function of the doping concentration, the temperature and grain size. Numerical calculations and experimental results have shown that the potential as a function of the doping concentration N d (for n-type crystals for instance) has a maximum at Nd ~ NB/L where N8 is the number of grain boundary states per unit area. Correspondingly, the mobility has a minimum, whereas for low doping levels ~B remains constant. (56'57~ For high doping levels and larger grain sizes the mobility increases with the doping concentration and approaches the single crystal values. The mobility in fine grained polycrystalline semiconductors is usually greatly reduced compared to single crystals and can thus be a major limitation for the efficiency of a polycrystalline material.
228
PROGRESS
IN MATERIALS
SCIENCE
The mobility is frequently determined by Hall and conductance measurements. If only one scattering mechanism is dominant the Hall mobility #H differs from the drift mobility # by a constant factor only. The proportionality constant depends on the scattering mechanism and for acoustic and non-polar phonon scattering it is /~H//~= 1.18, for polar phonon scattering 1.11, and for ionized impurity scattering 1.93, respectively. In heavily deformed and polycrystaUine semiconductors the crystals become rather inhomogeneous, and the interpretation of Hall effect measurements is less straightforward as has been discussed in Ref. (54). Therefore the relationship between Hall and drift mobility is more complicated, so that in many practical cases the difference is neglected completely. This may be justified if the electrical properties of the semiconductor or the efficiency of a solar cell are determined by other more important factors. Hall mobility measurements can be very useful in the characterization of new semiconductor materials where informations about the defect structure and impurity concentrations are very limited. If it can be established experimentally that one scattering mechanism is dominant in a certain temperature range, the doping and temperature dependence can be used to determine some of the properties of the scattering defect, e.g. the concentration of ionized impurities or dislocations. 2.4.2. Bulk recombination processes Whenever the thermal equilibrium condition of the semiconductors is disturbed by the generation of electrons and holes (pn # n 2) recombination processes exist to restore the system to equilibrium (pn = n2). The recombination process, where an electron recombines with a hole, is characterized by the lifetime of the charge carriers, which can vary over several orders of magnitude--from several hours to 10-Ss or less-~lepending on the physical process. Recombination also occurs in thermal equilibrium and balances the thermal generation of electron-hole pairs. The thermal equilibrium is in fact characterized by the condition that the rate of thermally stimulated transitions between any two energy levels is equal to the rate of the inverse recombination process (principle of detailed balance). It is useful to distinguish two classes of recombination mechanisms: the band-band recombination and recombination through the participation of crystal defects. Generally, all processes occur simultaneously and the total recombination rate is the sum of all individual processes. The generation and recombination processes between two energy levels are described by the transitions rates, the number of transitions of electrons or holes per unit time and volume. If fin and @ are the excess concentrations of electrons and holes as defined in eq. (10) their time dependence is described by the continuity eq. (13). For the case that no current flows in the semiconductor one can write d~p
0t = Gp - R n;
~n 0t = Gn -
Rp.
(46)
Usually the lifetimes zn and Zpfor electrons and holes, respectively, are introduced here instead of the recombination rates Rp and R~ which are determined by the difference between the thermal generation rates of the charge carriers, gnT and gpx, and the capture rates, rn and rp, by On R~ = --zn= r~ -- gnT; 6p
R p - - ~ p = rp -
gpT"
(47)
The calculation of the recombination velocities and of the lifetimes requires a detailed analysis of the different recombination mechanism.
SEMICONDUCTORS
FOR SOLAR
CELL APPLICATIONS
229
2.4.2.1. B a n d - b a n d recombination. Analogous to the band-band transition of electrons during the absorption of a photon the total energy and momentum of the system have to be conserved when an electron descends from the conduction to the valence band. Several processes have to be distinguished dependent on how the excess energy is dissipated and the momentum exchanged. The energy can be released by the emission of a photon (radiative recombination), or by the excitation of a second free electron or hole (Auger recombination). In the latter case the excited electron (or hole) finally transfers its energy to the lattice by the interaction with phonons. In indirect semiconductors the electron transition from the conduction to the valence band requires a change in the momentum through the participation of phonons so that part of the energy is exchanged with the lattice directly. The number of transitions, r, (per unit time and volume) of electrons or holes between two groups of energy levels, which are characterized by their density of states distributions NI (E) and N2(E), is proportional to the number of electrons in state 1 and proportional to the number of unoccupied states in group 2. Hence it is given by
r
= f s ( E , E ' ) N1 (E) f(E) N2(E' ) (1 - f ( E ' )) dE dE'
(48)
where s(E, E ' ) is the probability (per unit time) for the transition between the energy levels E and E' which can be calculated by quantummechanical methods. In thermodynamical equilibrium f(E) is the Fermi-Dirac distribution. For band-band recombination the expression can be simplified and expressed by the concentrations of electrons and holes in the conduction and valence band, respectively. In addition, it is assumed that the transition probability s(E, E ' ) for the Auger process depends on the number of collisions with free electrons (or holes) in the conduction and or valence band and is proportional to their concentration. Evaluating the integral one can show that the total recombination rate for minority electrons, R,, and holes, Rp, in a p-type or n-type material, respectively is given then by
Rn = snnp + Sanl n2P + Sa.2 p2n -- g.~, Rp = spnp + S~plp2n + Sap2n2p -- ggT.
(49a) (49b)
The transition probabilities, sn, Sp, Sanl, S~.2, Sapl, and S~p2summarize all remaining factors of the integral, and are usually determined empirically. The thermal generation rates can be derived from the condition that in thermal equilibirum R. = Rp = 0 and n = no and p = Po. Expressing the non-equilibrium electron and hole concentrations, n and p, by eq. (10) and assuming fin = 6p <<no (for n-type material) or fin = 6p <<Po (for p-type material) one can calculate R, and Rp. For the total lifetime of minority charge carriers one obtains then (45)
1 1 1 - = S.po + s~,~P~o + San2noPo = -- + - "~ "~n Tan
(p-type), (n-type).
(50) (51)
1 1 1 - = Spno + Sapl n~ + Sap2nop o = -- + - "IS Tp
Tap
The equations show that the lifetime decreases with the doping concentration. For low doping levels the probability of the Auger recombination (z,n, Zap) is usually small compared to the radiative recombination but due to the quadratic term increases rapidly with the majority carrier concentration. In fact in silicon it becomes the dominant recombination mechanism in heavily doped material (58) as can be seen in Fig. 7 where the theoretically
JPMS 35-3/4~C
230
lllll I I lllllll I
PROGRESS
I I Illlll I
IN MATERIALS
SCIENCE
dllll] I I Itllll[ 1 IIIIIII I I | [ 111111 I
1 I III111 I
0 10 s
o
&
p-Si a 77 K O 300K 6 400K
o
&
n-Si
r.i 771~ o 300K A 400K
a
w
P,\~
\
10_s
-_,_
'~ 10 -7
..~ 10 -8
\
\ \ \a
aA ~ oA~
***
"\
10 -S
"]LltlJ
1017
~ Jtlntd
1018
~ LJINdl
1019
I JlIHn[
1020
I l
till
I I IlJ~tlJ
i lU Illld
I I llllll[
1017
1018
10 TM
1020
Doping concentration,NO (cm -3)

(a)
Doping concentration, N O (cm -3) (b)
FIG. 7. Experimental minority carrier lifetimes in high quality p-type (a) and n-type (b) silicon as a function o f the doping concentration at different temperatures. The dashed lines represent the quadratic dependence predicted theoretically. (58)
predicted quadratic dependence can be observed for doping concentrations N t> 1018 c m -3 which are typical for doping levels in the emitter region. Radiative recombination lifetimes are much larger in indirect than in direct semiconductors such as most of the compound semiconductors. In pure intrinsic silicon it can be as long as 3 hr compared for instance with 6 x 1 0 - 7 s for InSb. The band-band recombination is a process which cannot be removed and therefore gives an upper limit for the lifetimes of charge carriers in semiconductors. However, it is not the only possible recombination process and in semiconductors with a large band gap the recombination is usually dominated by recombination at impurities and other lattice defects. For example the recombination through traps reduces the lifetimes in doped silicon to 100 #s or less. 2.4.2.2. Recombination at point defects. Point defects and extended lattice defects can introduce energy (trap) levels in the forbidden band gap. The investigation of the electronic properties of crystal defects in semiconductors is a major area of semiconductor research and it will be the subject of the following chapters to review these properties of the crystal defects which are important for the performance of photovoltaic semiconductors. For the following considerations it suffices to describe the defects by the position of the trap levels they introduce. Basically three different types of defects can be distinguished: defects which introduce either a single trap level, ET, or several levels, Exi, which can be charged successively with electrons or holes, or defects which possess a continuous distribution of states. In the first two cases N(E) in eq. (48) reduces to one or several sharp peaks (delta-functions) at the trap level energies Eri. Typical impurities in the first group are mainly doping elements and examples for multi-level defects are transition metals impurities in silicon. Extended defects like dislocations, interfaces and the surface of the crystal usually introduce a band of trap states which has to be described by a continuous density of states distribution ArT(E).
SEMICONDUCTORS FOR SOLAR CELL A P P L I C A T I O N S
231
Conduction Band
l gn
ikn
gp Valence Band
lkP
FIG. 8. Schematic diagram of the electron and hole transitions for a recombination process at a trap level, defining basic parameters.
The recombination through a single level trap E T which can either capture or release an electron will be derived first. In thermodynamical equilibrium there are NTfo (ET) occupied trap levels and [1 - f o (ET)]NT unoccupied levels, if NT is the total concentration of traps and f0 the Fermi-Dirac distribution. If the thermodynamic equilibrium is perturbed, by illumination for instance, the distribution function is changed and has to be determined from the kinetics of the electron and hole transitions at the level. The occupation of the trap is determined by four different processes (Fig. 8). The capture of electrons and the thermal generation into the conduction band, the capture of holes and the thermal generation into the valence band. The starting point for the calculation of the capture rates is eq. (48) which reduces for a single level to rn = STnNT [1 -- f(ET)]n; rp = STpN T f ( E T ) p . (52)
The thermal generation rates are proportional to the number of occupied (for electrons) or unoccupied states (for holes) and can be expressed by gnT = gn NT f(ET); gpT = gp ArT[1 -- f(ET)](53)
In thermal equilibrium the principle of detailed balance requires rn = gnT and rp = gpT which allows to determine the coefficients gn and gp. These parameters give the probability (per unit time) that an electron or hole, respectively, is thermally emitted from the trap level. Analogous one can define the probability for the capture process of an electron, kn, or hole, kp, by
kn = STnn k o = STpp
(capture) (emission)
(54a) (54b)
gn = STnnTl
gp = STpPTI
where nTt = nl (ET), PT1 = P l (ET), and nl (E), Pt (E) are defined by
n~(E) = Nc exp E -Ec KT
Pi (E) = Nv exp
Ev-E
K~
(55)
The constants nT~ and PT~ are determined by the position of the trap level ET and are related by the expression nTlPT1 = noPo = n~. Instead of the capture coefficients one can also define the capture cross sections, an and trp, of the trap level which are obtained if one divides by the thermal velocity of electrons, Vn, and holes, vp, respectively. STn = an Vn STp = % Vp. (56)
232
PROGRESS
IN MATERIALS
SCIENCE
The four parameters, k , , kp, g~ and gp depend on the doping concentration and temperature and characterize the trap center, as will be discussed below. Inserting the expressions (52) into the continuity eq. (13) gives a set of differential equations which can be solved for fin and @. In the steady state, for low excitation rates, low defect concentrations Nr, and Gn = Gp = G one obtains for the lifetime of electrons or holes the Shockley-Read-Hall (SRH) relationship
=
%o (no + nTl ) + Zno(Po +PTI) no + Po
(57)
The constants Zno and %o are related to the concentration and capture coefficients of the trap center and are given by
1 no = - - ;
a n Vn N T
Tpn=
1
a p Vp N T
(58)
The lifetime ~ depends on the doping level, the temperature and the nature of the impurity (through the capture cross sections and trap level). It is the quantity which enters into the eqs (26) for the photocurrent in the p--n junction solar cell. The cross sections reported in the literature can vary over several orders of magnitude, 10 -'2 cm 2 to 10 -22 cm 2. The large cross sections 10 -12 cm 2 to 10-'5 cm 2 are associated with attractive centers (positive charge states). The cross sections 10-~5cm 2 to 10-17cm 2 are usually found for neutral centers, and the small cross sections 10 -22 crn 2 with repulsive (negatively charged) centers. The nature o f a trap level can be further analyzed if one considers the relative magnitude of the transition probabilities k~, kp, g, and gp. Four different cases which are depicted in Fig. 8 can be distinguished if one considers the transition probabihties for electrons and holes between the trap level and the conduction and valence band. The ratios g,/kp and kdgp can be expressed by gA = exp EFp -- 2 E______ kp KT ' with
El = Ec + Ev - Er + K T In Nv Op~)p
N c o"n u n " (60)
k~ _ exp EF, -- E___ ___~ gr, KT '
(59)
For instance a defect is said to be a recombination center if in the occupied state the transition probability for the emission of an electron, gn, to the conduction band is much smaller than for the capture of a hole, kp, from the valence band. Similarly, in the empty state, kn must be much larger than gp. However, if the transition probability for the emission of an electron, g~, to the conduction band is much larger than the capture of a hole, kp, from the valence band, the defect is said to be an electron trap. In n-type material an electron trap is a majority trap, in p-type material a minority trap. The different cases are summarized in Table I. It is important to realize that the same defect characterized by the trap level and the capture cross sections for electrons and holes can be either a trapping, a recombination or generation center dependent on the magnitude of the parameters and the positions of the quasi-Fermi levels which determines the occupation and charge state of the center. It appears from these expressions that a trap center near the conduction band edge in n-type material is likely to be an electron (or majority) trap, and correspondingly a hole trap in p-type material will be located near the valence band edge. Trap levels near the middle of the band gap are more likely to be recombination centers. They can also be generation centers if EFp lies above EF,,
SEMICONDUCTORS
233
Table I. Characterization of a trap center with an energy level at E t by the capture (kn,kp) and emission rates (gn, gp) for electrons and holes. EF~ and EFp are the quasi-Fermi levels for electrons and holes
gn kn
g
Electron trap Hole trap Recombination center Generation center >>1 <<1 << 1 >>1 >>i <<1 >> 1 <<I
e,p-e, eFo-e,
>0 <0 <0 >0 >0 <0 >0 <0
a situation which can occur in space charge regions of a p-n junction for example. <59)The corresponding analysis of the SRH expression shows that the lifetime is mostly affected by (deep) trap centers in the middle of the band gap (n~ and p~ are large), whereas shallow level impurities reduce the lifetime less efficiently. The SRH lifetime z is a function of the doping level and decreases with increasing doping concentration first. For higher doping level the lifetime remains constant and is determined by the presence of trap level impurities and their corresponding lifetimes z ~ "[noor %0, respectively. Deep trap levels are frequently introduced by impurities such as transition elements which can be occupied by several electrons or holes. Their level structure is characterized by separate energy level ETi for each charge state (see Chap. 4.1.3). The occupation of these multilevel impurities is not described by the usual Fermi-Dirac statistics anymore and requires a more general statistical approach. If an impurity introduces several trap levels at Exi (i = 1 to m, with i = 1 being the lowest level) the probability fi that the level at ET~ is occupied is given by f =
j=l
gi exp{(iEF -- Ei)/KT} gj exp{ (jEF - EJ)/KT}
with
E i = ~ ETs
(61)
"='
where gi are the degeneracy factors for each level, and EF the Fermi energy. Generally, for the recombination the participation of all energy levels of the impurity has to be taken into account, which leads to a more complicated expression for the lifetime. (45) However, if the different energy levels are separated by more than a few KT, which is usually the case, mainly one level (close to the Fermi level) participates in the recombination process and one can apply approximately the SHR statistics for a single level. The occupation of this single level (at ETi for instance) can be expressed then by a simplified expression - - = exp - Noi KT
N|i EF
-
ETi
(62)
where N~i is the number of occupied states, Noi the number of occupied states for the level ETi_ 1. The other trap levels above at Er~ are mainly empty then whereas levels below the Fermi energy are predominantly occupied. The main difference to a single-level trap is that the active recombination level of a multilevel impurity changes as a function of the Fermi energy, which depends on the doping level and temperature. Hence the influence of the impurity on the lifetime in the semiconductor may change considerably as a function of these parameters. 2.4.3. Surface recombination The discussion of the electrical characteristics of a p-n junction solar cell has shown already that recombination at the surfaces of the device is important for the efficiency of the cell. The
234
PROGRESS
IN MATERIALS
SCIENCE
enhanced recombination is due to additional electronic states in the band gap which are introduced by the termination of the crystal lattice and the occurrence of broken ("dangling") bonds at the surface. The position and distribution of these energy states is related to the atomic structure and the reconstruction of bonds. Surfaces are usually also covered with an oxide layer or with adsorbed foreign atoms. Dependent on whether it is energetically more favorable to supply or attract electrons, these atoms can exchange electrons with the crystal and become positively or negatively charged. A similar role play structural defects like vacancies etc. in the semiconductor-oxide interface. It is therefore not surprising to observe that the surface properties can be greatly modified by surface preparation techniques like polishing, etching, ion beam bombardment etc. or also by exposure to different ambients. Therefore the control of the surface processing techniques and protection of the surface to maintain the desired properties is crucial for any semiconductor device, in particular for thin film semiconductors where the surface to volume ratio is greatly enhanced. For the following it is only necessary to describe the electronic states connected with a surface by the density of states distribution (per unit area) Ns (E). For a periodic surface structure the energy states form a two dimensional band. Though in reality the surface is rarely completely periodic the notion of a surface energy-band remains still meaningful. The band may be partly filled with electrons (Fig. 9) up to a level Eso, which is characteristic for the neutral surface and determined by the atomic structure and chemistry of the surface. If E~o differs from the Fermi energy of the crystal E F the surface states will be either filled with electrons or holes. Since these charge carriers are localized at the surface in either case a surface charge Q, builds up which is compensated by an equivalent space charge in the semiconductor to maintain neutrality. The corresponding electrical field and potential ~0~(x) leads to a band bending which shifts the surface states until the occupation limit of the charged surface Es equals the Fermi energy (or Quasi-Fermi level) of the crystal E s = E F. In the space charge region the crystal is depleted from charge carriers and assuming total depletion within a characteristic width, w, of the depleted zone the potential ~0~ can be calculated from Poisson's equation. ~2~ 1 dx 2 = E -~{ -- Q~6(X) + e[Nd -- N~ -- n ( x ) + p(x)]}. E(63)
Nd and Na are the concentrations of ionized donors and acceptors in the bulk, n ( x ) and p ( x ) can be determined from eqs (11) and depend on the position through the potential qh (x). The more general case that in addition deep level impurities are also present in the bulk will be
....................
FF
.Ns~ EvEsX>
FIG. 9. Schematic diagram o f the band structure at a surface with a continuous density o f states distribution N,(E) o f surface states. E~o occupation limit o f the uncharged surface states.
SEMICONDUCTORS
235
discussed in the next section when internal interfaces like grain boundaries are discussed. The charge distribution p ( x ) perpendicular to the surface is usually approximated by a step-like function and for instance for a n-doped crystal given by
p ( x ) = - Q ~ f ( x ) - e No ~9(x - w)
(64)
with the boundary conditions ~0s(~)= 0. 6 ( x ) and 69(x) denote the Delta and Heaviside function, respectively. The solutions of the problem gives the following expression for the surface potential ~O~o q~(0) = Q~ ~0~o- 8EEoe~" (65)
The potential barrier at the surface depends strongly on the surface charge which can be calculated from the density of states distribution of the surface N~(E) Qs = e
v+Vc
Ns (E)f~ (E + Vo) dE - eNs..
(66)
N~, is the electron concentration (per unit area) of the neutral surface, Iio = eq~so, and f~ (E) is the occupation statistics for the interface states. In thermodynamicalfs (E) is equal to the Fermi-Dirac distribution but has to be determined in a non-equilibrium situation for instance under illumination. The equations show that the surface charge depends on the potential barrier, since the position of the surface states relative to the band edges is assumed to be fixed. An additional charge at the surface changes the potential barrier and shifts all levels until the equilibrium is reached for E~ = Ev. Thus the barrier height and the surface charge have to be determined self-consistently from eqs (64) to (66). (It should be noted here that the assumption of charge-independent surface states may not be correct in general. Since additional charge is accumulated at the surface the Coulomb interaction between the electrons may change the position and density of states distribution of the surface states, analogous to the different charge states of a multi-level impurity. However, so far this possibility has not been taken into account.) The potential barrier in this approximation is positive for an n-doped crystal and negative for p-type material. In general however, the situation can be more complicated. If for instance in a n-type crystal the surface potential is very large the hole concentration could exceed the electron concentration near the surface and an inversion layer would form. The barrier could also become negative in a n-type material dependent on the position of the Quasi-Fermi level of the neutral surface, thus an accumulation layer would form. In these general cases the depletion approximation for the charge distribution is not correct anymore and one has to calculate the potential using the full expression in Poisson's equation, which usually requires numerical methods. It is evident that the calculation of Q~ requires the density of states distribution of the surface N~(E) which contains all the information about the electrical properties of the surface, but is usually not known and has to be determined experimentally. It also requires the determination of the non-equilibrium occupation functionf~ (E) which will be calculated next from the recombination statistics for the surface states. The starting point is the expression (48) for the recombination probability between two different groups of energy states. Since a planar defect is considered here the recombination rates for electrons, R~,, and holes, R~p, are defined per unit time and area. If one assumes that the principle of detailed balance can
236
be applied to each trap level separately one can integrate over the conduction and valence band states and obtain
=f
Ns(E)(1 -U (E + Vo))
+ 1Io)}dE,
(67a)
R~p = f {Sp(E)p~ Ns (E) (1 -f~ (E + Vo)) - gyp (E) N~ (E)f~ (E + Vo) } dE,
(67b)
where ns and ps are the concentrations of electrons and holes at the surface. The dimension of the capture coefficients, s.(E) and sr,(E), is that of a velocity in this case, therefore they are usually denoted as the recombination velocities of surfaces, or in general interfaces. The thermal emission of electrons and holes from the trap level into the conduction and valence band, respectively, is characterized by the thermal emission rates, g~, and gTp, which can be determined from the thermal-equilibrium condition Rs. = Rsp = 0 and f~ (E) =fo (E), where fo is the Fermi-Dirac statistic. Assuming an equal generation rate for electrons and holes under illumination for instance, or G. = Gp = G, and steady state conditions one obtains from the continuity eq. (13) analogous to the calculation in the previous section the condition Rs. = R~p = R which determines the occupation function f~(E) for the surface states under non-equilibrium conditions. Assuming again independence for each energy level one obtains
fs (E) =
s. ns + Spp,
Sn(ns + hi) + sp(Ps + P l )
(68)
where nl (E) and pm(E) are functions of the energy, E, and defined in eq. (55). Inserting the expression into eq. (67) yields
R=f{s.(E)sp(E)Ns(E)
nsPs--nlpl }dE. s.(n~+nl)+sp(p~+p~)
(69)
The recombination rate, R, of electrons and holes at the surface is completely determined, if one knows their concentrations n~ and Ps at the surface. Their calculation requires in general the solution of the continuity equation. An approximate solution can be derived if one considers for instance an n-type crystal with holes as minority carriers. Since in many cases the width, w, of the depleted region near the surface, which is of the order of the Debye length, Lo, is small with respect to the diffusion length of electrons, Ln, and holes, Lp, one can neglect the recombination in the space charge region. Furthermore, if the recombination current is small compared to the drift and diffusion currents one can assume an equilibrium between the electron and hole concentrations at the surface, ns and p~, and outside the depleted region, nw and Pw (at x = w) and obtains from eqs (11) n~=nwexp -~ , Ps=pwexp ~-~ .
This approximation is used in the thermionic emission model for the surface (and grain boundary) recombination. In general the surface charge and potential Vo are changed under illumination when additional charge carriers are generated and become trapped at the surface. They have to be calculated self-consistently from eqs (63) to (66) but for low generation rates it can be considered to be constant. Assuming that the majority electron concentration, nw,
SEMICONDUCTORS
237
at the edge of the depleted zone for an n-doped crystal does not differ too much from the bulk concentration, n, far away from the surface the integral reduces then to
nwpw - n 2 Vo nw exp~-~
sp(E)Ns(E)f~(E)dE
(71)
with the occupation function fs(E) for the surface states

1
f~(E) = 1 + exp{(E + V o - E v , ) / K T } "
(72)
This shows that the occupation function for the surface traps reduces to the Fermi-Dirac function in this case and the occupation limit of the surface states is equal to the Quasi-Fermi level of electrons in the bulk, Ev,. For low generation rates one can assume Ev. ~ EF, if EF is the Fermi level for the majority carriers without illumination. Under illumination n and p in the bulk differ from their equilibrium values no and Po- If one assumes nw ~ n and Pw ~ P and expresses n and p by eq. (10) one obtains for low generation rates 6n = 6n <<no
R =fin e x P ~ K
fsp(E)N~(E)f~(E)dE.
(73)
In most cases the assumption is made that the capture coefficients are constant sp (E) = const. so that the integral becomes equal to the total number of occupied trap states at the surface Nso. The recombination rate is determined by the capture of holes which are the minority carriers in the n-type crystal considered here. A corresponding equation can be derived for a p-type crystal where the integral becomes equal to the number of unoccupied states N~u. If one introduces the previously defined surface recombination velocities for minority carriers, S, = R/fin, in a n-type crystal, and the energy-independent capture cross sections for the trap levels from eq. (56), the following simple expressions are obtained
S n = trpvpNso exp K ~ '
Sp = anvnNsu exp VO.KT
(74)
The recombination velocities depend on the capture rates of minority carriers, the densities of occupied or unoccupied surface states, respectively, and the surface potential. The latter two parameters are both related to the density of states distribution of the surface levels as has been discussed above [eq. (66)]. Most effective surfaces are those with a Quasi-Fermi level near the middle of the band gap which is frequently the case for surface and interface states. Finally, it should be noted that in practical cases only a fraction of the surface states which determine the surface potential may be recombination centers. The remaining traps can only exchange carriers with one of the bands and are thus trapping centers either for electrons or holes according to the definitions which have been introduced in the previous chapter. 2.4.4. Recombination at extended bulk lattice defects Point defects influence the bulk properties, lifetime and mobility, of a semiconductor whereas extended lattice defects such as surfaces, dislocations, or grain boundaries change the lifetime and mobility of charge carriers only locally. In the following we will consider the electrical properties of a single extended defect first. If the density of the extended defects is so high that they cannot be considered independently from other defects a separate discussion is required. This situation can occur in crystals with high densities of grain boundaries or dislocations, for instance in polycrystalline semiconductors, which are of particular importance for solar cell applications. The electrical properties of this class of semiconductors
238
will be discussed later in Chap. 4.6 from the perspective of the properties of the individual defects. The electrical activity of an extended lattice defect in the bulk is like for the surface due to energy levels which are introduced in the forbidden band gap. Despite an extensive research, which will be reviewed in the case of grain boundaries in Chap. 4.4, still little is known about the microscopic origin of the trap levels. Basically two different possibilities are discussed: (1) On the atomic level a lattice defect is a strong perturbation of the lattice and of the covalent bonds between atoms. Though theoretical calculations show that dangling bonds can give rise to energy levels in the gap, it is not clear yet whether dangling bonds actually occur in a real structure since they are energetically very unfavorable. (2) Experimental observations indicate that the electrical activity can be related to the presence of impurity atoms at or near the defect site in some cases and it appears likely that both the structure and chemistry of the defect have to be taken into account. 2.4.4.1. Recombination at grain boundaries and interfaces. Analogous to a surface, grain boundaries and interfaces introduce electronic states in the band gap, which form a two dimensional band with a density of states distribution (per unit area) NB(E). These energy states can capture electrons and holes and lead to the formation of a grain boundary charge. The compensating space charge lies symmetric to the boundary plane for a homogeneously doped crystal (Fig. 10). The resulting potential barrier (sometimes described as a double Schottky barrier) and the corresponding depleted region in the vicinity of the boundary plane are regions of enhanced resistance and thus particularly important for the electrical transport properties of a polycrystalline semiconductor. Under applied voltage, U, the occupation of the grain boundary states changes and the recombination and electrical transport across the interface as well. Therefore it is important to understand the response of the grain boundary charge and potential to an applied voltage first. There are technological important semiconductors like ZnO for varistor applications for instance where the device properties are only determined by the electronic transport across the boundary. It has been pointed out previously that commercial semiconductors for solar cells are likely to contain a high density of impurity atoms and intrinsic point defects (particularly in compound semiconductors)
VB~ ~ . . . ~
VB+eU
FIG. 10. Band structure and potential barrier V~ of a grain boundary under bias voltage U. Density of states distribution NB(E) of grain boundary states in the insert.
SEMICONDUCTORS
239
which can introduce deep levels in the bulk. The general problem is therefore to calculate the potential barrier under a bias, U, and the presence of a high density of deep trap levels in the adjacent crystals. Similarly to the previous section one has to calculate the potential q~B(X) from the charge density in the vicinity of the grain boundary using Poisson's equation. The expression for the barrier height q~Bo= q~B(0) as a function of the applied voltage U has been derived by Greuter and Blatter. ~6) They have treated the general case that one donor level (trap level Ed, concentration Nd) and several deep traps (ETi, NTi, i >~ 1) are present which contribute to the screening of the grain boundary charge. The main results of the calculation will be discussed in Chap. 4.4. In the case of one deep trap level the following expression for the barrier height ~%o= +p,(0) as a function of the applied voltage U is obtained. q3Bo=~ - 1-where the critical voltage is given by Q~ Vc = 8EEoe(Nd + NTi)" (76)
Vc(
~c) 2
N T ( E T - - EFB)
(75)
+ e(Nd+NT)
The expression for the potential is identical to eq. (65) for the case that no deep traps are present and U = 0. The grain boundary charge QB can be calculated from eq. (66) inserting the density of states distribution NB(E) and the occupation statistics fB (E). In the approximations of the thermionic emission model fB has the form of the Fermi-Dirac statistics with a Quasi-Fermi level, Eva, for the grain boundary states given by 2
EFB = EF -- K T In 1 + e x p ( - e U / K T ) "
(77)
Similar to the calculation for the surface potential the grain boundary potential has to be determined self-consistently from eqs (63) to (66). The recombination velocity of the grain boundary under illumination can be calculated in the same way as for the surface using the Shockley-Read-Hall statistics. Under bias U the electron and hole concentrations, nB and PB, at the boundary are modified, however, compared to eq. (70) and given by nB= nw exp -- ~ B,, PB = Pw exp ~-~ Op, (78)
where VB= etp,o is the grain boundary barrier, B. = 1 + exp(-eU/KT)} and Bp = 1/B,, nw { and Pw are the electron and hole concentrations at the edge of the unbiased side of the depletion region (Fig. 10) and differ in general from the bulk concentrations n and p. In the n-doped case the majority concentration nw is essentially equal to the bulk concentration nw ~ n. Since for low illumination intensities n = n o + fin ,.~ n o one can assume that nwis equal to the equilibrium concentration nw ~ no. The minority concentration at the edge of the depletion zone, Pw, however, may be changed considerably. A consequence is that the grain boundary charge and hence the barrier VB under illumination can differ from their equilibrium values and have to be recalculated self-consistently. Several treatments of the recombination process at grain boundaries under illumination have appeared in the past few years.~61'62)A comprehensive analysis has been given recentlyC63)and is based on the following assumptions. Electrons and holes are generated uniformly in the crystal. Since holes, which reach the depletion zone edge by diffusion in the quasi-neutral bulk region are immediately
240
driven by the electrical field to the interface where they can recombine, the concentration Pw at the edge is determined by the generation of holes, the recombination at the grain boundary and in the bulk, and the diffusion current from the bulk to the boundary. A slightly simplified calculation as in the previous case will be given in the following, assuming that even under illumination the electron concentration nB still exceeds the hole concentration PB at the grain boundary. One obtains for the recombination rate R at the grain boundary the following expression then R = nwPw--n~ exp ( v B ) f -n--~ ~-~
sp(E)NB(E)fB(E)dE
(79)
where fB(E) is given by eq. (72) if one replaces the surface potential by the grain boundary potential VB. This equation is thus essentially identical to the expression (71) for the surface recombination for U = 0. For higher illumination levels the hole concentration PB at the grain boundary cannot be neglected anymore and one has to use the general expression for the occupation function given by eq. (68). The result yields a rather complex expression which has been derived in (64, 65). The barrier height can be calculated then self-consistently from eq. (75) when the boundary charge is expressed by
QB = e
n I and Pl are given in eq. (55), and NBn is the number of states (per unit area) of the neutral grain boundary. A numerical calculation of the grain boundary potential under illumination is given in Fig. 11. In this calculation it is assumed that U = 0 and no deep trap levels are present in the vicinity of the grain boundary. The results show that under illumination levels which are typical for photovoltaic applications the grain boundary potential decreases considerably. In order to estimate the effect on the recombination rate R the low illumination conditions are assumed for simplicity. The calculation ofpw yields the following expression for R = Sn gn with the recombination velocity S, given by 1
a n
;,v
NB (E)
+vB
s~nB + SpPl dE - eN~ sn(nB + n]) + Sp(Pn + Pl)
(80)
Bn exp - ~-~ O'pUpNBo f
VpLp + Dp 2 VpDp (81)
Jo" l
GENtler,oN ~ T E , ~ I S ~ S J
,a"
~E~E~.T,ON
,~1 r~l NArt, cs~.s,
io~
io*
(a)
(b)
FIG. 11. Measurements of the grain boundary potential upon illumination for a silicon bicrystal. (a) Computedvariationof the barrier height as a functionof the illuminationlevelfor differentcapture cross sections ~n of the grain boundary states (b). Parameters arc the doping concentration Nd=l.3 1016crn-3, Ex=O.ImeV, ap=2 10-14cm2, NB=2.0 1012ClT1-2, Lp=50#m and T = 300 K (1 sun = 1000W/m2).(63)
241
NBo = QB/e is the number of occupied grain boundary states, Dp and Lp are the diffusion coefficient and the diffusion length, respectively, of the minority holes. It should be mentioned here that without illumination or other generation processes 6n = 0 and hence the recombination rate for electron-hole pairs at the grain boundary becomes zero (R = 0). This result is important in view of a description of the electrical transport over the grain boundary barrier, which will be discussed in Chap. 4.4. Without electron-hole recombination the current which flows through the grain boundary under an applied bias is totally dominated by majority carrier transport then. The recombination velocity at the interface can be simplified in the two limiting cases of a high or low recombination rate of holes.
S. = O'pVpNBoBpexp VB (10W O'p),
(82a) (82b)
Sn =
2vpDp VpLp + Dp
(high ap).
A criterion for high and low capture cross sections ap of the holes can be derived from eq. (81) by comparing the two terms. Corresponding equations for the recombination velocity in n-doped crystals can be easily obtained. Since the Quasi-Fermi level for the majority carriers at the grain boundary is a function of the applied voltage, the potential lib (U) and the number of occupied states NBo(U), and hence the recombination velocity depend in a complex way on U in the first case. It is also evident from eq. (82) that the recombination velocity of individual grain boundaries may vary considerably dependent on their charge state. In a polycrystalline solar cell the voltages applied across the boundaries are usually small so that in this case EFB ~ Ev. Therefore both quantities VB(U) and NBo(U) remain almost constant and can be replaced by their equilibrium values. For high illumination levels the grain boundary potential decreases and the recombination velocity will decrease as well and approach a constant value according to eq. (82). Although the general expression for the dependence on the illumination level is more complex and requires numerical calculations the result here is essentially the same as in Ref. (63). 2.4.4.2. Dislocations. The previous calculation of the recombination velocity at grain boundaries can also be extended to recombination at dislocations, as well, and yields the same expression for the recombination rate. Since an applied voltage does not change the dislocation barrier one obtains for minority electrons in a p-type crystal
S n = O'pVpNDo exp V T
(83)
where NDo is the density of occupied dislocation states (per unit length). A corresponding equation exists for minority holes. Since dislocations are one-dimensional defects the recombination velocities, Sn or Sp, are defined for the surface of a cylinder with radius ro corresponding to the radius of the dislocation core around the dislocation line (Fig. 6). NDo has to be calculated from the occupation of the energy states NDL(E) given by
Noo = NDL(E)fDL (E + VD) dE
+re
(84)
wherefDL (E) is the corresponding occupation function given in eq. (68). Noo is related to the dislocation charge Qo = e ( N o o - No,), where No, is the density of states of the neutral
242
i(
(a)
;,
-, (b)
;,
",
F i t . 12. Effective lifetime Lefr o f polycrystalline solar silicon as a function o f the dislocation density NI); (a) for different values of the dislocation recombination velocity Sn: (1) l0 3 cm/s, (2) 104 cm/s, (3) 10s cm/s, (4) l0 6 cm/s, and (b) for different values of the bulk diffusion length L n with Sn = 103 cm/s. Black circles are measured data/~)
dislocation. Similar to the case of the grain boundary and surface the charge and potential have to be determined self-consistently. The particular features of the dislocations, such as their line character and strain field, enter the final equations therefore only in the calculation of the potential barrier VD and the dislocation charge QD. I'D has been calculated from Poisson's equation by various authors. (66'67) An approximate solution for small potential barriers is VD= QD in 2
2~E o
(85)
where the Debye screening length 2 and the line charge QD are in general a complex function of the doping concentration and temperature. The approximations which have already been given in Chap. 2.4.1 [eq. (44)] for a p-type crystal yield the following dependence on the doping concentration and dislocation density: NDo = N,/ND + ND, for a low occupation of the dislocation states. Since the presence of dislocations alters the carrier mobility # and the lifetime z in a crystal the effective diffusion length Leer= (KT#z/e) m is also changed. Several models have been developed to describe the influence of dislocations o n t e (68'69) and some of the numerical and experimental results for silicon are shown in Fig. 12. One can see that for dislocations densities above 104 cm -2 the diffusion length decreases and recombination velocities between S. ,~ 103 to 106 cm/s are obtained from the comparison with experimental measurements. The influence on Leeris more pronounced if the diffusion length of the undisturbed bulk Ln is high. So far, it has been shown that grain boundaries, and dislocations as well, influence the electrical properties of a semiconductor in two ways: the trapping states introduced by the lattice defects enhance the recombination velocity locally, and the scattering of the charge carriers at the charge of the lattice defect reduces the mobility. (Because of the long-range stress field of dislocations compared to grain boundaries there is an additional scattering component to the mobility, as has been shown in Chap. 2.4.1). Consequently the reduction of lattice defects and the use of a high quality material is one possibility to enhance the performance of the solar cells. However, under the constraint of low production costs, expensive material purification and processing steps may be prohibitive in the future. Therefore to improve the performance of a solar cell made from a semiconducting material of lower quality, the majority goal must be to limit the number of electrically active lattice defects and to reduce the recombination activity of these defects for instance by passivating them. In silicon for instance dislocations densities ND < 106 cm -2 and grain sizes larger than about 1 mm are considered as tolerable for a solar cell with an efficiency of about 10%. In
243
many semiconductors the dislocation densities can be kept below this value, as will be discussed in later chapters, however, in polycrystaUine thin film materials the grain sizes are usually in the micron range. In this case only semiconductors with high absorption coefficients are suitable so that electron-hole pairs can be generated close to the p-n junction then. 2.4.4.3. Experimental techniques. It is evident from the preceding analysis that the density of states distribution NB(E) and the distribution and concentration of trapping centers in the vicinity of the grain boundary are most important for the recombination behavior. Several experimental techniques like admittance and deep level transient spectroscopy (DLTS) have been proven to be very useful for the characterization of single grain boundaries in semiconductors like silicon, ~7-72) germanium,~73'74) o r Z n O (75'76) and for the determination of these quantities. They are increasingly applied to solar cell semiconductors and the particular problems due to the usually high defect concentrations will be discussed in Chap. 4.4. A direct determination of the recombination velocity of a grain boundary (or of dislocations and large precipitates) is possible with the Electron (or Light-) Beam Induced Current Technique (EBIC, LBIC) which is schematically depicted in Fig. 13. Electron-hole pairs are generated by an electron (or light beam) in a Scanning Electron Microscope (SEM) for instance and separated in a built-in electrical field of a Schottky contact or a p-n junction. The resulting current is amplified and used to modulate the SEM monitor. The collected current is reduced at defects because of the locally enhanced carrier recombination, therefore the defect shows up as a dark feature in the SEM image (Fig. 14). This technique allows a direct and immediate observation of the recombination behavior of the grain boundaries and other lattice defects and is very useful for the characterization of electrical active defects in solar cell semiconductors. For a quantitative determination of the recombination velocity the recombination current is measured in a single line-scan across the boundary. The analysis of the data requires the solution of the diffusion problem for the minority carries in the vicinity of the boundary which has been given by Donolato. ~77)The results for the grain boundary contrast allow to relate the recombination velocity Sn or Sp and the diffusion length of the minority carriers Lp or Ln, respectively, to experimental parameters which are obtained from the measured line-scan I(x), namely the total profile area given by A = ~I(x) dx and the variance tr 2 = ~ x2I(x) dx. The couple of values A/R and tr2/R determines a point in the calculated rectangular diagram (Fig. 15) and for instance in an n-type semiconductor can be related to a point (Lp/R, SnLp) in the curvilinear coordinate system from which S, and Lp can be calculated. The parameter R characterizes the electron excitation and is equal to the depth of penetration of the primary electron beam, which also corresponds to the maximum depth of the pair generation. R depends on the electron energy
Laser or Electron Beam
FIG. 13. Schematic diagram of the electron beam (or light beam) induced current method (EBIC, LBIC). The recombination can be enhanced at lattice defects such as a grain boundary.
244
FIG. 14. EBIC image o f grain boundaries in polycrystalline silicon. Dark spots are due to the enhanced recombination at dislocations and precipitates. (Scale marker ~ 200/~m).
and has to be determined for each material. In silicon the following relation has been found in the energy range 5 < E < 25 kV (R in /~m, E in keV) (78) R = 0.017IE ]'75.
10
(86)
. . . .
. . . .
"
. . . .
5
20 50 100 =sL
1
AIR
L/R= I
0.5
0.1 0.05
i I . . . . I . , I . . . . I , , I . . . . I
0.5
10 o/R
50
100
FIG. 15. Calculated diagram relating the profile area A and standard deviation tr 2 of the EBIC contrast profile I(x) to the grain boundary recombination velocity S and minority carrier diffusion length L. A and a 2 are defined in the text, R is the average penetration depth of the electron beam given by eq. (83). (77)
SEMICONDUCTORS
FOR SOLAR
CELL APPLICATIONS
245
Typical values which have been obtained for electrically active grain boundaries in germanium and silicon vary between Sn = 102- 10 6 c m / s . (79) Experimentally one finds a dependence of the recombination velocity on the injection current. This can be understood if one considers the dependence of the recombination velocity on the grain boundary charge which is increased with higher injection levels. The potential decreases then as can be seen from Fig. 11 and because of the exponential dependence on VB the total recombination velocity decreases, too. For the comparison of different data it is therefore necessary to measure at low excitation levels where VB remains essentially unchanged, though in this case the signal to noise ratio becomes a limiting factor. The EBIC (or LBIC) technique is useful for the determination of the total electrical activity and recombination velocity, however, since this is an integral parameter for the characterization of the distribution of the defect states of a grain boundary, a deconvolution of the results in terms of the actual trap levels is usually not possible. 2.5. Efficiency Requirements for the Solar Cell Material In Chap. 2.3 it was described that three parameters: the open-circuit voltage, Vow, the short-circuit current, Isc, and the fill factor, FF, could be used to describe the performance of a p--n junction solar cell. Since the maximum value of the fill factor can be expressed by the open circuit voltage, only the ideal limits of Vocand Is need to be examined. It is relatively easy to calculate the upper limit of the short-circuit current I~ = IL for any semiconductor from the light generated current, IL, given in eq. (20), if one assumes that any generated electron-hole pair can be collected in the external current. Assuming ideal values for the absorption and surface recombination the current mainly depends on the photon spectrum for the sunlight So(v). Integration from the lowest possible photon energy which can generate an electron-hole pair one obtains the maximum current as a function of the band gap of the semiconductor. Figure 16 shows the result for the spectral distributions AM0 and AMI.5.
_
t~
60
/ 1
~so 4o
30
i
a
.~
._E
10
0 0.5
I 1.0
I 1.5
Semiconductor band gap leVI
1 2.0
2.5
FIG. 16. Calculated upper limits of the short-circuit current density as a function of the energy hand gap of the solar cell semiconductor for a photon flux corresponding to A M 0 and AMI.5. ~
JPMS 35-3/~-D
246
It is quite evident that the current increases with decreasing band gap, since more photons have enough energy to generate charge carriers. The limitations on the open-circuit voltage of the p-n junction solar cell are less clearly defined. For an ideal cell (no surface recombination, infinite dimensions of the cell) the following expressions for Vo and the saturation current Is were given in Chap. 2.3 K T l n { ~ + 1}, Vo= e Dp Is=eANcN,,{-~+L-~D}exp(-~-~). The diffusion constants are proportional to the mobilities Dp., = KT#p,./e the diffusion and length Lp,. = ('Cp,nOp, n) 1/2 depends on the lifetime of minority carriers which is intrinsically limited by the radiative recombination in each semiconductor. It is evident that the saturation current/~ needs to be as small as possible for a maximum Voc. One can assign favorable values to the mobility and lifetime within the range of values that are required to produce a good solar cell. The parameter which depends most strongly on the choice of the semiconductor material is the band gap energy and the expression given above shows that with increasing Eg the saturation current decreases and the open-circuit voltage increases. This trend is opposite from that observed for I=, therefore there will be a maximum in the efficiency as a function of Eg. Calculations for two different sun spectra are given in Fig. 17(80)and show that the optimum band gap occurs between 1.4 and 1.6 eV. The near-optimal efficiency for AM1.5 (29%) occurs for GaAs (1.4 eV) whereas the peak efficiency for silicon (1.1 eV) of
50[ 40
Si J GoAs CdTe G ~ ~ J ALPSb

InP
AM
500
1.5 K
30
>(0 Z uJ LL L~.
,,, 20
!
-j
GaP
~
Eg (ev)
CdS
FIG. 17. Ideal solar cell efficiency at 300 K and AM1.5 as a function of the band gap energy for 1 sun and I000 sun concentration.O 8)
SEMICONDUCTORS
247
about 26% is lower than optimum but still relatively high. The corresponding open-circuit voltage is about 0.7 eV and the fill factor 0.84. The major reasons for the efficiency losses are that the energy of photons above Eg is wasted in the form of heat. Furthermore, the output voltage is only a fraction of the maximum potential by which the electron-hole pairs could be separated, because a p-n junction is inherently incapable of fully utilizing the maximum voltage which corresponds to the band gap energy. Several technological concepts have been developed in the past to enhance the ideal efficiencies. A simple idea is to enhance the intensity of the incident sunlight, which is most effective near the maximum. For instance for 1000 suns (i.e. 844 kW/m 2) the maximum efficiency increases to 37%. This increase is primarily caused by the open circuit voltage, Voc, since the ratio I~/Pin in eq. (34) is essentially independent on the incident power as can be seen from eq. (26). This concept is realized in concentrator solar cells, which need, however, direct sunlight and a tracking system to follow the path of the sun on the sky. It is also obvious that the energy of the incident photons can be used more efficiently if the band gap of the semiconductor could be adjusted to the different wavelength ranges. This idea has lead to the development of multiple cells with different band gap energies, as will be discussed in the next section. For instance calculations for a two band gap system with Eg = 1.56 eV and Eg = 0.94 eV under 1000 suns illumination have yielded a theoretical value of about 50%342~ The ideal limits of the efficiency have been calculated assuming optimal parameters for the material quality and the technical design of the solar cell. It has been shown that the important material parameters are the lifetime and mobility of the minority carriers in the bulk, and the recombination velocity at the front and back surfaces of the cell. Though for some semiconductors like silicon and gallium arsenide a high quality material can be produced, it is generally difficult to retain the quality of the material during processing of the cell especially under the constraint of low production costs. Since the material parameters are closely linked to the technical design and the fabrication of the solar cell, the actual device characteristics of the p-n junction will be lower than its ideal values. The limitations effect the three parameters in different ways, as will be discussed next. 2.5.1. Short-circuit current The short-circuit current is reduced by losses in the light-generated current. It is evident from eq. (26) that the incident light cannot be fully utilized because of the finite reflectivity, R, which in the case of bare silicon is about 30%. Antireflection coatings on the front surface are therefore usually necessary which can reduce such losses to about 3%. Furthermore, the necessity of making electrical contacts on both sides of the p-n junction generally results in a metal grid on the front side which is exposed to the sunlight. This blocks about 5-10% of the incoming light. Finally, to collect the light inside the cell a certain thickness of the material is required which is usually larger than the width of the p-n junction, particularly for indirect semiconductors with a lower absorption coefficient. For instance for silicon a thickness of about 400-1000/~m is needed to absorb most of the light. The optical thickness of a semiconductor can be reduced by "trapping" the light inside the crystal, so that it is reflected several times between front and back surface before it is finally absorbed. A fraction of the charge carriers is, however, always generated far away from the junction and some losses occur because minority carriers recombine before they can diffuse to the device terminals. The main recombination centers are all electrically active defects in the bulk of the semiconductor, the surfaces and interfaces in heterostructure solar cells. The material quality has therefore an important influence on the magnitude of the light-generated current.
248 2.5.2.
Open-circuit voltage
The recombination at the surface and in the bulk is also the fundamental process that determines Vow.Generally, the recombination processes in the entire cell should be minimized as much as possible. As can be seen from eq. (57) for the bulk recombination the lifetime is especially low if the carrier concentrations, n and p, and the parameters, n~ and p~, are small which occurs for trap levels near the middle of the band gap and carrier concentrations near the intrinsic concentrations. These conditions are especially met in the depletion region of the p-n junction if appropriate trap levels are present. Therefore recombination in the space charge region, which has been neglected in the calculation of the ideal current-voltage characteristics of the p-n junction can actually become very important. The addition of the recombination current yields modified current-voltage characteristics given by I=Isn expn~-I --IL (87)
where the parameter, n, is known as the ideality factor which varies in general with the current. For a single-level center one obtains, however, n = 2 and a saturation current, Isn, which is given by Isn = ~p),/--'--------~ (~n
enWA
(88)
where W and A are the width and total area of the junction, respectively. The non-ideal current-voltage characteristics have reduced open-circuit voltage compared to the ideal case and the efficiency usually decreases with increasing ideality factor, n. The current-voltage characteristics of the p-n junction are further modified because of a parasitic series, Rs, and a shunt resistance, Rsh, associated with the solar cell. The origin of the series resistance are the bulk resistance of the semiconductor and the resistance of the contacts and interconnections. The shunt resistance can be caused by lattice defects, like grain boundaries or large precipitates, in the depleted region or leakage currents around the edges of the cell. To evaluate the effect of these parameters one can describe the current-voltage behaviour of the p-n junction in the equivalent circuit shown in Fig. 18 and obtain I = I s n exp
e(U-IRs) n~'T
1 +
U-IR~ IL. R~----~
(89)
Plots of the equation for various combinations of the series and shunt resistance show that essentially the shape of the current-voltage characteristics and hence the fill factor FF is changed. Whereas a shunt resistance as low as 100 f~ does not appreciably change the power output of the device, it can be seen that a small series resistance of only 5 t~ changes the efficiency by 30% already. For a typical n-on-p type silicon solar cell the series resistance is about 0.7 f~. 2.5.3.
Temperature effect
A considerable fraction of the incident light is transformed into heat and hence the operating temperature of a solar cell can vary over a wide range of temperatures, especially in the case of concentrator cells. It is therefore necessary to understand the effect of temperature on the performance of the cell. The material parameters which mainly change with the temperature are the band gap energy which usually decreases and the minority lifetime which increases with increasing
249
RS
80-
\Rsh tOOR
60-
20 E
H
///i
:i
0 -20
-40
RS=0 / -8O -100

-08 -0.6 -0.4
R .t" 100~.
i
-/
-0.2 0
- -- I
0.2 0.4 0.6 0.8
1.0
V {VOLTS} FIG. 18. Calculated I-V characteristics for various p-n junction solar cells under illumination that include series Rs and shunt resistance R~. The insert shows the equivalent circuit. (3g)
temperature. This will increase the light-generated current and thus I~ slightly due to the increased light absorption and the increase in minority-carrier diffusion length. However, the open-circuit voltage will decrease more rapidly because of the exponential dependence of the saturation current on the temperature eq. (30), and correspondingly the fill factor will degrade, too. Therefore, the overall effect causes a reduction of the efficiency as the temperature increases. One can approximately estimate the change A Vo~if one differentiates eq. (32) with respect to the temperature and considers only the significant terms
A Vo~ ,,,~ e Vo - y K T -
Ego A T.
(90)
eT Ego is the band gap at T = 0K, and 7 a parameter which summarizes the temperature dependence in the prefactors ( ~ T -y) of IL/Is of eq. (32) and varies approximately between 1 and 4. Inserting appropriate values for silicon one obtains that the open circuit voltage decreases with temperature ( ~ 2.3 mV/C at room temperature) and the efficiency is reduced by about 0.5% per one degree Celsius. This effect is reduced for larger band gap materials, such as GaAs where the sensitivity to increasing temperature is about half as much compared to silicon. 3. TECHNOLOGY OF SOLAR CELL DEVICES In the previous chapters the standard type of a p-n junction solar cell and the main parameters which are determined by the properties of the material have been discussed. In
250
PROGRESS
IN MATERIALS
SCIENCE
order to fully utilize the potential of a material the device structure has to be optimized, too. Important factors which can be controlled by the device configuration are for instance the position and doping concentrations of the p-n junction, the geometry of the ohmic contacts which contribute to the series resistance and block part of the incident light on the front surface, and the surface reflectivity. Many variations of the basic solar cell concept have been proposed for achieving higher conversion efficiencies and some modifications shall be considered next. There also exists a large variety of device concepts which are not based on the p-n homojunction cell. They are developed to utilize different and frequently novel materials and may offer the potential for higher efficiencies at lower costs in the future. Some of these concepts will be discussed in the second part of the chapter. 3.1. Single-Junction Devices The standard p--n junction cell has been depicted previously in Fig. 3. The doping profile and the distance of the p-n junction from the front surface determine the built-in voltage, which finally determines the open-circuit voltage, and the current which is generated by the absorbed light. The constraints are to keep the material parameters in a range which gives a maximum efficiency. Generally numerical programs are used to optimize all parameters, however, it is possible to derive some guidelines for the design of the solar cell from the analysis of the basic eqs (26) and (32) which describe the p-n junction characteristics. In the following discussion the given parameters will refer to a silicon n-on-p type solar cell, but the conclusions are valid for other semiconductors as well. First, the optimum conditions for the short-circuit current which is given by eq. (20) will be analyzed. It turns out that the key parameters are the width of the n-type (emitter) and p-type (base) region, W~ and Wp, and the corresponding diffusion lengths, L~ and Lp. The width of the depletion region, W, will be neglected in the following. For a given frequency, v, of the incident light and the corresponding absorption coefficient, a(v), the total current collected from photons with the energy hv has been calculated in Chap. 2.3 and is given by the expressions (21) and for the diffusion currents of minority carriers at the edges of the depleted region, where it has its maximum value. The current decreases towards the front and back surface and the decay is determined by the recombination velocities at the surfaces. Both currents have a maximum value for a certain width of the emitter and base region, which can be determined by differentiating the expressions with respect to these parameters. The maximum value for the base width can be approximated for large and low surface recombination velocities by
wPax=l_ L. C in { I ~ L ,
S.L./Dn>>I '
(91)
Ln In{ 2 } W pax = 1 - ctL. OtL. (1 -+ ~L,)
S . L . / D . <<1.
(92)
For a large recombination velocity at the back surface and ~L. ~< 1 (longer wavelength) the optimal width becomes approximately equal to the diffusion length of the base region WpaX~ 0.7 Ln. The base current in this approximation is given by Jn = e Sn(1 - R)~tL. exp(-~t(W + W,)) S.L./Dn>> 1 (93)
and is proportional to the diffusion length in this region. IV, and W are the width of the emitter and depletion region, respectively. Since the photons with the lowest possible energies
SEMICONDUCTORS
251
also have the lowest absorption coefficients they should be absorbed in the base region of the p-n junction. For silicon one has near the absorption edge Ct(Vmin)~10-2--10 -3 #m -1, hence a total thickness of the device of about 100-1000 #m is required to absorb the main fraction of the incident light. Typical silicon solar cells have a thickness of about 300 to 400 #m which corresponds mainly to the width of the base region. On the other hand, it is evident from the equation for the current that the diffusion length of the base region should be as large as possible, whereas Wn and W have to be kept small. Therefore diffusion length and width of the base region have to be approximately equal for a maximum performance which yields values for the diffusion length in silicon > 100 #m. This is a large value and requires a high quality of the semiconductor material. The design of the front region of a solar cell is partly determined by the condition that the current which has to be extracted from the cell, has to be collected at contacts which partially cover the top of the cell. Therefore current must flow laterally through the top layer of the semiconductor whereas generally it flows perpendicular to the surface. It has already been noted that the series resistance of a solar cell is a critical parameter for the performance, therefore the resistance of the top layer must be kept small by doping. The minimization of the power losses due to shading of the front contacts and the resistance of the top layer determines the range of the doping level in the front region. In practical cases the doping levels are kept as high as possible, which also determines the diffusion length for this region as will be discussed below. Typical values for the doping concentration are Nd ~ 1019cm -3 and for the diffusion length Lp ~ 5-10/zm in the case of a n-on-p silicon solar cell. These constraints have to be considered if one discusses the total light-generated current in the cell. The shorter wavelengths are absorbed in the front region of the cell and the best parameters are determined from the optimization of the total current. In Fig. 19 the normalized light generated current (internal spectral response SR) collected from the n- and p-type region (for a given wavelength) is plotted as a function of the normalized emitter width, Wn/Lp.Optimum values for the base width and current have been assumed here and the parameter, aLp, is varied. For low absorption coefficients (aLp < 1, curve 1 and 2) the optimum value for the emitter width is about twice the diffusion length of the n-type front region, whereas the maximum for high absorption coefficients (aLp > 1, curve 3 and 4) lies at the front surface. The blue part of the spectrum (here aLp > 1) is mainly absorbed in the front region, hence
JL/Jo
0_8 0.6 0.4 0.2 1
2.0
4.0
6.0
W n /Lp
FIG. 19. Internal spectral response SR = JL/Jo (normalized light generated current) as a function of the normalized emitter width Wn/L_ for an n-on-p base solar cell diode. The calculation is for an opUmlzed base current and a neghglble current from the depletion region ( W = 0). Surface recombination Sp = S n and diffusion length Lp = L n are assumed to be equal in the n- and p-region. Parameters are SpLp/Dp = 10 and the normalized absorption coefficient ~Lp = 0.1 (1), 0.32 (2), 1.0 (3), 3.2 (4).
. . Y. . . .
252
PROGRESS
IN MATERIALS
SCIENCE
the current decays vary rapidly with increasing width of the emitter region. In order to obtain the total current one has to integrate over the entire spectral distribution of the incident light eq. (26). It is evident that the position of the junction should be as close as possible below the front surface so that the blue and violet part of the sun spectrum can be utilized. This condition also ensures that the base current is optimal. Typically the junction depth is about 0.2-0.4/~m below the front surface. The standard type of solar cell is therefore a n+-p junction for a p-type substrate with a highly doped front emitter region. Once the doping and width of the emitter region are specified the key parameter for the light generated current 1Lis the diffusion length of the substrate material as has been discussed above. The diffusion length Lp=(Dp%) 1/2 depends on the mobility and lifetime of the minority carriers in the base region. It has been discussed in Chap. 2.4.1 that the mobility decreases slightly with increasing doping concentration above 10~6cm -3, whereas the lifetime depends on the doping in a more complicated way. In Chap. 2.4.2 it has been shown that several recombination mechanisms can contribute to the lifetime. In a pure monocrystalline semiconductor radiative and Auger recombination determine the lifetime which depends on the doping concentration of the p-type side for instance by the approximate expression(5) Zl = . SuN, + SanlNZ~
1
(94)
For silicon the Auger recombination dominates for concentrations above 1018cm -3 as has been discussed previously. If deep trap impurities or other lattice defects are present the lifetime for this process takes the form for the p-type side
,
r2 O.nvnN v
1+
(95)
where the parameters have been defined in Chap. 2.4.2. Corresponding expressions are given for an n-type region. The total lifetime is determined by the sum of the inverse lifetimes for the individual recombination mechanism 1/z = E (l/z0 and in general decreases with the doping level. Therefore the diffusion length and the short-circuit current I~ = IL decrease with the doping level in the base region. To estimate the efficiency of the solar cell one has also to consider the open-circuit voltage which depends on the light generated current IL and the saturation current Is of the diode as given in the expression. One can see how the doping concentrations on both sides of the junction effect the saturation current if Is is re-written in the form / K T \ 1/2 f-~/2Fp ,,l/2Fn ) / Eg \ t'~n t~p I s = A [ ~ - - } NcNv~l/-iT~+ 1/--7~T-v e x p ( - - p ~ } . ~ (96)
The parameters Fn and Fp are also a function of the doping levels as can be seen from eq. (3 l) but in general vary only weakly. Considering the doping dependence of all parameters in eq. (96) it is evident that the saturation current decreases with increasing doping. Since the high doping level for the emitter region is required for other reasons, the doping level of the base region has to be determined from the condition of maximum efficiency. Although the light generated current decreases with the doping the open circuit voltage tends to increase with increasing (moderate) doping levels of the base region. Because of the opposite dependencies of I~ and Vo upon doping there will be an optimum substrate doping for maximum efficiency of the cell. Typical acceptor concentrations for a high efficiency silicon cell are Na ,~ 1016cm -3 for the substrate material.
SEMICONDUCTORS
253
Several device modifications of the standard n+-p (or p+-n) junction solar cell have been developed to improve the efficiency and three different types will be discussed. So far it has been assumed that the recombination velocity at the back surface is very high which is usually the case at ohmic contacts. The recombination at the front surface is generally lower because of the smaller contact area. Typical values for a silicon/metal interface are Sn = 106 c m / s . An analysis of the base current which contributes to I~ and the saturation current I s which contributes to Voc shows that a reduced recombination at the back side can increase I~ and reduce Is, so that the overall effect is an increased efficiency. One way of creating an effective low recombination velocity is by using a back surface field (BSF) which can be formed by a junction between highly and lightly doped material of the same dopant (see Fig. 20). Such a back surface field causes a repulsion of minority carriers from this highly doped region near the back surface and can be described by an effective recombination velocity Seer. A numerical calculation for the p + - n - n solar cell shows the dependence of Seeron the BSF thickness in Fig. 21) 813 A further reduction of the saturation current and hence an increase of Voc is possible if one forms a p-n junction with a non-uniform base doping/81) A typical example is the drift-field p-n(x) silicon solar cell which consists of a heavily doped p-type emitter and a n-type base where the doping increases towards the back surface. Similarly to the cell with a back surface field the influence of the surface recombination is smaller when a drift field is present. Non-uniform doping is usually also used for doping of the emitter region. The dopant is diffused from the surface into the moderately doped substrate which produces a concentration gradient in the emitter region. As will be discussed in Chap. 3.1 very high dopant concentrations can occur near the surface which may give rise to variety of detrimental effects for instance the formation of metal precipitates in phosphorus doped regions/82) Another ramification of the high dopant concentration is the lowering of the diffusion length which may be so severe near the surface that no current can be obtained from photons with high energies which are preferentially absorbed near the surface. Such "dead layers" have been produced in earlier silicon solar cells when high dopant concentrations were used to reduce the top layer resistance. The problem can be solved by decreasing the doping level below and reducing the width of the emitter region/83~ These much shallower solar cells are known as "violet" cells because of their better response to the shorter wavelengths (hv > 2.75 eV). The doping levels are typically about N d ~ 1018cm -3 and the junction depth IV, = 0.2/zm. For a numerical analysis of the high doping region one has to account for the narrowing of the band gap which has been mentioned before and the enhanced reduction of the lifetime due to Auger recombination.
Sn
[cm/s]
Dopingconcentration ~
n(x)
I~,//~1
10 3
In-ease
.SF
_V/'AI
I'"
10 2
Porltlon x
FIG. 20. Schematic diagram of the doping profile at the back side contact o f a p - n junction solar cell with a back-surface field (BSF).
I I I I I I t 1.0 2.0 3.0 4.0 BSF region thickness [I.tm]
FIG. 21. Calculated surface recombination velocity S,, as a function of the thickness of the BSF region/8P'
254
Besides the configuration of the emitter region the design of the top surface is an important area of cell design. It has already been mentioned that the development of the metallic grid on the surface has to be optimized with respect to the sheet resistance of the highly doped emitter region and the shadowing of the cell by the contact fingers. The various techniques and problems connected with the grid design have been described elsewhere ~84)and shall not be considered here. Another important problem is the optical loss due to the finite reflectivity, R, of the semiconductor material. Basically two approaches are used to decrease R. Most commonly used are antireflection coatings on the top surface of a material. The idea is to adjust the thickness of the layer so that reflected light at the top surface is out of phase with reflected light from the interface to the substrate. Extinction of the reflected light occurs for a phase difference of 2/4 which can naturally only be fulfilled for a single wavelength. A minimum of reflection occurs for Rmin
=
{n~--nn2~ In ~ n 2 + non2 }
(97)
where no, nl and nz are the refractive indices of air (or glass), coating and substrate, respectively. The reflectivity is zero if n 2 = nonz which in practical cases is adjusted to the wavelength at the maximum of the sun spectrum. The weighted average of sunlight which is reflected from the coated silicon surface can be kept at 10% compared to 30% for bare silicon. Antireflection coating materials have to be transparent and must have the correct refractive index. Typical materials for silicon are amorphous layers of SiO2, AI203 and Si3N 4. The use of multilayer coatings with different indices can further reduce the reflectivity. Another possibility of changing the reflectivity is by texturing the surface (Fig. 22c, d). This can be produced by particular etchants which preferentially attack inclined crystallographic planes so that pyramidal structures form385~ With the combination of several techniques it is currently possible to keep the reflectivity well below 3%. The conventional p-n junction device concepts which have been described so far yielded efficiencies of less than 17% for mono-crystalline silicon solar cells. With the subsequent availability of very high purity, float zone material and the increased research in cell modelling and design, efficiencies up to 22% have evolved during the last years.(86-89)These achievements in cell performance could be obtained by more complicated cell designs and structures. The main purpose is to collect the incident light better by minimizing the front contact area and improving the anti-reflection coating, and by an improvement of the junction properties. Some of the recent developments are illustrated in the device cross-sections in Fig. 22. The
Light
cntact-'~/
AI--~"
h*-bus
bar
~ I ~
t-...... .
--..-.....-. -....~- ....~..\\..~__ back contact
~ ~ , Light
FIG. 22(a)
FIG. 22(b)
SEMICONDUCTORS
255
Laser groove
xlde
::.,,o,
..,,.co,.o,
,,/
Back contact
FIG. 22(c)
Fig. 22(d) FIG.22. Schematicrepresentations of recent high efficiencydesigns for silicon solar cells. (a) Passivated emitter solar cell (PESC) design,(86)(b) back-side point contact cell,(88)(c) buried contact cell with laser grooved surface, (d) SEM image of a textured (etched) GaAs surface (courtesy of S. Bailey). (Scale marker _-_.2/~m). "passivated emitter solar cell" (PESC) design (Fig. 22a) by Green e t al. (s6) has an especially high open-circuit voltage of about 0.66 V and an efficiency of 20.5%. A further evolution of this cell concept is the laser grooved, buried contact silicon cell shown in Fig. 22c. This approach uses junctions perpendicular to the incident light and has thus only a minimal front contact area. In addition, the cell has front and back surface texturing and a back surface field. Another possibility to eliminate the shadowing losses of the front contacts is to bring both contacts out to the rear of the cell. Figure 22b presents a configuration of a point contact cell for which efficiencies of about 22% have been reportedfl 8) The major drawbacks so far
256
are the more sophisticated cell design and hence the higher production costs. However, the excellent performance demonstrations have provided the knowledge and the basis for the further development of less expensive high efficiency cells. Though most p-n homojunction solar cells utilize single or polycrystalline silicon they are not restricted to this material. In fact compound semiconductors such as GaAs, InP, CdTe and various ternary/quaternary compositions of these compounds such as GaA1As, GaAsInP have been receiving considerable attention for photovoltaic applications (22'9)because of their excellent photovoltaic characteristics. GaAs for instance has an almost perfect band gap energy for optimal efficiencies (see Fig. 17). The cost of the high quality single crystals is a major contribution in these technologies and solar cells utilizing compound semiconductors are usually considered for space applications or terrestrial concentrator cells (see Chap. 3.5), in which the cost of the cell is less significant compared to other requirements such as higher performance and long term stability under the intense irradiation in space (irradiation resistance). Especially much of the work on InP has recently been directed towards space applications, primarily because the cells offer a better irradiation resistance compared to silicon and GaAs solar cells. ~) Most compound semiconductors are direct band gap materials and have therefore higher absorption coefficients. The devices require therefore less material compared to silicon and problems with high surface recombination velocities are even more serious for the conventional homojunction structures. The technique to reduce the recombination influence of the top surface was to make the top layer as thin as possible. For a highly absorbing semiconductor like GaAs the emitter region in this case has to be an order of magnitude thinner ( ~ 500/~) compared to silicon. Therefore for fabricating cells on the basis of GaAs and other compound semiconductors epitaxial growth techniques such as Metalorganic Chemical Vapor Deposition (MOCVD), Liquid Phase (LPE) or Molecular Beam Epitaxy (MBE) are usually preferred. A common p-n homojunction structure for a GaAs based solar cell is shown in Fig. 23. Efficiencies of about 24% for GaAs and about 19% for InP (92)single crystal cells have been reported. A summary of the status of current (1989) cell structures and parameters for a variety of technologies has recently been given by Kazmerski. (s) Material related problems for some semiconductors may require other device structures which are more suitable for the particular material. A device type which has recently been used for CdTe (93) and amorphous silicon (a-Si) is the p-i-n diode structure (Fig. 24). This design differs from the conventional n-p homojunction by the extended depletion region
l i
AR n +- GaAs p -
coating
V//////.~///////////2/~////2/,~
p+ - GaAIAs p - GaAs
(emitter)
(window) (emitter) (base)

(buffer)
GaAs (base)
(buffer)
n +- GaAs
p+- GaAs n - GaAs
p+- GaAs
(substrata) Back contact
n +- G a A s
(substrata)
(a)
(b)
FIG. 23. Cross sectional representations of two high efficiency structures for GaAs-based solar cells: (a) shallow homojunction n-on-p cell with a thin (,~ 500 A) n+-GaAs emitter region, (b) heteroface p-on-n cell with a thin ( ~ 500 A) p-A1xGa 1_ xAs window region. Al xGa I _ xAs has a large indirect band gap for x > 0.8 and thus a low absorption in the window.
257
I
k
Glass ITOorSn0
I
~;~i~3 ]
p-doped IPShH n-doped IPSi:H
(at)
(b)
FIG. 24. Cross sectional representation of two p-i-n cell design for an amorphous Si: H thin film cell; (a) films on glass substrate, (b) on steel substrate, which also serves as a back side contact.
where the semiconductor is basically intrinsic. Most of the photocurrent for this structure is generated in the depletion region as given by eq. (25) and the design of the intrinsic region offers a great flexibility for the cell optimization. The thin film is either deposited on a glass substrate which covers the front surface of the cell (Fig. 24a), or on a metal (steel) substrate, which can be utilized as a back side contact (Fig. 24b). 3.2. Heterojunction Devices The previous discussion has shown that the energy of the incident light is only partly utilized for a particular semiconductor. Theoretically one can expect an improvement of the efficiency of a solar cell device if it consists of materials with different band gap energies which match different parts of the solar spectrum. This concept is realized in a heterojunction device which is formed between semiconductors with different band energies. A typical n-on-p heterojunction band diagram is shown in Fig. 25, The material with the larger band gap Egt is on the top. Light with energy less than the band gap energy Eg] but greater than Eg2 passes through the first semiconductor, which acts as a window, and will be absorbed by the second semiconductor. Carriers generated in the depletion region and within a diffusion length of the junction are collected similar to a p-n homojunction solar cell. However, light with photon energies larger than E~] will be more efficiently utilized by the first semiconductor. The advantages of heterojunction solar cells over conventional cells include enhanced short wavelength response, lower series resistance, if the first semiconductor can be heavily doped, and higher irradiation resistance. The band diagram of the heterojunction and the depletion region are determined by the work function (the difference between the vacuum level and the Fermi energy) of the two semiconductors ~bi, the electron affinity Xi (the difference between the vacuum level and the
Ec
j
f
~ Eg2 p - GaAs
Ev
LighL
n -
AIGaAs )
FIG. 25. Schematic band structure of an n-on-p AIGaAs-GaAs solar cell showing a spike in the conduction band at the interface due to the difference in the electron affinities AXi of the two semiconductors. The internal potential barrier is determined by the difference in the work functions A~bi. The light enters through the semiconductor with the larger band gap first.
258
PROGRESS
IN MATERIALS
SCIENCE
conduction band edge) and the band gap energies E~. The built-in potential tpnp is simply the difference between the work functions Pnp= ~b2- tk~ and the spatial distribution on the space charge can be calculated similar to the p-n homojunction. The main difference for a p-n heterojunction is the discontinuity in the conduction band energies which is equal to the difference in the electron affinities AEc = ~(2 ~1 and the band offset AEv = Eg2 Eg1-~-~2 ZI" A negative AEc (or AEv for a p-on-n heterojunction) produces a spike in the conduction (or valence band, respectively) which is undesirable for photovoltaic applications. The spike impedes the flow of minority carriers across the junction from the p-type to the n-type region and the photocurrent will be reduced. Such spikes can however be avoided by a suitable combination of electron affinities and doping levels.(94)For the ideal case with negligible spikes the expressions for the photocurrents in heterojunctions are essentially the same as in heterojunctions. For an n-on-p heterojunction, the minority electron and hole photocurrent densities are given by the expressions (21) and (23), except that the absorption coefficient and the diffusion lengths Lp and L, are replaced by the corresponding quantities for each semiconductor. In addition, the photocurrent density from the depletion region is given now by
-
Jdp ---- eSo(1
- R)[exp(-~,xn)(1 - - e x p ( - ~ n W l )
) d- e x p ( - - c t p X p ) ( 1 - - e x p ( - - ~ p W 2 ) ) ]
(98)
where W~ and WE are the widths of the depleted regions on either side of the heterojunction and the total width W = W~ + I412. In the latter expression it has been assumed that no recombination takes place in the depletion region. For a normal homojunction in a high quality single crystal the recombination losses in the depletion region can in fact be neglected if the concentration of recombination centers is negligible. For heterojunctions, one has the inherent problem that the crystal structure changes across the junction and an interface is formed between the two semiconductors. It has been mentioned in Chap. 2.4.4 that interfaces can be efficient recombination centers because they can introduce deep trap levels in the band gap. They can also provide sites for quantum mechanical tunneling processes for current transport across the junction. In any case the interface traps degrade the performance of the solar cell and it becomes essential to produce heterojunctions with a low density of interface traps. Though the origin of the interface states is not completely clear yet and an area of intense research, as will be discussed in Chap. 4.4, it seems to be that the density of interface traps is related to the degree of mismatch between the crystal lattices of the two semiconductors. Therefore the requirements for a good n-on-p heterojunction solar cell are a small AEc and a good lattice mismatch (for a p-on-n heterojunction AEv has to be small). A system that fulfills these requirements is for instance n-AlxGa~ _xAs-p-GaAs in which the A1GaAs is the semiconductor with the wider band gap. As the composition, x, of the A1GaAs layer increases the band gap energy Egl increases and the spectral response extends to higher phonon energies. The AIGaAs layers can be deposited on single crystal GaAs substrates by MOCVD or MBE. These are standard techniques used for microelectronic devices and produce high quality, single crystalline epitaxial layers. A derivative of the heterojunction cell is the heteroface cell, which is a p-n homojunction to which a semiconductor having a larger energy gap has been added. The most common cell structure for the A1GaAs~3aAs system is represented in Fig. 23b. Because of the good lattice match of the A1GaAs layer it can be grown epitaxially with a low density of interface states. For compositions of x > 0.4 the AlxGa~_xAs becomes an indirect semiconductor and the absorption therefore low. In essence, this top layer will absorb only little light and act as a window for the underlying p-n homojunction. The main advantage is that the A1GaAs
259
layer passivates the surface of the underlying GaAs and effectively reduces the surface recombination velocity. In fact, this device structure has resulted in the best efficiencies reported for single junction cells so far with values of about 24%. A class of materials which are frequently used in heterojunction devices are conducting glasses (oxides). They serve not only as part of the heterostructure but also as an antireflection coating and contact material. The oxide semiconductors include ZnO (Eg = 3.35 eV), SnO2 (Eg = 3.5 eV, electron affinity Z = 4.8 eV), InO (Eg = 3.5 eV, Z = 4.45 eV), and ITO, a mixture of SnO 2 and In203 (Eg = 3.7, Z = 4.2 to 4.5 eV) and have good electrical conductance and optical transparency in thin film form which makes them useful for photovoltaic applications. 3.3. Schottky Barrier and MIS Solar Cells A Schottky barrier is formed when a metal and a semiconductor are brought into contact. In a Schottky barrier solar cell the light-generated charge carriers are separated by the electric field that is created by the space charge near the interfacial region. A schematic band diagram is shown in Fig. 26. The difference of the work functions of the metal and the semiconductor gives the contact potential q~sc= ~ M - q~s. More important is the barrier height q~Bs which determines the current transport over the Schottky barrier and is given for an ideal diode by the difference of the metal work function and the electron affinity of the semiconductor ~%s = ~bM-- Xs. Under applied bias the barrier height changes slightly because of the image force induced lowering of the potential energy A~%s = (eE/4nEsEo)1/2 for charge carrier emission into the semiconductor (Schottky effect), but this effect is small for low electric fields (E < 105 V/cm) applied in solar cells. In practical cases, however, it is frequently found that the barrier height is independent of the work function of the metal. This has been explained by the presence of electronic states in the band gap which are introduced by the metal-semiconductor interface. Though their origin can generally be attributed to the atomic defect structure of the interface and the effect of impurity atoms, many details are still not clear and primarily determined experimentally. The barrier height in the presence of interface states can be approximately (95~expressed by q)BS=
C(~t)M --
ZS) "[- (1 -- C) 1 e
(Eg --
Ex) - AtPBs
(99)
where c is given by c = Ei/(E~+ e 26Nss)'Ei and es are the permittivity of the interface layer and the semiconductor, Nss is the density of states of the interface levels and Ex the occupation level of the neutral interface, 6 is the thickness of the interface layer. For many semiconductors experimental results have shown that the metal-semiconductor interface introduces a
11,
hv
FIG. 26. Schematicband diagram of a Schottky barrier solar cell under illumination. The insert shows the density of surface states.
260
distribution of trap levels located in the lower half of the band gap approximately JET,~ Eg/3. The density of states distribution is in the range of Nss ~ 1013 eV -l c m 3. For photovoltaic applications the metal must be thin enough to allow a substantial amount of light to penetrate into the semiconductor. Typically the metal layers are about 100 tl, thick which gives for instance for the transmission coefficient, D = 1 - R, of a gold film with antireflection coating D ~ 0.9. The advantages of Schottky barriers include (1) low temperature processing, so that the starting properties of the substrate material can be maintained, (2) adaptation to polycrystalline and thin film solar cells, (3) high current output and good spectral response because of the presence of the depletion layer right at the surface, and (4) high radiation resistance due to high electrical field near the surface. The short-wavelength light entering the semiconductor is mainly absorbed in the depletion region, whereas long-wavelength light is absorbed in the neutral base region. The major contributions to the photocurrent IL is therefore from the depletion and the neutral base region and can be calculated similar to the p-n junction. The basic current-voltage characteristics of a Schottky barrier solar cell under illumination is given then by eU
Is=AA*TZexp(~-~)
(100)
where U is the applied voltage, n the ideality factor, A the total illuminated area, and A * the effective Richardson constant (,~ 30 to 120 A/cm 2K 2). The conversion efficiency given by eq. [34] increases with the barrier height. Taking PBs= Eg as the limiting case, the maximum efficiency is about 25%. However, for most metal-semiconductor systems made on uniformly doped substrates, the maximum barrier height is only about 2Eg/3 which reduces the efficiency considerably, about 10% for Si and 15% for GaAs for an ideal solar cell. The performance of Schottky barrier solar cells is therefore limited. These practical limitations can be partly removed by inserting a thin insulating layer between the metal and the semiconductor. The current-voltage characteristics of such a MIS solar cell device is similar to the Schottky diode (Fig. 27). The saturation current contains an additional term due to tunneling currents through the insulating layer and is given by /' etpBs I s = AA*T2exPl--K-~lexp[--(eqgx)\"~ ,] [/2
Wi]
(10l)
~+~/~.+"~'Y~'+~/#.+'/~+wa+~//~+~/~.+~%~#,+~ ~ ~ ~ ~ ~ ~%~ ~:~ ~ ~ erslon layer /
20 - 30 A ~ m m ' ~ l I
EF
FIG. 27. Cross section o f a solar cell having a metal-insulator-semiconductor (MIS) structure (a). Corresponding band structure of the MIS cell (b). The thickness of the oxide layer below the contacts is about 20-30 ~ .
261
where ~o is the barrier height presented by the insulating layer and wt the insulator thickness. T The open circuit voltage can be easily determined from eq. (29) and is given b y (96) Voc=
nKT{ In ~-g--T5+ --k--~IL eCPos_k(ecPT),/2wi}
(102)
This equation shows that Voc increases with the insulator thickness. On the other hand the short-circuit I~ will decrease, causing a degradation of the conversion efficiency. An optimum oxide thickness for a metal-SiO2-Si system is found to be about 20/~. A recently developed MIS solar cell has shown efficiencies of about 15%.
3.4.
Thin Film Solar Cells
In thin film solar cells the active semiconductor is a polycrystalline or amorphous thin film which has been deposited on a supporting substrate such as glass, ceramic, metal, plastic or another semiconductor. Various deposition techniques such as evaporation, chemical vapor deposition (CVD) or sputtering are available today and offer a great flexibility to form semiconductor films of various compositions. The basic requirement is that the thickness of the film is larger than the inverse of the absorption coefficient so that most of the light can be absorbed, and that the diffusion length is larger than the film thickness so that most light generated carriers can be collected. Thin film solar cells are therefore mainly made from compound semiconductors with direct band gaps and high absorption coefficients. The main advantage of thin film solar cells is their promise of lower costs since less energy for processing and relatively lower costs for the materials are required, and large scale production becomes feasible. The flexibility in the deposition techniques also allows development and utilization of novel semiconductors which otherwise might be difficult to produce. The deposition of semiconductors on foreign substrate usually results in polycrystalline or amorphous films with optical and electrical properties which can be substantially different from the single crystal behavior. This is mainly due to the large number of grain boundaries and other lattice defects as will be discussed in Chap. 4. One of the major problems with thin film cells is however that in many cases the higher defect density also reduces the efficiency and stability of the semiconductor compared to the single crystal cells. Great efforts have therefore been made to understand the influence of lattice defects on the photovoltaic properties of the semiconductor. Though remarkable improvements have been obtained in particular cases, some fundamental problems which will be addressed in the following chapters still need to be solved. An interesting alternative approach to thin film cells has been to start with a high-efficiency single crystal technology and use it to make large area, low cost thin film cells. It makes use of the CLEFT process (97)which allows to separate mechanically single crystal layers of GaAs which have been grown epitaxially on a reusable GaAs substrate by the MOCVD technique. This method produces a high quality thin film which can be used for fabricating the solar cell either before or after the separation. Efficiencies exceeding 21% have been reported for a AIGaAs/GaAs double heterostructure solar cell. The first polycrystalline thin film cells were fabricated with an active layer of a polycrystalline cuprous sulphide (Cu2S) film of about 0.1-0.3 pm thickness on a ~ 2 0 # m thick polycrystalline layer of CdS. The Cu2 S layer can be easily formed by reacting the CdS film in a cuprous solution (Clevite process). (9s)A schematic diagram in Fig. 28 which is also typical
JPMS 35-3/4~E
262
P R O G R E S S IN M A T E R I A L S S C I E N C E I~C
CdS
2i3~J Ilnt~cestates
(b)
(a)
FIG. 28. Schematic diagram of a polycrystalline Cu2S/CdS thin film solar cell (a) and corresponding
band structure (b). for other thin film cells shows basically a heterojunction structure for the cell design. The Cu2S material is p-type with a band gap of 1.2 eV, whereas the CdS is n-type with a gap of 2.3 eV. Most of the light-generated current comes from the thin Cu2 S layer and relatively little from the CdS layer with its larger band gap. The prime advantage of the Cu2S/CdS solar cell is the ease of fabrication and the low processing costs. However, reliability problems virtually eliminated all work on this particular cell although efficiencies of about 10% were reported. The solar cells degraded under high humidity and at higher temperatures (> 60C), or when the load voltage exceeded 0.33 Vfl 9) These problems could be overcome with the further development of copper based ternary semiconductors whose properties will be discussed in Chap. 7. Among these CulnSe2 has received the most attention because of its extremely high absorption coefficient and stable electro-optical properties. Other advantages include a bandgap that is suitable for either homojunction or heterojunction device types and a lattice structure and electron affinity that matches with common n-type window materials such as (Cd,Zn)S, ZnO or ITO (In203-9 tool% SnO2). Generally the device structures suitable for maximum performance vary with the semiconductor materials which are used in the solar cell. They are determined by the specific properties of the material and the fabrication techniques that can be applied. Two typical device concepts are shown in Fig. 29 for a CuInSe2 solar cell but devices have also been modelled as homojunctions, p-i-n structures or Schottky diodes. The design is more complex because if low resistivity CuInSe2 and CdS are used throughout the device to minimize the series resistance, copper precipitates tend to form at the hetero-interface destroying the junction properties. (1)The structures have also been designed to increase the low open circuit voltage (and hence the low efficiency) which has been shown to be partly due to the recombination at the CuInSe2/window interfaces. The best devices reach efficiencies now up to 12.5% and have the potential for further improvements. Many thin film deposition techniques produce an amorphous semiconductor film if the processing temperatures are kept low. Despite this possibility only amorphous silicon (a-Si) has attracted wide interest as a photovoltaic material. The demonstration of the ability to dope controllable a-Si n- and p-type and to fabricate a Schottky barrier was an important first step for electronic and photovoltaic applications. Since the first discoveries, amorphous silicon has been thoroughly investigated and been utilized in solar cells which power watches and calculators. Amorphous silicon cells now account for about one-third of the world's photovoltaic market. Amorphous silicon usually contains significant concentrations of hydrogen (10-40%) which is mostly bound to silicon atoms and necessary to saturate free covalent bonds in the disordered crystal structure. The difference in the crystal structure compared to the single

AR Coating AI Grid
263
2.0 #m 1.71Jm 0.8 #m 0.81~m 2.0 #m 0.31zm
Glass
I p-doped a-Si:H
CdS
c.t.so 2 Sputtered Mo Alumina Substrate
~\\\\\\\\\\\\\\\\\\\~ UndopedB-ShH ]
~\\\\\\\\\\\\\\\\\\\~1
~'//////////////////////////////////~
n-doped B-ShH p-doped a-ShH

n-doped a-ShH p-doped B-Si:H n-doped a-Si:H
Undopeda-Sl:H / ~\~\\\\\\\\\\\\\\\\\N
~/////////////////////////////////A
(a)
(b)
FIG.
29. Device cross sections for current designs of CdS/CulnSez (a) and (CuZn)S/CulnSe solar cells (b). 2
FIG. 30. Cross sectional representation of a tandem p-i-n
structure for an amorphous Si:H solar cell.
crystalline state leads to new electro-optical properties as will be discussed in Chap. 4.6. In contrast to crystalline silicon as an indirect semiconductor hydrogenated a-Si:H has an optical absorption characteristic that resembles the behavior of a crystal with a direct band gap of 1.6 eV, which is somewhat larger than the bandgap width for optimal performance. A film thickness between 1-3/~m is sufficient to absorb the light and a-Si:H films can be easily grown by plasma deposition techniques onto metal or glass substrates. The minority carrier diffusion length is estimated to be below 1 #m therefore the depletion region forms most of the active carrier collecting region. Another consequence is a rather high bulk series resistance which is undesirable for a solar cell as has been discussed before. Most current cells are based on the heteroface p-i-n structure depicted in Fig. 24 but cells have been also designed as Schottky barriers. A very undesirable property of the a-Si : H solar cells is their instability--the efficiency degrades by about 15-20% when exposed to light usually within the first 48 hr of operation. This light induced degradation problem has become one of the most investigated problems in recent years because of its impact on this promising technology. Although this material related (Staebler-Wronski) effect has not found a generally accepted explanation it could be shown that special cell designs can help to reduce the degradation. Further improvements in the efficiency and stability can be obtained for multi-junction cells which will be discussed in Chap. 8. A cross section of multi-junction p-i-n cells is shown in Fig. 30 and recent measurements suggest that they are more stable compared to single junction cells. Materials and device studies have recently focused on the development of a-SiGe: H and a-SiC compounds for high performance multi-junction cells. {~') The alloying of a-Si with germanium and carbon offers the possibility to vary the optical properties over a wider range, as will be discussed in Chap. 8, so that the device designer has greater flexibility in constructing devices. The bandgalzs of a-SiGe ( ~ 1.4 eV) and a-SiC (1.85 eV) make these compounds suitable for triple-junction devices and recent results have shown efficiencies exceeding 10%. Stacked junction devices degrade more slowly primarily because thin stable films are introduced and the overall degradation rate is approximately an average over the single cells. {ll) Polycrystalline and amorphous thin film solar cells especially in connection with tandem device structures have a great potential for further development because of the flexibility in the utilization of novel materials and processing techniques but they present also a great challenge for a fundamental understanding of the physical properties of the semiconductors and their relationship to crystal defects.
264
3.5. Optical Concentration The concentration of sunlight by lenses and mirrors offers an attractive and flexible approach to reducing cell costs. In general, the higher the concentration ratio, C, the smaller is the range of angles of the incident light rays that can be accepted by the system. Once the concentration ratio exceeds about 10 the system can utilize direct sunlight only and a tracking system is required to follow the sun across the sky. The diffuse part of the sunlight which amounts to about 20% on a clear sky is wasted in such a system. For low concentration ratios ( < 20) it may be sufficient to adjust the concentrator only periodically. A novel approach to concentrate the sunlight uses luminescent concentrators. These are glass or plastic sheets which are doped with luminescent dyes that absorb sunlight and re-emit it in a narrow frequency band in all directions. The light is partly trapped in the sheet by total reflection and can be collected at the edges by the attached solar cell. Such a concentrator can accept light from all directions and therefore does not require a tracking system. The concentration of the light is determined by the absorption and the losses due to partly trapping in the sheet. Concentrating the sunlight on a cell will also raise the operating temperature which will decrease the efficiency as has been discussed in Chap. 2.5.3. Active cooling is therefore required for concentration ratios above 50. It is entirely feasible that a photovoltaic system in the future uses both the electrical and thermal energy collected by the cooling system. The geometrical concentration ratio C of an ideal lens or mirror system as a function of the range of angles, 0, is different for a linear- and point-concentration system and given by
1 1
C2 = sin(0/2)'
C3D = sin2(0/2 ).
(103)
The three-dimensional concentrators give a higher concentration ratio for the same range of angles and allow a maximum concentration of 45,000 for an angle of about 0 = 0.5 which is determined by the finite size of the sun's disk. Almost ideal concentration ratios are obtained by the compound parabolic concentrator (CPC). This system consists of two parabolic reflectors which are separately aligned with their focal points as shown in Fig. 31a. Other systems use standard planoconvex or Fresnel lenses (Fig. 31b).
\
Acceptance angle
\ /
/ :
Light
I',/I
FOCUS
(a) (b)
FIG. 31. Schematic diagram o f a non-imaging compound concentrator with parabolic reflecting surfaces (a), and a solar cell with an integrated concentrator lens (b).
SEMICONDUCTORS
265
The higher light intensity primary changes the light generated current, IL, which increases linearly with the intensity according to eq. (26). The total current,/, under these conditions is still approximately given by eq. (29) with a saturation current given by eq. (30).
Is= l, T3exp(- -ffT).
(104)
If the minority carrier density approaches that of the substrate doping high injection condition prevails and the saturation current is given by2~
where I~ and/2 are constants which combine all transport and geometrical parameters, and the doping concentrations of the device. They are usually obtained empirically. The open-circuit voltage becomes Voc=
2KTel n[IL+ 1) ~s
(106)
and increases logarithmically with the light intensity. Thus for a fixed temperature the efficiency under light concentration increases since both open-circuit voltage and short-circuit current increase. Since the open-circuit voltage decreases as a function of the temperature the efficiency is reduced if the solar cell operates at elevated temperatures, as has been discussed in the previous chapter. With increasing current densities the series resistance becomes increasingly important and eventually decreases the efficiency again due to a reduced fill factor as has been shown in Chap. 2.5. Solar cells which are intended to operate in a concentrator system have to be designed to minimize the series resistance. This requires the use of low-resistivity substrates, minimization of the sheet resistance of the top emitter layer and reduction of the contact resistances. The general behavior of a solar cell is therefore an increase in the efficiency at lower light intensities and a decrease for higher values. In practical systems the peak efficiencies occur at concentration ratios between 20 and several hundred. Efficiencies of about 20% have been reported for concentrator cells on the basis of silicon ~) and about 25% for GaAs (14) which is still well below the peak efficiency limit of about 37% (see Fig. 17).
3.6.
Multi-JunctionStructures
In connection with the heterostructure cells it has already been discussed that the spectrum of the incident sun light can be utilized more effectively if semiconductors with different band gaps are employed. Practical systems differ in the method by which the light is directed towards the semiconductor with the appropriate band gap. One concept separates the light by using spectrally sensitive mirrors before it is directed to the appropriate cell, whereas in the second case, the tandem cell approach, a series of cells is stacked upon each. The cells are arranged from the top side with decreasing band gaps, so that light with longer wavelengths and lower absorption coefficients passes through the upper cells and is absorbed in the bottom cells. The latter concept seems to be more compatible with todays semiconductor thin film technologies and has been mainly pursued in the past. Since the design of a cascade system adds extra complexity to the solar cell it is better suited for concentrator concepts.
266
(a) Front contact
p/n junction
Connecb'ng
PROGRESS
IN MATERIALS
SCIENCE
II
~ ~ ~ ' ~ Cell I
(b) Front contact

p/n junction Transparent Adhesive p/n junction Low bandgap ~ ~ ~ ................ Cell 1 High bandgap
......
Junction p/n junction
Cell 2 Low bandgap Rear contact
Rear contact
FIG.32. Cross-sectionalrepresentations of tandem cell structures: (a) two terminal, monolithic cell, the connecting junction is transparent and of low resistance; (b) four terminal, mechanicallystacked, the connecting material is a transparent adhesive. T a n d e m configurations can be completely integrated with two terminals or connected only mechanically, with four terminals, as shown in Fig. 32. In the first case one has the problem to fabricate a conductive and optically transparent connection between the top and bottom cell. A usual technique is the use of a tunnel junction which sometimes has stability problems. Numerical programs which incorporate individual cell performance parameters, e.g. current-voltage characteristics and spectral quantum efficiencies, are currently applied to calculate optimal configurations for tandem solar cells. Some calculations of the efficiencies of a tandem structure as a function of the bandgaps for the top and b o t t o m cells are shown in Fig. 33. Four terminal designs offer a wider range of bandgaps for m a x i m u m efficiencies. The bandgap of the top cell can vary between 1.5-1.9 eV and for the b o t t o m cell between 0.95-1.2 eV. Recently, a tandem cell using a silicon bottom cell and a G a A s top cell was reported with a terrestrial concentrator efficiency of 31% at C = 350, which is the highest efficiency reported so far for a photovoltaic device. (t5) The number of available semiconductors is significantly larger in the second case and thus allows greater flexibility for this structure. The tandem concept seems especially suitable for thin film solar cells and currently a number of systems such as a-Si: H/a-Si-Ge: H (efficiency 12.7%), a-Si:H/CulnSe2 (efficiency 14.6%) or CuGaSedCulnSe2 are being considered. (8) Calculations of the efficiencies of cascade systems with more than two semiconductors show that for an ideal system the efficiency saturates for about n > 7. (328) In reality one has to consider however additional unavoidable optical losses when connecting the different cells
2t
L ' I'
J ' J ' i
!~,'l'J-Af r . , -
',I
I~
25~C
....
.-"
-1
j
L
~ E'~ 1.5 ].4
13
l""r"-1""~i
~I
Bottom Cell E n e r g y G a p (eV)
FIG. 33. Calculated isoetticiency contours for a voltage matched two-junction tandem structure as a function of the band gaps (AM0 spectrum at 1 sun, and for 25C and 80C). The vertical axis is the bandgap of the upper and the horizontal axis the bandgap of the lower cell. (327)

I I I
267
! l
65 60 55 5O 45 40 u> 3S
1000 runs
= splitting
30
25 20 15 10 5
I 1 I 2 J l 3 4 I 5 i 6 i 7 I 8 I I i I l 1 9 10 11 12 13 14 AM1 illumination
iF`
'\, \
\ Tandem
AM1 spectrum C =1000
L 2
i 3
I 4
l 5
1 6
i 7
I 8
L i I i i 1 9 10 11 12 13 14
Number of celli (a)
Numll~" of eelts (b)
FIG. 34. Calculated maximum efficiencyfor multigap cells for concentrated (1000 sun) and unconcentrated light (1 sun) and an AM 1 spectrum (a). Effect of optical losses for concentrated light (b)O28).
which will eventually decrease the total efficiency of a tandem system (Fig. 34). An optimized system will therefore most likely consist of about two to four cells stacked upon each other.
4. STRUCTURALAND ELECTRICALPROPERTIESOF LATTICE DEFECTS IN SEMICONDUCTORS In the previous chapter it was shown that recombination of electrons and holes at lattice defects is a major source of losses in a semiconductor solar cell. The efficiency is limited by the number of trap levels which are introduced in the band gap by the crystalline defects. The electrical and other physical properties (e.g. the mobility of point defects) are closely related to the atomic structure of the lattice defect and both aspects have been studied thoroughly for m a n y defects during recent years. For the understanding of these properties it is therefore also important to know the crystallographic structure. In a crystalline semiconductor the lattice defects consist of point defects and the multi-dimensional defects like dislocations, grain boundaries, interfaces (including surfaces) and precipitates. In a m o r p h o u s semiconductors in addition unsaturated covalent bonds have to be considered. These defects are either introduced during the preparation of the starting material or later during the various steps of the solar cell processing, as will be discussed in Chaps 5-8 for different semiconductor material. The general structure and physical properties of crystalline defects as far as they are important for solar cell materials will be discussed in this chapter.
268
4.1. Diffusion and Electrical Behavior of Point Defects

The properties of defects which are important from the material perspective of solar cell semiconductors are their intrinsic electrical properties, their diffusion and solubility behavior, and their interaction with other lattice defects. One can distinguish intrinsic point defects if they are formed from the same species as the atoms of the lattice and extrinsic point defects if they are of a different nature. It has long been recognized that the fundamental intrinsic point defects participate in many physical processes such as precipitation or diffusion of impurity elements. For a crystal consisting of one species of atoms two basic types of intrinsic point defects can occur, namely vacancies, V (one missing atom at a regular, substitutional site) and self-interstitials, I (one extra atom squeezed into the crystal). For a binary compound semiconductor with the sphalerite structure and two species of atoms occupying different sublattices a third type of intrinsic defect can occur, the antisite defect, an atom occupying a substitutional site of the other sublattice. One has to specify the sublattice for each vacancy and interstitial (e.g. for a two component compound 1I,, Vb, la, Ib) here since the atomic environment and therefore the physical properties are different. In ternary compound with the chalcopyrite structure for instance (see Chap. 7) where two kinds of atoms occupy one sublattice in an ordered arrangement it is also possible to form an antisite defect on the same sublattice which leads to an even higher multiplicity of intrinsic defects (12 in this case). In addition complexes can be formed like divacancies, vacancy self-interstitial pairs (Frenkel pairs), etc. or aggregates of more than two species, which shall still be considered point defects. The definition and identification of intrinsic point defects in amorphous semiconductors requires a different approach which will be discussed in Chap. 8. Intrinsic point defects play a rather important role in compound semiconductors because they are formed in large concentrations in non-stoichiometric compounds and then determine the electrical and optical properties. Typical examples are the thoroughly investigated antisite EL2-defects in GaAs or the intrinsic defects which are responsible for the doping of Cu-ternary compounds. The behavior and contribution of intrinsic defects will therefore be discussed separately. The most important extrinsic point defects in semiconductors besides the doping elements are carbon, oxygen, hydrogen (mainly in silicon), and transition metal impurities. In addition to their fundamental physical properties one has to consider the formation of agglomerates, precipitates, or complexes with other point defects, which may lead to different electrical properties, and the interaction with extended lattice defects (e.g. dislocations, grain boundaries) which can change the electrical activity of this defect significantly.
4.1.1. Intrinsic point defects

In thermodynamical equilibrium a crystal contains a certain concentration of intrinsic point defects since the Gibbs free energy is lowered by their formation. The thermal equilibrium concentrations (in dimensionless atomic fractions) of vacancies C~, and of self-interstitials, q, Cp, are given by
sF Cv = exp - ~ exp --K-'T ' ~ v g 1 C~q = exp S[ exp -~--~ (107)
where K is Boltzmann's constant and T the absolute temperature. Sv and H F are the entropy v and the enthalpy of formation of a vacancy, which is equivalent to the removal of an atom from its substitutional site and the deposition on the surface, S~ and H F are the corresponding V
SEMICONDUCTORS
FOR S O L A R CELL A P P L I C A T I O N S
269
quantities for self-interstitials. If both vacancies and self-interstitials are present it can be expected that they react with each other according to the reaction
I + V*--,0
(108)
where 0 represents the undisturbed substitutional site in the lattice. This reaction, which has been proposed for silicon by several authors, describes the balance between the recombination of vacancies and self-interstitials and the spontaneous thermal creation as (Frenkel) pairs in the lattice. In most cases it is assumed that the binding energy between vacancy and self-interstitial is small and they can be treated as independent point defects at least at high temperatures. Measurements in silicon have shown that it takes about one hour to establish the equilibrium at 1100C and several days at 900C. The point defect concentrations can be disturbed locally by a number of processes e.g. in silicon by surface oxidation, which injects interstitials into the lattice. For sufficiently long times the local equilibrium will be established between vacancies and interstitials according to reaction (108), which can be described by the law of mass action as
C v C l = C~/q C~ q = exp
KT
(109)
where Cv and C~ are the vacancy and interstitial concentrations, respectively, and G F = H E - T S F and G F = H F - T S v are the Gibbs free energies. In compound semiconductors reaction (108) can only occur between a vacancy and an interstitial of the atomic species that belongs to the same sublattice, otherwise an antisite defect would result. Therefore one has to formulate an equivalent equation for the vacancy and interstitital concentrations in each sublattice. In addition because of the conservation of the stoichiometry of the compound the concentrations of vacancies and interstitials in each sublattice are not independent from each other, which leads to supplementary constraints. 4.1.1.1. V a c a n c i e s a n d s e l f - i n t e r s t i t i a l s in s i l i c o n . In general the thermally generated concentrations of vacancies and interstitials in silicon (and probably in other semiconductors as well) are very low (<1016cm -3) even at the melting point and the experimental determination of the formation enthalpies and entropies are difficult and indirect. Although there exists a long standing controversy in the literature (17) whether self-interstitials are present at all in silicon in concentrations comparable to the vacancies, most authors seem to agree now that both species co-exist and have to be taken into account for understanding diffusion and other processes in silicon such as impurity precipitation or point defect complex formation. (18)The most direct information about the behavior of vacancies and self-interstitials have been obtained from investigations of self-diffusion. The self-diffusion coefficient D so is given by
DSD = "-'V/'~SD ~t-" ----xJV/'~C ~ q -~- Dl C~q !-1-IF)SD
(110)
where Dv and D 1 a r e the diffusion coefficients of vacancies and self-interstitials, respectively. The temperature dependence of the individual terms is given by with r4SD _* * r4r **V - - V
f HISD) D I C ~ q = DlSo e x p ~ - - ~ - ~ - ; D
"{-
Hv M,
(111)
with H so = H~ + H M, v
(112)
270
w h e r e H F and
H Mare the formation and migration enthalpies, respectively. The self-diffusion coefficient D sD as a function of the temperature has been determined by several techniques and the results indicate that a nonlinearity in the Arrhenius plot for D sD occurs. At temperatures near the melting point the activation energy is around 5 eV with a prefactor D sD ,~ 103 - 102 cmZ/s which decreases below 1270 K to about 4.0 - 4.5 eV and D sD ~ 1 cm2/s. This change in the parameters indicates a change in the self-diffusion mechanism and it has been suggested that it is the vacancy mechanism (D sD) that dominates at low temperatures and the self-interstitial diffusion (D sD) in the high temperature regime. Further experimental evidence comes from the analysis of a large number of impurity diffusion experiments such as doping or some transition elements where it is assumed that the impurity diffusion requires the participation of vacancies or self-interstitials. 7) Particularly the analysis of the gold and platinum diffusion (via the kick-out mechanism) has led to activation enthalpies and entropy factors(19) which seem to agree quite well with the self-diffusion data obtained by tracer measurements. Much less is known however about the concentration and diffusion coefficients of vacancies and self-interstitials. The analysis of the diffusion profiles after Au and Pt diffusion has yielded the following results for the self-interstitial (atomic) concentration and the diffusion coefficient C~q 2 x 104 exp 3.8eV'~ KT J D~ = 0.2 exp { 1.2eV'l 2 ~ - j~ [cm/s].
(113)
Annealing experiments following electron irradiation at helium temperature also indicate a rather low migration energy both for the vacancy and the interstitial. Although there is no direct evidence for the self-interstitial (e.g. an electron level) and one has to admit that the indirectly obtained data remain controversial and reliable information on C~q and C~ or Dv and D~ are not known, it seems to be that the activation enthalpies for vacancies and self-interstitials are rather similar between 4 to 5 eV (being lower for the vacancy) and the migration enthalpies between about 0.1 to 1.2 eV. A summary of some of the currently available results and parameters is given in Table 2. It should be mentioned explicitly that the quantities given so far are for intrinsic silicon. In doped crystals the formation and migration enthalpies depend in general on the position of the Fermi level EF. It turns out that it is the formation energy of the point defect which is changed in a doped crystal. This can be understood if one considers that the defect introduces a trap level in the band gap, so that the system can gain energy by the occupation of the trap level ET with an electron or hole. For instance for a vacancy one has the reaction
Table 2. Activation enthalpies (H sD) and prefactors (Do sD) for the diffusion of vacancies and self-interstitials as determined from measurements of self-diffusion (SD) or foreign atom diffusion in undoped silicontl7,19)
Defect Vacancy
D sD [cm2/s] H so [eV] 20.0 154.0 0.57 40.0 1460 9000 914 4000 4.4 4.65 4.03 4.6 5.02 5.13 4.84 5.0
T [K] 1103-1473 1128-1448 1173-1373 973-1523 1320-1573 1373-1573 1173-1373 973-1523
Remarks (SD) (SD) (Au) (Au/Pt) (SD) (SD) (Au) (Au/Pt)
Interstitial
SEMICONDUCTORS
271
V*-*V + + e for a donor level and V*--~V + h + for an acceptor level. One obtains for the formation energies HW+ and HW- of the charged defect in the two cases
HV+ = H~o -- (E.r - E v ) HW = HWo + (E-r - EF)
(1 14)
H~o is the formation energy and ET the trap level of the neutral vacancy. The concentrations of the charged vacancy c~q+ and C~q_ are given then for a non-degenerate trap level by c~,q+ = C~, exp q C~q = C~ exp q
KT
where C~, is the concentration of the neutral vacancy given by eq. (107). These equations q can also be derived from the general expression for the occupancy of a trap level which has been given in Chap. 2.4. Equation (115) is in fact completely general and applies to any defect which introduces a trap level in the gap. The total concentration of the defect is the sum of charged and neutral defect. A thorough analysis of the doping dependence of the diffusion of shallow and deep level impurities which involve the participation of vacancies or self-interstitials has shown that it can be correctly described if one assumes an acceptor level in the lower half of the band gap for the vacancy, and a donor and acceptor level in the lower and upper half, respectively, for the silicon-interstitial. (I7) Experimental and theoretical investigations, have revealed a rather complex level structure for the vacancy which shows the behavior of a negative U-center [110]. The donor level V /+ (Ev + 0.057 eV) lies below the level V +/++ (Ev + 0.13 eV) therefore the charge state V + is unstable and V + + changes directly to V when the Fermi level is raised. Since both donor levels lie near the valence band edge and a further acceptor level near the conduction band edge (E c - 0 . 0 9 eV) this result cannot easily be reconciled with the doping dependence of impurities diffusing by the vacancy mechanism. Much less is known about the level structure of the self-interstitial and further investigations are necessary to confirm the conclusion from the diffusion experiments. Despite the remaining controversy about the concentrations of vacancies and selfinterstitials in thermal equilibrium it is generally agreed that these native defects in silicon can be generated in excess by a variety of processes. For instance thermal oxidation (m) of a silicon surface or the intrinsic formation of oxygen precipitates (see Chap. 4.1.2) produces an excess of self-interstitials whereas the addition of a chlorine-containing compound in the oxidizing atmosphere (e.g. HC1, C12) may lead to an injection of vacancies into the crystalline silicon. (1~2) Although it has been mentioned above that according to reaction (108) the equilibrium between vacancies and self-interstitials will be restored experimental results indicate that this process can be rather slow. Measurements at 1100C have shown that it takes about one hour to reach thermal equilibrium. Considering the estimated diffusion coefficients of the two defects the extrapolation to 900C leads to times of several days. This means that for experiments performed in this temperature range vacancies and self-interstitial can be treated as independent species and can exist in supersaturation for not too long times and provided other sinks for the annihilation of these defects such as dislocations are not present. Several typical device processing steps such as doping diffusion lie in this temperature regime. The supersaturation of intrinsic point defects can have a strong impact on the diffusion behavior of impurities which require the participation of either vacancies or interstitials as a diffusion vehicle. For substitutional elements such as group III to V elements in silicon the diffusivity D~ is composed of a contribution D v involving vacancies (vacancy mechanism) and a contribution D~, involving self-interstitials (interstitialcy mechanism) where the relative
272
contributions depend on the specific element and the temperature. In a non-equilibrium situation where the equilibrium concentrations of vacancies (C~, and interstitials (C~q) are q) disturbed the diffusion coefficient is given by CI ~ Cv D~ =~-~ DV +~-~-~DI (116)
where Cv and Cx are the actual vacancy and self-interstitial concentrations. Generally, atoms smaller than silicon such as boron or phosphorus diffuse preferentially via the interstitialcy mechanism and their diffusion can be enhanced by a supersaturation of self-interstitials. From a combination of experiments at 1100C the following values have been obtained for D~/D~, namely: 0 . 9 - 1.0 (B, P), 0 . 2 - 0.4 (As), and 0.01 - 0 . 0 2 (Sb). A process which produces an even higher supersaturation of self-interstitials compared to the surface oxidation is the diffusion of phosphorus from the surface into silicon. 13) This doping process can lead to an enhanced diffusion of substitutional dopants (e.g. boron) in front of the in-diffusing phosphorus, known as the "emitter push effect", and is clear evidence for the interstitialcy mechanism for some of the doping elements. An influence of the intrinsic point defects on the diffusion behavior is also known for the transition metals which can occupy both substitutional and interstitial sites. Their diffusion mechanism involves the transition between substitutional and interstitial sites with the participation of intrinsic defects. These more complex mechanisms will be discussed in Chap. 4.1.3. Non-equilibrium concentrations of intrinsic point defects can also be produced during radiation with high energy particles or photons. This offers the possibility of studying the properties of intrinsic defects which otherwise may be too difficult to observe because of their low concentration. In this way many results have been obtained for instance on vacancies and vacancy related complexes in silicon and other semiconductors. For solar cell applications these results are particuarly useful since exposure to high energy radiation occurs for solar cells which are used in space. Radiation induced defects are the dominant cause for the degradation of the solar cells in space and the investigation of the damage mechanism are therefore important. This particular problem will be addressed in Chap. 4.2.
4.1.1.2. Intrinsic defects in binary compounds. Some of the binary compounds such as GaAs and CdTe have also attracted much interest for photovoltaic applications since they are direct semiconductors with a high optical absorption and an almost ideal band gap energy. Mainly because of their utilization for electronic devices much progress has been made during the last decade in the investigation of their structural and electronic properties and the fabrication. Intrinsic point defects are even more important in binary (and multinary) compounds since they are introduced to compensate deviations from the ideal stoichiometry of the compound. Their concentrations can by far exceed the thermally generated defects in the compound with ideal composition particularly in multinary compounds. The most prominent and thoroughly investigated native defect is the EL2 center which is the dominating electrical defect in semi-insulating GaAs. A discussion of the intrinsic defects in compound semiconductors is therefore necessary to understand many of the fundamental electronic properties. Most of the binary AB compound semiconductors crystallize in the cubic sphalerite structure where the two atomic species occupy the different fcc sublattices (Fig. 35). This ordered arrangement of atoms allows a larger variety of intrinsic defects. In addition to the vacancies (VA, VB) and the self-interstitials (Ai, Bi) which can be formed for each species,
273
(a)
(b)
FIG. 35. Schematic representation of the diamond cubic (a) and sphalerite lattice (b). The A- and B-atoms in a binary AB compound occupying the two fcc sublattices are shown by black and white circles. The hexagonal (a) and two equivalent tetrahedral interstitial sites Ta (b) in the sphalerite lattice (and diamond cubic) are indicated by dotted circles. Small circles show equivalent C-sites, the center of a rhombus formed by three adjacent atoms and a Td site. The mid-position between two adjacent C-sites are M-sites, the equilibrium positions for hydrogen in silicon, for instance.
Table 3. The majority intrinsic defects for a binary AB compound with a given stoichiometryAs = [A]/[B].The donor (D) or acceptorscharacter (A) is derivedassuming a compound with predominantly ionic character Defect type As > 1 As < 1 Schottky Va (D) VA (A) Frenkel (A) Ai (D) VA (A) Frenkel (B) VB (D) Bi (A)
Antisite
Aa
(A)
BA
(D)
antisite defects (AB, BA) c a n occur where an A atom occupies a B-site on the second sublattice (or vice versa). In thermal equilibrium the concentrations of these defects are determined by their formation entropies and enthalpies, but as mentioned above are more likely to be present because of an excess of A or B atoms. Because of the different number of valence electrons of the A and B atoms only charge compensating defects can occur simultaneously for a given composition. The general theory which has been worked out for non-stoichiometric ionic crystals by Kr6ger Or4) gives for a binary AB compound the following defects which have been summarized in Table 3. For a given stoichiometry which can be characterized by the ratios As = [A]/[B] either one or the other defect occurs in excess. The concentration of these defects in thermodynamical equilibrium is determined by the stoichiometry and their formation entropies and enthalpies. In non-equilibrium situations which are the more frequent cases the kinetics such as the diffusivity are important, too. The influence on the electronic properties of the semiconductor is determined by the acceptor and donor character of the defects which in an ionic model can be estimated from the excess or deficiency of the valence electrons for the particular defect. The character that can be expected for the different defects is indicated in Table 3, however, no information can be obtained about the position and the particular level structure of the defects. It has been discussed in the previous section that the formation energies of a charged defect differs from the neutral one, hence the position of the trap levels are also important for the determination of the defect concentration. Generally p-doping lowers the formation energies of defects with donor character and n-doping favors acceptor defects. The self-interstitial defects generally seem to have the highest formation energies in many semiconductors therefore one is mainly left with the vacancy and antisite defects. (Whether the rather high self-interstitial concentration in silicon which is due to the unusual large entropy factor is an exception remains to be seen.) In agreement with this general rule one has observed in undoped (LEC grown) GaAs that the main defect in As-rich material is the ASGa antisite and in Ga-rich material the GaAs antisite. In the first case the material is semi-insulating, mainly because of the donor character of the Asca antisite which compensates residual acceptor impurities (Zn, C), in the second case the crystal is p-type. '5~ The ASGa antisite defect has attracted considerable interest in
274
PROGRESS
IN MATERIALS
SCIENCE
the past because it seems to be related to the major electrically and optically active defect in semi-insulating GaAs. The thorough investigation of the properties of the so called EL2 defect and the Asia antisite has yielded enormous insight into the structural and electronic properties of GaAs and will be discussed in Chap. 6. In doped crystals the defect situation is different because the shift of the Fermi level can favor the occurrence of the vacancy defects in either case. In fact one observes that in the As-rich n-doped crystals the VA~becomes the dominant defect whereas in the Ga-rich material the Asoa acceptor remains dominant even for p-doping, m6~ The VGa vacancy which might be favored by p-doping seems to be a rather unstable defect which disappears by the reaction VG~--~Asc~ + VA~. 4.1.1.3. Intrinsic defects in ternary compounds with chalcopyrite structure. Some of the Cu-ternary compounds which crystallize in the chalcopyrite structure are currently the most promising candidates for photovoltaic applications (see Chap. 7). Similarly to binary compounds intrinsic defects are introduced to maintain the crystal structure for non-stoichiometric compounds. A major difference is, however, that some of the intrinsic defects are shallow level defects which can be used to dope the crystal. In fact the conductivity type of these semiconductors is usually determined by the non-stoichiometry of the crystals instead of an external doping by shallow level impurities. Despite the importance of the intrinsic defects for the Cu-ternary compounds their structural and electronic properties are still poorly understood. Some of their structural aspects which can be derived from geometrical considerations shall be discussed here and results about the electronic properties will be presented in Chap. 7. Since CulnSe2 is the material which has received the most interest results will be mainly given for this compound. Of particular importance for solar cells are polycrystalline thin films, therefore the role of intrinsic defects in a semiconductor containing a high density of grain boundary shall be discussed separately in Chap. 4.6. There exist 36 known ternary (I-III-VI2) chalcopyrites (see Table 16 in Chap. 7) which can be considered as isoelectronic analogs to binary II-VI semiconductors with the zinc-blende (sphalerite) structure. Despite the electronic similarity the substitution of two column II elements by a column I and III element is accompanied by significant changes in the lattice structure. The differences can be understood by considering the chalcopyrite unit cell (Fig. 36). Anions (X) and cations (A, B) occupy the two different sublattices as in the zinc-blende structure. However, an ordering of A and B atoms in the cation sublattice leads to a doubling of the zinc-blende unit cell. In addition, the crystals often show a tetragonal distortion where the ratio between the lattice parameters, r/= c/2a, often differs from 1 by as much as 12%. For a random distribution of the A and B atoms on the cation sublattice the chalcopyrite again reduces to the sphalerite structure and some ternary compounds seem to prefer the disordered structure 17)at room temperature. A third difference in the ordered structure is the displacement of the anions from their zinc-blende sites. Each anion (X) has two A and two B cations as nearest neighbors, whereas in the zinc-blende structure each anion has four cations of the same kind as nearest neighbors (and vice versa). Because of the different bond strength the anion X usually adopts an equilibrium position that is closer to one pair of cations than to the other (Fig. 36), which explains the anion displacement, u, and an unequal bond length RAX ~ RBx (bond alteration) related to u, ~/by
R A x = a [ u 2 + I ( 1 +r/2)] 1/2 RBx=a u --
+ l ( 1 +~/2\-il/2.
(117)
275
J
FIG. 36. Tetragonal unit cell of the chalcopyritelattice and projection of the upper and lower cell on a (001) plane. For an ABX2 compound the positions of the cations (A, B) are represented by large circles and squares, and of the anions (X) by small circles. The atomic positions along the c-axis are at (n/4) c(n = 0 to 8) and indicated by different shadings at the right side. The influence of the structural peculiarities of the chalcopyrite lattice on the band structure and the bonding has been thoroughly discussed by Zunger and Jaffe. ~HS) The theoretical calculations have determined the bonding between next neighbors which can be described as a mixture between covalent and ionic bonding. Although the lattice structure is still very similar to the sphalerite structure which indicates a major contribution of the covalent bonding the electronic and other properties have been frequently discussed by considering these semiconductors as ionic compounds. For the following structural considerations, however, the covalent character will be emphasized. There are twelve intrinsic point defects which can be formed in the ABX2 chalcopyrite lattice, three vacancies (VA, VB, Vx) and interstitials (Ai, Bi, Xi) for each atomic species, and six antisite defects. In addition to the antisite defects which can be formed by an exchange of anions (X) and cations (A, B) like in binary compounds (Ax, Bx, XA, XB) one can also form antisite defects on the same sublattice by an exchange of the cations (AB, BA). Structural models for some of these defects which show the next neighbor environment are depicted in Fig. 37. Since nothing is known about the structural properties of any of these defects one can only describe some of the apparent differences in comparison with similar defects in the diamond cubic or sphalerite structure. Similar to the vacancy in the dc-lattice one can assume that the unsaturated bonds reconstruct (Fig. 38). For the cation vacancies (VA, VB) only X-X bonds can form whereas contrary to the anion vacancy (Vx) in binary compounds two types of bond reconstructions are possible here, between A - A and B-B, or two A - B bonds. The structures of the other defects show similarly the occurrence of cation-cation and anion-anion bonds with the
X~
,X FIG. 37. Structural models for two antisite defects in a ternary ABX2compoundwith chalcopyrite structure, Bx antisite (a), and BA antisite (b).
FIG. 38. Structural models for two vacancy defects in a ternary ABX2 compound with chalcopyrite structure, (a) vacancyat an anion site Vx, (b) vacancyat a cation site VA. Possible bond reconstruction is indicated.
276
exception of the antisite defects on the cation sublattice AB and Bh. Because of the polarization of the bonding and the ionic charges at the atoms in these compounds one can assume that bonding between atoms which carry like charges (such as A-A or X-X) are rather unfavorable. One may therefore expect that the formation energies of the antisite defects AB and BA which avoid these bonds are lower compared to the other defects. Recent calculations of the formation energies ~9.~20)in CulnSe2 indeed show that these two antisite have the lowest energy (see Table 17) and hence are likely to be the most dominant intrinsic defects in this ternary compound. The electronic character of these defects has been determined from a simple calculation of the number of valence electrons and the assumption that the crystals are ionic compound. Considering the bond reconstruction for the vacancies this approach may be too simple to actually predict the character of these defects. In fact different conclusions have been drawn as to the acceptor or donor character of the selenium vacancy.(121) The results are also summarized in Table 17 and show that both donor and acceptor defects occur. The positions of the energy levels remain, however, undetermined. At present no conclusive experimental or theoretical results exist which allow to determine the structural and electronic properties of any of these defects in CuInSe2 or any other Cu-ternaries, therefore this issue remains a challenge for the future. It has been mentioned above that a difference between binary and Cu-ternary semiconductors is the possibility to determine the conductivity of the ternary compounds simply by changing the composition. The experimental results indicate that at least some of the intrinsic point defects which are produced by the non-stoichiometry are shallow level impurities. The same behavior has not been observed for binary compounds where for instance in As-rich GaAs a deep level defect, namely the ASGaantisite (EL2 defect) is formed. One might explain this difference between the binary and ternary compounds if one assumes that the antisite defects on the cation sublattice (AB, BA) are the shallow level impurities, which cannot be formed in a binary system. Photoluminescence and Hall effect measurements in CulnSe2 which will be discussed in Chap. 7 indeed show the occurrence of shallow level impurities although no defect could be conclusively related to a particular energy level. One of the problems is which of the possible defects actually occur for a particular composition. The parameters which are usually used to characterize the stoichiometry of a ternary compound are the molecularity Am and the stoichiometry As which are defined b y (122) Am=[A]-I [B] AsSl[X] s2 [A] + s3 [B] 1 (118)
where si are the numbers of excess and deficiency electrons for cations (A, B) and the anion (X), respectively. Since the formation of structural defects has to maintain the neutrality of the crystal only charge compensating defects can be introduced simultaneously. General models have been developed for ionic crystals to relate the composition described by these parameters to the formation of the defect pairs which can be present. The majority defect pair that actually occurs is the one with the lowest formation energies. These arguments have been applied to analyze the electrically observed behavior of CulnSe2 and to relate it to the stoichiometry of the crystals as will be discussed in Chap. 7. Unfortunately, these results are still inconclusive partly due to the fact that the formation energies of these defects and their electrical character are not known with sufficient accuracy. Another problem that will be addressed in the next Chap. 4.4 is the fact that most of the experimental results have been obtained for polycrystalline compounds. It has been suggested recently02) that the formation energies of intrinsic point defects in compound semiconductors
SEMICONDUCTORS
277
which contain a high density of grain boundaries may be lower because of a preferential formation of these defects at particular grain boundary sites. A numerical calculation has estimated that these energies which are included in Table 17 can be significantly lower compared to the bulk formation energies. Therefore, the identification of the main defect pair in a polycrystalline ternary compound for a given composition is further complicated. More fundamental studies of point defects in bulk and polycrystalline crystals are certainly needed to understand the behavior of intrinsic defects in ternary compounds.
4.1.2. Oxygen and carbon

Besides the doping elements, which are usually intentionally introduced, oxygen and carbon are the most abundant impurities in single crystalline silicon. Apart from the introduction through the starting material they are incorporated in the melt during crystal growth through the gas ambient and the contact with the crucible. The amount of an impurity which is finally incorporated in the solid is determined by the melt concentration and the segregation coefficient ko = cct/CL (crt and C L a r e the concentrations in the solid and the liquid). The rather high segregation coefficients for carbon (ko = 6 10 -2) and oxygen (ko = 1.25) and their relatively high solubility in the solid compared to the other elements are unfavourable for a refinement during solidification. Other elements such as nitrogen (ko = 7 x 10 -4) or the transition elements (ko < 5 x 10 -a) which may also be introduced at the same time can be more easily retained in the melt. Dependent on the growth technique the pollution of the melt is unavoidable to some extent so that it becomes important to understand and reduce the influence of these impurities on the electronic properties of the material. Dislocation free Cz-silicon crystals contain oxygen and carbon in concentrations of about 10 ~8 c m - 3 and 1017 c m -3, respectively, which is far above the solubility at room temperature. Due to the slow diffusion at lower temperatures and the relatively fast cooling rates during crystal growth the solution remains supersaturated. However, annealing an as-grown material leads to enhanced oxygen and carbon diffusion and finally the precipitation of the equilibrium phases SiO2 and SiC. Since annealing steps are part of the device processing it is important to understand the precipitation behavior and their impact on the material properties. Cz- and Fz-silicon crystals are mainly used for electronic devices but also for high efficiency solar cells and the results which are obtained mainly from measurements on high quality crystals are thus immediately relevant for the solar cell fabrication, too. More important for the photovoltaic applications is, however, low cost polycrystalline silicon which can be prepared by a variety of techniques (see Chap. 7). These are materials with high concentrations of carbon, oxygen and other impurities, and a variety of other lattice defects such as dislocations and grain boundaries. Therefore they may respond differently to annealing treatments. Since much less is known about the carbon and oxygen behavior in these materials one has to rely on extrapolations from the results which have been observed in single crystal silicon and will be discussed next. Carbon and oxygen are also incorporated in other semiconductors but have not been considered important impurities for the electronic properties, partly because other impurities or defects occur with higher concentrations and are thus more dominant. Therefore in the following the properties of these elements are mainly discussed for silicon. A particular case which requires a separate discussion, however, is the intentional oxidation of compound semiconductors such as CulnSe2 which is a necessary step to improve the solar cell performance. This process will be discussed in Chap. 7. The properties of carbon and oxygen have been presented in numerous review articles and conference proceedings and for a detailed information one should refer to these papers. 23 ~25)
JPMS 35-3/4~F
278
T [K] 1800 1600 1400 1200 1000 800 600 400
PROGRESS
IN MATERIALS
SCIENCE
SI:OLiquid
F'Z
J
CZ/
L + SiO 2
SIO 2 Precipitates Polyhedra
Si:OSolid/
New donors
Platelets Ribbons
Thermal donors O-VsI complexes
I
10
12 14 16 18 20
LogN[cm"3]
FIG. 40. Schematic representation of the off-center bondcentered position of an interstitial oxygen in the silicon lattice. Black circles indicate equivalent oxygen sites.
FIG. 39. Phase diagram for oxygen in silicon and typical temperature ranges for the occurrence of oxygen related defects after annealing. Shaded area indicates the narrow liquid-solid two phase regime.
4.1.2.1. Atomic oxygen in silicon. Oxygen in silicon forms a peritectic phase diagram with the SiO2 phase (Fig. 39) which explains the segregation coefficient ko > 1. The maximum solubility in the solid is about 2 x 10 ~8cm -3 and the solubility as a function of the temperature above 750C is given by 026)
C~=C~oexp
KTJ
with
C~=1.8
Hg=1.52[eV]
(119)
similar values C~o = 0.52, H~ = 1.4 eV have been given in Ref. (127). Oxygen occupies an interstitial position between two silicon atoms in a slightly off-center position as indicated in Fig. 40. Because of its incorporation in an interstitial position dissolved oxygen leads to an increase in lattice constant. The main interest in the investigation of oxygen in silicon is that it can give rise to a number of electrically active defects which will be discussed in the following. It should be mentioned, however, that no electrical activity has been observed for oxygen in the interstitial position itself. The diffusive jump of oxygen requires the breaking of Si-O bond without the participation of an intrinsic point defect. The interstitial migration through the silicon lattice described by the diffusivity D O has been measured over several orders of magnitude and is given by 027) Do = Doo exp
-- KTJ
with
Doo = 0.11 [cm2/s]
H o = 2.51 [eV]. D
(120)
In agreement with these considerations it is observed that an excess of vacancies or self-interstitials has no influence on the diffusion behavior of oxygen. 4.1.2.2. Primary oxygen agglomeration at low temperatures (below 650C). Depending on the annealing temperature (and times) different stages of oxygen participation have been observed for as-grown oxygen rich silicon. The first stage occurs at temperatures around 450C. The diffusivity is so low that only the agglomeration of a few oxygen atoms can occur. The defects which form after annealing times of several hours introduce two shallow donor levels in the band gap at Ec 70 meV and Ec 150 meV which are considered as two different charge states ( o / + and + / + + ) of the same defect. These so called thermal donors (TD) in fact consist of a variety of different defects (TD1 - 9) which can be monitored separately by
SEMICONDUCTORS
FOR SOLAR
CELL APPLICATIONS
279
IR measurements. The concentrations of the different T D centers increases first with time up to values of about 1016cm -3. Since the oxygen concentration in the bulk decreases at the same time with an activation energy which corresponds to the oxygen diffusion it has been assumed that the T D are different stages of small oxygen agglomerates which develop into each other by the addition of more oxygen atoms. Concentration measurements of the different T D centers as a function of temperature show that the same mechanism limits the rates of their formation. A major problem is, however, that the activation energy for the formation of the T D centers is only about 1.7eV. Several models for the T D generation have been developed (128-13) based on the oxygen aggregation model with the common necessity to assume an enhanced oxygen diffusion at low temperatures which has to be about 3 to 4 orders faster than the extrapolation for the normal oxygen diffusion at high temperatures. Recently a new approach has been tried to solve the dilemma with the suggestion that the TD's are in fact self-interstitial aggregates and only the first step is an oxygen complex which serves as a nucleus. The nuclei may be small oxygen complexes (presumably 03) which are formed at higher temperatures and exist already in the as-grown material. Self-interstitials may be produced during the formation of 02 complexes at lower temperatures following the reaction ~30) Oi + Oi~--~O2 + I. (121)
Although this model can account for the main experimental observation no direct proof for the nature of the T D exists and further investigations are necessary to solve the controversy whether oxygen or self-interstitial aggregates are the observed T D centers. After annealing above 600C the T D disappear or do not form at all. Instead a new type of defects appears in the temperature range around 600-700C with a rod-like morphology and an average size of about 20/~. High resolution TEM revealed crystalline precipitates which were at first identified as a high pressure SiO2 phase (coesite). 3 Later measurements indicated a rather low oxygen concentration and the features were re-interpreted as hexagonal silicon3132) The exact nature of these very small crystalline precipitates thus remains controversial. 4.1.2.3. Oxygen precipitation above 800"C. The annealing treatment at temperatures above about 800C finally shows larger amorphous precipitates with average sizes up to 1000 ,~ dependent on the annealing temperature and time. The precipitates are identified as the SiOx phase with x ,~ 2. An important aspect is that the formation is associated with a volume change A V given by
A V = Vsio2 Vsi -
2 Voi.
(122)
Considering the position of the oxygen interstitial one can assume Vo~ ~ 0 and since Vsio2 ,~ 2 Vsi the volume change corresponds approximately to the volume of one silicon atom AV ~ Vsi.~133~As soon as the volume increase of the growing precipitate can no longer be elastically adapted the surrounding matrix is either plastically deformed or self-interstitials are emitted (about one silicon atom for two oxygen atoms). Alternatively the absorption of vacancies is also feasible. The self-interstitials emission or vacancy absorption can finally lead to the formation of dislocations loops or interstitial type stacking faults which has actually been observed experimentally. ~131'134) The development of the precipitate sizes can be explained by theories for diffusion limited growth ~j35) (see Chap. 4.5) assuming the normal oxygen diffusion. The morphology of the oxygen precipitates depends on the annealing temperature and roughly three regimes can be
280
PROGRESS
IN MATERIALS
SCIENCE
distinguished (see Fig. 39). Precipitates in form of ribbons occur below 800C. Between 800C to 1050C plate-like precipitates dominate which are often accompanied by prismatic dislocation loops. Above 1050C the precipitates take the shape of a regular or truncated octahedron. The platelets and ribbons have the (001) habit plane, while the octahedral precipitates have (1~1) bounding faces. The main features of the precipitation process can be explained on the basis of conventional nucleation theories and are noncontroversial. It has been found that precipitates formed between 600 and 900C are electrically active and associated with donor states3 ~36-~3s)They are commonly termed "new donors" in order to distinguish them from the thermal donors formed at low temperatures. It is assumed that these donor states are related to the SiOx/Si interface states and probably result from incomplete oxidation (excess silicon) in the SiOx precipitate near the interface. These donor states may be destroyed by a high temperature heat treatment when the composition at the interface is restored. The main aspects of the oxygen behavior described so far are valid for crystals with low carbon concentrations. Experimental results frequently show that carbon can have a significant influence on the precipitation behavior at all stages. Since carbon is frequently present in comparable concentrations especially in solar cell silicon these interaction processes are very important. The basic properties of carbon in silicon and the interaction with oxygen will thus be discussed next. 4.1.2.4. Atomic carbon in silicon. Carbon dissolved in silicon is in equilibrium with SiC and forms an eutectic phase diagram (Fig. 41). In thermal equilibrium the maximum solubility in the solid is about 3.5 x 10~7cm -3 (7ppm) and the eutectic concentration about 5 x 10 ~scm -3 (100 ppm). The solubility as a function of the temperature is given by (139) C~q
=
C~o exp
-- KTJ
with
C~,
= 10 2,
H F = 2.3 [eV].
(123)
Contrary to oxygen a carbon atom occupies a substitutional position. Because it is a smaller atom compared to silicon it leads to a decrease of the lattice constant corresponding to a lattice contraction of about one atomic volume for each carbon atom. This is contrary to the lattice expansion of oxygen, therefore one can expect that an attractive interaction between oxygen and carbon atoms can occur which may help to reduce the strain. If both carbon and oxygen are present in silicon in high concentration which is frequently the case in silicon for terrestrial solar cells the solid solubility is less defined presumably because of this interaction between carbon and oxygen. 4) In fact dissolved impurity concentrations of up to 2 x 1018cm -3 (40 ppm) have been measured exceeding the maximum solubility by an order of magnitude. There is therefore evidence that high concentrations of oxygen enhance the solubility of carbon considerably. Carbon on a substitutional site as an isoelectronic element to silicon is electrically inactive. It should be mentioned, however, that carbon can also occupy an interstitial site where it becomes electrically active and introduces two trap levels ( E c - 0.12 eV, Ev + 0.27 eV) which can be observed for instance by DLTS314o This defect can be produced by low temperature electron irradiation but anneals out at room temperature. The diffusion coefficient has been measured in the temperature range 1050 to 1400C and is given by 42) Dc = Dco exp - KTJ
with
Dco = 1.9 [cm2/s], H~ = 3.2 [eV].
(124)

T [K] 1800 1600 1400 1200 I000 8OO 6OO 4OO
281
Si:C Liquid
L + SiC
SiC Precipitates
C-Icomplexes
I I I I I I I I si
10 12 14
16 18 20 LogN[cm-3]
~ I f
si
[11 o]
FIG. 41. Phase diagram for carbon in silicon and temperature range for the occurrence of SiC precipitates. Shaded area indicates the narrow liquid-solid two phase regime. The CI-complexes occur after irradiation at low temperatures (below room temperature) but may also exist at higher temperatures in crystals with a high supersaturation of self-interstitials.
i"-"
FIG. 42.
~ Sl
Structural model for the carbon-siliconinterstitial (C-I) complex in silicon.
Since carbon occupies substitutional sites the diffusion requires intrinsic point defects. In general the diffusion can be described by eq. (116) involving vacancies and self-interstitials. The rather high activation energy suggests a vacancy mechanism, but it has been observed that a supersaturation of self-interstitials after in-diffusion of phosphorus (143) enhances the diffusion of carbon diffusion at 900C. Goesele and Tan (144)have proposed that the diffusion mechanism should involve a silicon self-interstitial (Sii or I) through the reaction Cs + I,-, (CI) where a highly mobile carbon self-interstitial is formed. Such a complex has indeed been observed after low temperature electron irradiation and a split-interstitial position as shown in Fig. 42. has been suggested for this defect. Despite the question whether the diffusion actually occurs through the CI-complex the importance of the substitutional carbon is also related to the fact that it can selectively trap a mobile self-interstitial and thus affect processes which require self-interstitials as will be discussed in the next section. 4.1.2.5. Carbon related microdefects. If carbon is present in concentrations exceeding the solubility a driving force exists for precipitation in the form of SiC. However, even in carbon rich (> 5 x 1016 c m - 3 ) Cz- and Fz-crystals silicon-carbide precipitates are rarely observed. 023) This is partly due to the fact that carbon has a diffusion coefficient which is about an order of magnitude lower compared to oxygen so that for typical cooling conditions diffusion becomes too slow for the precipitation of the SiC particles. Another reason may be the lack of suitable nuclei for heterogeneous nucleation. Instead, it has been shown that the presence of carbon can induce a variety of secondary defects which have been studied extensively. (~23) Although the nature of these defects has not been identified in all cases the investigation of the conditions of their formation has lead to the following model which can account for many observations. In a typical cooling situation after crystal growth for instance of float zone (low oxygen content) and dislocation free silicon crystals a supersaturation of self-interstitials can build up depending on the growth condition (cooling and growth rate). Since in silicon the formation energy of interstitial-type dislocation loops is fairly high the formation of carbon-self-interstitial complexes and CI-agglomerates can be a more favorable alternative
282
because of the volume contraction. Without plastic deformation the growth can only proceed by the further absorption of self-interstitials (or emission of vacancies). It has been proposed that these possible agglomerates are identical to the "B-swirls", a defect which is observed in Fz-grown silicon crystals under certain growth conditions (see Chap. 5.1). More sophisticated arguments by F611 et al. ~145)lead to the conclusion that B-swirls may eventually collapse and transform into the so called A-swirls which have been identified as interstitial-type dislocation loops with an average size of about 0.5 to 10/zm. Within this model one can expect that a reduction of the self-interstitial concentration should also retard the carbon agglomeration or precipitation. Such a suppression of carbon precipitates is actually observed in carbon-rich (> 1017 c m -3) Fz-grown silicon after annealing between 550 to 1300C when dislocations are present which can easily absorb self-interstitials and reduce the driving force for the precipitation. (123) The investigation of carbon-rich Cz- and Fz-silicon has clearly shown that the carbon is involved in the formation of microdefects and many results can be explained by assuming the formation of CI-agglomerates as an early stage of carbon precipitation. The nature and the mechanism for the formation of these micro-precipitates is, however, not yet completely understood. 4.1.2.6. Co-precipitation of carbon and oxygen. The previous models relate the formation of oxygen and carbon precipitation to the excess or deficiency of self-interstitials. The equilibrium concentration of self-interstitials can be disturbed in several ways. A supersaturation can occur simply by lowering the temperature because of the lower solubility or by the oxygen precipitation itself for instance through the reactions 4.15 or 4.16. The concentration can also be changed by external processes such as surface oxidation or phosphorus diffusion. Considering the second external processes and only one impurity in excess a supersaturation of self-interstitials will promote the carbon precipitation and retard the oxygen precipitation. An undersaturation has the opposite effect. These trends are in general agreement with the observation as has been discussed before. If the two impurities are present in comparable concentrations their interaction has to be taken into account. A supersaturation of self-interstitials will certainly again promote the carbon diffusion and the formation of Ci-agglomerates. Considering the different volume changes it is likely that the CI-agglomerates may act as nucleation sites for the SiO2 precipitates and hence enhance the oxygen precipitation, contrary to the situation where the carbon concentration is low. Experimental results in fact indicate that the level of carbon concentration may be essential for the different behavior of oxygen precipitation in cases where self-interstitials are externally generated. At lower temperatures where thermal donors form the situation is different. It has been observed that carbon reduces the TD formation rate considerably. This has been used as an indication that the thermal donors may actually be self-interstitial agglomerates and hence the process competes with the carbon-self-interstitial formation. In view of the controversy about the nature of the TD the influence of the carbon on the nucleation rate remains an open problem. A different situation is the first case when no external generation processes occur but a supersaturation of self-interstitials is present after cooling down the crystal. If the carbon concentration is sufficiently high CI-agglomerates can form as explained above. For high oxygen supersaturations the driving force for precipitation may be high enough so that precipitation occurs independent of the carbon concentration. For lower oxygen
SEMICONDUCTORS
283
concentrations the presence of CI-agglomerates can facilitate the nucleation of oxygen precipitates. This is in general agreement with the experimental observations, ~46) however, more refined considerations of the detailed balance between self-interstitials, CI-agglomerates, oxygen precipitates or dislocations are required. Detailed models have been developed to account for all experimental situations and for a discussion of these details one should refer to the original papers. 4.1.2.7. Silicon-carbide precipitation at high temperatures. One of the puzzling results for carbon in Cz- and Fz-grown crystals is that in general the SiC precipitation is suppressed. In fact, it seems to be rather difficult even in carbon-rich single crystals at high annealing temperatures ( > 1300C) to obtain SiC precipitates. On the contrary one can observe SiC precipitates quite frequently in carbon-rich silicon for solar cells. 047,148) Typically SiC precipitates with sizes between 0.1-1/~m can be observed (see Chap. 5). TEM and H R E M investigations ~49) have identified the cubic 3C (fl-SiC) and the hexagonal 15R (~-SiC) phase (Fig. 82). Since SiC can occur in a variety of different phases (polymorphism) other structures may occur as well but have not been identified so far. The 50 known SiC phases differ by the stacking sequence of the close-packed planes. The various SiC phases are semiconductors with band gaps between 2.2 and 3.2 eV. One of the reasons for the more frequent occurrence of SiC precipitates in cast or ribbon grown solar silicon may be the usually higher carbon concentrations compared to electronic grade silicon. Another proposal has been given recently to account for the apparent difference of SiC precipitation at high temperatures in these materials. (~5'~5~)It is assumed that even for high carbon concentrations the SiC precipitation is hampered in the solid but under particular conditions can occur directly from the melt. Since the most likely place for heterogeneous nucleation is the melt-interface it has been suggested that the SiC precipitates nucleate at the solidification front and grow until they are incorporated by the advancing interface. Based on this idea an analytical solution of the diffusion problem has been derived assuming growth of the precipitates by diffusion of carbon in the melt. The results show that the maximum radius is proportional to the ratio AC,m/V, where AC,m is the supersaturation of carbon at the interface and v the interface velocity. Slow interface velocities and high supersaturations of carbon can thus lead to large precipitates. Typical values for directionally solidified solar silicon are v = 0.001 #m/s and C,m "~ 150ppm which yields precipitate sizes in the micron range. These results indicate that larger SiC precipitates ( > 0.1 #m) may actually have formed at the melt-interface. For higher interface velocities of about 0.01 #m/s or lower supersaturations of about AGain ~,~ 100 ppm precipitate sizes below 10 nm result. These conditions occur for some ribbon or sheet growth techniques ( E F G silicon) or in Cz- or Fz-crystals (low carbon concentrations). Whether these nanometer size SiC precipitates actually form, or if carbon stays in solution and gives rise to the CI-agglomerates in the solid state has not been decided yet, but the result can explain why the SiC precipitation is less frequently observed in these materials. 4.1.2.8. Electrical activity of carbon and oxygen related defects. There is clear evidence that oxygen forms a variety of defects which are electrically active. At low temperatures ( ~ 450~C) the thermal donors are formed with maximal concentrations of about 1016 c m 3. Since they anneal out at higher temperatures they are probably not important for solar cell silicon. At higher temperatures (600-900C) new donors appear which are likely to be related to SiOx precipitates, whereas SiOx precipitates formed above 1000C are less electrically active. The origin of the electrical activity is certainly not understood yet but may be related to the
284
Si/SiOx interface. Another possibility is that the electrical activity of the SiOx precipitates is determined by impurities which are gettered, a process which is used to remove electrically detrimental impurities. Still much less is known about the electrical activity of the carbon related defects. Since there is clear evidence that carbon has an influence on the oxygen precipitation which can lead to electrically active defects the main influence on the electrical properties of silicon may be indirect. It has also been observed that A- and B-swirls change the lifetime of single crystal silicon but there is no conclusive evidence yet that the precursors, possibly the CI-agglomerates, possess an electrical activity, although it has been speculated that they may be responsible for the reduced lifetime in EFG-silicon (see Chap. 5.2). Another possibility is the occurrence of very small (nm size) but electrically active SiC precipitates which have not been found yet as discussed in the previous section. The electrical properties of SiC precipitates if they occur are likely to be determined by their semiconducting bulk properties. One can expect that the electrical conductance is strongly affected by the contamination of impurities (doping) from the surrounding silicon matrix. The interaction with impurities can be seen in Fig. 87 for instance which shows the precipitation of fast diffusing copper at a SiC precipitate in cast solar silicon. Contaminated SiC precipitates in the micron range will be rather conductive and thus detrimental for the performance of a p-n junction if they lie in the depletion region for instance.
4.1.3. Transition metal impurities

Deep level impurities with high capture cross sections for electrons and holes are efficient recombination centers in semiconductors. They are of special concern for devices and photovoltaic applications. Transition metal impurities in silicon are known to form deep trap levels in the forbidden gap. The most extensively studied impurities are the 3d transition elements from Ti to Cu, and some of the 4d and 5d elements such as Mo, Pd, Ag and Pt, Au, respectively, in silicon. Their electrical properties and the diffusion and solubility behavior have been reviewed and discussed in a large number of articles [e.g. Ref. (152)]. A summary of the most important properties from the perspective of the photovoltaic application will be given in the following sections mainly for silicon. Transition metals are also present in other semiconductors but are usually less investigated because they are not the most important deep level defects which can be observed. For instance in GaAs the dominant deep level EL2 center which occurs in semi-insulating LEC-material at a concentration of about 1016 c m -3 has attracted the most interest in recent years. The concentration of transition metal impurities generally can be kept below this concentration level and hence they play a greater role in rather pure semiconductors such as electronic grade silicon. The situation in low cost solar cell silicon is different since higher concentrations of impurities have to be tolerated. Therefore the transition metals contribute significantly to the total recombination behavior of the material. The properties of most of the transition metals which can occur in solar silicon are listed in Table 4. 4.1.3.1. Solubility. The transition metals (except the noble metals) form complex phase diagrams with a number of intermetallic phases in each case. Some of the room temperature equilibrium phases between silicon and the metals are also summarized in Table 4. For most of the elements a silicon-rich MeSi 2 phase is formed in equilibrium with the ~-phase (silicon with dissolved impurity). Exceptions are the 5d elements osmium and iridium which occur in the MeSi 3 phase and the elements with a complete (or nearly complete) d-shell which form
SEMICONDUCTORS
285
Table 4. Experimental data for the diffusivity and solubility of 3d and some 4d/5d transition elements in silicon. D Oand HD are the prefactors and activation enthalpies for the diffusivity, and H i , S] the formation enthalpies and entropies for the solubility. The diffusion coefficients at l l00C, the maximum solubility ( C ~ ) , and the room temperature equilibrium (silicide) phases with silicon are also included 52'326~
3d Elements Silicide phase H i [eV] Ti TiSi 2 3.05 4.2 1 x 1015 1.79 1.5 x 10 -2 4 x 1 0 -9 Zr ZrSi 2 5 x 1014 V VSi 2 2 x 1016 Cr CrSi 2 2.79 4.7 1 x 1016 1.0 10 2 2 1 0 -6 Mo MoSi 2 2 x 1014 Mn MnSi 2 2.81 7.3 2 x 1016 Fe FeSi 2 2.94 8.2 2 x 1016 0.87 6.2 x 10 -3 4 x 1 0 -6 Ag 5 X 1017 1.6 0.002 2.7 x 10 -9 Co CoSi2 2.83 7.6 3 x 1016 Ni NiSi2 1.68 3.2 7 x 1017 0.47 2 x 10 -3 4x10 5 Pt PtSi 1.9 0.1 1.I x 10 -8 1 X 1017 0.39 2.4 x 10 4 9.0 x 10 -6 Cu Cu3Si 1.49 2.4 2 x 10 Is 0.43 4.7 x 10 -3 lxl0 4 Au
S~/K
C m~'x[cm -3] H D [eV] D O [cm2/s] D1100 [cm2/s] 4d/5d Elements Silicide phases C max [cm -3] H D [eV] D O [cm2/s] Dil00 [cmZ/s]
~ 1 0 -7 Nb NbSi 2 5 X 1014
3 x 1 0 -6 Pd Pd2Si 3 1016
~ 1 0 -5 W WSi 2 9 x 1013
a metal-rich phase (Cu3 Si, PdSi and PtSi) or no silicide at all as for the noble metals Ag and Au. Solubility curves for the a-phase of the 3d transition elements are shown in Fig. 43 and are characterized by an increasing solubility of the elements with increasing atomic number. In general the solubilities of transition metals are several orders of magnitudes lower compared to the doping elements. The solubility C eq below the eutectic temperature is exponentially dependent on the temperature S~ C eq = exp -~ exp {
-- KTJ
H~'~
(125)
where S~ and HS~ are the formation entropy and enthalpy of solution for the metal atom in solid silicon in equilibrium with the metal-silicide. Figure 43 shows that for the 3d-elements the solubility of Cu and Ni is significantly higher compared to the other elements. This is also reflected in the values for the formation enthalpies which are summarized in Table 4 and allows to separate the elements into two subgroups 3dI (Ti, Cr, Mn, Fe and Co) H~ ~, 2.85 eV 3dII (Cu, Ni) H s ,~ 1.6 eV.
There is little information available for vanadium which belongs to the 3dI group and most of the 4d and 5d transition metals so that a comparison of these elements is currently not possible. An important issue regarding the solubility of transition metals is whether they occupy substitutional (C~q) or interstitial sites (c~q). From a careful analysis of solubility measurements for the 3d elements it has been concluded that at high temperatures they preferentially occupy tetrahedral interstitial sites (see Fig. 35) in silicon with a smaller fraction of atoms occupying substitutional sites. In general one has to assume that for each element the concentration ratio ceq/Ces q is different and varies with the solvent and the temperature. For instance diffusion measurements for gold in silicon can only be reconciled with theoretical models by assuming that a large fraction of Au occupies substitutional sites in silicon3146) The solubility for the transition elements above the eutectic temperature is retrograde. Empirically it has been found that the maximum solubility of the impurities (see Table 4) is related to the distribution (segregation) coefficient, ko, by the Fischler relation ko ~ 1.92 x 10 -22 Cm~ax The distribution coefficients for transition elements are generally very .
286
D
c [cm'3 ]
10 18 10 17 10 16
10 -4 10 -6 10-8
--
NI,Co
Cu
--
"'*'-....
Cr,Mn
10-10 --
cu
10 15 10 14 10 13 10 12 10 -12 - 10 -14 10-16
', T m
'1
0.6
I
0.7
I
0.8
I
0.g
I
1.0
I
1.1
I
1.2 1000/T [K "11
I
0.6
I
0.7
I
0.8
I
0.9
I
1,0
I
1.1
I
1.2 1000/I" [K -1]
FIG, 43. Experimental solubilities of 3d transition metals in silicon, collected from Ref. (152).
FIG. 44. Temperature dependence of the diffusion coefficients of 3d transition metals and of the selfinterstitial in silicon, collected from Ref. (152) and (329) for titanium.
low therefore it is possible to refine silicon crystals during solidification and keep the metal impurity concentrations in the solid low. The solubility of the transition elements at room temperature is also low because of the strong temperature dependence of the solubility, less than about one atom/cm 3. Therefore to a large extent precipitation of the metal-silicide phases will occur during cooling. 4.1.3.2. Diffusion. Transition metals are fast diffusers in silicon at high temperatures (T > 0.7 Tm, Tmmelting temperature). The diffusion constant as a function of the temperature can be expressed by D = Do exp { - K ~ } (126)
and is shown in Fig. 44 for the 3d elements. In Table 4 some of the activation energies HD and prefactors Do in the temperature range 900-1350C are listed together with the diffusion coefficient at 1100C. The activation energies for the 3d elements vary between 0.5 to 1.0 eV and the diffusion coefficients at the melting point are in the range of 10-Scm2/s. The comparison with the diffusion coefficients for self-diffusion (Hso ~ 4.86 eV) and some of the doping elements (HD ~ 3.7 eV for P or B) suggests that in the high temperature regime these high diffusion coefficients can only be explained by an interstitial diffusion mechanism. Although titanium has a rather large activation energy compared to the 3d impurities EPR results indicate that the atoms also occupy interstitial positions at high temperatures. The few results for most of the 4d and 5d impurities are less clear so far and certainly require more investigations. A summary of the various experimental results for silicon is that a significant fraction of the transition impurities occupy interstitial sites, and the high diffusivity is due to a high mobility of the interstitial species in the lattice. In general, the diffusion of such impurities involves an exchange of atoms between substitutional and interstitial sites. Basically two mechanisms have been proposed in the literature for the diffusion of these impurities in semiconductors: the dissociative and the kick-out mechanism [107] both of which require the
SEMICONDUCTORS
287
participation of native point defects. The dissociative mechanism describes the transition of a substitutional atom A s to an interstitial site A i by the reaction As*--,Ai + V (127)
which involves the generation of a vacancy, V. Alternatively, the kick-out mechanism requires the presence of a self-interstitial, I, in the reaction As + I ~-~Ai. (128)
Both mechanisms are controlled by the concentrations and diffusivity of either vacancies or self-interstitials and hence lead to different predictions about the diffusion behavior. For instance, the basic equations for the diffusion via the kick-out mechanism describe the transport of the interstitial impurities (diffusivity Dj and concentration Ci) and of the self-interstitials (D~, CI), and the interchange between the different species by a law of mass action for the reactions (128).
Ot =DI--~-Yx2+ ~ - + K 1
C i ~2C i ~C s
CI
(129)
0t = D~ ~x 2
Cs Cseq C 1Ci C~ q c e q "
~t
(130)
(131)
Cs and Ds are the concentration and diffusion constant for the impurity on substitutional sites. The quantity K~ describes the strength of sinks or sources for self-interstitials, which are frequently related to dislocations by an equation
K~ = ?~ N o D ~C~q
(132)
where No is the dislocation density and 7t a parameter which characterizes the geometrical dislocation arrangement and is of the order of one. Numerical and approximate solutions of this set of equations have been derived and extensively discussed in various articles. (17) In solar cell semiconductors with a rather high density of dislocations or grain boundaries one can assume that the sink density is high enough so that the equilibrium of the self-interstitial concentration can be established almost instantaneously locally (C~ = c~q). In this case the eqs (129) and (130) lead to the normal diffusion equation for the total impurity concentration Ctot = Cs + Ci t3Ctot 02 Ctot C~q t~t = D e f r ~ Deer=Dic~ q + C ~ (133)
where Deer is an effective diffusion constant. The same equation is obtained for the substitutional and interstitial components since both concentrations are proportional according to eq. (131). If the impurity is essentially dissolved interstitially (C~q>>C~q) the diffusion constant is directly determined by the interstitial diffusion coefficient De~ ~ D,. The same result is also obtained for the dissociative mechanism if an equilibrium for the vacancy concentration is assumed. The diffusion profiles are well known error-function solutions (see for instance eq. (143). The diffusion behavior becomes more complex if vacancies and self-interstitials are not in equilibrium locally. This situation can arise in dislocation free silicon and has been observed
288
for the Au diffusion where for instance U-shaped diffusion profiles at the surfaces can occur which cannot be described by the error function solution. The analysis of the eqs (129) to (131) for this case shows that this behavior can indeed be predicted. Au is supposed to occupy mainly substitutional sites where it is rather immobile but diffuses rapidly in the interstitial lattice. The U-profile is due to the enhanced transformation of Aui to Aus near the surfaces, which act as sources or sinks for vacancies (dissociative mechanism) or interstitials (kick-out mechanism). The diffusion via the kick-out mechanism is described by an effective diffusion coefficient Deer=
C seq D I C~q
(134)
which is strongly dependent on the concentration of the substitutional component. This equation shows that it is the diffusion of the self-interstitial which dominates the Au transport. The quantitative interpretation of the experimental results in terms of the two mechanisms is related to the controversy about the concentration and diffusivity of the native defects, vacancy or self-interstitial, in silicon, but seems to favor the kick-out model for the Au diffusion.9~ For most of the 3d elements the experimental results at high temperatures indicate, however, that the interstitial diffusion is the dominant diffusion mechanism independent of native lattice defects. Nonetheless, one cannot rule out at present that for some elements (such as Au) the more complex diffusion processes have to be considered to account for all possible diffusion conditions. At room temperature the impurities are still mobile to some extent. Considering their low solubility it is evident that during or after cooling the major fraction of the dissolved impurities will disappear and preferentially precipitate as the pure metal for Ag and Au or forming the metal-silicide phase for the other elements. Alternatively the elements can associate with other impurities or intrinsic point defects or may form small clusters. The morphology of the defects which actually nucleate at low temperatures is determined by the kinetics of the precipitation and complex formation process and hence dependent on the thermal history of the specimens and the presence of other defects. Of particular interest is the interaction with interfaces, dislocations or other lattice defects which will be discussed separately for the grain boundaries in Chap. 4.4. In some cases the various stages of precipitate formation have been studied by HREM 54'155) and the results usually show a complex morphology (size, phase and structure) of the precipitates (see for instance Fig. 86). After rapid cooling (quenching) metal atoms may also remain on interstitial or substitutional sites and be quite stable if the number of nucleation sites for precipitation or the concentration of point defects for complex formation is low such as in high purity single crystals. In solar cell semiconductors with lower perfection and purity one can, however, expect that after cooling the transition metals are either precipitated in the bulk or at extended lattice defects or have formed complexes with other point defects. The ramifications of these interaction processes on the electronic properties of the semiconductors will be discussed in the next sections. 4.1.3.3. Energy levels of the transition elements. Transition metals are the most important impurities which form deep energy levels in the forbidden band gap. Because of their usually high mobility they are easily introduced during device processing and because of their electrical activity can be rather detrimental for the device performance. Although they have been studied for several decades and the results are described in hundreds of publications
SEMICONDUCTORS
289
conclusive identifications and established energy levels are in many cases not available. The complexity of the issue becomes evident if one considers the results of the previous sections. A rather consistent picture has emerged for high temperatures. The results of the solubility and diffusion measurements for silicon show that the 3d transition elements diffuse interstitially and stay preferentially on interstitial sites. The high diffusion coefficient for the 4d and 5d elements indicate a similar behavior but more experimental data are certainly necessary to establish the diffusion mechanism and the ratio of atoms which occupy substitutional and interstitial sites. The situation at low temperatures is less defined since the elements can occur in a large variety of different defect states which are determined by the experimental conditions as discussed before. It is evident that the electronic levels which are associated with a complex defect such as a three dimensional precipitate is difficult to predict and can vary with the morphology. Therefore it is not surprising that the variability of the defect structure is reflected in the large number of different trap levels which have been associated with a particular impurity. A comparison of different results is thus only possible if the defect configuration at the atomic level can be established and related to a particular trap level. This problem has been solved to some degree for the electronic levels of impurities when they occupy interstitial sites. Some of the transition elements with lower diffusivity can be kept in the interstitial position after quenching. Well established are the electronic levels of the interstitial 3d elements (Ti to Fe) in silicon which are given in Table 5. The results show that in general the impurities introduce donor and acceptor levels which are associated with different charge states. The lowest state is the double-positive donor state which lies in the lower half of the band gap and for iron and chromium is shifted into the valence band. In p-type silicon at low temperatures the interstitial metals are therefore positively charged (MiU). A particular important process which can occur then is the formation of a shallow acceptor-metal pair. The mobile ion can be captured by a negatively charged shallow acceptor (A?) and form a (M~+Au) pair simply by Coulomb attraction. These acceptor~lonor pairs introduce a new donor trap level which have been observed for a number of transition elements 52) and are also listed in Table 5 for boron doped silicon. Acceptor-metal pairs have also been observed for Ga and A1 doped silicon. The binding energies are usually small and close to 0.5 eV so that the pairs dissociate at elevated temperatures above about 200C. Although in general fast cooling or quenching is necessary to keep the metal on interstitial sites it is observed that a larger fraction of the impurities such as Ti, V or Cr may still remain dissolved even after slow cooling because of the decreasing diffusion coefficient with decreasing atomic number of the 3d metals. In fact it is observed that even small
T a b l e 5. D o n o r and acceptor levels of 3d transition elements on interstitial lattice sites. For other transition elements the occupation of an interstitial site has not been observed or could not be confirmed yet. The energy levels of the metal-boron pair are included in cases where they occur. Data collected from Re (152)
Impurity Ti V Cr Mn Fe
Acceptor level M -/ E c - 0.08 eV E c - 0.16eV E c - 0 . 1 1 eV
D o n o r level M /+ E c -- 0.28 eV E c - - 0.45 eV E c - - 0.22 eV E c - - 0.42 eV Ev + 0.39 eV
Double donor level M +/+ + E v + 0.27 eV Ev + 0.30eV Ev + 0.25 eV
Pair donor level MB /+
E v + 0.28 eV E c - - 0.55 eV Ev + 0. I0 eV
290
PROGRESS
IN
MATERIALS
SCIENCE
Table 6. Various donor and acceptor levels of some transition elements which are experimentally observed by Hall effect or DLTS measurements. The atomic nature of the defects is in most cases unknown or controversial~59'332) Impurity Acceptor levels Donor levels
Co Ni Cu Pd Ag Pt Au
Ev + 0.35 eV Ev + 0.22 eV Ev+0.18eV Ev + 0.24 eV Ev + 0.20eV Ev+0.21 eV Ev + 0.37 eV Ev+0.35eV Ev + 0.34 eV
E c - 0.53 eV Ec - 0.35 eV Ev+0.33eV Ev + 0.52 eV Ev + 0.53 eV
Ev + 0.36 eV
Ec - 0.54 eV Ec -- 0.33 eV E c - 0.54eV Ec - 0.25 eV Ec - 0.28 eV E c - 0.54eV Ec - 0.54 eV
Ev + 0.32 eV Ev + 0.32eV Ev + 0.33 eV Ev + 0.29eV Ev + 0.35 eV Ev + 0.35 eV
concentrations o f about 1 0 t4 c m - 3 of Ti or V (156) can severely degrade the solar cell efficiency likely because o f the remaining interstitial component (see Chap. 5.1). E P R and D L T S measurements indicate that Co, Ni, and Cu in silicon cannot be kept on interstitial sites even by rapid q u e n c h i n g Y 2) They precipitate or form other defects and introduce a variety of levels as discussed above. Some results which have been reported in agreement by a number of authors are compiled in Table 6. In most cases there are a number of results published which cannot be reconciled with other measurements. The concentrations of defect levels associated with precipitates or complexes are usually rather low compared to the impurity concentration most likely because the formation of the electrically active defects require the participation of many atoms. Therefore higher concentrations of the metal impurity are necessary to produce deep level concentrations which can become harmful for the efficiency of silicon solar cells (see Chap. 5). Since the metal impurity concentrations can be kept below certain levels ( < 1016cm -3) rather easily the slowly diffusing transition metals are more detrimental because even at low concentrations a substantial fraction can remain on interstitial sites and be electrically active. The situation for the 4d and 5d transition metals in silicon is less clear. Although some of these elements, such as gold, are thoroughly studied, the atomic configuration is in most cases not established. The main trap levels which have been frequently observed are given in Table 6 for the most c o m m o n of these elements. But similarly to the situation for precipitating metals such as Cu and Ni controversial measurements have also been reported for most of these impurities which suggests that the electrical activity of these elements is also due to complexes, aggregates or precipitates. In some cases it has been proposed that after cooling or quenching a fraction of the impurities may also occupy a substitutional site. For Ni for instance the dissociative mechanism was assumed to explain the diffusion behavior of the presumed substitutional nickel with an acceptor level at Ec - 0.41 eV. The results are, however, not conclusive as in other cases, so that at present the energy level of any substitutional impurity has not been established. Transition metals can also form deep trap centers in compound semiconductors but their concentrations are frequently lower compared to deep trap levels which are associated with other defects such as the antisite related EL2 defect in GaAs. For a compilation and discussion of these data one should consult the vast literature on this subject.
291
4.1.4. Hydrogen passivation of deep and shallow level defects

The behavior of hydrogen in semiconductors has been the object of intense experimental and theoretical investigations over the last decade (157-~59~and has been discussed in several review articles and conference proceedings. Besides the fundamental scientific interest the major reason is the importance of the role which hydrogen plays in modern silicon technology. It has been recognized that at almost every stage of the device processing hydrogen can be introduced into the silicon lattice and effect the device properties. In the fabrication of crystalline silicon solar cells hydrogen is intentionally incorporated because it can have beneficial effects on the performance and the efficiency.6) Hydrogen is even more important for amorphous silicon where it is usually introduced during the deposition of the films and is essential for the utilization of the material. (37'~6~)This particular aspect of the role of hydrogen will be discussed in Chap. 8. Hydrogen is also one of the four dominant neutral impurities in high purity germanium (besides Si, O and C) and has also been investigated in GaAs and some other compound semiconductors such as AIGaAs or GaP. Most of the information is available for silicon therefore the main results will be discussed for this material. A consistent picture of the state of hydrogen in the silicon lattice apparently has evolved during the last years and can be summarized as follows (see also Table 7). At low temperatures (< 500C) hydrogen can occur in three different states:
Table 7. Donor (D) and acceptor (A) levels of impurities and lattice defects in silicon (a) and GaAs (b) which can be passivated by hydrogen. For GaAs defect levels are compiled which can occur during epitaxial growth.(~59,~6t)The origin and nature of the defects is in most cases unknown or controversial. Ed is the dissociation energy for the defect-hydrogen complex as determined from annealing measurements. For extended defects a distribution of defect states N ( E ) has to be assumed
Lattice defects or impurities Au Ag Pd Pt Cu Ni Fe O~ (thermal donor) O-V (A center) V-V (divacancy) Plastic deformation Grain boundaries B Ga A1 Lattice defects or impurities LEC: EL2 LEC: EL6 LPE: - / E L 5 / E L 3 / MBE:-/EL6/EL5 Grain boundaries Si, S, Se, Te, Sn, Ge Zn, Cd, Mg, Be
Electron trap [eV] Ec - ET 0.54 (A) 0.54 (A) 0.22 (A) 0.28 (A)
Hole trap [eV] E v + Er 0.35 (D) 0.29 (D) 0.32 (D) 0.20 0.18 0.35 0.53(A) 0.21 0.53(A) 0.32 (D)
ED [eV] 2.3 2.2 2.4 2.3 2.5 2.5 1.1 1.9 1.9 3.1 2.5 1.1 1.6 1.9
0.07 0.15 (D) 0.18 (D) 0.22 (D) N (E) N (E) 0.045 (a) 0.065 (A) 0.057 (A) Electron trap [eV]
E c -- E.r E o [eV]
0.82 0.22 0.13 0 . 3 8 0.54 0.73 0.17 0.28 0.45 N(E)
2.5 2.2 2.2 2.9 2.1 2.1 1.6
292
(1) It can be trapped at unsaturated covalent bonds. This Si-H site has the lowest potential energy of all states. It is also clear that trapping can occur at point defects and extended lattice defects although it is not certain whether unreconstructed bonds or chemically induced reconstruction of bonds is the cause. (2) In the absence of defect sites to which hydrogen can be bound molecular hydrogen (H2) is the stable configuration. Molecular hydrogen occupies the tetrahedral interstitial site (Ta) in the diamond cubic lattice (Fig. 35), is electrically and optically inactive, and almost immobile. (3) Atomic hydrogen occupies the M-site which is the lowest energy site for unbound hydrogen. The diffusivity for atomic hydrogen is high even at room temperatures. The calculated and experimental activation energies are 0.3 eV and 0.5-1.2 eV, respectively for migration from M to M-sites across the C-sites. The three forms of hydrogen are in equilibrium amongst themselves, and the bound and molecular forms dominate at low temperatures which explains why hydrogen is rather immobile then. At higher temperatures the bounded configurations break up and the atomic form dominates, therefore the effective diffusivity becomes high. Theoretical calculations show that atomic hydrogen should have a ground state in the valence band but there is no direct experimental evidence that a donor level exists in the upper half of the band gap.
4.1.4.1. Passivation of deep levels. The particular property of atomic hydrogen, namely that it reacts with certain point defects, impurities or lattice defects such as grain boundaries and dislocations in crystalline silicon and other semiconductors (GaAs, A1GaAs, GaP, etc.) and passivates their electrical activity is the driving force for the utilization of hydrogen in the device and solar cell technology. Typically the passivation is achieved by exposing the specimen to a low pressure (0.1-0.5Torr) rf hydrogen plasma at temperatures between 100-400C. The depth to which the hydrogen penetrates and neutralizes the defects depends on the density of all sites to which the hydrogen can bind (especially the shallow level impurities), the temperature and duration of the plasma treatment. It can vary between about 1-100 #m. An alternative method is the hydrogen incorporation by ion implantation at low energy and high currents. This technique is usually used for passivating amorphous and crystalline silicon for solar cell applications. The commonly called Kaufmann ion sources allow a very good control over the penetration depth and other parameters but have the major disadvantage of an extensive damage of the near surface region which can be detrimental for the device performance. Besides lattice defects such as dislocations and grain boundaries many of the metal-related deep trap centers can be neutralized by atomic hydrogen. The most troublesome impurities in silicon are the transition metals which can be easily introduced during device processing. They have high capture cross sections for electrons and holes and hence degrade the performance even at low concentrations. The mechanism by which hydrogen passivates deep levels is rather clear in cases where the origin are dangling bonds. The bonding of the hydrogen atom to the dangling bond forms a bonding and antibonding state which are pushed into the valence and conduction band, respectively, and hence become electrically inactive. The mechanism for the passivation of transition metals presents a different problem. Since the microstructure of all deep-level centers in semiconductors is largely unknown, one can only speculate about some aspects of the deactivation process. Some impurities such as Pt and Pd in silicon are known to distort the lattice, so that they may be bound more strongly to some of the next neighbor atoms, whereas the remaining bonds are to some extent dangling
293
bonds which may bond with the hydrogen. In other cases the presence of hydrogen may well cause a distortion or even a reconstruction of bonds so that a stable bond with the hydrogen atom can be formed. In the following sections the origin of deep trap levels associated with grain boundaries (and dislocations) will be discussed. Despite the prevailing discussion in the literature about the electrical activity of these lattice defects in terms of dangling bonds the experimental and theoretical results seem to indicate that impurity related trap levels may play a greater role. Corresponding to the uncertainty about the origin of the electrical activity of extended lattice defects the mechanism of the passivation also remains unsolved, although it is evident that in both cases (impurities and/or dangling bonds) a passivating effect can be expected. Certainly a better understanding of the structure and electrical properties of grain boundaries and dislocation is necessary to understand the observed passivation effects in silicon and other semiconductors. Table 7 compiles some of the impurity and lattice defect states in silicon and GaAs which can be passivated. They include the lifetime controllers Au, Pd and Pt, (159)and impurity levels which are related to Cu, Ni, Ag, Fe complexes. The nature of the metal-related traps are in most cases not clear. The observation that the Fe-interstitial donor (at Ev + 0.39 eV) and the impurities V and Cr which are more likely to remain on interstitial sites at low temperatures cannot be passivated may suggest at least for the 3d transition elements that only metal-complexes can be deactivated. Since most of the deep trap levels associated with transition metals have not been identified yet this issue still remains open. In undoped Liquid Encapsulated Czochralski (LEC) grown GaAs the main defect is the EL2, center, as has been discussed in Chap. 4.1. This ASGa related antisite defect can be passivated by hydrogen. Similarly other deep level defects which are commonly observed in liquid phase epitaxy (LPE) and MBE grown material but have not been identified can also be deactivated. It is a general observation that a passivated defect can be re-activated by annealing. The thermal stability of the deactivated defects is indicated by the dissociation energy ED for the hydrogen-defect complex. In a simple kinetic model the fraction of donor-hydrogen complexes N / N o remaining after a time, t, at an annealing temperature T is given by No ED In ~- = tv exp KT" (135)
Usually an annealing treatment above 400C (for an activation energy of about 2.5 eV) is sufficient to reactivate the trap level. Generally the most stable levels are dislocation-related defects and grain boundaries, followed by deep level and shallow level impurities, as will be discussed next. After a passivation treatment deep and shallow levels are passivated likewise. From the perspective of practical applications where it is required to retain the doping concentration, the lower thermal stability of doping elements offers the possibility of reactivating them at low temperatures without loss of the deep trap passivation. 4.1.4.2. Shallow level passivation. It is well established now that shallow level impurities in a variety of semiconductors can be passivated by hydrogen. In Si, GaAs and A1GaAs both the deactivation of acceptor and donor impurities occurs and in CdTe, ZnTe and other semiconductors the passivation of acceptors has been observed. In silicon the acceptors B, Ga, A1 etc. are almost completely passivated (over 99%) within the penetration depth of the hydrogen. By sharp contrast the shallow donors P, As and Sb are only weakly deactivated and in most cases the change in carrier concentration is negligible. A further interesting result
JPMS 35-3/4~G
294
is that a thin n-layer in p-type silicon (e.g. in an n+p diode) can severely impede the diffusion of hydrogen into the semiconductor. These and other observations have led to the following model for the deactivation reaction of shallow impurities at least in silicon and GaAs. The basic assumption is that hydrogen has a donor level in the upper half of the band gap. In p-type silicon this level would be depleted and the hydrogen atom positively charged. The acceptor passivation can thus be described as a compensation reaction A- + H+ ~-~(AH) (136)
The ion pairing occurs as a result of the Coulombic attraction. Although this model describes the essential features of the shallow level passivation there is still some controversy about the nature of the hydrogen-dopant bond which is formed. Most commonly accepted is the formation of a Si-H bond along a (111 ) direction which interrupts one of the acceptor-silicon bonds while the acceptor moves off center in the opposite direction. This configuration can be viewed as a SiH-ASi3 system with a three-fold coordination of the acceptor atom and a weak interaction between H A . (162)In contrast, theoretical calculations also propose a possible interaction between the acceptor and hydrogen atom and the formation of a H-ASi4 configuration while the hydrogen is bound to the acceptor opposite a Si-A bond. The mechanism of the donor passivation is less clear. In n-doped silicon hydrogen will be neutral and passivation may occur by the reaction D + + H + e-~-~(DH) . (137)
The electronic character of the bonding and the position of the hydrogen atom in the lattice is however unclear. Neutral atomic hydrogen H in n-doped silicon can also form the stable and rather immobile molecule H2 via the reaction H + H*-* HE. (138)
This reaction competes with the previous reaction and may in fact dominate at low temperatures and prevent a substantial passivation of donors in agreement with the observation. This would explain for instance the barrier effect of a highly n-doped layer in p-doped silicon. A comprehensive discussion of the experimental and theoretical results is given in Ref. (159). 4.1.4.3. Diffusion of hydrogen. The understanding and interpretation of the diffusion behavior of hydrogen in semiconductors is complicated by the fact that trapping at impurities and lattice defects, and the molecule formation can occur. Earliest high temperature measurements (between 970-1200C) determined the diffusion coefficient in silicon to b e 063) DH=DHexp --
KTJ DHo=9.4 X 10-3[cm2/s] HD=0.48[eV].
(139)
The activation energy is close to the calculated value for the atomic hydrogen migration from an equilibrium M-site to another of about 0.32 eV. Therefore one can assume that at high temperatures diffusion occurs through interstitial migration of atomic hydrogen. For a hydrogen trap level in the upper half of the gap most of the diffusion is by the neutral species since the Fermi energy at high temperature lies in the middle of the gap and the semiconductor becomes intrinsic. Later studies (164)in a lower temperature range (400-500C) gave different
295
values, namely DHo = 4.2 x 1 0 - S c m 2 S-t and H~ = 0.56 eV, which showed a much slower diffusion. Extrapolation to low temperatures (<400C) gives diffusivity values which are orders of magnitudes higher than what is usually observed. This impeded diffusion at low temperatures can be understood if one considers the trapping at defect sites and the formation of molecular hydrogen which is much less mobile (activation energy 2.7 eV for migration from an interstitial Td site via a hexagonal site). The effect of the defect and self-trapping processes leads to an effective diffusion constant Ca Deer= ~TT Drt
CT = CH "3!-CH 2 "~ CH T
(140)
where Ca, CH2, CH_ T are the concentrations of mobile atomic hydrogen, molecular hydrogen, and trapped hydrogen, respectively. CT is the total hydrogen concentration. In p-type silicon hydrogen will be present as a mixture of H and H +, whose relative concentrations depend on the temperature and the doping concentration. The charged atom can interact with acceptors at low temperatures (< 300C) whereas at higher temperatures the neutral species dominates which favors the molecule formation. In both cases the diffusion will be impeded. In n-type silicon only the neutral species is present and a significant pairing of molecules will occur which leads to a much slower permeation than in p-type silicon. At temperatures below 150C donor-hydrogen bonds can form, which passivates the electrical activity and impedes the diffusion further. In general, all processes take place simultaneously and the resulting diffusion coefficient and profile are difficult to predict. One can obtain some idea about the influence of the trapping effects if one considers limiting cases for steady state diffusion with a constant surface concentration of CHo. If molecule formation is the dominant process and dissociation is negligible one obtains a diffusion profile as a function of the penetration depth x given b y ~16t)
C.(x) = C.o
,+
1 c~ De~ = - // 6Dn ,~,/212 C. + CH2
(141)
where Km is the rate constant for molecule formation and DH the diffusion constant given in eq. (139). The effective diffusion constant becomes independent of the atomic diffusion constant. In the case that trapping at impurities dominates the diffusion process and dissociation can be neglected the steady state solution is given by
CH(x)=CHoeXp - - X \ DH /
where NT and KT are the trap concentration and rate constant for trapping, respectively. Without trapping the diffusion would be described by the normal profile
CH(x) = CHoerfc 4DH---~"
(143)
It is evident that experimental diffusion data differ considerably dependent on the experimental variables such as surface condition, trap concentrations and plasma parameters. A summary of the more reliable diffusion data is given in Fig. 45, along with the calculated diffusion constant using the experimental values mentioned above.
296
PROGRESS
IN
MATERIALS
SCIENCE
TEMPERATURE (*C) 700 500 3O0 2O0 IO0 50 27 ,~0-5/" I " t I I I I

/
t0"6~"
/
HYDROGEN IN Si
10-7k 10-0 ~ . , ~ ~" o" 4.2 x 10"5m2 S"1 ~'~u 10_s -~.-0 56.V i.~ ,10-10 (j h u I 0 -11 o Ae
"*
O0
eA
o
~, 10-13 10-14
to-V,
~0-~
1.5
2.5 3 3.5 IO00/T(K)
FIG. 45. Diffusion coefficientof hydrogen in crystalline silicon, collected in Ref. (159). Solid line calculated from the Ichimiya and Furuichi relation.<]64) Finally it should be mentioned that the solubility of hydrogen is also greatly affected because of the trapping o f hydrogen which can lead to much higher hydrogen concentrations as one can expect from the extrapolation of high temperature data given by
CH = Cno exp ~ _ H ~ J [ KT ~,
(144)
with C , o = 2 . 4 x 1 0 2 ~ c m -3 and H [ t = l . 8 8 e V , ('63) or C a o = l . 6 x 1 0 2 c m -3 and H h = 1.46 eV. ('~) Extrapolation to 300C yields in the first case CH ~ 2 X 107 crn -3 which has to be compared with measured data o f about 10 's to 102ocm -3.
4.2. Radiation lnduced Defects Radiation induced damage in semiconductors occurs through ionization and the displacement of atoms. The incoming beam of high energy particles or photons creates ionization along the path in the solid. Electrons are excited into higher energy states and descend after sufficient time into the ground state and release the energy mainly in form of heat to the lattice. Some electrons may be trapped at defects and alter their charge state which can change the recombination behavior of this defect or lead to the dissociation of complexes. The ionization may also cause redistribution o f defects by a process known as ion-enhanced diffusion. 65) Despite these secondary effects which can have an influence on the lifetime o f
SEMICONDUCTORS
FOR
SOLAR
CELL
APPLICATIONS
297
the semiconductor, ionization is primarily the cause for the energy loss of the radiation in the solid and transient in nature. Only the remaining 0.1% of the particles energy are responsible for the permanent damage in the crystal structure. The radiation damage depends on the energy distribution and intensity of the particle beam. For solar cells used in spacecraft details of the space radiation environment and its effect on the solar cell performance are important. They are compiled in Ref. (166) and have been discussed in a recent review. 067) Radiation induced atomic displacement by collision results, initially, in the formation of a vacancy-interstitial (Frenkel) pair. The energy which is transferred to the struck atom depends on the velocity of the beam particles. The maximum energy Tm for relativistic and non-relativistic particles with velocity v which can be transferred is Tm
=
E (Ml + M2) 2 2Mr (2 + ~--~E2~
4MIM2
(v/c << 1) (v/c>>l)
(145a) (145b)
Tm=EM2 \ -2
Mlc ]
where E and M, are the kinetic energy and mass of the incident particle, and M2 the mass of the struck atom. Kinetic energies of protons in space range between 1 to 100 MeV with high energies occuring during sun flares. Electrons may have lower energies of about 0.1 MeV but because of the lower mass can have high velocities so that relativistic effects have to be considered. The threshold displacement energy Td is defined as the minimum energy to cause displacement in the solid. Higher energies in excess of Td are transmitted to the displaced atom which in turn can displace further atoms then and lead to a more complicated defect structure. There seems to exist a relation between the threshold energy and the lattice constant of the semiconductors which is shown in Fig. 46. Although the threshold energy is useful to determine whether radiation induced Frenkel pairs can occur it is usually not possible to relate them to the final defect structure. The primary reason is that these defects are mobile even at low temperatures and can thus migrate and combine with other lattice defects or impurities. They are the primary cause for the degradation of the solar cell performance in space. The defects which are recombination active contribute to the reduction of the lifetime of the material and thus the degradation of the solar cell performance. According to eq. (36) the different lifetimes Ti which are related to a particular radiation-induced defect with the concentration Nri can be summarized and yield the total lifetime r of the material
1 1 1 1
Uth N T i -- = - - + 2 "17 To - - = - - "q- 2 0 " i Ti To
(146)
where To is the lifetime before irradiation. The individual lifetimes Ti are expressed here by the eqs (57) which have been derived in the Shockley-Read-HaU theory in Chap. 2.4.2. tri are the capture cross sections of the defects and Vth (=Vn or vp) the thermal velocity of the minority carriers. This equation holds for low injection and defects concentrations which are significantly less than the majority carrier concentrations. In many cases the concentrations NTi are proportional to the fluence q~ of the irradiation NTi = Ci 4~ so that eq. (140) and the corresponding equation for the diffusion length L reduce to
-=--"]-(~2ffiVthCi---q-g~T %
TO --
1
+KL (147)
- q = ~o L
where K~ is the lifetime damage coefficient, KL = K,/D the related diffusion length damage coefficient of the minority carriers. The use of K L rather than K~ implies that the diffusivity
298
100
PROGRESS
IN MATERIALS
SCIENCE
I
C
80
60
40
20
Ge~.
"D
Si----
10
cdse-~
0-- ~6aAs
S
CdTe--. "~'\ ~ lnAs ~x \ ~ - GaSb ~- ZnTe
1
0,15
I
0.25 0.30
0.20 I/a o, iN. /~-1
FIG. 46. Correlation of mean threshold energy, Td, with lattice parameters (a -~) of several semiconductors.O 651
does not vary with the fluence. Since diffusion lengths o f a semiconductor are relatively easy to measure these constants are extensively used to characterize radiation damage in solar cells. In general, they depend on the energy of the incoming radiation and also vary with the majority carrier concentration (or conductivity) of the semiconductor. Most o f the solar cells used for spacecraft are based on silicon and recently also on GaAs. A third cell type based on InP and currently in the research and development phase shows very good radiation resistance compared to Si and GaAs and may once replace these semiconductors for space applications. (24'91) For 1 MeV electron radiation with fluences between 1013 to 10]6cm -2 a typical value of KL for a p-based 0.1 ~ c m silicon solar cell is KL ~ 8 x 10 -1 and a corresponding value for a p-based GaAs cell is KL ~ 4.5 x 1 0 - 8 . (169) InP solar cells with similar cell characteristics are significantly more radiation resistant than both silicon and GaAs which is demonstrated in Fig. 47 where normalized efficiencies are depicted as a function of the fluence. (17) In silicon and GaAs the radiation resistance decreases (or KL increases) with increasing base doping concentration or conductivity, whereas the reverse is observed for InP cells.
SEMICONDUCTORS
FOR SOLAR
CELL
APPLICATIONS
299
/ - InP, DARK
,'- lnP. ILLU/'~INATED
.7 - -
5x 10 fl-ca
"51013
1014 1015 1 I~V ELECTRONFLUENCEor.~"2
1016
FIG. 47. N o r m a l i z e d efficiencies o f InP, G a A s a n d silicon solar cells after 1 MeV electron radiation. (171
Since the damage coefficients give only an average description of the radiation resistance of a semiconductor no insight into the damage mechanism itself is obtained which would allow the better understanding of the obviously complex behavior of the semiconductors. "['he most valuable tools for an identification of the radiation induced defects are the ESR and DLTS methods which in principle allow structural and electronic properties of the defects. Most of the experimental results are again available for silicon where many of the defects which are produced during irradiation could be identified. On the other hand, relatively little is known about the nature of radiation-induced defects in GaAs and InP. Whereas for InP the lack of sufficient data is due to the fact that the radiation damage in this material has been studied for only a few years, the situation is different for GaAs. Whereas many trap levels could be observed by DLTS after irradiation the structural identification by ESR is in most cases not possible because of the insufficient fine structure of the signals. The only defects which have been identified so far in as-grown GaAs are the arsenic antisite Asia which is either identical or at least very similar to the EL2 defect (see Chap. 4.1.1) and the defect C r 2 in intentionally doped GaAs. ('7') The identification of trap levels observed by DLTS measurements is therefore in most cases a matter of intuition and speculation. As has been discussed in Section 4.1.1 for multinary semiconductors one can infer to some extent from the stoichiometry of the crystals which intrinsic defect pairs may occur preferentially and thus obtain some clues about the nature and behavior of the intrinsic defects. In combination with other techniques such as IR-absorption, photoluminescence etc. some ideas can be developed about the nature of the defects. However, since irradiation produces non-equilibrium concentrations of intrinsic defects a priori one has to take into account all possible defects which can be formed in the compound. Since their number is comparatively higher than in elemental semiconductors and additionally because of the tendency of many defects to form complexes the identification frequently leads to contradictory results and remains to some extent speculative. Therefore the interpretation of the radiation-induced electrically active defects observed by DLTS is far from completed yet and one should refer to the original articles and reviews for an extensive discussion of the results for GaAs and InP. Without a discussion of the various DLTS traps which have been observed after radiation in n- and p-type GaAs the following general conclusions can be drawn. The major radiation induced trap levels seem to be associated with intrinsic defects on the arsenic sublattice. Since
300
PROGRESS
IN MATERIALS
SCIENCE
the EL2 which is the major defect in semi-insulating GaAs is likely to be the AsGa ( - X ) antisite it does not occur after irradiation and is therefore of less importance for GaAs space solar cells. An important observation is an enhanced annealing of the radiation induced defects around 100C under forward bias conditions in n-type GaAs and around room temperature under reverse bias in p-type GaAs. Several mechanisms have been proposed to explain the effect such as recombination of electron-hole pairs (recombination enhanced annealing(m) or alternating trapping of electrons and holes which changes the charge state and the migration energy (Bourgoin mechanism(m)). Although the exact nature of the mechanisms responsible for the enhanced annealing are not quite yet clear the experimental result is of particular importance for GaAs solar cell since it offers the possibility of extending the useful lifetime by annealing treatments at rather low temperatures. The experimental situation for silicon is more advanced since many of the intrinsic defects observed by DLTS have also been detected by ESR and other methods which can give information about the structural properties. The primary defects after irradiation are vacancies and self-interstitials. Since the self-interstitial appears to be mobile even at helium temperature (174) and recombines with other defects it has never been observed directly. Therefore very little is known about the isolated defect and most of the properties which have been attributed to the self-interstitial are obtained indirectly as discussed in Chap. 4.1.1. A silicon di-interstitial is, however, identified after neutron irradiation and consists of two self-interstitials which are separated by a substitutional atom and oriented along the (001) axis. This defect anneals out at 170C. It has been suggested that the disappearance of the silicon interstitial after irradiation in p-type silicon is due to trapping at shallow acceptor impurities which exchange their site then so that the interstitial impurities and the vacancies remain in equal proportions. (175) An example in p-type silicon is the formation of the boron interstitial which can occur in different charge states and may be a negative U-center. The boron interstitial is unstable at room temperature but appears to be stabilized by the complex formation with interstitial oxygen in Cz-grown crystals: B i - - O i . This defect which introduces a trap level at E c - 0.27 eV and is stable below 200C has been tentatively assumed to be the predominant cause for the radiation induced degradation in boron based n-p silicon solar cells. The majority of the radiation-induced defects are related to the vacancy. A partial list of the vacancy related defects relevant to solar cell silicon, their trap levels, and the annealing temperatures above which the particular defect disappears is given in Table 8. The properties of the single vacancy have already been discussed in Chapter 4.1.1. In addition, divacancies and higher-order complexes (three and four vacancy atoms) have been observed in irradiation silicon. The divacancy can either be formed directly when the energy of the irradiating particle is high enough to displace two neighboring atoms or by migration and association of single vacancies. The divacancy in its different charge states introduces three trap levels in the band gap and is stable below about 300C. The vacancy can also form a considerable number of complexes with other impurity atoms particularly the doping impurities.(176)In fact, in phosphorus doped Fz-grown silicon the P-V complex (E-center) is the dominant defect observed by ESR which is stable below 130C. In boron doped silicon the corresponding B-V complex occurs although it could not be confirmed directly by ESR measurements. Besides the doping elements carbon and oxygen particularly in Cz-silicon are the major impurities and a number of complexes can occur. In Czochralski grown silicon a vacancy-oxygen complex is observed which in its negative charge state is often referred to as the A-center. A single oxygen atom occupies a slightly off-center bond centered site (see Fig. 40) and in the V-O complex bridges two of the four
SEMICONDUCTORS
FOR
SOLAR
CELL
APPLICATIONS
301
Table 8. Electronic levels of point defect complexes in silicon which can occur after electron irradiation and are related to vacancies, interstitials, oxygen, carbon and some doping elements. Included are the annealing temperatures above which the complex dissociates or the defect levels disappears (323)
Annealing Temp. Defect

V /+ V ++/+ V /+ ( V - V ) +/ ( V - V ) /(V V ) -/ (O-V)(O-V) O V2 O-V 3 I-I P-V As-V Sb-V B-V B+ B?
Trap level
leVI Ev+ Ev+ Ec Ev+ Ec g c -Ec -Ec Ev+ EcEc Ec Ec E~ -E c -E E Ev+ Ev + Ec E 0.05 0.13 0.09 0.22 0.42 0.23 0.17 0.17 0.40 0.49 0.4 0.47 0.44 0.13 0.45 0.27 0.12 0.30 0.36 0.034 0.039
Remarks
Vacancy
[K] 170
Negative
U 60 570 610 570 620 100 620 720 440 420 450 460 260 ~ 50 470 320 340 600
Divacancy
A-center (Si20 molecule)
Di-interstitial E-center E-center E-center
Negative
Bi~) i
Ci C-I C~42 s Li i Lii~) i
Donor Donor
broken bonds of the vacancy thus forming a Si~O-Si "molecule" (Fig. 48). The negatively charged complex in n-type silicon is stable below 620 K whereas the neutral complex is only observed below 100 K. Multivacancy-oxygen defects such O-V2, O-V3, O2-V2 and O3-V3, have been identified by ESR where in all cases the oxygen is trapped at the vacancy in the Si-O-Si molecule configuration. Carbon is usually present in various degrees even in high quality silicon and one can expect that radiation induced defects containing this atom are also of primary importance. In p-type silicon the formation of the carbon-silicon interstitial (C-I) has been reported(177) which has already been mentioned in Chap. 4.1.2. This defect seems to be rather mobile and has been proposed to be responsible for the enhanced diffusion of carbon at 900C in the presence of a supersaturation of self-interstitials. 46) After irradiation the defect is only stable below 65C, whereas above this temperature a new defect occurs which is supposed to be the carbon-carbon interstitial Cr-Cs. This complex seems to disappear at temperatures around
[1~] - -
FIG. 48. Atomic configuration of the vacancy-oxygen complex (A-center), the oxygen atom is slightly displaced off the substitutional position.
302
PROGRESS
IN MATERIALS
SCIENCE
300C. Further complexes which have been reported involve the participation of oxygen presumably C<Oi. (178) Despite the many defects which can be observed after irradiation it has been assumed that the boron-oxygen complex Bi---O i is the primary cause of the radiation-induced degradation in silicon solar cells. It has also been observed that doping with lithium can increase the radiation resistance. 79) Therefore lithium doped p+n and counterdoped p+n solar cells have been mainly used for space applications. Lithium is a shallow donor and occupies interstitial sites in the silicon lattice. It has been inferred from infrared spectroscopy that interstitial lithium forms a complex with interstitial oxygen. It was therefore argued that this process competes with the boron-oxygen formation process and hence reduces the formation rate for this complex. This tentative conclusion about the beneficial influence of lithium on the radiation resistance is supported by DLTS measurements on irradiated lithium-counterdoped cells. 4.3. Dislocations Today's single crystal growth techniques for silicon and gallium arsenide allow the elimination of dislocations or a reduction in their density to levels where they are usually not detrimental for device properties anymore. Accordingly, the general interest in the investigation of their properties also declined. Nonetheless, a few groups continued to study the behavior of dislocations in covalently bonded semiconductors particularly in germanium and silicon for several decades. The quality and purity which can be achieved today for these semiconductors exceeds all other crystals and one can intentionally introduce just a few dislocations and study their properties under well defined conditions. These investigations have provided a very thorough understanding of the structural, electronic and mobility properties of dislocations. Although there are still a few fundamental problems unsolved a rather comprehensive picture has emerged. Since there are excellent reviews and numerous papers available on the subject~54'18,181) the results shall not be presented here again. From the perspective of the solar cell application there are, however, particular technological processes where dislocations play an important role. As has been mentioned before the requirement of low cost techniques for terrestrial solar cells has lead to the development of unconventional crystal growth techniques. Of particular interest are ingot and ribbon growth techniques for silicon which will be discussed in Chap. 5. These methods can produce polycrystalline crystals with high densities of dislocations and grain boundaries. Dislocations are also incorporated during hetero-epitaxial growth particularly for materials with a large lattice mismatch. Not surprisingly, a large number of papers have recently focused on the dislocation nucleation in epitaxial films. Since this is a general problem more relevant for the electronic device technology it shall not be discussed here either. Nonetheless, these developments are certainly also important for thin film solar cells which are grown by epitaxial techniques. In the following the discussion will be limited, however, to dislocations in silicon because it appears to be the most promising low cost solar cell material so far. Polycrystalline crystals produced by ingot and particularly ribbon growth techniques can contain so many dislocations that they become the most detrimental defects for the solar cell performance. Like many other lattice defects dislocations are efficient recombination centers and their recombination behavior is similar to grain boundaries as discussed in Chap. 2.4.4. Since the dislocation nucleation in low cost processes is to some extent unavoidable it may be important to understand how they nucleate and multiply in order to reduce their densities. Fortunately, the dynamics of the dislocation motion in silicon has been studied for decades
SEMICONDUCTORS
303
FIG. 49. Projection of the sphalerite structure on a (01 I) plane showing a dissociated glide set dislocation with a stacking fault between the 30 and 90 partials. A and B sites are indicated by dotted (or black) and white circles, respectively. Core atoms with dangling bonds are emphasized by black circles. Exchange of A and B atoms leads to a second type of dislocation because the core atoms either terminate with a row of A (~ dislocation) or B atoms (/~ dislocation).
and is well known. The main results which are considered to be essential for the dislocation generation during crystal growth shall be presented. 4.3.1. Dislocation velocity The atomic structure of dislocations in semiconductors has been studied extensively both experimentally and theoretically. The essential results are as follows. The dislocations glide on { 111 } planes and are dissociated into partial dislocations which bound a stacking fault (Fig. 49). The dissociation width varies between about 10 to 100/~ depending on the material. Since the { 111 } glide planes consist of a double layer of atoms in the diamond cubic and sphalerite lattice the extra half plane of the dislocation can terminate between the widely (shuffle set) or narrowly (glide set) spaced planes. There is evidence now from H R E M in silicon and germanium that dislocations usually belong to the glide set although one cannot completely rule out that shuffle set configurations occur, too. In AB compound semiconductor with the sphalerite structure the core atoms of the partial dislocations can either be all A or B atoms which leads to two different types of dislocations (~ and fl) with usually different mobilities. The core region of the partials is a few atomic distances wide and geometrical models show that it contains a number of unsaturated bonds. It was a long standing controversy whether these dangling bonds are reconstructed or not but there is general agreement now that the atomic structure is mostly reconstructed. A dissociated dislocation contains, however, even in equilibrium a number of defects such as kinks, constrictions which are connected with jogs of various heights, or reconstruction defects. Dangling bonds may thus exist locally at these point like defects or also segregated impurities and introduce deep levels in the forbidden gap. Besides these details which are important for the electrical properties of the dislocations it is evident from the microscopic picture that the motion of a dislocation is connected with the breaking and reconstruction of bonds, a process which requires high energies. This is described in terms of a high periodic (Peierls) potential parallel to the (110) direction which has to be overcome during motion. Therefore in contrast to dislocations in metals the motion of dislocations in covalently bonded crystals is a strongly thermally activated process. Experimental results have confirmed that the velocity, v, is a function of the stress z acting upon the dislocation in the glide plane and the temperature T, and is given by the general relationship V=Vo~ ~exp -~-~
304
V [crn/s]
10 -2 10 -3 10 -4 10 -s 10 -s 10-7 10-8 I 1.0 I 1.1 I 1.2
60 o dislocation
Table 9. Experimental results for the activation energies Q, stress exponents m (stress r in MPa), and prefactors vo defined in eq. (148) for the dislocation velocities of 60 dislocations in p- and n-doped silicon. Data are taken from Ref. (181)
Doping [ e l -3] 2.0 1.0 1.0 1.0 1.2 8.0 1.4 1.0 2.7 6.8 8.0 3.0 x 1013(P) x I014 (P) X lOiS(p) x 1017(As) x lO is (P) x 10 Is (Sb) x 1019(P) x 10is (B) x 1017(P) x 1017(B) x 10 Is (B) x 1019(B) v o [cm/s MPa]
- -
m 1.5 1.35 -1.35

- - - -
Q[eV] 2.20 2.15 2.16 2.05 1.67 1.4 1.41 2.15 2.34 2.20 1.85 1.6
9.0 x 105 1.1 106

- -
2.9 x 103
- -
I 1.3
I 1.4
I 1.5
I 1000/T
[K "1]
5.3 9.0 9.3 1.5
FIG. 50. Experimental dislocation velocities of 60 dislocations as a function of the temperature for a shear stress z = 120MPa in pure and doped silicon. Curve (1) 8 10tScm -3 (Sb); (2) 6 101Scm -3 (As); (3) 6 x 10 t7 cm -3 (P); (4) 8 x 101Scm -3 (B); (5) 2 x 1017cm -3 (A1); (6) 2.5 x 1014cm -3 (B); from Ref. (181).
x
X 102 x 105 x 106 x 106 ---
1.4 -- -
1.38 1.38
where Q is the activation energy for the dislocation motion. In most cases the stress exponent, m, is slightly temperature dependent and the activation energy Q(T) a function o f the applied stress. The prefactor is a material constant. Equation (148) has been found for all covalent semiconductors which have been studied so far. The velocity depends on the character of the dislocation and usually experimental results are presented for the screw and 60 dislocation which are the fundamental dislocation types o f covalent semiconductors. A general curved dislocation can be considered a mixture of screw and 60 components and their velocity may be estimated from the two velocity values. As mentioned above, in compound semiconductors one has to distinguish further between ~- and fl-type dislocations. In general, during plastic deformation both types are formed, however, their velocities are usually very different. The experimental results which have been obtained for the dislocation velocities in silicon are summarized in Fig. 50 and Table 9. The velocities of screw and 60 dislocations are usually measured in the temperature range between 450 and 800C and turn out to be rather similar. For undoped silicon and high stresses an activation energy of about Qo ~ 2.2 eV and a stress exponent m ~ 1.3 to 1.5 have been determined. For stresses below about 20 MPa the activation energy increases as shown in Fig. 51 for the 60 dislocation and reaches values of about 2.5 eV at 2 MPa. The dislocations are therefore slowed down in the low stress regime.
Q leVI
60 o dislocation
2.6
2.4
2.2
1.1
m-~
2.0
I 25
I
50
I
75
I
1O0
I
x [MPa]
FIG. 51. Activation energy Q(z) of the dislocation velocity as a function of the shear stress for a 60 dislocation in pure silicon. D a t a summarized in Ref. (181).
SEMICONDUCTORS
(a) &Q
FOR
SOLAR
CELL
APPLICATIONS
305
[ev]
0.0 -0.2 -0.4 -0.6 -0.8
I I I I
AA
(b) AO leVI
0.0 -0.2
As Sb P
B
AI
-0.4 -0.6 -0.8

I I I
1014 10 is 1016 1017 1018 1019
p[cm -3]
I I I I I I I 1014 10 is 10 is 1017 10 18 10 19 n[cm-3]
FIG. 52. Change of the activation energy AQ of dislocation motion with doping for p-type silicon (a), and for n-type silicon (b), data collected from Ref. (181).
The dislocation velocity is considerably changed in doped crystals. Whereas isoelectronic impurities such as Sn and Ge in silicon do not influence the velocity the typical doping elements such as boron or phosphorus etc. increase the velocity if the doping concentrations exceed the intrinsic free carrier concentration for the particular temperature where the velocity is measured. Since most of the measurements are carried out in the temperature range below 800C the doping concentrations have to exceed about 10~7-10~8cm-3 in order to observe the doping effect. The experimental results which are depicted in Fig. 52 and summarized in Table 9 show that both for n- and p-doping the activation energy Q decreases in the extrinsic doping range with increasing doping level whereas the stress exponent and the prefactor remain essentially unchanged. Since the effect is independent of the doping element it is evident that it is the change in the Fermi level which causes the enhanced mobility of the dislocations. There are only a few results on the stress dependence of Q but they indicate that it resembles the stress dependence of the undoped crystals so that one can summarize the doping dependence of the dislocation velocity by the relationship
Q = Qo(z) -
AQ(EF)
(149)
where Qo is the activation energy of the undoped crystal. It should be mentioned that impurities can also interact with dislocations via the elastic strain field which has the opposite effects and slows down the dislocation motion. This usually occurs for rather high impurity concentrations, or low velocities when impurities can accumulate at or near the dislocation and form a Cotrell cloud or even precipitates. For instance a Cotrell cloud can develop during crystal growth when impurities are rather mobile and in fact dislocations tend to be "pinned" in the as grown state. A starting stress is necessary then to break the dislocation away from the pinning impurities, but afterwards the dislocation moves with the normal velocity. This pinning effect depends on the thermal history of the crystal and can be avoided if dislocations are introduced from surface or internal dislocation sources at lower temperatures. Pinning effects have also been observed for slow moving dislocations at rather low stresses. Several theoretical models based on atomic models of the dislocation structure have been developed to explain the thermally activated motion and the doping dependence. Dislocations in covalent structures have to overcome the rather high Peierls potential during motion which would require a very high activation energy if the entire dislocation would proceed in a single activated step into the next Peierls valley. The basic assumption is therefore that the dislocation moves by the local formation of double kinks on the dislocation line and the sideways motion of single kinks under the applied stress. The thermally activated steps are thus the double kink formation and the kink migration and the activation energy Q for the
306
PROGRESS
IN MATERIALS
SCIENCE
dislocation velocity is thus a combination of the activation energies for these two processes Hdk and Hkm, respectively. The doping dependence is explained by assuming that either Hdk or Hkm are changed when the Fermi energy shifts towards the band edges. Most of the models assume that in thermal equilibrium point-like defects at the dislocation such as the kinks, jogs or reconstruction defects in the core of the dislocation may introduce energy levels in the band gap. If for instance the concentration of these defects determines the double kink formation rate, the energy for this process can be reduced by the gain in electronic energy when these levels become charged. The energy change for the formation energies can be described by the same formalism which has already been used for the calculation of the formation energies of charged vacancies [eqs (114 and 115)]. For a single donor (charge state E~ ) or acceptor level (E~-) which is connected with the dislocation defects one obtains H~k = H~k - (g~- --
EF) ,
Hdk = H~k + (E~ -- EF)
(150)
where HJ-k and H ~ are the new double kink formation energies when the donor or acceptor level become charged. Similar arguments have also been used to explain a reduction of the migration energy. In any case the change of the double kink formation or kink migration energies translates directly into a change of the activation energy AQ of the dislocation velocity. Although the details of the model for the dislocation velocity are important here (see references) one obtains in most models for either an acceptor or donor level AQ(EF) = (E{ - EF); AQ(EF) = (EF -- E i ) . (151)
Since the charged point-like defects are distributed along the dislocation and lead to a charged dislocation line one has also to include the electrostatic energy between the double kink and the rest of the dislocation line. This contribution is neglected here for simplicity and the reader is referred to the original papers for a full discussion of the doping effect.81) The comparison with experimental results shows that for silicon a donor E~ and an acceptor level E~ have to be assumed to account for the observed decrease of the activation energy in n- and p-type silicon. Although there is still some controversy about the identification of the defects which introduce the trap levels and whether Hak o r akm are reduced the essential features of the thermally activated motion and the doping dependence can be described quantitatively within these models. 4.3.2. D&location multiplication The doping effect is of particular importance for the nucleation of dislocations during growth of low cost solar cell silicon since the crystals are usually doped. The enhanced mobility of the dislocations when the material becomes extrinsic can lead to dislocation multiplication even at lower temperatures. Although so far no attempts have been made to describe the dislocation nucleation during growth of polycrystalline crystals quantitatively, the theoretical background for these processes has been developed and shall be discussed next. In fact the elemental semiconductors and to some extent a few III-V compounds were the first materials whose plastic properties were described quantitatively from a microscopic point of view using the knowledge about the mobility of dislocations [eq. (148)]. The investigation of the plastic deformation of these crystals in combination with the analysis of the dislocation microstructure which develops led to a description of the multiplication processes of dislocations and their mutual interaction with increasing dislocation density.
SEMICONDUCTORS
FOR
SOLAR
CELL
APPLICATIONS
307
At the early stages of deformation when slip takes place predominantly on a single glide plane the average stress ze~ acting upon a dislocation is the macroscopic shear T reduced by a backstress which is due to the presence of other dislocations and depends on their mean distance according to the relationship
Tefr = z - ( A N ) 1/2
(152)
where A is a proportionality factor which can be calculated approximately for a given arrangement of dislocations. With increasing dislocation density the effective stress for dislocation motion decreases thus continuously. For higher stresses when a denser network of dislocations develops other interaction mechanisms begin to operate which shall not be taken into account here for simplicity. During deformation the density of dislocations increases and several mechanisms have been proposed by which dislocations can multiply. They can be summarized by the following multiplication law
dNm
dt
- BNm v
(z~) n
(153)
where B is a constant and n an exponent which depends on the model for the multiplication process. Values for n can be n = 0, 1. The experimental observation shows that in general only a fraction of all dislocations contribute to the shear and the plastic deformation %~ of the crystal. Therefore Nm is the density of mobile dislocations here, but at the early stages of deformation one can assume that Nm = N. The second basic equation for the plastic deformation expresses the total strain rate by the sum of the elastic and plastic strain rates and is given by de 1 dr (154)
dt - G dt + Nmbv
where b is the Burgers vector of the dislocations and G the elastic shear modulus on the glide plane. The second term is the plastic strain rate dEpddt which is related to the mobile dislocation density and velocity by the Orowan relationship. The expressions ~52 ~54~together with expression for the dislocation velocity48~ are the fundamental equations which describe the plastic deformation at the early stages of deformation. For higher deformations when slip occurs on several glide planes additional assumptions have to be made about the deformation process. These equations have been solved for two standard deformation conditions, namely the dynamical deformation where the strain rate is kept constant (dE ~dr = const) and the creep test experiment where the stress is kept constant (z = const) and good experimental agreement has been obtained. The situation during crystal growth is similar to the creep test condition since it is the stress which is controlled by the experimental conditions. The major problem is to calculate the thermal and local stresses in a real growth situation which requires in general sophisticated numerical simulations. Once the thermal stresses have been determined one can apply these equations iteratively and determine the dislocation density as a function of time. Computer simulations of the crystal growth process have been used already to calculate the temperature profiles during single crystal growth, however, realistic simulations of the developing dislocation structure during growth particular in ingot and ribbon grown material are not available yet. There is no doubt, however, that numerical simulations will be used more extensively in the future and make use of the basic knowledge which is available for the dislocation motion and multiplication in covalent crystals.
308
4.4. Grain Boundaries Grain boundaries are the main crystal defects in polycrystalline thin films with grain sizes in the micron range. They also play an important role in polycrystalline bulk and sheet semiconductors where the grain sizes are usually larger (0.1-10 mm). Many of the electrical and optical properties of these materials are determined by the corresponding properties of the grain boundaries. The interactions with free charge carriers, intrinsic point defects such as vacancies and interstitials, and impurities are the most important physical processes that have to be considered. They are closely related to the atomic structure and defect chemistry of the grain boundaries. The structural aspects have been studied experimentally and theoretically in recent years primarily in silicon and germanium whereas the most advanced concepts for electrical transport phenomena at grain boundaries have been developed for compound semiconductors like ZnO or SrTiO3 which are used for varistor devices. Despite the structural and compositional differences between these materials a rather general description of the grain boundary properties has emerged which has been summarized in various reviews and conference proceedings36'182-185) The general picture allows a qualitative and quantitative understanding of the electrical behavior of grain boundaries in a variety of polycrystalline semiconductors although in special cases the particular properties of the material have to be taken into account. It appears that the designers of solar cells using polycrystalline semiconductors have not always taken full advantage of the knowledge about grain boundaries that has been accumulated in the past, partly due to the fact that the common material problems are attacked from the perspective of very different research areas. The basis for an understanding of the behavior of grain boundaries is the description of the atomic structure and chemistry. The structural and chemical disorder determines the electronic states in the forbidden gap which are primarily responsible for the electrical activity of the grain boundaries. Therefore the main experimental and theoretical results on the atomic structure and the distribution of the electronic states will be discussed first for elemental and compound semiconductors. The grain boundary states interact with the free charge carriers in the bulk and two different processes can be distinguished. The permanent trapping of charge carriers leads to a grain boundary charge and the formation of a potential barrier which hampers the flow of free carriers through the grain boundary. The electrical transport properties at single grain boundaries will be discussed in the second part of the chapter. In polycrystalline semiconductors with grain sizes below a few microns which is typical for most polycrystalline thin films the influence of the grain boundaries becomes so dominant that the conductance properties may differ considerably from a single crystal. The electrical transport properties in polycrystalline semiconductors will be reviewed in Chap. 4.6. The electronic states can also enhance the recombination of electrons and holes. Since this process is of particular importance for the photovoltaic properties of semiconductors the recombination behavior has been discussed separately in Chap. 2.4.4. The atomic structure of grain boundaries is also important for the interaction with intrinsic point defects, impurities and dislocations. Since these defects determine the electronic properties of the bulk such as the resistance or diffusion length their interaction with grain boundaries can have a profound impact on the bulk properties of the material. For instance the gettering of undesirable impurities may have a beneficial effect on the electronic properties of the semiconductor. At the same time the electronic properties of the grain boundary can also be changed. Of practical importance is the interaction with hydrogen or other elements which can passivate the grain boundary states to some extent. Though these interaction
309
processes are rather important in polycrystalline semiconductors they are less studied so far and will be discussed in Chap. 4.5. 4.4.1. Atomic structure and electronic states Grain boundaries separate regions of different crystallographic orientation. In general a grain boundary is defined by the orientation relationship between the two adjacent crystals and the orientation of the grain boundary plane. The geometrical relationship between the two crystal lattices can be described by a rotation about a common axis and the rotation angle. A general description of the geometrical aspect of grain boundaries has been given in Bollmann's O-lattice theory080 and recently by Pond. ~t87~For cubic lattices the rotation can be described two dimensionally if one projects the lattice on a plane perpendicular to the rotation axis [hkl] and selects a suitable reference coordinate system given by al = a[k 2 + F, hk, hi], a2 = a[0 / k], and a3 = a[h k l] parallel to the rotation axis. The rotation of a vector, x, of lattice 2 in the plane with respect to lattice 1 can be described by { cos0 - R sin 0 lsin01 cos 0 J
y=Ax
A=
(155)
with the rotation angle 0 and R = (h 2 + k 2 + 12)1/2. In the case of non-cubic lattices such as the chalcopyrite structure for many ternary semiconductors these equations have to be modified. Experimental evidence shows that in general it is necessary to include a relative translation, d, of the two lattices with respect to each other: y = A x + d. This is equivalent to a rotation about an axis which is parallel to the a3-axis but displaced with respect to the origin by
u = (E -
A ) -~ d.
(156)
This equation can be used to determine the translation invariance of the atomic structure of a particular grain boundary and is important if dislocations in grain boundaries are considered. Additional degrees of freedom are required to determine the orientation of the grain boundary plane. Particular orientations occur for twist boundaries where the grain boundary plane is perpendicular to the rotation axis and tilt boundaries with a plane parallel to the rotation axis. In the latter case one degree of freedom remains for the orientation of the plane: a rotation about the a3-axis. Symmetric tilt boundaries occur for orientations where each lattice is rotated by +0/2 with respect to the grain boundary plane (Fig. 53). Geometrically a general grain boundary can be considered to consist of a mixture of a tilt and twist component. In fact the experimental observation shows that in many cases this separation is a useful concept for the description of a random grain boundary.
Angle 0
FIG. 53. Schematic diagram of a tilt grain boundary defining rotation axis, rotation angle 0, and grain boundary plane.
JPMS 35-3/4--H
310
FIG. 54. TEM micrographof a small angle grain boundary in solar cell silicon showinga network of dislocations. (Scale marker ~0.1/~m). The microscopic structure of grain boundaries in semiconductors has been most thoroughly studied by conventional (TEM) [e.g. Ref. (188)] and high resolution transmission electron microscopy (HREM) in germanium and silicon [e.g. Ref. (189, 190)]. For small angular deviations (0 < 5 ) the grain boundary structure consists of an array of dislocations (Fig. 54) whose character and arrangement is determined by the misorientation. For a simple tilt grain boundary a set of edge dislocations is sufficient for the angular deviation. The Burgers vector, b, and the mean distance, d, of the dislocations determine the misorientation angle 0 = The character of particular small angle grain boundaries has been studied in germanium and silicon 88) and confirmed the relationship. For larger misorientation angles the mean distance of the dislocations becomes so small that a description of the grain boundary structure in terms of a dislocation model is not useful anymore. A typical example of a random large angle grain boundary in silicon is given in Fig. 55 which shows an irregular atomic structure. On a macroscopic scale a characteristic feature of random grain boundaries is frequently the occurrence of facets and straight segments of the boundary plane. In general, these facets may be only a few atomic steps wide and only be resolved by HREM. The facets and straight segments lie along particular crystallographic (low-index) directions which indicates a lower grain boundary energy for these particular orientations. It has been shown experimentally that in many cases these low energy boundaries occur for particular orientations where the two crystal lattices form a superlattice: the coincidence site lattice (CSL). (It should be mentioned that in many cases the two crystal lattices are also slightly shifted with respect to each other which actually destroys the CS-lattice, but the concept is nonetheless very useful in terms of a geometrical description of the orientation relationship between the two lattices.) Geometrically, the CSL is characterized by the relationship that some lattice sites are coincident in both lattices which can be described formally by the relationship y - x = t, where t is a translation vector of the reference lattice. Inserting this into eq. (155) yields a set of vectors which form the CS-lattice.
1
1
b/d.
y=Bx
B=~
RctgO/2
1 tgO/2t
(157)
S E M I C O N D U C T O R S FOR SOLAR C E L L A P P L I C A T I O N S
311
FIG. 55. High resolution TEM micrograph of a large angle boundary in polycrystalline solar cell silicon. (Scale marker ~ 50/~).
The particular misorientations for which a CSL occurs can be easily derived then from the equation Rkl 0 = arctan ~ T
~=R2k2+k~
(158)
where k~ and k 2 are integer numbers. The CSL boundaries are usually characterized by the rotation axis and the Y,-value which is a measure of the number of coincidence points within the unit cell of the CSL. (Low values indicate a higher density of coincidence points.) The orientation of the straight segments of grain boundaries planes frequently follows a direction with a high density of coincidence points which suggest a good atomic lattice match in these cases. Consequently the research has focused on low-energy coincidence grain boundaries which can be grown intentionally in bicrystals with a deviation from the coincidence orientation of about A0 < 0.01 . The most thoroughly investigated grain boundaries are coincident tilt grain boundaries in germanium and silicon with a (001), (011), and (111) rotation axis. Some of the CSL grain boundaries such as twin boundaries (Y,3 or 9 (011) tilt boundaries) also occur naturally in polycrystalline semiconductors because of their rather low energies. The directional covalent bonding in semiconductors which crystallize in the diamond cubic, sphalerite or chalcoyprite structure suggests to determine the atomic structure of CSL boundaries by using simple geometrical stick and ball models. Despite the early successes of the modelling to demonstrate that low energy CSL grain boundaries are characterized by a good reconstruction of the covalent bonds across the boundary plane, this geometrical approach is not unambiguous. In fact experimental HREM results have shown that except
312
PROGRESS
IN MATERIALS
SCIENCE
(a)
(b)
F1~. 56. Geometrical models o f the atomic structure o f a ~3 and ~9 grain boundary in the diamond cubic lattice for rotation angles 0 = 70.5 (a) and 0 = 38.9 (b). In the projection on a (011) plane white and black circles indicate atoms on the paper plane and above at a distance a/x/2, respectively. (a, lattice constant).
for the E3 and Z9 grain boundaries (Fig. 56) the real atomic structures differed in most cases from the simple geometrical models. A typical example is the E 11 (011) tilt grain boundary for which a structure with a doubling of the CSL periodicity has been found. 091) In addition the two lattices are shifted relative to each other which obviously lowers the total energy. These translations are usually of the order of one Angstrom. These results show that the geometrical models are of limited use for the determination of the actual atomic structure and relaxations at the grain boundary, and have to be supported and confirmed by experimental method. A general picture that has evolved so far from the investigation of CSL grain boundaries is that the atomic structure can be described in terms of small structural units. A systematic characterization of these building blocks has been given for (001) and (011) tilt grain boundaries in elemental semiconductors.(192-n94) They are characterized for instance for the (011) boundaries by five, six and seven-membered rings of fourfold coordinated atoms (Fig. 57) which can occur in various arrangements in the different structures. A fundamental issue from the perspective of the electrical activity of grain boundaries is the question whether unreconstructed covalent ("dangling") bonds remain in the grain boundary structure which are supposed to introduce electronic levels in the band gap. Geometrical models suggested that atomic structures can occur where a reconstruction of all bonds is not possible. In order to solve the problem computer calculations of the total energy of particular grain boundary structures have been applied. In most instances valence force fields or interatomic potentials(19s-t97) have been used to determine the grain boundary energy for a particular structure. However, it turned out that the accuracy of the calculations is not good enough to distinguish between reconstructed and unreconstructed atomic structures and to reproduce the actual observed structures. Another drawback of these earlier methods is that no insight into the electronic structure of the boundaries is gained. Furthermore, there is no account taken in these models of the change in bonding at the boundary and how that influences the boundary structure and energy. This has prompted several authors to turn to theoretical calculations which are quantum mechanically based. Simple structures such as the incoherent 5.3 tilt grain boundary and a 5.5 (001) twist boundary(19s,199)have been calculated recently and the structural results are in agreement with the experimental observation. An important finding is that the low energy structures are totally reconstructed and no deep electronic levels are introduced in the forbidden gap.
313
(a)
(b)
(c)
(f)
(g)
(h)
FIG. 57. Geometrical models of five-, six- and seven-membered structural units which can occur in grain boundaries with diamond cubic, sphalerite and chalcopyrite lattice3 t2~ Projection on a (011) plane (a-g) and on a perpendicular (0I 1) plane (h). In ABX2 compounds with the chalcopyrite structure the positions of the cations (A, B) are represented by large circles and squares and the anions (X) by small circles. Incorrect bonds are emphasized by bold lines and dotted atoms if connections between widely spaced planes occur. In a binary compound AX with the sphalerite structure the A and B sites are occupied by the same kind of atoms. In the diamond cubic structure all atoms are identical.
Numerical calculations and H R E M investigations so far have been limited to particular misorientations and simple grain boundaries. Therefore an important question remains how the results can be extended to random grain boundaries. As mentioned previously, it can be observed experimentally that a grain boundary tends to facet into segments which have a low energy. Although in most cases the structure of these grain boundaries has not been studied it is generally assumed that the faceted segments can be described by the same structural units which have been derived from the atomic structures of low energy grain boundaries. A different situation may occur at the edges of steps and facets. The atomic structures of these parts have not been analyzed systematically so far although in some cases the incorporation of a dislocation has been observed. Another possibility to maintain a low energy structure is the incorporation of dislocations which can occur for small angular deviations A0 from a low energy misorientation. This is analogous to the formation of a small angle grain boundary. An essential difference in this case is, however, that the possible Burgers vectors are determined from the constraint that the atomic structure of the large angle CSL boundary has to be maintained. The general theory for this situation has been derived in Bollmann's O-lattice theory which shows that the Burgers vectors have to belong to the DCS lattice which can be defined for each CS-lattice. The set of DSC vectors can easily be derived from the translation invariance of the CS-lattice by using eq. (156). Any translation vector u of the CSL yields a possible DSC vector. A typical example of a dislocation network in a large angle grain boundary in silicon is shown in Fig. 58. In general a set of three screw and edge dislocations is necessary to account for the twist and tilt deviations. The Burgers vectors of grain boundary dislocations have been analyzed in a few cases and been shown to belong to the nearest DSC lattice/lss) The Burgers vectors, the mean distance of the dislocations and the angular deviation AO are related by
314
FIG. 58. TEM micrograph of a dislocation network in a large angle grain boundary in polycrystalline silicon. Bright contrast shows twin lamellae bound by coherent E3 boundaries which occur frequently in this material. (Scale marker g0.1 #m).
the same relation as for a small angle grain boundary. The results also indicate that not always the smallest possible Burgers vectors occur, but no general rule seems to exist so far which would allow to predict the actual DCS-vectors. A numerical calculation of the energy of DSC dislocations in a Z9 grain boundary in silicon (~95~has shown that an important contribution to the total energy of grain boundary dislocations is their core energy, which suggests that dislocations are preferred which minimize both the core and strain field energy. The experimental and theoretical results which have been obtained so far show the following common features for large angle grain boundaries in elemental semiconductors. (1) Random large angle grain boundaries tend to facet over a distance of a few nanometers into segments which have a preferential orientation and low energy. (2) Low energy grain boundaries occur for CSL orientations and small angular deviations are compensated by the incorporation of dislocations with Burgers vectors from the corresponding DSC lattice. (3) Low energy grain boundaries and segments consist of small structural units of a few atoms with a considerable amount of atomic and bond relaxation. (4) The atomic structures of low energy boundaries which can be totally reconstructed are favored. The distorted region is only a few angstrom wide and crystallinity is maintained right up to the interface. (5) No deep electronic states are introduced in the band gap by reconstructed grain boundaries.
SEMICONDUCTORS
1315
Experimental investigations of the electrical activity of some low energy CSL boundaries which will be discussed in the next section indeed show that they are electrically inactive which is in agreement with the structural investigation. However, in many cases random grain boundaries show electrical activity which raises the questions about the origin. At the present level of structural characterization the microscopic origin of the electronic origin of the electronic properties of grain boundaries is still far from being understood. Basically two possibilities are feasible. First, the atomic structure of random grain boundaries cannot be described completely by reconstructed structural units. Possible sites for dangling bonds could be in the core of dislocations and at edges of steps and facets. Electron spin resonance (ESR) measurements on small-grained polycrystalline silicon ~2) have shown a spin density of 1012 cm 2 which has been attributed to dangling bonds in the grain boundaries. It should be noted however that other defects near the grain boundary may also be responsible for the dangling bond states. Second, impurities (and intrinsic point defects in non-stoichiometric compound semiconductors) segregate or precipitate at grain boundaries and are responsible for the electrical activity. The experimental difficulties in the determination of the usually very low point defect concentrations in semiconductors and particularly at grain boundaries have prevented a systematic investigation but there is increasing experimental evidence of the influence of point defects on the electrical activity as will be discussed in the last section. In polycrystalline semiconductors the majority of grain boundaries grow randomly and are usually low-symmetry, large-angle boundaries. Depending on the growth conditions preferential orientations of the grains may occur with the possibility of the formation of coincidence (e.g. twins) and near-coincidence grain boundaries. Nonetheless, the problem remains that the numerous results on special grain boundaries so far are of limited use for the understanding of the electrical activity of random grain boundaries in polycrystalline semiconductors. 4.4.1.1. Compound semiconductors. Systematic studies on the structure and electrical properties of grain boundaries in compound semiconductors are not available. Scattered results have been reported for GaAs ~2~,22) GaP, InSb or SiC. The most thoroughly investigated systems are ZnO and SrTiO3 which are important materials for commercial varistor applications. Grain boundary structures in these ceramic semiconductors (or insulators) are largely determined by the interaction with impurities and show particular features such as amorphous interface layers. These phenomena are not observed in other (solar cell) semiconductors, therefore grain boundaries in ceramics shall not be considered here. A review of the electronic properties of grain boundaries in these materials has been given by Greuter and Blatter recently.~6) Almost no experimental results are available for grain boundaries in CulnSe2 or other ternary compounds with chalcopyrite structure which are important for photovoltaic applications. Taking the various experimental observations together several analogies to grain boundaries in elemental semiconductors are evident, so that in a first approach the grain boundaries in Ge and Si may serve as model systems for covalently bonded compound semiconductors. Nonetheless, the atomic structures of grain boundaries in compound semiconductors show features which are particular for these materials, although it is unclear how they influence the electronic and other properties of these semiconductors. In addition to the deformation of the covalent bonds (bending and stretching) at the interface (structural disorder), one has to consider in a compound crystal the different occupation of lattice sites at the interface (occupational disorder). The differences become evident if one analyzes geometrical models of atomic structures of CSL boundaries which have been developed for instance for (011)
316
(a)
(b)
FIG. 59. Geometrical model o f the atomic structure o f a ~19 grain boundary in the sphalerite lattice for rotation angles 0~ = 26.53 (a) and 180 + 0~ (b). In the projection on a (011) plane small and large circles indicate different atoms (A, B). Black and white circles indicate sites at 0 and a/~/2, respectively, above the image plane. Incorrect bonds between A - A and B-B atoms are indicated by bold lines.
tilt boundaries in the sphalerite (26) and chalcopyrite lattice. (12) In stoichiometric AB compounds the A and B atoms occupy the different fcc sublattices of the diamond cubic lattice. The basis for the geometrical modelling is the assumption that the bonding between neighboring atoms is mainly covalent and hence directional. Although the bonding also has an ionic contribution in compounds the possible influence on the grain boundary structure is neglected. A typical example for the atomic model of a El9 boundary with the two rotation angles 01 = 26.5 and 02 = 180 + 01 is given in Fig. 59 in the projection on a (011) plane. The atomic structure consist essentially of the same structural units as for a El9 boundary in silicon for instance, some of which are depicted in Fig. 57. A characteristic feature is, however, the occurrence of a number of incorrect bonds which can be either between A - A or B-B atoms. In the first structure two physically distinguishable forms of the grain boundary structure can exist, whereas in the second case the boundaries are indistinguishable. The same type of wrong bonds occurs in other grain boundary structures as well although their number and distribution varies with the misorientation. It has already been pointed out that atomic structures derived from geometrical models have to be considered with caution. Since these wrong bonds are likely to possess a high energy the grain boundary structures in compound semiconductors may differ considerably from the corresponding structures in elemental semiconductors, Since no experimental and numerical investigations of grain boundary structures are available yet for compound semiconductors these issues remain open so far. Grain boundary structures in ternary semiconductors ABX2 with the chalcopyrite structure have recently been analyzed in a similar w a y . (12) Geometrical models have been developed based on the crystal structure which has been described in Chap. 4.1.1. The differences to the sphalerite structure can be understood by considering the chalcopyrite unit cell (Fig. 56). Because of the doubling of the unit cell compared to the sphalerite structure and the tetragonal distortion along the c-axis the (011) and (012) planes are not equivalent as in cubic lattices and rotations about an axis perpendicular to these planes lead to different grain boundary structures. For a comparison with the previous case a model for a E9 (011) tilt boundary in CuInSe2 is shown in Fig. 60. Four different structures have to be distinguished now which occur for the rotation angles 0~ = 39.7 , 02 = 141.7 and 180 + 0~, 180 + 02. In general, these angles depend on the c/a ratio of the chalcopyrite unit cell. In the diamond cubic lattice the corresponding structures occur for the misorientation angles 0~ = 38.5 , 02= 1 8 0 - 0 1 ,
SEMICONDUCTORS
317
01
02
180 + 01
180 + 02
FIG. 60. Structural unit models for a Y~9grain boundary in the chalcopyrite lattice. Four non-equivalent variants occur for different rotation angles. Half of one period is shown in all cases. Bold lines indicate incorrect cation-cation or anion-anion bonds.
180 + 02, and are all identical. The complete models consist of a period of four equidistant layers and are also characterized by the occurrence of wrong bonds. Because of the two different cations A, B in general three different cation-cation and three anion-anion bonds have to be considered some of which are depicted in Fig. 61. It has been pointed out in 4.1.1 that the incorrect bond sites might be favorable nucleation sites for the formation of intrinsic point defects. This is important for non-stoichiometric compound semiconductors where the lattice structure is maintained over a wider range of compositions by the formation of intrinsic point defects such as vacancies, interstitials and antisite defects. A simple calculation of the formation energies of the twelve possible intrinsic point defects in CulnSe2 (see Table 17) has shown that they are lower if they occur at incorrect bond sites in the grain boundary. Hence in a polycrystalline compound with a high density of grain boundaries and wrong bonds intrinsic point defects may preferentially nucleate at these grain boundary sites instead of in the bulk. Since intrinsic point defects can introduce deep and shallow levels in the band gap their modified occurrence can have an important influence on the electrical behavior of the polycrystalline semiconductor. This is especially important for ternary compounds where some intrinsic point defects introduce shallow levels and the doping level is usually determined by the composition of the crystals. Experimental results mainly for polycrystalline compounds clearly indicate the importance of the grain boundaries for the electrical properties of the materials. However, at the present level of investigations the knowledge about the structural and electrical properties of individual grain boundaries in compounds is still insufficient to understand the behavior of polycrystalline compounds.
(a)
(b)
F]6.61. Tetrahedral coordination of a cation-cation (a) and an anion-anion bond (b) which can occur at grain boundaries in ABX 2 compounds with the chalcopyrite structure. The positions of the cations (A, B) are represented by large circles and squares, and of the anions (X) by small circles.
318
PROGRESS
IN MATERIALS
SCIENCE
4.4.2. Electrical activity of grain boundaries The previous discussion has shown that still little is known about the origin of the electronic levels in the band gap by the presence of grain boundaries in semiconductors. Theoretically the electrical activity of a grain boundary is not clearly defined and is best described by the techniques which are applied to probe the grain boundary behavior. They can roughly be separated into two groups which either probe the recombination behavior of electrons and holes at the grain boundary states and those which determine the grain boundary charge due to the permanent trapping of charge carriers. The first category includes photoand cathodoluminescence spectroscopy, and the electron (EBIC) or light beam (LBIC) induced current method. These techniques yield information locally about the recombination velocity at grain boundaries but it is difficult to extract information about the density of the grain boundary states in the band gap. The methods have been applied to a variety of grain boundaries in silicon and germanium and have shown that low energy grain boundaries with reconstructed structures are generally inactive in agreement with the structural investigations. Electrical activity is more likely to occur for random large angle grain boundaries and after annealing treatments which promotes segregation or precipitation of point defects.(71) While an enhanced recombination of electron-hole pairs at extended lattice defects is a clear sign of electrical activity it does not imply that the imperfection is also charged. In silicon, for example, enhanced grain boundary recombination has been observed whereas the electrical transport properties across the boundary were identical to the single crystal region which indicates a neutral grain boundary and no potential barrier. In the following only charged grain boundaries will be considered. The main experimental techniques which have been applied to investigate the grain boundary states are electrical conductivity measurements across the boundary, admittance and deep level transient (DLTS) spectroscopy3z7) There are also a few conductivity measurements parallel to grain boundaries which indicate that for large contamination with impurities an enhanced conductivity along the boundary can be observed. From the results of these measurements a model for the electronic properties of grain boundaries has emerged that can account for most of the experimental data and allows to determine the intrinsic electronic properties of the grain boundary such as the density of the interface states NB(E). This conceptually simple model has found wide acceptance for the description of transport properties at grain boundaries and in polycrystalline semiconductors06) and shall be discussed here. The basic idea is that the grain boundary is charged when the interface states are occupied and thus represents a repulsive potential for the majority carriers in the bulk. In the bulk a compensating screening charge occurs which corresponds to a band bending and grain boundary potential q~B(x) as shown in Fig. 62. The calculation of the potential barrier depends on the details of the electronic structure at the grain boundary and in the adjacent bulk of the semiconductor. Since the origin of the electronic interface states is not known some assumptions have to be made for the modelling which are however plausible considering the experimental observations. (1) The grain boundary is a plane two-dimensional interface. The electronic states in the band gap thus form a two dimensional band which in general may be distributed over the whole gap and be described by an a priori unknown density of states NB(E). The occupation of the states yields an interface charge Q~.
SEMICONDUCTORS FOR SOLAR CELL APPLICATIONS VB~~-E+ + + E1+ ++

i
319
VB+eU l" ~ ~
+
c
++ + +
eN~
-XL "XLI O
'"
-QS(x)
XR]
X~O
I,x
FIG. 62. Energy band diagram and spatial charge distribution for a potential barrier under bias forming at a grain boundary. Eo, No and El, Nt are the energy levels and concentrations of a shallow and deep level impurity, respectively. The density Na(E) and occupation of the interface states is shown in the insert.
(2) The grains are homogeneously doped with one shallow impurity (for instance a donor with a trap level Ed and a concentration N0) and contain in addition several deep bulk traps (gTi, NTi). (3) The occupation of the traps is described within the depletion approximation. The calculation of the potential barrier ~%(x) from the charge density in the vicinity of the grain boundary requires the solution of Poisson's equation. In the depletion approximation a trap in the band gap is charged between the "screening lengths", Xli and xn, and neutral outside dependent on the position of the trap level Exi and the Quasi-Fermi energy ErB. Assuming for instance an n-type semiconductor with i = 1. . . . . n + 1 bulk traps where the first level (i = 1) refers to the donor level the charge distribution is given by
p(x) = - Q B ~5(x) + e Z NTi { O ( X
"~ Xli ) -- O ( X -- Xri)}
(159)
with the boundary conditions ~oa(x~i) = 0 and p. (x.) = - U . The expression for the barrier height ~Oao= q~B (0) as a function of the applied voltage U has been derived by Greuter and Blatter. (6) If one separates the contribution of the shallow donor level from the sum one obtains
~%o = ~ 1 +
e (No + Y~NT~)
(160)
where the critical voltage is given by Vc = Q~ 8eEoe(Nd + E Nvi) (161)
The potential barrier is positive here in an n-doped and negative in a p-doped semiconductor. The expression for the potential is the generalized form ofeq. (75) in Chap. 2.4.3 when a larger
320
number of deep traps is present. The potential barrier at the interface depends strongly on the grain boundary charge QB which can be calculated from the density of states distribution of the interface NB(E)
QB=e f NB(E)fB(E + VB)dE--eNBn.
(162)
NB, is the electron concentration (per unit area) of the neutral grain boundary, VB= eq~Bo, andfB (E) is the occupation statistics for the interface states. In thermodynamical equilibrium fB(E) is equal to the Fermi-Dirac distribution but has to be determined in a non-equilibrium situation for instance under illumination. It can be shown that under applied bias in the thermionic emission model fB has the form for the Fermi-Dirac statistics with a Quasi-Fermi level, EvB, for the grain boundary states given by
Eva = EF- KT In
1+
exp(-eU/KT)"
(163)
(A slightly different expression has been derived by various authors ~22)using the diffusion approximation.) Similar to the calculation for the surface potential in Chap. 2.4.3 the grain boundary potential has to be determined self-consistently from eqs (160-163). Figure 63 shows calculated barriers as a function of the applied voltage for a single level, a Gaussian and rectangular density of states distribution/6) The dominating term in eq. (160) is the first one which depends strongly on the grain boundary charge QB and the bias U. As the bias increases the barrier height decreases. The increase of the boundary charge with increased voltage counteracts and slows down the decrease of the barrier, thus for small voltages the potential barrier is pinned by the grain boundary states. A highly localized density of states can thus stabilize the barrier for moderate voltages. The barrier breaks down rapidly when all grain boundary states are filled and QB essentially remains constant. The presence of deep bulk defeqts generally decreases the barrier height because of the additional screening. For typical values for a deep trap with a concentration of NT = 1017era-3 and NT/Nd= 0.1 the barrier can be reduced by as much as 0.1 eV which is large compared
1.8
I I I ' I
Q~
.m
>1,o ....
>
z. V (V/Ge)
10
FIG. 63. Calculation of the barrier height and interface charge as a function o f the applied voltage for a single level, a Gaussian and a rectangular density o f states distribution. The influence of the presence of a deep level defect is shown for the Gaussian distribution (dashed-dotted line)/2~4)
321
to K T at room temperature. A second effect is that with increased voltage more ionized bulk traps are shifted below the Fermi level and do not contribute anymore to the screening. A deep trap level can become completely neutral if the potential q~Bo (U) ~ ETi -- EFB which leads to a transient stabilization of the barrier with increasing voltage, as can be seen in Fig. 63 by the shoulders in q~Bo(U). 4.4.2.1. Time-independent electrical transport (current-voltage characteristics). The barrier height VB determines the current flow of majority carriers through the grain boundary. Assuming steady-state conditions the total current density, JDC, for majority carriers (e.g. electrons) is determined by the current from the left and right hand side into the barrier, JL and JR, the electron current into the grain boundary states, JTL and JTR, and the current emitted from these states into the conduction band, 2J E. For symmetric emission into the left and right grain the total current on each side J1 and J2, respectively, is given by ~28~
J t = J L - - J R + J+R--JE; J 2 = J L - - J R - - J T L + JE .
(164)
The trap and emission currents determine an interface current JB into the grain boundary states JB = JTL + JTR -- 2JE. (165)
JB is related to the previously calculated recombination rate Rsn (see Chap 2.4.4) for the grain boundary by the relationship JB = eva Rsn. It has been shown there that without illumination or other hole generation processes no electron-hole recombination occurs at the interface therefore the interface current is zero and the contributions of minority carriers to the total current through the barrier can be neglected here. It should be noted that under non-steady state conditions the interface current JB becomes non-zero and is accompanied by a change in the grain boundary charge and barrier. This situation will be discussed below in more detail. The total dc-current JDC = J l = J2 can be calculated using either one of the eqs (164). In most cases the barriers are so wide that tunnelling of carriers through the potential does not contribute significantly to the total current. Smaller barriers can however occur for high doping levels or in polycrystalline materials as will be discussed in Chap. 4.6 and tunnelling currents have occasionally been included in the calculations. Generally it is assumed that the currents can be calculated within the thermionic emission modeF28) and the models yield for Joc
JDc=enov.(1 --c)exp --~--~
VB)(
eU) 1--exp ~--~
(166)
where no is the electron concentration which is determined by the doping in the vicinity of the grain boundary N d = n o = N c e x p { - ( E c - E v ) / K T } . The prefactor is also frequently expressed by the effective Richardson constant A* with e Ncvn = A T 2. The constant c is the capture probability for electrons at the interface and is given by
C =~
1 fe
__v+V
S.(E)NB(E) {1 --fB(E)} dE
(167)
fB(E) is the Fermi-Dirac distribution with the Quasi-Fermi level EvB for the grain boundary states given in eq. (163). Usually it is assumed that the capture cross section S,(E) is constant so that c becomes proportional to the number of empty grain boundary states, NB, c = S. NB/2. Typical values for the capture cross section a r e Sn ~ 10-16-10-~4cm-2 and for the density of states NB ~ 10~2-10~3cm -2 so that the factor (1 - c) ~ 1.
322 E
I <
==
r<
E tO
~9
ta-a
,It
1 B- 4
+<>
X O
+0o -101
-2 -1
1 I] -6
i..,I...,I .... I,,,,I.,,,f
E.5
3.5 I/T
4.5
[K^-1.10,,-3]
0 10g(Vm=(V/groin))
FIG, 64. Zero-bias conductivity crs as a function of the temperature for a near-coincidence Z9 grain boundary (0 = 35 ) in n-type germanium. The different symbols represent doping concentrations between 2.3 x 1014 to 5.8 x 1014cm-3. (29)
FIG. 65. Calculated I - V characteristic for a grain boundary barrier. The density o f states distributions are the same as in Fig. 63. Deep bulk defects in the space charge region of the grain boundary cause a strong increase in the leakage current. (6)
For small voltages e U <<KT the potential barrier is almost constant and the current density JDC linearly dependent on the voltage. Therefore one can define the zero-bias conductance, trB, for the grain boundary by trB = JDc/U which is given then by o" = ~e2 noVn(1 -- c)exp ( V B ) a --~-~ (168)
aB as defined here has the dimension f~-] cm -2 and is exponentially dependent on the grain boundary potential. Temperature dependent measurements of the conductance have been used to determine the potential barrier. A typical conductance measurement for a grain boundary in germanium is given in Fig. 64. (29) A grain boundary potential of about VB = 0.3 eV has been obtained from these measurements. Under the assumption that the grain boundary states are localized around a single level Ex and no deep traps near the interface are present one can use the eqs (160-163) to determine ET from the potential barrier. Values of about ET ~ 0.1 eV above the valence band have been obtained for germanium. For larger voltages eU>>KT the current density Joc depends critically on the voltage dependence of the grain boundary barrier liB. Since Va always decays rapidly at higher voltages when all traps are filled the current shows a sharp rise above a certain bias as can be seen from the calculated curves in Fig. 65. The steepest increases occur for single level boundary states. Deep traps in the space charge region tend to smooth the decay of the barrier and hence the increase of the current. It should be mentioned that with increasing decay of the barrier at high voltages the finite conductivity of the grains has to be taken into account which can be seen for U > 4 V. The current-voltage characteristics is critically dependent on
SEMICONDUCTORS
323
the electronic structure of the grain boundary and hence can vary from sample to sample. The developed model describes most of the aspects of the electronic properties of grain boundaries in a variety of materials correctly. However, a few cases in compound semiconductors are known, where an additional mechanism is needed for describing the breakdown behavior. (~6) In the technological important material ZnO it is observed that the switching between low and high resistance proceeds much faster than predicted from the above calculations. A consistent explanation of this and other observations can be given when minority carriers are taken into account. For larger applied voltages and an optimum doping range electrons which are injected into the positively biased grain can pick up enough kinetic energy, so that minority carriers can be generated by impact ionization. The holes can drift-diffuse back to the barrier and take part in the screening of the interface charge. These additionally generated holes can also lead to an electron-hole recombination at the interface state which is zero otherwise. The calculations for this situation show that this process lowers the potential barrier faster and leads to an accelerated increase of the current. The previous discussion has shown that current-voltage measurements can be used to determine electronic properties of grain boundaries. The current-voltage method has the disadvantage, however, that it is often not sensitive enough to study all details of the grain boundary states such as the density of states or the capture cross sections. One reason is that the parameters which are actually measured like for instance the grain boundary potential depend only on an integral containing the density of states NB(E) [see eqs (162) for the grain boundary charge]. One way to overcome the difficulty is to probe the interface states by changing the Fermi level. The basis for this deconvolution procedure is a relationship which can be derived from this equation (for constant temperature)
OEFB = e f v+vc NB(E) ~ B(E) dE ~ eNB (EFB). Of OQr~
(169)
Measurements of the grain boundary as a function of the Fermi level which can be charged by doping or an applied voltage can thus provide information about the density of states distribution. Results by this deconvolution technique have been obtained for grain boundaries in silicon (2~), GaAs (211) or ZnO. 4.4.2.2. Time-dependent electrical transport (admittance spectroscopy). Time-dependent experimental methods such as admittance and deep level noise spectroscopy (DLNS) (2)2) or DLTS offer further possibilities to study the emission and capture process of electrons or holes and the distribution of the interface states and therefore have been increasingly applied during recent years. In dynamical measurements capture and emission processes of electrons or holes at the grain boundary states are stimulated. Several models have been developed and are based on the double-Schottky barrier theory which has been discussed above for the steady-state situation. A realistic description of the dynamical electrical activity of a grain boundary has to calculate the rearrangement of electrons and holes at the different trap levels. The corresponding charges which have to be taken into account are the grain boundary charge QB, the screening charge of shallow and deep level traps on either side of the interface QT~ = QRi + QLi, and eventually the minority carrier charge trapped at the interface. A particular stable method is the admittance spectroscopy which has been successfully applied to grain boundaries (213) and shall be discussed briefly here. In this technique the response of the system to an applied time-dependent voltage U(t) = Uo + U~ exp(iogt) is
324
Ec E
F1G. 66. Representation of the different trapping and emission processes occurring at a charged grain boundary. For the interface states an electron and hole trap is assumed and represented by their density of states distribution.
measured, where the modulation eU~<<KTis small. The resulting current over the barrier, JAC, is in general phase-shifted with respect to this applied bias, thus JAC(t) = Jo + J1 exp(icot) where Jz is a complex quantity. Usually the complex admittance is measured as a function of the frequency co, the bias Uo, or the temperature T. It can be separated into the real and imaginary part a(co) = J~/Ut = G(co) + icoC(co)where the conductance G(co) and capacitance C(co) contain the parameters of interest for the deep bulk and interface traps. The current and hence the admittance have been calculated within the linear response approximation. The calculations involve the description of the capture and emission process of carriers in the traps which are characterized by typical time constants (Fig. 66)
V.o KT Za=S, noV, (_eUo) 1 -- exp KT]

1 exp for the grain boundary states and
"l~Ti = 6TigTi no Vn e x p
(170)
(171)
for the deep bulk traps. VBo is the equilibrium interface potential, aTi the capture cross section of the traps and Gzi the degeneracy factor which is 0.5 for a simple donor. The time constants for the shallow traps can be neglected since they respond much faster compared to the deep trap levels. If minority carriers are trapped which occurs in the high field regime as has been mentioned above an additional time constant for the electron-hole recombination is necessary. It is evident from the previous discussion that the time-delayed rearrangement of the charges modulates the barrier height VB(t) = V,o + VB1exp(icot) and the interface charge QB, which have to be calculated self-consistently. The time dependence of QB determines an interface current ors = OQB/Otwhich was zero in the stationary conditions and is approximately given (for eUo>>KT) now by JB =/co
CBCg
C R + C L --~ C B
U 1 exp(icot)
(172)
SEMICONDUCTORS
FOR SOLAR
CELL
APPLICATIONS
325
where CL and CR are the high frequency capacitances of the left- and right-hand space charge. They are related to the total capacitance of the barrier by CHF = (C R C L ) / ( C a --[- CL). CB((D ) is the interface capacitance which is given by CB(m) = - S. K T 1 + icozB
c e2 1
(173)
where c/S. is given by eq. (167) and equal to the number of empty interface states. The modulation of the barrier is out-of-phase with the applied voltage and acts in two ways on the system, it modulates the over-barrier current, as can be seen from eq. (166), and leads to a displacement current JD because of the corresponding change of the screening charge QTi = --QB- The interface current acts similar to the base current in a transistor why the model is described as the trap transistor model. The calculation of the displacement current for instance for the left-hand side JoE yields the following appropriate expression
JDL =
ico
CR C L
CR + CL + CB
U~ exp(icot).
(174)
Equations (172-174) show that for a small interface capacitance the interface current is small and the displacement current determined by the high frequency capacitance of the space charge, whereas for a large interface capacitance the situation is reversed. The total current JAC can be calculated on either side of the barrier from eq. (164) including the displacement and interface current JAC = Jl + Ja + JDL- The linear expansion of this equation and the self-consistent solution of the barrier height from the coupled set of equations allows to determine the admittance and finally the expressions for the frequency dependent parts: the capacitance C(oJ) and conductance G(co). The complete expression in the general case is rather complicated and a detailed discussion has been given in Ref. (214). For simplicity only two special cases shall be discussed here. In the first case a single deep trap level is assumed with a low concentration so that the screening charge is mainly provided by the shallow impurity. For zero bias Uo = 0 the potential barrier becomes symmetric and the conductance and capacitance are approximately given by Go(CO) = orB+ CHFO) 2
1 + ((/)~'TI) 2
nl "CTI
(175)
Co(~) = CHF 1 -~ 1 +~TT1) 2
(176)
with n~ ~ NT~/Nd, ~T~ the relaxation time for the trap given in eq. (171), and aB the zero-bias conductance from eq. (168). Whereas for nl<< 1 the capacitance is essentially constant over the entire frequency range, the conductance is more sensitive to the presence of deep bulk trap levels and will show a resonance at ~O~T~= 1. The location and height of the resonance can be used to determine the trap level ETI, the defect concentration NTI, and the capture cross section eT~. Essentially the same result is obtained if more than one trap is present. Numerical calculations for the case of several bulk levels yield that each trap can show up in the conductance as a pronounced resonance and can be analyzed separately. In the second case a finite bias Uo > KT/e is considered and deep trap states are neglected. The total admittance is approximately (for c <<1) given now by CR a(O)) = (~rB+ i~OCL) CR + CL + CB(co)"
JPMS35-3/,~-1
(177)
326
P R O G R E S S IN MATERIALS SCIENCE
The separation of tr (to) into real and imaginary parts yields the conductance and capacitance. The characteristic time constant zB in eq. (173) for the interface capacitance CB(to) is usaUy large compared with conventional measuring frequencies so that for a small bias the total capacitance C(to) is mainly dominated by the space charge capacitance and equal to the high frequency capacitance CHF. With increasing bias the barrier decreases rapidly, however, and the time constant becomes comparable to the experimental frequency range so that a strong interface component shows up in the total capacitance. This anomalous behavior due to the trapping of majority carriers at the grain boundary has first been studied at grain boundaries in silicon (213~and is observed in other semiconductors such as GaAs, as well (Fig. 67). It has been mentioned above that in the high field regime minority carriers can be generated and the hole dynamic has to be included in the calculation. This adds a large negative contribution to the junction capacitance (inductive behavior) for large biases. Such a capacitance behavior has been observed at grain boundaries in GaAs bicrystals (Fig. 67) and Z n O . (21) Negative capacitances seem to be unique, however, to compound semiconductors, since they have never been observed for germanium and silicon. The interface resonance also shows up in the conductance G(to). While the positions of the deep trap resonances do not depend on the bias (eq. (171) for the deep trap relaxation
10~ Hz 103 Hz
~ C ~
20
T : 281 K
~ -20
-601
0
I
1
I
2 Bias voltage (V)
I
3
FIG. 67. Capacitance C(V, ~) for a grain boundary in a n n-type GaAs bicrystal for two different frequencies, The time-delayed modulation of the barrier height VB by electrons or holes trapped at the interface is the reason for the large positive or negative capacitance values. The negative capacitance for voltages V > 2.9 eV indicates inductive behavior.~6~
, I , i ' y / I
327
10-'
~ 1 0 "~' 10-6 w:103 S-, ~
vo4v/ /= l V //Vo
/ J // u=lO 2 s -~
IO-B
10-l I~ I
200
,
l
~00
I
600
T (K}
FIG. 68. Calculated temperature dependence of the grain boundary conductance G(~o, T) for different bias voltages. The deep traps show frequency-dependent resonances at low temperatures36)
time), the resonance due to the grain boundary trapping shifts to higher frequencies with increasing bias. The bias dependence can thus be used to distinguish between the two processes. Instead of changing the frequency one can also vary the temperature and hence the relaxation times. A numerical calculation of the conductance as a function of the temperature is given in Fig. 68 and shows several peaks due to deep trap relaxation whenever the resonance condition e~Tri= 1 is met. From the shift of the peak positions for different frequencies one can determine the trap levels ETa, the defect concentrations NTi, and the capture cross sections aTi. The advantage of the admittance technique is that it provides a direct method to identify deep traps near the grain boundary and the interface states itself. Due to a limited range of frequencies and temperatures the method is usually limited to trap levels with ETi < 0.6 eW. Since in the earlier measurements little attention has been given to the ionized deep bulk traps near the interface this method (as well as DLTS) has been a very useful tool to study the electrical properties of grain boundaries and its local environment. 4.4.2.3. Grain boundary states. The most comprehensive electrical measurements on individual grain boundaries (usually in bicrystals) are available for germanium, (73'74'29) silicon, (2~3)and GaAs321) The results on grain boundaries in germanium and GaAs indicate that the interface states are rather localized. Electrically active <011> tilt grain boundaries in germanium analyzed by the I-V deconvolution technique show a trap level in the lower half of the band gap at about EB ~ Ev + 0.1 eV, whereas the well reconstructed twins (E3 and Z9) are electrically inactive. In n-doped GaAs <011> tilt grain boundaries with a variety of misorientation angles have been studied by DLTS and I-V analyses. Independent of the orientation two grain boundary trap levels at about E m ~ E c -- 0.65 eV and Ea2 ~ Ec - 0.8 eV have been observed, whereas the density of states was reported to depend on the misorientation angle particular for the 0.65 eV level. In addition to the grain boundary states deep trap levels in the vicinity of the boundary have been observed, too. The most comprehensive measurements are available for polycrystalline and bicrystal silicon, however, a consistent description of the grain boundary states has not been obtained yet. Since measurements of different authors are usually carried out at different grain boundaries and materials a comparison of data is in general difficult. Localized grain
328
PROGRESS
IN MATERIALS
, , , ,
SCIENCE
, , ,
1013
a
IE U "7 > h~ v 1011 " l
E. le
tb
~ .
.
~cm . ~ " a " * * ~ ' ~

~Oo
,~ ~
IZ
O. 3 5
O. 4 5
O. 5 5 EB x ( E V )
O. 6 5
O. 7 5
FIG. 69. E x p e r i m e n t a l d e n s i t y o f s t a t e s d i s t r i b u t i o n f o r a single g r a i n b o u n d a r y in F z - s i l i c o n b i c r y s t a l w i t h d o p i n g c o n c e n t r a t i o n s o f 5 1012 c m -3 (a, b), 1.4 x I0 t4 c m -3 (c), 1.4 x 10 t5 c m -3 (d). S p e c i m e n s c a n d d w e r e a n n e a l e d a t 7 3 0 C f o r I hr. (73)
boundary states have been observed in n-type Fz and Cz bicrystals of particular misorientations by DLTS and I-V-deconvolution at EB~ ~ Ev + 0.35 eV, EB2 ~ Ec - 0.29 eV or EB2 ~ Ec - 0.42 eV. In addition most measurements show a rather broad distribution of states in the middle of the band gap as can be seen in Fig. 69. The values for the density of states near midgap vary between NB~ 10 - H - 10-~3cm-2eV-L The results are compared with measurements on polycrystalline silicon (see Chap. 5) which reflect an average over a large number of grain boundaries of various orientations. A rather localized state appears in the lower half of the band gap and agrees with results on particular bicrystals. The second maximum at EB2 ~ Ec - 0.3 eV may be similar to the localized states observed by some DLTS measurements but has not been seen by measurements at other grain boundaries. The occupation limit of the neutral grain boundary is usually found to be at the midgap EB0 ~ Ev + 0.55 eV. Therefore the charging of the grain boundary states is almost the same in n- and p-doped crystals. Some authors claim that grain boundaries in silicon introduce band tails at the band edges similarly to amorphous silicon or MOS interfaces. (2~5) These results have been obtained from polycrystalline silicon and the investigation of one particular grain boundary. The measurements are however contradictory to results from other authors (2~6) so that this issue cannot be considered as settled yet. The quantitative analyses of admittance measurements at grain boundaries also permit the study of the spatial distribution of interface states. (~'3)The frequency dependence of G(co) can be compared with the model calculations and gives a distribution of relaxation times of the interface states. Results have been obtained for grain boundaries in silicon and could be explained by assuming fluctuations of the grain boundary potential due to an inhomogeneous distribution of the interface charge in the boundary plane. An additional TEM analysis of the grain boundary structure in one casd 2~7) suggested that the spatial distribution of the charge correlated with microstructural features such as dislocations or steps. Recent DLTS results on copper decorated grain boundaries (7~)in silicon have also been explained with an inhomogeneous current transport through the boundary. It is interesting to note that the localized states which are observed for the grain boundary states are also frequently found for other lattice defects such as dislocations in germanium or for heterostructures and surfaces in GaAs. Whereas in elemental semiconductors this coincidence is not understood, in compound semiconductors this may support the "unified defect model" proposed by Spicer e t al. (218)which assumes that the pinning states are point
SEMICONDUCTORS
329
defects of the host lattice rather than structural imperfections at the grain boundary. Candidates for the pinning states are anion or cation vacancies, antisite defects or impurities. From the perspective of the structure of grain boundaries in compound semiconductors which has been discussed in the previous adapter it may seem reasonable to assume that the wrong bond sites which attract point defects play an important role for the electrical activity of the grain boundaries in these materials. The analogy between the electronic states at grain boundaries, heterostructures and surfaces suggests to generalize the results to all compound semiconductors and to use the extensive knowledge from surface and heterostructure investigations to predict the positions of grain boundary states in polycrystalline samples or bicrystals. Since an important class of photovoltaic materials are polycrystalline compound semiconductors such as the copper-ternaries for which almost no electrical properties of the grain boundaries are known this generalization may be of particular importance. One can summarize these results which characterize the electrical activity of grain boundaries in compound semiconductors. The grain boundaries show (1) narrow localized states at energy levels typical for the lattice; (2) a density of states which depends on the processing and is affected by lattice point defects; (3) deep bulk defects near the grain boundary which can reach rather high concentrations and thus contribute to the electrical activity. Finally it should be noted that many of the basic concepts which describe the electrical activity of grain boundaries, like for instance the charge trapping and recombination behavior, are also applicable to interfaces, surfaces and dislocations. The different crystallographic nature of these extended lattice defects adds of course particular properties which also distinguishes them from grain boundaries. Since the presentation and discussion of the advances in these research fields would exceed the frame of the review here, the reader is referred to recent review papers on this subject.
4.5. Interaction of Impurities with Grain Boundaries Experimental results on the electrical activity of grain boundaries and other lattice defects such as dislocations or the surfaces show that the segregation of impurities plays an important role. Unlike semiconductors used for microelectronic devices where defect-free, high purity material is used solar cell semiconductors are usually less defect free and contain higher concentrations of impurities because of the constraints of low production costs. The impurities which are of main interest in silicon are the doping elements (mainly B, A1, P), carbon and oxygen which can occur in rather high concentrations, and fast diffusing transition elements (for instance Cu, Fe, Ti) which introduce deep trap levels and are thus electrically very active. In compound semiconductors one has to consider in addition intrinsic point defects such as antisite defects which can act as shallow and deep level traps. Of particular importance are also impurities like hydrogen which can passivate the electrical activity of lattice defects to some extent. Since hydrogen passivation is a standard process in the solar cell technology it will be discussed separately. The interaction of the impurities with lattice defects affects the electrical properties of the semiconductor in several ways. First, it changes the distribution of the impurities in the crystal. This can be beneficial if detrimental impurities are removed from the electrically active regions of the device ("gettering"). Typical examples are the intrinsic gettering at oxygen
330
precipitates in silicon or back surface gettering in microelectronic devices. On the other hand the electrical activity of the lattice defect itself can be changed which may be detrimental for the performance of the solar cell if for instance the recombination velocity of interfaces and the surface is increased. The utilization of gettering techniques in solar cell devices is thus limited since all regions of the semiconductor (bulk and surface) contribute to the performance of the cell. Nonetheless it has been recognized that the investigation of the interaction of lattice defects with impurities is important from the perspective of the optimization of the solar cell performance. Although it has been frequently observed qualitatively that impurities interact with other lattice defects and change their electrical behavior(7~)little is known quantitatively so far about these processes. Since most experimental results in semiconductors are available for grain boundaries in silicon the discussion of the fundamental mechanism will be mostly related to this material. Grain boundaries also affect the diffusion behavior of impurities since they offer a favorable diffusion path along the boundary plane which enhances the diffusion coefficient. This aspect is important since diffusion processes are standard technological steps in the solar cell fabrication and the control of the diffusion profiles are essential. Grain boundary diffusion will be discussed in the second part of this chapter. 4.5.1. Segregation and precipitation 4.5.1.1. Segregation. An inhomogeneous distribution of impurities in the solid due to segregation is the result of a difference of the impurity solubility for instance between the solid C~ and the liquid CL, or between the solid and the interface CG. Segregation during solidification is described by the segregation coefficient /Co= C~/CL which is more or less correlated with the solid solubility of the impurity. For ko < 1 the solid is purified and the melt enriched with impurities. Typically in silicon transition metals have very low segregation coefficients and can efficiently be removed from the solid, whereas doping elements and carbon and oxygen are less affected during solidification. For interfaces a thermodynamical description of the impurity segregation has been given in several articles. (219'22)The basic idea is that the interface can be considered a separate two dimensional system for which thermodynamical variables analogous to the bulk can be defined. For several impurities (i = 1 to n) these variables are in particular the atomic concentrations in the interface Cci, the chemical potentials/~oi, and the interaction coefficients ~ij which are related to the enthalpy of mixing by AHr~ = E ~ijCci C~j. The corresponding variables in the bulk are C~ a n d / ~ . Although in semiconductors the bulk concentrations are generally low the concentrations in the interface may become quite high, therefore the interaction of impurities in the grain boundary has to be considered. The coefficients ~ij are positive for an attractive and negative for a repulsive interaction. The enrichment of an impurity, i, at the grain boundary is then given by C,i exp - coi =
AGi
l+
f A6j ) Z C,i ~exp ~ - 1
KT
(178)
where the free enthalpies are AGi = AGioq- AGiehemand AGio=/Zc, - #~i +/~Go -/~,o (i = 0 for the solvent atom). AGio accounts for the change in elastic and vibrational energy of an
SEMICONDUCTORS
FOR SOLAR
CELL
APPLICATIONS
331
impurity in the interface. In the case of one impurity (i = 0) and a low bulk solubility (C,~ << 1) which is typical for most impurities in semiconductors one obtains (index omitted) AG C~ exp KT CG = I+C~ where the free enthalpy has the simple form AG --- AGo + ~0 (1 - 2Co) (180t exp ~ - ~ - 1 (179)
~0 is the interaction coefficient between the impurity and solvent atoms. Similar to the solid-liquid segregation a segregation ratio ko = Co/C, can be defined. At a given temperature the interface concentration Co increases with the bulk concentrations Ca and reaches a saturation value C~, the maximum solubility in the interface. The saturation value may not be reached if the bulk solubility is low. In this case the maximum solubility limit at the interface becomes temperature dependent. For attractive segregation (AG > 0) the enrichment at the interface, k~, increases with decreasing temperature, and increasing the temperature leads to impurity desegregation. In the case of two impurities (i = 1, 2) where both impurities can occupy the same atomic interface site the expressions for AG~ assume the following form AGI = AGlo + 0q0 (1
- - 2 C G I ) -Jr- (0~12 - - 0~20 - - ~t0) GG2 (~12 - - ~20 - - (~10) C G I "
(181a) (181b)
AG2 = AG2o + ~20 (1 - 2C02) +
For segregation on different sites without competition slightly different expressions are obtained. These equations for a ternary system describe the influence of a second impurity on the segregation behavior of impurity 1 (cosegregation). An attractive interaction between the impurities increases the segregation of impurity l, but it can also lower the maximum interface solubility. Cosegregation phenomena have been obsex:ved in metal alloy systems. 4.5.1.2. Precipitation. The thermodynamical description of the interface as a two-dimensional phase also includes that the interface has a limited solubility C~ for a particular impurity. Precipitation at the interface will occur in thermodynamical equilibrium if the solubility limit C~ for a given temperature is exceeded. Since according to eq. (179) the interface concentration is a function of the bulk concentration the solubility limit is reached for a particular bulk concentration C max. The interface solubility is also exponentially dependent on the temperatue. If C m'x < C eq for a given temperature these two values define a concentration range for which precipitation occurs at the interface but the impurity is still dissolved in the bulk. An example may be the formulation of oxide micro-precipitates in the cores of dislocations of small angle grain boundaries which has been observed in silicon and germanium. ~2221In the second case (C max> C eq) if the maximum solubility in the interface has not been reached precipitation may already occur in the bulk if the concentrations exceed the solubility limit whereas the impurities are still dissolved in the interface. In semiconductors the solubilities of most impurities are rather limited and strongly temperature dependent as has been discussed for instance for the transition metals [eq. (125), Chap. 4.1.3]. Therefore it is likely that the solubilities limits both in the bulk and the interface are exceeded during cooling and precipitation will occur in the bulk and at the interface. Experimental situations are usually characterized by a thermodynamical non-equilibrium,
332
therefore it is more likely that the precipitation behavior is determined by the kinetics of the segregation and precipitation processes rather than by considering the equilibrium situation. 4.5.1.3. Segregation and precipitation kinetics. The thermodynamical equilibrium which has been described so far may not be reached in experimental situations, since atomic diffusion or other thermally activated processes are involved to reach the equilibrium. Several diffusion models have been developed to describe the segregation kinetics.(223) In the simple case of a single impurity in a binary system with no interaction the time dependence of the interface concentration C~(t) is given by CG(oo)- Co(O)
Cc(t) - C~(0) = 1 -- exp(s 2) erfc(s); s =
2(D,
t)l/2C~
(182)
dCG(oo)
where d is the width of the interface region and D~ the diffusion coefficient of the impurity in the volume. Since in semiconductors the amount of segregated impurities at grain boundaries is probably less than an atomic monolayer in most cases one can assume that the interface width is of the order of the lattice constant. For short times and CB(O)= 0 the equation reduces to
CG(t) = 4 ~
--
(183)
More general models have been proposed, (224) taking into account the interactions between different atomic impurities in a binary or ternary system and the different values for the diffusion coefficients of the various species. The segregation kinetics become rather complex then and may lead to an accelerated or retarded segregation. The segregation kinetics change when the solubility limits in the bulk (C~, or at the q) interface (G~) are exceeded and precipitation occurs. Once the precipitates are nucleated the growth of the second phase particles (composition Ca) can be controlled by several rate limiting steps. In many cases it is assumed that the growth is limited by the atomic diffusion of the impurity to the precipitate. In general, one also has to consider the volume change between the particle and the matrix which may require the transport of vacancies or self-interstitials towards or away from the precipitate. The particle growth may also be controlled by a surface reaction rate when the impurity atoms at the particle-matrix interface must surmount a barrier in order to enter the precipitate. A number of theories have been developed to account for the different situations. The main results of the diffusion-limited growth for simple geometrical situations shall be mentioned here for a supersaturated solution with concentration C > C~q. A supersaturation can occur for instance by quenching a specimen or if the nucleation of the second phase during cooling is hampered by a lack of nucleation sites. Solutions of the diffusion problem have been derived by Ham (t35) under the assumption that the nucleation is not rate limiting and that a periodic array of spherical particles has formed in the bulk. If the radius r(t) is small compared to the mean distance rm the time dependence of the particle radius and the maximum radius are approximately given by
r(t)=(
rma mr--C7j x=r
(184)
where AC~ = C - C ~ . The first equation is valid for short times where the bulk concentration C~ remains essentially constant and the critical radius of the nucleated particle rc = r(0)
333
is negligible. The initial growth follows a t 1/2law and is determined by the diffusion coefficient of the impurity and the supersaturation AC,. The expressions for the time dependence become more complex for longer times and particles with non-spherical shapes. The equations are also applicable for precipitation at interfaces if the bulk diffusion is rate limiting and interface diffusion can be neglected. More refined theories have been developed for situations where the bulk concentration between precipitates decreases or smaller particles disappear at the expense of larger ones (Oswald ripening)3225-227) In semiconductors with low impurity concentrations compared to metals and large distances between precipitates these models may not be required to describe the precipitation kinetics. It has been discussed in Chap. 4.1.3 that the diffusion of transition elements in general may require the participation of native point defects. Since the equilibrium concentrations of vacancies and self-interstitials can be disturbed significantly during the precipitation one cannot rule out that the growth rate of precipitates is controlled by the transport of these intrinsic defects. In order to estimate the influence of the intrinsic defect transport on the growth rate the following simple model is assumed. A supersaturated solid solution with a homogeneous distribution of the impurity concentration C = C + C > C~- where C~q is the q, solubility limit for the impurity, (the subscripts, s and i, denote substitutional and interstitial sites for the impurity). At high temperatures the interstitial impurity concentration shall exceed the substitutional component Ci >>Cs which is typical for most transition elements. For the initial growth phase (short times) the total concentration remains constant therefore one can assume that C i ~ C =constant. During precipitate formation self-interstitials are generated and injected into the lattice with a generation rate Kc. In addition it is assumed that in the bulk self-interstitials disappear at sinks such as dislocation (or grain boundaries) with a rate K~.K~ and K~ [see also eq. (132)] are given by
K G = ] : v N p D I C e a q', K 1 = 71NDDIC~ q .
(185)
7v is a geometry factor, and Np the density of precipitates with Np = 1/r 3 where rm is the mean distance of the precipitates. The generation of self-interstitials at the precipitate is proportional to the number of impurity atoms which enter the precipitate, thus proportional to the supersaturation AC, = C - C ~ q. It is assumed now that the impurity diffuses via the kick-out mechanism so that one can use the eqs (129-131). The results are the same if one assumes the dissociative mechanism and applies the corresponding equations for the vacancy concentrations. Neglecting the spatial variation for the concentrations the set of equations can be simplified and leads to a single differential equation for the substitutional concentration Cs(t) O C =s K,<--~- -s q }- { Kc C - 1 . ( (C s LC 1 The short time solution for this equation is C _ Cs(t) = ypNpDIAC~t. (187) (186)
The conservation of matter requires that the radius of a growing spherical precipitate is given by Cp{r(t) - r(0)} = rm{C~(t ) - C} t/3 = rm{Cs(t)
-
C} 1/3.
(188)
334
PROGRESS
IN MATERIALS
SCIENCE
Combining eqs (187 and 188) and assuming r(0)= 0 one obtains for the initial growth
r(t) = [~TpAC~DIc# t} 1/3"
(189)
The comparison with eq. (184) shows that the time dependence for the self-interstitial controlled growth is weaker, t 1/3, and determined by the diffusion coefficient of the self-interstitial D~. Although this quantity is not well known as has been discussed in Chap. 4.1.1 estimations show that it may be much smaller than the diffusion coefficient for many transition metal impurities. Therefore the precipitation process should proceed slower as one might expect from the high diffusivity of these elements if the growth rate is diffusion limited. Recent experimental results of the copper precipitation at grain boundaries in silicon (Fig. 86) indeed suggest that the growth of the precipitate is not determined by the diffusion coefficient of the copper according to eq. (184) but by a slower process. Since the formation of the Cu3 Si phase is accompanied by a large volume change the injection of self-interstitials (or alternatively the absorption of vacancies) is necessary. Although the simplified calculation given here explains a slower rate, for a quantitative comparison more experimental data and a refined calculation which includes the particular geometrical situation at a grain boundary for instance are necessary. The precipitation processes which have been described so far are kinetic reactions in the solid state which occur during cooling when the solubility limit of the impurities are exceeded. For some impurities such as carbon and oxygen in metallurgical silicon for terrestrial solar cells the solubility limits may already be exceeded in the melt. In this particular situation precipitation may occur in the melt or more likely at the solid-melt interface. It has been mentioned in Chap. 4.1.2 that heterogeneous precipitation from the melt at the solidification front may be responsible for the precipitation and growth of SiC in carbon-rich silicon. Since the interface moves during solidification and impurity atoms are continuously rejected into the melt due to the lower solubility in the solid the solution of the diffusion problem becomes more complicated. An analytical solution of the diffusion problem has recently been derived for diffusion limited growth of the precipitates3228) The results show that the maximum radius is determined by the ratio of the interface velocity and the melt diffusion coefficient, DJv, and given by
rma x - -
Om~mf~a.n~
v(c~)
- -
(190)
with the supersaturation in the melt given by AC~m= Cm- C~ and C~ the eutectic concentration for the impurity. Slow interface velocities and high supersaturations of the impurities can lead to large precipitates. Typical values for directionally solidified silicon are v = 0.001 cm/s and AC~m,~ 100ppm which yields precipitate sizes in the micron range (Chap. 5). Precipitates of this size are detrimental for the p-n junction region of a solar cell since they lead to short circuit currents and a deterioration of the device performance. The melt-interface mechanism may be thus important for carbon, since high concentrations of carbon are quite common in solar cell silicon and precipitation reactions in the solid are slow because of the low diffusion coefficient for carbon in the solid. The ramifications of the melt-interface mechanism on the formation of carbon related defects and the electrical properties of silicon are discussed in Chap. 4.1.2. 4.5.1.4. Experimental results for grain boundary segregation. The experimental investigation of impurity segregation and precipitation at grain boundaries in semiconductors is difficult
335
and hampered by the fact that generally the impurity concentrations in semiconductors are low. Direct methods such as AES, EELS, SIMS or XPS are frequently not sensitive enough to detect the rather low impurity concentrations at interfaces. Despite the difficulties there are a few AES results on fractured surfaces of polycrystalline silicon ~229~which show an enrichment of C, O, F, AI and Fe. It is assumed that the exposed surfaces are grain boundary regions and hence the results may indicate segregation effects. Further developments are the application of mapping and scanning techniques using SIMS or AES with a spatial resolution of about 0.5-0.8/~m (SIMS) and 300-500/~ (AES). They give information about the distribution of impurities in the volume and at grain boundaries if the impurity concentrations are high enough. A typical example for segregation at a grain boundary in silicon is shown in Fig. 70 for boron, aluminium and hydrogen.~23~It demonstrates the segregation of the doping impurities at the grain boundary and the different behavior of hydrogen when forming complexes with these elements. In these methods the spatial distribution of the impurities is obscured by the resolving power of the techniques, however, and does not reflect the actual impurity distribution. Analytical techniques which have a higher lateral resolving power are mainly based on transmission electron microscopy (TEM), but also require a strong enhancement of impurities at the interface (> 1019 c m - 3 ) . This can be achieved for instance in highly doped polycrystalline silicon where segregation at grain boundaries can accumulate enough impurities to be detected by STEM analysis. Segregation of AI, As(23t) and p~232~has been observed in this way, but there are only a few systematical quantitative anlayses which give information about the segregation behavior. Some experimental and theoretical data are summarized in Table 10 Typically one obtains interaction energies of about 0.2-0.6 eV for the doping elements in silicon and an enrichment corresponding to 0.1 to 0.3 atomic layers. High resolution TEM which can resolve individual atomic columns is usually not sensitive enough to detect a few segregated impurity atoms. The recently developed chemical imaging technique attempts to overcome this limitation and uses image processing techniques to analyze HREM images. It has been demonstrated that one can distinguish between different atomic species in particular cases for instance when the interdiffusion at GaAs/AIGaAs interfaces is studied. A further improvement can be expected from the Scanning Tunneling Microscopy and Spectroscopy (STM, STS) technique~234~which has already been applied to show the segregation of boron, hydrogen or aluminium at a grain boundary in silicon or the enrichment of oxygen at a grain boundary in CulnSe2 (Fig. 71). One can certainly expect that in the future more results from the application of these methods will be obtained. Experimental techniques which have a high chemical sensitivity are radioactive tracer measurements or Mrssbauer spectroscopy but are limited to a few suitable radioisotopes such as lac, 3H, 57Fe or 57C0. Qualitative results can be easily obtained by autoradiography as has been demonstrated for instance in the case of the cobalt accumulation at a grain boundary. Qualitative tracer measurements require mechanical or chemical sectioning techniques with a rather low spatial resolution. This technique has recently been applied to investigate the cobalt segregation at a single grain boundary in silicon. (234) Despite the low cobalt volume concentration of about 1013cm 3 the quantitative analysis enabled it to measure an enrichment of about kG ~ 50-100 and the depletion of the cobalt concentration in the vicinity of the grain boundary. The width of the depleted zone is about 50-100/~m and depends on the cooling rate of the specimens (Fig. 72). Denuded zones with comparable extension have also been observed for other impurities usually by etching. These results are particularly important for polycrystalline semiconductors because they show that the interaction of impurities with grain boundaries can change the distribution of electrically
336
AI
Si-H
Si-H
B-H
AI-H
..................
!!!!!!!!!!!!!!!
il:illlI!llll!!i
ii!iIlliiliilil
II!!!!i!!!1!!!~
~*lp#ttt~tti~
IIiIHalttiil|
FIG. 70. Volume-indexed AES mapping of the distribution of hydrogen at a grain boundary in silicon. (a) p-doped crystal with boron segregation, (b) with aluminium segregation at a grain boundary.(23)
active defects in the v o l u m e within a r a n g e which is c o m p a r a b l e to the g r a i n size in m a n y cases. A f u r t h e r result o f the c o b a l t e x p e r i m e n t o b t a i n e d b y M 6 s s b a u e r s p e c t r o s c o p y was t h a t the e n r i c h m e n t o f c o b a l t is due to p r e c i p i t a t e s o f C o S i 2 which are r a t h e r i n h o m o g e n e o u s l y d i s t r i b u t e d in the b o u n d a r y plane. R e c e n t s y s t e m a t i c a l T E M studies o f the i n t e r a c t i o n o f c o p p e r with g r a i n b o u n d a r i e s have also s h o w n t h a t c o p p e r p r e c i p i t a t e s (Cu3 Si) nucleate at
Table 10. Experimental and numerical results for the energy difference AG between bulk and grain boundary sites (interaction energies) of some doping and transition elements in silicon<231,232,236)
AG (calc) AG (exp)
B 0.18 <0.05
P 0.27 0.33 0.44
As 0.2 0.43 0.42
3d elements 0.6 --
337
FIG. 71. Spectroscopic scanning tunneling microscope (STM) image of a grain boundary in boron doped and hydrogen passivated silicon. Silicon atoms (black), boron atoms (white), and hydrogen atoms (dashed) (from Ref. (233)).
the interface during cooling (Fig. 86). The distribution and size of the precipitates is rather inhomogeneous and depends primarily on the cooling rate, and to some extent on the structure of the boundary. Random boundaries tend to accumulate more precipitates, partly due to the fact that a preferential precipitation occurs at dislocations and steps which are more frequent in random grain boundaries. The quantitative analysis of the growth of the precipitates indicates that the rate limiting step for the growth is not the impurity diffusion. Since copper silicides have a large lattice mismatch with silicon which requires the participation of vacancies or Si-interstitials during precipitate formation it has been speculated that Si-interstitial injection in the volume may be the rate limiting step. ~235) There are numerous electrical measurements now which indicate the influence of segregated impurities on the electrical properties of grain boundaries and interfaces. These indirect techniques are mainly resistivity measurements, admittance spectroscopy, DLTS and LBIC or EBIC which have been partly described in previous chapters. In combination with analytical techniques the segregation of oxygen, carbon and hydrogen in silicon and various other impurities in Si and GaAs has been studied qualitatively. A careful quantitative analysis of the interaction of copper with a grain boundary in silicon by DLTS (71) recently has shown that copper precipitates can be responsible for the electrical
c(x)/co
1.oo 0.75 -0.50-0.25 -I 50 I lO0 I 150 I 200 870 C
[i.Lm]
FIG. 72. Experimentally determined concentration profiles of cobalt near a grain boundary in silicon after segregation. The specimens are quenched after in-diffusion of radioactive cobalt at 870C and 970oc3234)
338
activity and that the distribution of the density of states of the grain boundary is related to the inhomogeneous distribution and size of the precipitates. Finally it should be mentioned that theoretical calculations which have been very useful already for the investigation of the atomic structure of grain boundaries are also applied now to the interaction of the impurity interaction. One can expect that with the further development of the theoretical methods the calculated results become more reliable. A summary of experimental and calculated results (for a X9 grain boundary) is given in Table 10 for some doping and the 3d transition metals (Ti, V, Cr, Mn, Fe, Co, Ni, Cu)3236) For the 3d elements the segregation energy is almost the same AG ,~ 0.6 eV since no volume changes are assumed. These results indicate an attractive interaction for all elements which is smaller for the doping elements and almost negligible for boron in agreement with the experimental observation. Most of the other analytical techniques which have been applied so far cannot distinguish between the accumulation of an impurity due to segregation or precipitation. Therefore at present the number of experimental data for semiconductors are not sufficient and too scattered to develop a consistent picture of the impurity-grain boundary interaction for the different impurities. Qualitatively the interaction of doping elements (B, P, As), carbon and oxygen, and of the 3d transition elements has been observed in silicon. Whereas the doping elements seem to segregate and remain dissolved at the grain boundary even at room temperatures, the other impurities have only been observed in the form of precipitates. The question remains open whether there is also a fraction of the impurities dissolved in the grain boundary and how they contribute to the electrical properties. Much less is known about the interaction of impurities and intrinsic point defects in compound semiconductors. For instance the segregation of oxygen at grain boundaries has also been observed in GaAs and CulnSe2. However, the results are only qualitative and not related to the electrical properties of the semiconductors. No experimental results are available about the interaction of intrinsic point defects which are especially important in compound semiconductors because they can be dominating electrical defects. The discussion in this chapter has shown the significance of the impurity-interface interaction for the electrical properties of the material. Although the discussion has mainly focussed on the structural and electrical changes at interfaces one has also to consider that the segregation and precipitation processes leads to an inhomogeneous distribution of the impurities in the bulk. The enrichment of impurities at grain boundaries and other interfaces can be accompanied by a depletion of the impurity in the vicinity of the lattice defects up to 100 #m wide which may be a considerable fraction of the grain size. In general the width and depth of the denuded zone depends on the thermal history of the specimens such as diffusion temperature or cooling rate and the interaction behavior between the interface and the impurity. The investigation of these interaction processes and the ramifications on the electrical properties of solar cell materials has not received the necessary attention in many cases but is important considering the far reaching consequences. 4.5.2.
Interface diffusion
The preferential diffusion along grain boundaries or other interfaces is a well known phenomenon which has been investigated and described in a large number of articles. Grain boundary diffusion is important for diffusion processes in polycrystalline semiconductors during device fabrication, however, surprisingly few experimental results are available about diffusion at particular grain boundaries which would allow study of the process in detail.
339
Most of the experimental results have been obtained for metals which has lead to a rather detailed description of the grain boundary diffusion. A relatively simple model has been developed by Fisher (23v) and is used in almost all investigations. The grain boundary is considered a thin slab of a finite thickness d = 2a, perpendicular to the surface, where the diffusion coefficient D~ differs from the bulk value D,. The concentration profile in the bulk C, (x, y, t) and in the interface Cc (y, t) is calculated from the standard diffusion equations (Fick's law) for two dimensions (x-axis parallel, y-axis perpendicular to the interface)
dt = D ~ s y :
~C G
+~Wx2 j
for I x l > a
(191)
~2C G . D~ OC~ for Ixl = a (192) 8t - - D G ~ - t a Ox where it is assumed that Cc (y, t) is constant inside the interface layer in the x-direction. The flux of atoms between the interface and bulk region is described by the second term in the last equation. For self-diffusion it is assumed that the edge of the interface region [xl = a the concentration profile is continuous thus C~ = C,. Refinements of the model have assumed that for foreign atom diffusion at [x[ = a the ratio of bulk and interface concentration is determined by the equilibrium segregation ratio kc = CUC~. Exact solutions of the diffusion equations have been given for two boundary conditions. (1) the constant source model, where C, (a,0, t ) = C (Whipple (z38)) (2) the finite source model, where a thin layer of material at the surface is assumed which can be exhausted and decays with time (Suzuoka(239)). The solutions involve non-resolved integrals which have been tabulated. A simple analytical solution for the constant source case has been obtained by Fisher under the approximation of a very large interface diffusion coefficient D G
x-a
with
/3- D~ (D~t) 1/2

From this solution it appears that the CJC ratio varies with t ~/4which is different from pure volume diffusion where the time dependence of the average diffusion length is described by t 1/2. In many experimental techniques (for instance sectioning parallel to the surface) an average penetration depth is measured and the concentration profile in x-direction is integrated (ca(y, t)= <C~(x,y, t ) > ). From the exact solutions it can be shown that for /3 > 5 the average concentration Ca is proportional to the exponential term in eq. (193).
Ca(y, t) ~ Cexp
DG kGa
nl/4(/3D~t)l/2 .
(194)
A plot of In Ca vs y should give a straight line then. Provided that the volume diffusion coefficient D~ is known one can determine the parameter PG = Dckca from the slope of the plots. The diffusion experiment does not allow separation of the parameters k G, a and DG so
340
that additional assumptions have to be introduced to determine the interface diffusion coefficient. Considering the atomic structure investigations of grain boundaries and interfaces it is reasonable to assume that the interface width is of the order of the lattice constant. In fact the width d is an arbitrary parameter without a physical significance. Other models which have been developed to describe the interface diffusion such as atomic jump models, molecular dynamic studies, or others do not need this parameter but introduce at least one other structure related parameter so that the problems remain the same. It is generally observed that the temperature dependence of the diffusion parameter PG(T) follows an Arrhenius law P c ( T ) = P c o e x p { - ~Qc} . (195)
In the approximations for low interface concentrations given above the activation energy is then given by Qc--AG + H~ where AG is given by eq. (180) and H~ is the migration enthalpy of the impurity in the interface. More complex diffusion models include the participation of intrinsic defects so that in general the activation energy is a sum of the formation and migration energies of these grain boundary defects. Usually in metals values for H~ ,,~ (0.4 to 0.7) x HSv are found (H so volume self-diffusion enthalpy). In polycrystals the preceding solutions are only valid if the grain boundaries are perpendicular to the surface and independent from each other. If the grain boundary is inclined by an angle ~oone measures an apparent diffusion coefficient by Dc cos2tp. If the grain size is smaller than (D~ t) ~/2the diffusion profile is typical for volume diffusion with an effective diffusion coefficient Doff=fDc wherefis the fraction of the interface region which is calculated on the assumption of the interface width d. In a polycrystalline material with random grain boundaries one usually considers an average diffusion coefficient which is determined empirically. The prediction of the diffusion in polycrystalline materials from measurements at single grain boundaries is further complicated since the structure of the grain boundary has a strong influence on the diffusion coefficient. For instance it is observed that for twin boundaries in silicon which have a rather perfect atomic structure no accelerated diffusion occurs but can be measured in more random large angle grain boundaries. The microstructure of a polycrystal (the distribution and type of the grain boundaries) is thus important, which may explain that diffusion data obtained for various polycrystalline materials are often rather controversial. In semiconductors only a few experimental data are available for grain boundary diffusion. Because of the technical importance these are primarily measurements of the diffusion of doping elements in polycrystalline silicon thin films which have grain sizes in the micron range. Activation energies Qc (including the segregation energy AG) between 1.5 and 1.95 eV have been determined~24~for phosphorus whereas antimony seems to diffuse more rapidly with a lower activation of about 0.83 e V . (241)
4.6. Electrical Transport Properties of Polycrystalline Semiconductors In the previous chapter it has been shown that many of the structural and electrical aspects of grain boundaries in elemental and compound semiconductors are well understood. Although many experimental results have been obtained for special boundaries it was possible to draw some general conclusions particularly for compound semiconductors. In fact it appears to be that the electronic trap levels of random grain boundaries are rather similar
SEMICONDUCTORS
341
and the main differences in the electrical activity lie in the concentration of the trapping states. Despite the remaining uncertainties about the origin and the distribution of trap levels for random grain boundaries the knowledge about the electrical activity of individual interfaces suffices in many cases to describe the electrical behavior of polycrystalline semiconductors as well. For photovoltaic applications polycrystalline materials are produced by a variety of techniques which will be described in the following chapters. Casting and sheet growth techniques primarily used for silicon yield polycrystals with grain sizes in the range of about 0.1-10 mm, whereas thin film techniques such as chemical vapor deposition (CVD) typically produce grain sizes in the micron range (0.1-10 # m). Considering the width of the space charge region (typically about 0.01-1 #m depending on the doping level) grain boundaries in the first class of materials are electrically separated and the electrical properties of the semiconductors are to a large extent determined by the bulk properties. The grain boundaries can be considered as individual defects imbedded in large regions of single-crystalline material with properties which are determined by the doping impurities and other electrically active defects. We have seen, however, that grain boundaries interact with impurities and dislocations and thus influence the bulk properties within a much larger range of about 10-100 # m. These are indirect ramifications on the electrical properties of the semiconductors which are much less studied yet are presumably important for these large grained materials. Thin film polycrystalline semiconductors have grain sizes comparable to the space charge width of the grain boundaries so that they interact electrically. This can lead to electrical properties of the material which may be very different from the behavior of single crystals. Contrary to the previous description of the electrical behavior, the grain boundaries cannot be treated as individual defects anymore and the situation requires a new approach to the problem which will be discussed in this chapter. The high density of grain boundaries in thin film semiconductors also affects other physical properties such as light absorption, recombination or diffusion of impurities but they are usually explained in terms of the corresponding (averaged) properties of the individual grain boundaries which have been the subject of the previous chapters. A large number of electrical measurements on polycrystalline semiconductors were carried out during recent years because of the numerous technical applications of these materials in integrated circuits, varistors, sensors, and photovoltaic devices. The electrical conductance of polycrystalline silicon [e.g. (56, 183-185, 242-245)] and ZnO ~6) as a function of the doping concentration, the microstructure, the temperature, and various other processing parameters is one of the most systematically investigated properties. These measurements have confirmed that the electrical properties are related to grain boundary phenomena. Many results are also available for polycrystalline compounds for solar cells such as Cu-ternaries or CdTe, however, these results are rarely discussed from the perspective of the individual grain boundary behavior, although it seems quite evident that grain boundaries are most relevant in these materials as well. A typical example is the electrical conductivity in CulnSe2 and the influence of oxidation treatments to enhance the efficiency of solar cells which will be discussed in Chap. 7. Electrical transport properties. The important electro-optical parameters of semiconductors for photovoltaic applications are the carrier concentration and mobility which determine the conductivity, and absorption and recombination processes. The understanding of the electrical transport in polycrystalline semiconductors thus provides the basis for an
JPMS 35-3/4~J
342
P [t3crn] 10 4 10 3 10 2 10 1 10 0 10 -1 10 -2 10 -3
I
P [~crn] 10 4 10 3 10 2 M ~ ~~
10 1 10 0 10 -1 10 -2 10 -3
1015
1017
10TM N[cm-3]
10 TM
1016
10TM N[cm -3]
FIG. 73. Specific resistance of Po = 1/ao as a function of the doping concentration N [crn -3 ] for monoand polycrystalline silicon (open squares: B-doped, black squares: P-doped) ~245) and mono- and polycrystalline germanium, Sb-doped3246~
understanding of the electrical behavior in general and hence shall be discussed here. In solar cells the applied voltages are low and high-field effects which are responsible for the non-linear current-voltage behavior (the varistor effect) for instance in ZnO (see previous chapter) are not important, therefore the low voltage (zero-bias) conductance (or specific resistance) will be considered only. In photovoltaic and other devices the conductivity is usually measured parallel to the substrate. Since the microstructure of thin film semiconductors often shows a columnar structure the electrical transport is mainly perpendicular to the grain boundaries. A typical example of the specific resistance Po in a polycrystalline thin film as a function of the doping concentration is given in Fig. 73 for silicon and germanium. For silicon the main feature is a strong increase of the resistance (over several orders of magnitude) in a narrow range of doping concentrations (usually around 1018cm -3) and a constant value for low doping levels. The increase is observed both for n- and p-type films and is related to the grain size of the films. It disappears for larger grain sizes above about 100#m and the conductance approaches the single crystal behavior. The temperature dependence changes from a positive to a negative slope at high doping levels and is exponential at low doping levels with an activation energy of about 0.55 eV as in intrinsic silicon. For germanium a quite different behavior is observed. The specific resistance decreases for lower n-doping levels and the material becomes even more conductive compared to the single crystal behavior. For p-doping the resistivity changes corresponding to the single crystal but is on average lower. Hall measurements have confirmed that the change in the specific resistance or conductivity ~r = e (n#. +p/~,) is mainly due to a change of the majority carrier concentration (n or p) o and accompanied by an overall decrease in the Hall mobility (Fig. 74). The results for silicon show that for low doping concentrations and small grain sizes the grains are considerably depleted from majority carriers whereas the mobility has a minimum at about N ~ 1018cm -3. In germanium the increase in conductivity is due to an inversion in the conductivity from nto p-type below ~ 1018cm -3 and an increase of the hole concentration. These results indicate that a substantial amount of majority carriers are trapped in the grain boundary states. A similar behavior is also observed for non-stoichiometric polycrystalline CuInSe2 where the
SEMICONDUCTORS
343
n+p [em-3] 10 19 10 18 10 17 10 16 10 15 10 14 10 13
1) poly-germanium 2) poly-silicon /
1014
10 TM
10 ts N [cm"3]
0 F]G. 75. Schematic band structure of a polycrystalline semiconductor with a periodic array of negatively charged grain boundaries, E T is the occupation limil of the neutral grain boundaries.
FIG. 74. M e a n electron and hole concentration <n + p ) in the bulk for n-doped poly-germanium (a) and n-doped silicon (b). Poly-germanium shows an inversion from nto p-type conductivity below N ~ 10 m cm-3, (246)whereas poly-silicon remains n-type, respectively. (245) Solid lines are fitted theoretical curves.
doping level is determined by the stoichiometry of the crystals (Fig. 108 in Chap 7) and the conductivity thus changes with the composition of the crystal. In general, theoretical models are based on the assumption that grain boundaries are identical in their electrical behavior and distributed perpendicular to the current direction at an equal distance L corresponding to the average lateral grain size (Fig. 75). Different conduction models have been developed based on the grain boundary barrier model. Although in many cases the experimental results can be explained qualitatively a number of modifications have been proposed to obtain also numerical agreement. These refinements of the models include assumptions about additional conduction mechanisms such as tunnelling, modifications of the shape of the grain boundary potential, or an additional amorphous grain boundary layer. Some of the earlier modifications seem obsolete in view of the current knowledge about individual grain boundaries and therefore shall not be discussed further here. The calculation of the conductance requires the determination of the carrier concentration n + p and the mobility/~. The main difficulty for the calculation of the carrier concentration arises when the space charge regions and the potential barriers overlap. The band structure in the entire crystal is distorted then and no undisturbed bulk region exists anymore. As for the case of the single grain boundary the grain boundary charge QB, the potential barrier VB and the compensating space charge Q~ have to be determined self-consistently for a given doping N and Fermi-level EF. However, since no undisturbed bulk region exists and the positions of all energy levels in the entire crystal depend on the distribution of the electrons and holes on grain boundary and bulk states the Fermi-energy is a priori not fixed anymore. Most of the calculations have neglected the problem and assumed that between the grain boundaries (at x = L/2 in Fig. 75) the original bulk state remains where the Fermi-energy can be determined from the doping level. This is a particular serious limitation in semiconductors where the presence of the grain boundary states inverts the conductivity type, a behavior that has been observed in polycrystalline germanium. In a situation where the position of all energy levels depends on their occupation itself the Fermi-energy has to be determined from fundamental principles for instance by the minimization of the free energy of the whole electronic system as has been shown recently.(246)
344
The results which will be briefly described here show that for the calculation of the electron and hole concentrations in the interface (nB) and bulk states (nTi), or the conduction and valence band a different Quasi-Fermi level EFi is required then. The first step in the calculation is the determination of the grain boundary potential VB as a function of the mean electron, n, and hole concentrations, p (averaged over the entire bulk region). These quantities can be measured experimentally which thus allows a direct comparison with the theoretical prediction. Since the potential barriers in fine-grained polycrystalline semi-conductors are usually rather small V(x)< KT, an analytical solution from a linearized Poisson's equation can be obtained for VB and VD (at x = L/2). VB= bz tanh ~ ;
z 7ta
I'D = bz
7ta
sinh(7~a) a 2 sinh(xa)
(196)
with the abbreviations

e2L 2
b = 4EEo
- -
a 2 = (n + p )
e2L 2
4eeoKT
where L corresponds to the average grain size. The other quantities have the same meaning as defined before, z = na/L is related to the number of occupied grain boundary states na
nB
f NB(E) fs(E - EFB)dE -- Nan
(197)
and the grain boundary charge QB = enB (eq. (162)). The occupation function fB(E -- EFa) as obtained from the minimization of the free energy has the form of the Fermi-Dirac function where the Quasi-Fermi energies Eva for electrons in the grain boundary states and EF = Evn = Evp for electrons and holes in the conduction and valence band are approximately given by
n EvB=EF+ I'D-- VB; EF=Ec+ K T I n ~ - z
OVD On "
(198)
Together with the neutrality condition N d - Na = n --p + z (Nd and Na are the donor and aeceptor conentrations, respectively) this set of eqs (196-198) allows to calculate the parameters n, p, nB and VB for a given density of states distribution NB(E). The main features of the numerical calculations are depicted in Fig. 76 which show the total free carrier concentration, n + p, as a function of the n-doping concentration Nd. A single grain boundary level at Ex, with NB = 1012cm -2, a grain size of 0.1/~m, and the material parameters of germanium (band gap Eg = 0.66 eV) have been assumed. For a trap level in the upper half of the band gap (ET >t Eg/2) the crystal remains n-type and the average electron concentration decreases rapidly in the doping range Nd ~ NB/L with decreasing doping level. This behavior is due to the trapping of carriers at the interface and the corresponding depletion of the bulk. At high doping levels the interface states are filled and the total carrier concentration increases almost linearly with the doping concentration. A different behavior is observed for a trap level in the lower half of the gap. The electron concentration now decreases even below the intrinsic level so that holes become dominant at low doping levels and the total carrier concentrations goes through a minimum. The Fermi level is pinned by the interface states in the lower half of the gap. The implication of the result is that in this ease a polycrystalline semiconductor can become very conductive again at low doping levels due to the inversion and the increase of the hole concentration. The dependence
SEMICONDUCTORS
345
Iog(n.+p) G
18
17 16 15 14 13 12
! I I i
3)
I
E, = 0.3 V
I
14
15
16
17
18
19 Io9 N [cm33
FIG. 76. Calculated average carrier concentration n + p in polycrystalline germanium as a function of the doping level. A single grain boundary level E r at different energies is assumed. Grain size L = 0 . 1 / J m and
NB= 10~2cm-2. t~)
on the position of the interface level JETis reversed for p-doping, where an inversion of the majority carrier type occurs for a trap level in the upper half of the band gap. Figure 77 shows the corresponding change of the grain boundary potential which goes through a maximum when the carrier concentrations decreases rapidly, however, is quite small at high and low doping levels (VB <~KT). The results remain qualitatively the same for other density of states distribution NB (E) where JETis the occupation limit of the neutral grain boundary then or for other materials such as silicon. The rapid change in the carrier concentration depends on the grain size. For larger grains the transition region is shifted to lower doping levels and finally does not occur at all. The conductivity is usually described within the thermionic emission model. Since for a polycrystalline semiconductor the applied voltage drops over a large number of grain boundaries the voltage across a single boundary is quite small. Therefore the conductivity can be described by the zero-bias conductance given in eq. (168). Although other models have
vo/e v]
0.06
GERMANIUM
0.04
I) ET= 0.1 eV 2) E = 0.3 eV T eV
0.02 -
0.0
0,02
! i I I I !
14
15
16
17
/9
ioq N [o~33
FIG. 77. Calculated potential barrier height of a grain boundary in polycrystalline germanium. The parameters used in the calculation are given in Fig. 76. cu6)
346
PROGRESS
IN
MATERIALS
SCIENCE
been developed as well they result essentially only in modifications of the prefactor A (T) so that one can summarize the results for instance for n-type conductivity by
tr=A(T) n exp(- K~ ) .
(199)
The mobility is determined by the prefactor and the exponential dependence on the potential barrier and hence goes through a minimum with decreasing doping level. The numerical results explain qualitatively the behavior of polycrystalline silicon and germanium. The experimental results for individual grain boundaries in silicon indicate a midgap trap level ET, therefore a strong decrease in the carrier concentration is expected both for n- and p-doping. This result is actually observed as can be seen from Fig. 74. The mobility also shows the predicted minimum in agreement with the model. (56'57'2451 The decrease of the carrier concentration and to a lesser degree the mobility are responsible for the sharp rise of the resistance of polycrystalline silicon. The experimental results have also been used to dete~rnine the density of states by the I-V deconvolution technique. The averaged density of states is depicted in Fig. 78 and compared with measurements of a single grain boundary. The maximum in NB(E) in the lower half of the band gap is also observed for a single low energy grain boundary which may indicate that it is also a characteristic feature of random boundaries. It should be noted that results from other authors (2151 have yielded different density of states distributions, in particular a rise towards the band edges (band tails) therefore the details of the density of states for grain boundaries in silicon remains contradictory. The conductivity measurements in poly-germanium are a typical example for a case where the grain boundary levels lie in the lower half of the gap. Since experimental results on single boundaries have shown a trap level at about 0.1 eV above the valence band (see Chap. 4.4) the total carrier concentration of poly-germanium should increase again with decreasing n-doping concentrations because of the increase of the hole concentration. This is indeed observed as can be seen from Fig. 74. The numerical solutions of the eqs (196-198) have been fitted to the data and yielded the following parameters E T = 0.07 eV, NB = 6 x 10~2cm -2, and L = 75 nm which is in agreement with grain sizes determined by TEM.
(a) Nb ~ E ~ poly - S i l i c o n IO 13
J
(b)
~ 1014
I0 12
Afromde-measurements
lO I] 1012
I0
II
Oooo,o ,
I0 I0
E v O.l o.3
= I I Z
,o,,
I Ev -o.,~
~A~
A
1 o.,~
o.5
0.7
0.9
levi
tu o
I I 1 -o.2 o 0.2 ENERGY E (eV}
Ec
FIG.78. Experimentaldensitiesof states distributionfor grain boundariesin variouspoly-siliconthin films. (a) Comparison of measurements with a single grain boundary in silicon as measured by admittance spectroscopy,(b) from Ref. (2151. (73,245)
347
The results show that it is possible to relate the electrical behavior of individual grain boundaries to the conductance properties of polycrystalline semiconductors and help to explain the essential features. The comparison with numerical results allows determination of an average value of the parameters which describe the electrical properties of single grain boundaries. This result may be of particular importance for compound semiconductors where grain boundaries seem to have rather similar electrical parameters compared with other defects but are difficult to study as a single defect. As has been discussed in Chap. 4.4 their properties may be deduced then from measurements at surfaces or heterostructures for instance which offers the possibility to predict some of the electrical properties of the polycrystalline materials. This approach has not been tried systematically so far for photovoltaic semiconductors but may be very helpful in cases where novel materials such as ternary compounds are studied.
5. SINGLE CRYSTAL AND POLYCRYSTALLINE SILICON
Silicon is still the most widely used solar cell material both for high efficiency cells but also for low cost applications. It is also the best studied semiconductor so that one can expect that in the future further improvements of solar cell efficiencies will become feasible particularly for the terrestrial solar cells. In this chapter the two main developments of crystalline silicon will be discussed. In the first section the most important material problems are addressed which are connected with the growth of single crystals for high efficiency cells. Although many of these issues are also important for the device technology and have been solved from the perspective of the device performance there are additional requirements for solar cell silicon which shall be discussed in particular here. In the second chapter some of the most promising low cost growth techniques will be presented. Since these techniques inherently produce a material with higher concentrations of lattice defects their investigation and the relation to the processing is essential for the improvement of the device performance. Silicon is also used in the amorphous state as a thin film material. Since amorphous semiconductors have quite different properties compared to crystalline semiconductors this subject shall be discussed separately in Chapter 8.
5.1. Cz and Fz Grown Silicon
High quality single crystal silicon is either grown by the Czochralski (Cz) or float-zone (Fz) technique. The growth methods and technical requirements have been described in numerous papers and shall not be discussed here. Today, about 80% of the single crystals are grown by the seeded Cz-growth method from a crucible. Because of the contamination of the melt from the wall of the crucible (usually high purity quartz) the purity of Cz-crystals is limited. The maximum resistivity of an uncompensated crystal can hardly exceed 20 ~ cm. Particularly oxygen is present in higher concentrations (1017-10TM cm -3) compared to the float-zone technique. The Fz-technique is a crucible free method where a free floating molten zone is passed through the hanging feed crystal. It is limited to the growth of crystals with smaller diameters (,~10cm) but of higher purity. Resistivities of about 100flcm can be easily obtained and by applying multi-pass zone melting resistivities close to the intrinsic level can be reached. Both methods have advantages for particular applications in the device technology. As far as purity is concerned the Fz-technique is far superior. In contrast, the Cz-method is advantageous if crystals with larger diameters and high doping concentrations
348
are required for instance as a substrate material for epitaxy. Such a material can hardly be produced by the Fz-method. The application for solar grade silicon requires, apart from a high quality, low production costs. Improvement of these existing crystallization techniques in terms of lower costs are feasible but certainly restricted. Limitations in the growth rate and other factors do not allow to substantially increase the throughput. For the Cz-method for instance reductions in the cost can be achieved by bigger charges, larger diameters, and a better utilization of the expensive crucibles by a continuous melt replenishment. With these improvements implemented the crucial point becomes the reduction of the material loss due to the slicing of the ingots. Although new slicing techniques have been developed with thinner blades or wires and an increasing number of slices per ingot there are certain limits on the cost reduction which can probably not be exceeded. Therefore it is evident that other single crystal growth techniques have to be developed in the future which are more suitable for the photovoltaic application. Material properties. It has been discussed in previous chapters how lattice defects such as many impurities, dislocations, or grain boundaries can reduce the lifetime of the semiconductor by recombination processes and hence limit the conversion efficiency of a solar cell. One might therefore expect that high purity, dislocation free Fz-crystals should be especially suitable for the fabrication of high efficiency solar cells. However, even for this material the lifetime is usually only about 1 ms which is still much lower than the theoretical limit determined by radiative and Auger recombination, or by recombination at the remaining impurities (see Chap. 2.4). The reason is that in these high purity crystals a variety of microdefects occur which nucleate during the growth and cooling process and reduce the lifetime. The types of defects which are typically observed under various conditions are so called A, B, D and I microdefects.(247) The most important growth parameters for the defect formation are the thermal gradient, G, and the growth speed, V. A- and B-swirls (because of their swirl patterns in the cross section of a crystal) form at lower growth rates (typically below about 4 mm/min) and for larger thermal gradients, a parameter which is mainly determined by the crystal diameter. The condition for a swirl free crystal is thus a large V/G ratio. D defects appear at higher growth rates ( > 4 mm/min) whereas the occurrence of I-defects is less affected by these parameters. Lifetime measurements on Fz-grown crystals show that there are two major reasons for the lifetime degradation. The first effect is shown in Fig. 79 and depicts the dependence of the lifetime on the cooling rate given by the product V G. This degradation is a general effect and not related to the distribution and concentration of swirls, D- or I-defects in the
100
"~ 10-
Dislocation-free,
Swirl-free
1.0
100 2;0 300' 4;0 5(30 Coo!i~g rate V = G (=C/rain)
600
FIG. 79. Lifetime vs cooling rate for a swirl-free Fz-silicon single crystal. (247)
349
crystal. The second influence is, however, clearly related to the formation and distribution of A and B swirls which act as recombination centers. D and I-defects which are more homogeneously distributed in the crystal have only a minor influence. The origin of the defects and especially the dependence of the cooling rate on the lifetime is not understood. It has been discussed in the literature (248) that fast cooling m a y produce a supersaturation o f vacancy clusters, clusters enclosing impurities or even impurity complexes which may act as recombination centers. It has also been suggested that D defects m a y be vacancy clusters (249) but they are apparently not related to the lifetime degradation. Therefore D-defects and the non-detectable defects which m a y form during fast cooling must be of different origin. The I-defects are thought to be micro-precipitates of trace oxygen in float-zone silicon and are even smaller compared to D-defects. (25) The more thoroughly investigated defects are the A- and B-swirls which have been discussed in Chap. 4.1.2. A-swirls have been identified as interstitial type dislocation loops in the range 1-3 # m ~251) whereas the nature of the B-swirls which have a size of about 600-800 A is not quite clear yet. It has been proposed that they are carbon-self-interstitial clusters (145) which eventually transform into A-swirls. EBIC investigations show an effective recombination radius of about 40/~m for A-swirls so that for an observed density of about 106-107 cm -3 about 5% to 20% of the crystal volume are affected. Since A- and B-swirls can be suppressed by the growth conditions other defects finally limit the lifetime but their nature and origin are basically unknown. Another material aspect during crystal growth is the homogeneity of the distribution of doping elements. Generally one observes a radial dopant variation (striations) in the crystal which is larger for group V than for group III elements. The variations depend on m a n y factors such as the growth rate, the rotation rate, the crystal diameter and orientation. Under optimum conditions a dopant homogeneity of about _+7.5% can be achieved325z) For high doping levels above about 1017 cm -3 the doping impurities also reduce the lifetime (see Fig. 7) according to the mechanism discussed in Chap. 2.4. 5.2. Polycrystalline Silicon The need to reduce the production costs of silicon for terrestrial solar cells has also lead to a reexamination of the primary steps for the preparation of the starting material from the primary source SiO2 in quartzite rocks or sand. Conventional reduction processes are the carbothermic and the aluminothermal reduction (252) SiO2 + 2C ~--~Si+ 2CO 3SIO2 + 4AI~--~3Si + 2A120 3 The chemical properties of the resulting silicon material vary with the applied process and the terminology given in Table 11 is used to characterize the impurity content. Depending
Table 11. Classification of different qualities of silicon material and corresponding concentrations of electrically active impurityo4s) Silicon categories Electrically active impurities (at %) 1 x 1 0 - 7 t o 1 x 10 - s Poly hyperpure 5 x 10-6 to 5 x 1 0 - 7 Cz monocrystals 1 x 10-3 to 1 x 10-5 (or better) Terrestrial solar grade 8x l0 -3 to l x l0 -4 High purity (HP2) 2 x l 0 -~ to 8x 10-3 High purity (HPI) Metallurgical grade 210 to 5x l0 -j
350
PROGRESS
IN MATERIALS
SCIENCE
on the purity of the starting material the reduction process yields either metallurgical-grade (MG) or high-purity (HP 1) silicon. There are a number of feasible routes to upgrade the purity of silicon which depend on the desired quality of the final product. High quality silicon for devices and high efficiency solar cells utilize the conversion into a volatile silicon compound such as SIC14, SiHCI3, or Sill 4 which can be refined in the gas phase. Low cost processes start from MG- or HP 1 silicon directly and use a number of refining procedures. Impurities with a low segregation coefficient such as the transition metals can be effectively removed during solidification of the melt (e.g. by multi-pass zone refining) below 1016cm -3, whereas this procedure has little effect for the doping elements (e.g. B, P or As) or oxygen and carbon which have unfavorable segregation coefficients. Since the following crystal growth techniques which are described below involve the solidification step anyway this process can be an integral part of the refining procedure. Some impurities like iron and aluminium are concentrated at higher levels at grain boundaries in MG-silicon. This offers the possibility to remove them by acid leaching after the MG-Si is crushed and milled to a powder size comparable to the grain size. The melt can be purified by blowing a reactive gas through the liquid which can remove A1, Ca, C or P depending on the gas that is used. Another technique applied is the liquid-liquid extraction by using a molten slag which has a higher solubility for the impurities but does not mix with the silicon melt. The final combination of these refining procedures to upgrade silicon will not necessarily involve all the steps discussed and will be selected from the perspective of minimizing the costs and the purity of the silicon which is desired. The terrestrial solar-grade silicon which can be obtained by these procedures has impurity levels which can range from about 10t5 to 1019 cm -3 depending on the impurity which is considered. Silicon with these impurity levels is usually used for the following low cost crystal growth processes. There are basically two approaches which both result in a polycrystalline silicon material: the ingot technology and the sheet growth techniques. They will be discussed in the following sections. A different approach is the utilization of vapor-deposition techniques (e.g. CVD) which result in either monocrystalline (epitaxial), polycrystalline or amorphous thin films. So far only epitaxial and amorphous films seem to be technically feasible solutions (see Chapter 8), since polycrystalline thin films have never displayed an adequate conversion efficiency. Despite these negative results, however, all possibilities which are offered by thin film technologies are currently explored. These developments are still at a very early stage and shall not be discussed here any further. 5.2.1. Ingot technique Several ways of making mould based ingots have been explored in the past. The most common variations are directional solidification and casting techniques3253) The main problem with these techniques is the high reactivity of the molten silicon with almost every other material. Therefore contamination of the melt from the wall of the mould can be quite severe and the choice of the container material is thus an important aspect of the ingot technology. Although dissolution of the crucible wall in contact with the reactive silicon melt can hardly be suppressed the material must exhibit a high degree of inertness and purity. There are only a few container materials which fulfill these requirements such as high density graphite, quartz ( S i O 2 ) , Si3N4, SiC, mullite (SiO2)x(A1203)l_x, or a combination of these. Both quartz and graphite are available today in high-purity grades so that contamination apart from carbon and oxygen is only a minor problem. The solidifying melts adhere to the container walls unless special precautions are taken. Since the thermal expansion coefficients of silicon and the container material are different either the ingot or the container will crack
351
43
]'m --Tempera t u r e ~
FIG. 80. Schematic representation of the temperature profile during ingot growth by directional solidification or the Bridgman-Stockbarger method. The crucible is moved in a fixed temperature field (from Ref. (253)).
during the cooling process. In addition silicon expands during solidification because the melt has a higher density. Some measures to solve the problem are either to use one-way moulds which is economically unfavorable or to protect the inside of the container by powders of S i 3 N 4, SiO2, or graphite. Directional solidification or Bridgman-Stockbarger growth is a simple technique which is schematically depicted in Fig. 80. ~254)Generally two different variants are used. In the first case the container is moved within a fixed temperature profile which causes the melt to solidify from bottom to top. Crystallization can be spontaneous or initiated by seeding, however, so far single crystal growth could not be maintained throughout the ingot. In the second case both container and heating system are fixed and the temperature profile is varied by reducing the heat. A particular variant is the Heat Exchanger Method (HEM) t255~where the reduction of the heat is combined with a simultaneous localized heat extraction through a seed crystal at the bottom of the mould by a helium gas stream. This method results in a rather homogeneous distribution of large grains (several cm). The solidification rates range from 0.7 to 10 mm/min for ingots with typical cross sections between 5 x 5 to about 20 x 20 cm 2 and an average weight of about 20 kg. Casting techniques are rather similar and involve the pouring of the melt into the mould prior to the subsequent directional solidification. The solidification process is also controlled by a sequence of different temperature steps during the process. Crystallization occurs randomly at the bottom and the walls of the container. Grains grow in competition with each other and a preferred orientation develops which is controlled by the heat flow direction in the melt. Increasing the growth rate results in smaller grain sizes. Usually the casting process is operated discontinuously but continuous casting may be an attractive alternative if the crystallite structure can be controlled sufficiently. Cast ingots typically weigh about 8 kg with cross sections of about l0 x l0 cm 2.
352
PROGRESS
IN MATERIALS
SCIENCE
The ingot technique has to be coupled with an efficient and cost effective slicing method. Recent developments include multiple blade and wire techniques which reduce the cost of the slicing process and the waste of silicon material. The ingots are usually cut into wafers of about 10 x 10 cm 2 and a thickness between 300 to 400/zm. These wafers are processed then into solar cells by techniques which have been described in Chap. 3. Materialproperties. Polycrystalline silicon produced by ingot techniques has been commercially available for more than a decade and used for solar cells. The conversion efficiencies of solar cells made from these materials lie currently (1990) between 13 % to 13.5 %. Applying new design concepts which have been used for single-crystal cells can further improve the efficiency but one has to consider the trade-off between performance and cost. Most of the wafers are produced by only a few manufacturers and sold under commercial brands such as SILSO. Therefore many of the investigations on polycrystalline silicon which are published in the literature have used these commercial materials. Although many aspects of these polycrystalline materials are thus investigated, the results are difficult to compare since the materials are rarely completely characterized with respect to their impurity content or the condition under which the crystals have been produced and treated after crystal growth. It has been realized in the meantime that more systematic investigations of the defect behavior are necessary for a further improvement of the solar cell performance and research programs in France and Germany have been initiated to support the research. The poly-silicon ingots have typical grain sizes between 1-10 mm, in some cases (e.g. H E M silicon) even larger (Fig. 81). This means that in a final wafer grain boundaries extend from the top to the bottom and a columnar structure results which reduces the influence of the
FIG. 81. Grey scale coded LBIC mapping of a polycrystallinesilicon wafer from cast ingot material showing the increased recombination of some grain boundaries (indicated by black lines). (Courtesy of H. Koch.)
SEMICONDUCTORS
353
grain boundaries in the final solar cell. The grain boundaries are mainly large angle grain boundaries but can be close to low energy orientation because of the directional solidification which favors growth of grains with particular orientations. For instance ( l l 0 ) twin boundaries and very thin (<1/~m) twin lamellae are frequently observed. Dislocation densities usually vary in the range between l03 to 104 cm -2 but may reach 108 cm 2 locally. The origin of the dislocations and possibly the twins is generally attributed to the local thermal stresses which occur during the growth, however, no detailed models are available for their nucleation. Depending on the purity of the starting material and the contamination through the crucible walls and ingots contain a variety of impurities at various concentration levels. In addition to the impurities which can be observed in "electronic-grade" silicon unconventional main group elements can be incorporated such as Mg, Ca or S, Se, Te. As has been discussed in chapter 4 the decreasing solubility during cooling for most of the impurities produces a high driving force for precipitation, complex formation with other point defects or segregation at grain boundaries and dislocation. Because of the large number of impurities which are usually present complex interactions may occur which for the most part are unknown so far. The impurities can also stay in solution or form only very small microdefects if their diffusion coefficients are low and precipitation is prevented during cooling. Small defects in the nanometer range are difficult to observe and therefore have been rarely analyzed. Larger microdefects (> 1000 A) have been studied in cases where the concentration of a particular impurity is relatively high and precipitation occurs during cooling. Besides the doping elements carbon and oxygen are generally the impurities with the highest concentrations. The analysis of their distribution in SILSO ingots for instance has shown that the highest concentrations for oxygen occur at the bottom whereas the carbon concentrations increases towards the top. (253) Investigations of silicon ingots from other manufacturers yielded different distributions, so that the segregation behavior is likely to depend on the solidification process. Whereas SiOz precipitates are usually not observed after the ingot growth because of the higher solubility and lower oxygen concentrations, silicon-carbide precipitates do occur with typical diameters of about 1 #m. They are distributed in the bulk with densities up to 104 cm -2 and do not preferentially occur at grain boundaries or dislocations. Recent TEM and SEM investigations have identified them as either the cubic fl-SiC or the hexagonal 15R phase (Fig. 82). (256) In ingots with very high carbon concentrations (> 200 ppm) and depending on the cooling rate very large precipitates can also occur with diameters exceeding l0 #m. Their analysis with TEM, EDS and Auger spectroscopy demonstrated that these precipitates consist of SiC and a second phase such as AIN (Fig. 83) or SiO2N. The occurrence of the second phase is clearly related to a high concentration of these impurities such as AI or N in this case or the presence of small nuclei of the compound. In view of the difficulties to form SiC precipitates in solid silicon as discussed in Chap. 4.1.2 it is possible that these SiC precipitates have already formed in the melt by heterogeneous nucleation either at the melt-interface or at (preformed) nuclei in the melt. A theoretical calculation of the diffusion limited growth for these two cases (228) indeed shows that the observed diameters can be explained quantitatively. The diameters depend on the carbon concentration and the growth rate (eq. (190) in Chap. 4.5.1), and in the case of a heterogeneous nucleation in the melt on the thermal gradient. An important result is that the reduction of the carbon concentrations below the eutectic level (about 50-60 ppm) is sufficient to suppress the SiC carbide formation from the melt which is in general agreement with the experimental observation.
354
FIG. 82. SEM image of large//-SIC precipitates in polycrystalline solar cell silicon. Precipitate size about 1/~m.
Although it is possible to prevent the oxygen and carbon precipitation by reducing their concentrations it is clear that the solidified silicon is a supersaturated solution. Since there are no systematic experimental observations available which show the occurrence of very small defects below about 100/~, it is unclear where and in which form these and other impurities are present in the as-grown crystals. It has however been observed by etching for
FIG. 83. Optical micrograph of the co-precipitation of SiC flight) and AIN (dark) in a single particle in ingot polycrystalline silicon3256~ (Scale marker ~ 10/~m).
SEMICONDUCTORS
FOR
SOLAR
CELL
APPLICATIONS
355
instance that annealing treatments of as-grown wafers at high temperatures (1200C) produce point-like defects. (253'254) Usually the concentration is reduced in the vicinity of grain boundaries (denuded zone) which indicates a segregation or precipitation at the interface, however, again not much is known about the nature of these defects. Therefore one must admit that the oxygen and carbon related defect formation in as-grown and annealed poly-silicon is almost unknown compared to Cz- and Fz-grown crystals (see Chap. 4.1.2 and 5.1). Since annealing treatments are an integral part of the solar cell processing it is certainly necessary to know more about the formation of these microdefects which can be electrically active and effect the lifetime of material and the solar celt performance. Another entire group of impurities are the transition elements which are present at various concentration levels, very mobile even at low temperatures and electrically active (see Chap. 4.1.3). Due to their small segregation coefficients (between about 10 -3 and 10 8) their concentrations in the solid can usually be kept below certain levels ( < 1016 c m 3) depending on the element, but it is still very important to control and reduce the impurity levels in the melt. The deep levels which are introduced by the transition metals and the concentration of these levels depend on whether the impurity occupies an interstitial or substitutional lattice site or forms a complex (e.g. cluster, precipitate) with other elements. Measurements of the influence of these trap levels on the performance of solar cells have shown that there are certain impurity levels (threshold values) below which the influence on the material can be neglected. These threshold values have been determined for a variety of transition and other elements in p- and n-base solar cells (produced from Cz-grown single crystals) and are summarized in Table 12(156) and in Fig. 84. The results indicate that the highly mobile elements (Cu, Ni, Co etc.) which tend to precipitate or form complexes introduce electrically active defects in concentrations much below the total impurity concentration, whereas slower diffusing elements (e.g. Ti, V, Mo) which remain dissolved to a larger extent introduce trap levels in concentrations closer to the
Table 12. Experimetnally determined threshold concentrations of impurities (Nth) above which the efficiency of p- and n-based silicon solar cells decreases. The solar cells are manufactured from single crystalline Cz-grown materiaD 156) Included are the maximum solubility and the diffusion coefficient at 1 1 0 0 C of the elements if available ~m)
Impurity Copper Nickel Cobalt Iron Manganese Chromium Vanadium Titanium Silver Palladium Molybdenum Niobium Zirconium Gold Tungsten Tantalum Carbon Oxygen Aluminium Phosphorus
P-base solar cell Nth [cm -3]

4.1 5.0 1.1 2.5 1.8 1.3 2.5 2.6 4.0 2.0 6.0 1.6 3.6 1.0 1.2 2.3 1 1 4.4 1.7 x 1017 10 I5 1015 1014 1014
N-base solar cell Nth [ c m -3]

8.0 --1.8 9.5 1.2 4.1 3.7 --1.3 -------8.5 -1016
Max solubility
N ~ ax [ c m -3 ] 2 7 3 2 2 1 2 1 5 3 2 5 5 1 9 1 3.5 2.0 1.5 1.8 x 1018 1017 1016 1016 1016
Diff. coefficient
D . 0 0 [cm2/s] 1 4 1 4 3 2 1 4 3 x x 10 4 10 -5 10 5 10 6 10 6 10 6 10 - 7 10 9 10 9
--
1014 1014
1014
1014
1016
x 1012
1012
1013 1013
x x x x x
1015 1015 I 0 I1 1012 1011 1013 1012 1011 1018 1019 1015 1018
1016 1015 x 1017

1016
1012
x x x x x x
1014 1014 1014 1017 1014 1017 1018 1021 1022
---I 10 6
--
1013
1016
4 8 9 2 3
x x x
10 14 10 12 10 -11 10 12 10 -13
356
(a)
Met~
Impurity Cor~ntr~ion
(~*1
1.0
0.!
-~ o.~
0.4
!
0.2 0.0
, . ,1
P-
TypeSilicon
10I1
, l0 B
Ill
i012
1014
Metal I m p u r i t y
1015 1016 Concentrllion (Jtocns/cm)~
1017
101I
(b)
:o- s lo- 4
Melal Impurity Concentration (ppma)

m- 3 lo- 2 1o 1 LO 1o
1.0 m
o
o.s
.~, 0.6 0.1
0.2
Ni
N - TypeSilicon
iO11
1012
1013
1014
Metal Impurity Concentrilion
1015 1016 latomslcrr )1
10I?
10TM
FIG. 84. Experimental solar cell efficiencies vs impurity concentration for a 4 f~'m p-base (a) and 1.5 flcm n-base device (b). 56)
total impurity concentration (see also Chap. 4.1.3). In some cases the ratio of electrically active centers NT with respect to the total impurity concentration Nx/Ntot has been measured and the following values 0.23 (Cr), 0.28 (V), 0.4 (Ti) and 1.0 (Mo) ('56) which were obtained show indeed the almost total electrical activation of these elements. Although in general the ratio Nr/Ntot should depend on the thermal history (cooling conditions during solidification, annealing treatments) of the solar cells, these values can serve as guidelines for the impurity concentrations which can be tolerated without loss in the efficiency. Rather detrimental are the slow diffusing elements like Ti, V, Mo, Nb etc. which degrade the lifetime already in concentrations of about 1011 to 1012cm -3. Most of these elements also have a low solubility so that their total concentrations can be kept low, but vanadium and titanium for instance can occur in concentrations of up to 1015 to 1016 cm -3 so that they can introduce rather high concentrations of trap centers. In practice one has to monitor all of these impurities and eliminate their source if they occur in higher concentrations. The threshold values have been determined in single crystal cells containing only one impurity. On the contrary a polycrystalline material contains a variety of impurities and their interaction with other point and lattice defects can lead to new electrical defects or even a
SEMICONDUCTORS
357
passivation of other impurity levels. For instance it has been observed that additions of copper can reduce the recombination centers which are introduced by Ti, V and Zr in Cz-silicon. The limited amount of experimental data which is available for threshold values for polycrystalline cells indicates that the results do not differ too much from single crystal cells, but one should be careful so far to generalize these observations. A further important aspect for the electrical behavior of polycrystalline ingot silicon is the influence of dislocations, grain boundaries, and larger precipitates (e.g. SIC). It is generally assumed that dislocations occur as a result of thermal stresses during solidification although the nucleation mechanism is not known yet. Although local densities above 1018cm2 can be observed which indicates the presence of a localized active dislocation sources none of these centers (for instance a precipitate) could be identified yet. As discussed in Chap. 4.3 dislocations are very mobile in silicon above about 500-600C, which facilitates their nucleation, but a high contamination with impurities during cooling can effectively reduce their mobility. In as-grown silicon many dislocations are thus effectively pinned and the reduction of their density by further annealing treatment is only partly successful. It is therefore important to minimize thermal stresses during growth and thus keep the dislocation density low. Dislocations have been shown to be very active recombination centers. Measurements of their influence on the lifetime which have been shown in Fig. 12 indicate that the dislocation density has to be kept below about 105 cm -2 for a high quality ingot material. Like grain boundaries, dislocations and large (e.g. carbon and oxygen) precipitates can also act as preferential nucleations sites for metal precipitates. Precipitation of copper-silicides during cooling has been observed in copper rich silicon (above 1016 cm -3) and is shown in Figs 85-87 for a dislocation, grain boundary, and a SiC-precipitate. In a typical as-grown ingot material with lower impurity concentrations precipitation is rarely observed but it cannot be ruled out that post-annealing treatments (e.g. during solar cell fabrication) can lead to precipitation and defect formation with an enhanced electrical activity. Whereas the already high electrical activity of dislocations may not be affected as much by further impurity contamination, it appears that SiC precipitates first become recombination centers after decoration with impurities. Although dislocations and large precipitates can be very detrimental for the device performance the investigation of their properties has not attracted as much interest as grain boundaries. A large number of investigations on single grain boundaries in bicrystals and in polycrystalline silicon has demonstrated their electrical activity as recombination centers. The main results of these investigations have been discused in Chap. 4.4. Topographic evaluations of the grain boundary activity in ingot silicon for instance by four-point probe resistivity mapping or the EBIC and LBIC technique (Fig. 81) show a non-uniform behavior: - - Small angle grain boundaries (0 < 10) which consist of an array of dislocations show an efficient current degradation (20%-30%). Random, large angle grain boundaries are usually also strong recombination centers but their electrical activity can vary considerably at the same boundary and for different boundaries. - - Coherent twins which occur rather frequently and other low energy (near-) CSL boundaries are usually not or only weakly active. Despite the rather sophisticated description of the electrical activity of grain boundaries which has emerged in the past few years the fundamental questions about the origin of the electrical activity have not been completely answered. The results about the electrical behavior of the grain boundaries in a polycrystalline ingot indicate both the importance of the structure as well as of the contamination with impurities. Segregation and precipitation of impurities at
-
JPMS 35-3/4~K
358
FIG. 85. TEM micrograph of copper precipitates at dislocations in polycrystalline silicon. Because of large mismatch stresses between the r/"-Cu3Si particle and the silicon matrix dislocations are also punched out during precipitation. The dislocation breaks away from the precipitate and offers further nucleation sites. The repetition of the process leads to the formation of precipitate colonies3253)(Scale marker ~ 2 # m).
grain b o u n d a r i e s has been d e m o n s t r a t e d by S I M S a n d A E S investigations.(233'257)The results o f the copper p r e c i p i t a t i o n at i n d i v i d u a l g r a i n b o u n d a r i e s have s h o w n the influence o f the structure o n the p r e c i p i t a t i o n b e h a v i o r (7~'235'25s) a n d the electrical activity. Since there is a d r i v i n g force for m o s t o f the elements to precipitate d u r i n g cooling one c a n expect that
359
Fro. 86. TEM micrograph of copper precipitates (Cu3Si phase) at a large angle grain boundary in polycrystalline silicon, also showing the punching out of dislocations. (Scale marker ~ 1/~m). segregation and precipitation occurs at grain boundaries. However, T E M investigations o f as-grown poly-crystals with low impurity concentrations rarely show precipitates at grain boundaries or dislocations. There is however an interaction with grain boundaries which is reflected in the reduced impurity levels and increased lifetimes in the vicinity o f grain boundaries (about 10-100 # m , see Figs 88 and 89). One must admit, however, that at present
FIG. 87. TEM micrograph of copper precipitates (Cu3Si phase) at a SiC precipitate in polycrystalline silicon. (Scale marker 20.5 #m).
FIG. 88. Contaminated polycrystalline (SILSO) wafer annealed at 1250C/1 hr. Etching reveals depletion of point-like defects (small pits) in the vicinity of large angle grain boundaries. <253) (Scale marker ~ 100 gm).
L~S
4.;P FIG. 89. LBIC mapping at a grain boundary in cast polycrystalline silicon showing an enhanced recombination at the grain boundary, and a reduced recombination in the vicinity (Courtesy of H. Koch). 360
SEMICONDUCTORS
361
a model of the electrical activity which would give guidelines for instance for the preferential growth of grain boundaries with lower electrical activity is not available. The concentrations of impurities and other point defects in the bulk determine the upper level of the lifetime or diffusion length which can be achieved in a particular material (currently about 100 #m). The grain boundary recombination contributes to the total lifetime reduction in a polycrystalline material and there is a trade-off between the effort to increase the grain size and the gain in the average lifetime. For the current impurity levels which can be achieved in a low cost ingot process grain sizes above a few millimeters appear to be sufficient to reduce the grain boundary activity to an acceptable level. Several ways have been proposed to reduce the influence of the grain boundary recombination after the ingot growth in the processed wafers. One possibility is to use the enhanced diffusion of impurities along grain boundaries. For instance in a p-base material phosphorous is diffused into the crystal preferentially along the grain boundaries. The resulting p--n junction around each grain boundary acts as a collector for minority carriers before they can recombine at the interface. This process has been shown to reduce the electrical activity of the grain boundaries. Another possibility is the passivation of grain boundaries with hydrogen which has been described in Chap. 4.1.4. The hydrogenation is carried out in a plasma or by ion implantation (Kaufman source). These processes have been shown to reduce the recombination at grain boundaries (and of other defects such as dislocations or precipitates) and improve the efficiency. For a polycrystalline material which has already a rather high efficiency (e.g. about 13%) there is only a marginal increase (0.1 to 0.5%, depending on the material) which in many cases does not justify the increased processing costs. One problem with the hydrogen passivation technique is the low penetration depth of hydrogen of only a few microns. Responsible for this behavior is the complex diffusion mechanisms which involves trapping at defects and the formation of the rather immobile H2 molecule. Therefore only near surface recombination centers are passivated and most of the defects in a 300-400 #m thick solar cell are possibly not affected. Since at the same time shallow acceptor impurities are also passivated which may change the properties of the near-surface p-n junction the influence of the hydrogen on the performance of a solar cell is complex and certainly not completely understood. 5.2.2. Ribbon and sheet growth techniques Despite the promising results for the efficiencies of solar cells which are produced from ingot silicon this technology still has the disadvantage that it requires high-efficiency slicing operations and wastes about 30 to 40% of the material. Many research efforts have therefore been directed towards the development of growth processes where the material is directly grown in form of sheets or ribbons. A large number of technical approaches have been developed which have been described in many review and proceeding papers. (e.g. Refs (253, 259-261). None of these techniques has reached a commercial level yet, but a few have been developed so far that solar cells with more than 10% efficiencies can be produced. Some typical technical solutions will be described next. Most of the sheet techniques require some guiding system to shape the solidifying melt. One way to distinguish them for instance is by describing them according to the various degrees of interaction between the solidifying melt and the guiding system. Here another aspect shall be emphasized which is related to the direction of the sheet growth and the direction of the proceeding solidification front.
362
PROGRESS
IN MATERIALS
SCIENCE
5.2.2.1. Sheet growth and solidification direction parallel The major techniques described in this section are characterized by the existence of a molten film which is shaped by a die. The solid-liquid interface proceeds parallel to the sheet pulling direction, therefore the solidification rate, V, determines the sheet growth velocity, Vs = V. In general the velocity with which the solidification front advances is determined by the thermal gradients and typically varies between 10-100 mm/min. There are several techniques which differ by the degree of wetting between the shaping die and the melt. In the "dendritic web growth" (WEB) the contact with the shaping die is limited to the edges of the sheet. They are stabilized either by dendrites which grow simultaneously with the ribbon or by graphite or quartz filaments which fed through the melt to the crystallization front ("edge-supporting pulling" (ESP) method). The contact between melt and crucible is also reduced in the "silicon sheet from powder" (SSP) method. In a first step a supporting ribbon is formed by sintering silicon powder on a high purity quartz or coated poly-silicon bed. In a second step substrate and pre-formed ribbon are separated, and the sheet is recrystallized by pulling a molten zone along the ribbon. Melting is achieved by focussing light from a line source on the sheet. The solidification proceeds in the direction of the ribbon and the rate is determined by the speed with which the molten zone can be pulled along the ribbon. ~26-262) Other techniques use a shaping die for the sheet growth. A schematic diagram of the "edge-defined film fed growth" (EFG) process is shown in Fig. 90. A modification of the process is the "capillary action shaping technique" (CAST) which differs from the EG-process primarily by the shape of die top which determines the meniscus shape and height. The CAST process allows a greater meniscus base and a looser coupling between the die top and solid-liquid interface which results in a smoother ribbon surface. In both cases the melt is supplied by capillary forces through grooves or channels inside the die and the meniscus is anchored at the outer edges of the shaped die. In order to illustrate some of the issues which are relevant for achieving a continuous, stable growth the EFG method shall be discussed further. The meniscus height adjusts its position with respect to the pulling speed as long as enough melt can be supplied by the capillary action. This adjustment is important since the process becomes more stable then against fluctuations in temperature or speed. The
(a)
(b)
FIo. 90. Schematicdiagram (a) of the "dendritic web growth" (WEB) and (b) the "edge-defined film-fedgrowth"(EFG) techniques.Similartechniquesare the "edgesupportedpulling"method(ESP) and the "capillaryaction shapingtechnique" (CAST), respectively. 0~
SEMICONDUCTORS
363
stability of the meniscus is determined mainly by the temperature gradient near the solid-liquid interface. Heat flow calculations for the meniscus shape as a function of the geometry of the die and the thermal conditions have established the parameters which are important for the process especially the thickness of the ribbons, d. The following approximate relation has been obtained from a one dimensional heat flow model (263) T,-- Tm
dhln(dt/d){(Cm~CdKTSm)'/z [1 --fl XL(4--d) " --
4(1
+41nfl)]'/2--pVL
(200)
where Cmis the emissivity of the ribbon surface, L the latent heat of fusion, p the density of the solid, Xs and x L are the thermal conductivities in solid and liquid, h is the meniscus height, dt the thickness of the die top, fl = Tm/To, and Tm, Tt, To are the melting temperature (at the melt-interface), the temperature at the die top, and the average temperature of the environment, respectively. The thermal gradient adjacent to the solid-liquid interface is given by
G = (I", rm)
d, - d
d h ln(dt/d)
(201)
Equations (200) and (201) allow to calculate the ribbon thickness as a function of the growth speed and the environmental temperature To. For the E F G technique typical values for a ribbon thickness of about 430 ~tm (typical for a solar cell) and To = 1100C yield a growth speed of 4.7 cm/min and temperature gradients between about 20-250C/cm at the ribbon edge and surface, respectively. The maximum growth rate is obtained for G = 0 which yields Vmax=
(.m xs KTSm'~'/2 ~ d-p~2~ j [1 -
fl-4(1 + 4 ln/~)] 1/2.
(202)
Figure 91 depicts the maximum growth rate for silicon as a function of the ribbon thickness, d, and Tm > To. The growth speeds are by about a factor of ten larger than for the ingot technology which demonstrates the advantage of these methods with regard to the throughput of material that can be produced. The width of the ribbons is determined by the width of the die and the growth speed and can vary between 7 to 10 cm. It is evident that in both techniques the die material is of great importance. High density graphite, vitreous carbon and silicon carbide are the most promising die material. The
1412 10
u~
8 6
0 4b
I i
0.02
0.C.
O. 0 . 0 8 0.10 0.12 P,!bt::~-',n tk'ici,~es.s (c.m)
0 %
0?6
FIG. 91. M a x i m u m growth rate vs thickness for an E F G silicon ribbon3264)
364
interaction with the silicon melt introduces carbon up to the solubility at the melting point (about 50-100 ppm) therefore carbon rich ribbons are grown with carbon concentrations in the solid of about 5-100 ppm. The enrichment of carbon at the melt-interface can lead to the growth of large SiC precipitates in the melt which disturb the meniscus and the smoothness of the resulting ribbon surface or can even be incorporated into the solid. Ribbons grown by the CAST method contained about 0.1 particles/cm2 compared to EFG material with 5/cm2.07,264) 5.2.2.2. Microstructure and electronic properties of EFG and CAST ribbons. EFG, CAST, WEB, ESP, SSP or related techniques in general produce high quality material. The ribbons are either single crystalline or characterized by a very low density of grain boundaries, most of them being coherent twin boundaries with (m) planes. For instance, in EFG ribbons the twins are arranged in thin lamellae which ran parallel to the growth direction. The other main crystallographic defects in this material are dislocations with average densities of about 104 cm 2 and local densities up to 106 cm 2, stacking faults and SiC precipitates. The precipitates (mostly near the surface) can become rather large between about 30 to 300/~m, however, their density can be reduced by elevating one side of the die over the other which causes a considerable reduction of the SiC precipitates on both side of the ribbon. All of these defects are electrically active to some extent, which is partly due to the contamination with impurities. TEM investigations of high quality ribbons in general reveal the absence of precipitates other than SiC which is comparable to the situation for ingot material, and in most cases it is unclear which defects limit the efficiency of the solar cells. Because of the high quality of the ribbons grown by these methods, the solar cell efficiencies range between about 14-17%. A disadvantage of these techniques is the still rather low growth velocity. Although the total throughput in a production facility can certainly be increased for instance by continuous melt replenishment and growth of longer ribbons, there are inherent limitations. Estimates of the total solar cell area which can be produced in a given period of time show that these sheet growth techniques are probably not superior to the ingot growth in connection with improved slicing techniques. 5.2.2.3. Sheet growth and solidification direction perpendicular. Higher ribbon growth velocities can be obtained if the sheet velocity, Vs, and the solidification velocity, V, are decoupled. A number of technical solutions have been proposed where the melt solidifies (almost) perpendicular to the sheet growth direction. The major techniques are characterized by a moving support ribbon (substrate) on which the silicon solidifies. The techniques differ basically in the way how the substrate is brought into contact with the melt. A schematic diagram is depicted in Fig. 92 for methods where the melt is cast on the substrate. In the "ramp assisted film transport" method (RAFT) the substrate ribbon is moved upwards inclined by an angle of about 45 , whereas in the "ribbon growth on substrate" (RGS) technique the substrate is moved almost horizontal. The width of the ribbon is shaped by
FIG. 92. Schematicdiagram of the "ramp-assistedfilm transport" (RAFT) and "ribbon-growthon substrate" (RGS) techniques.
SEMICONDUCTORS
FOR
SOLAR
CELL
APPLICATIONS
365
a die into which the melt can be poured. These techniques allow a continuous operation since the die can be replenished continuously and the substrate ribbon can rotate like a conveyer belt. The requirements for the substrate material are similar to other techniques: a high degree of inertness against the silicon melt and no adhesion so that ribbon and substrate can be separated easily. Therefore the most likely candidates are again graphite materials. The contact between solidifying melt and substrate is a crucial problem in this technique since it determines to a large extent the resulting microstructure of the ribbon and the contamination with impurities. This problem is reduced in another technique where the substrate is replaced by a graphite net (S-WEB method) which is pulled over the melt, so that capillary forces can hold the melt between the meshes of the net before solidification. In all methods the solid-liquid interface proceeds perpendicular to the sheet pulling direction, therefore the solidification rate is in general independent from the sheet growth velocity. In practice, however, there exist other constraints which usually limit the growth rate. For instance for the RSG technique, it is desirable to solidify the melt inside the die, therefore the sheet velocity, Vs, is to some extent determined by the geometry of the die and the solidification rate, V, which depends on the temperature difference between melt and substrate temperature AT = Tm- Ts. The sheet velocity has been calculated in a simple heat flow model yielding Vs =
4~Xs S AT pLd(2~c s - ~d)
(203)
where s is the length of the growth wedge, d the ribbon thickness, and ~ the effective coefficient of heat transfer between silicon and substrate or die material. Figure 93 shows the ribbon thickness as a function of the sheet velocity for different AT. Typical sheet velocities for a ribbon thickness of about 300 #m are between 500-2000 cm/min, which is about a factor 100-400 faster compared to the EFG or CAST technique. The growth of 10 cm wide ribbons with a homogeneous thickness has been reported. (265) 5.2.2.4. Microstructure and electronic properties o f R A F T or R G S ribbons. Only a few results are available about the quality of sheets grown by the RAFT or RGS method. A typical feature of these techniques is a small grain size between 100-1000 #m. Since the substrate provides a high density of nucleation sites for the solidifying melt and solidification is rapidly completed grain growth due to selective growth of larger grains is rather limited. The microstructure depends to some extent on the temperature profile which can however only be varied within certain limits. A further improvement of the grain size is thus only feasible through a reduction of the nucleation sites at the substrate for instance by a selection of a
[gm] 1000-800600-400-200--
m AT ~ 200 K
7
:.1
1.0 2.0 3.0 4.0 5.0
6.0 7.0 8.0 [m/min]
FIG. 93. T h i c k n e s s vs s u b s t r a t e v e l o c i t y f o r R G S r i b b o n s f o r different t e m p e r a t u r e g r a d i e n t s A T p e r p e n d i c u l a r to the s u b s t r a t e . (265)
366
suitable coating of the substrate material. It is apparent that the fundamental processes which determine the nucleation are not known so that an improvement can only be expected by empirical experience. Besides the high grain boundary density the ribbons can also contain high dislocation densities. Average densities can range from 105 to 10 6 c m 2, with local density up to 108 cm2. It is generally assumed that these high densities are caused by high thermal stresses in the ribbon during solidification. An optimization of the temperature profiles and post-annealing treatments may reduce the dislocation densities in the future to acceptable levels. SiC precipitates are also observed in these materials due to the intensive contact of the melt with the graphite die and substrate. Empirical observations yield precipitate diameters of about 1 pm. Due to the rapid cooling of the ribbon it is unlikely that these precipitates have formed in the solid. Instead one must assume that precipitation occurs at the melt-interface as described in Chap. 4.1.2. Considering the solidification velocity ( ~ 0.005 cm/s) and typical carbon concentrations AC = 100 ppm which can be expected in the melt in contact with a graphite die one can calculate the maximum diameter from eq. (190) and obtains about 0.5-1.0/~m which is in agreement with the experimental observation. No systematic results about the electrical properties of this type of sheet growth silicon have been published. Considering however the difficulties to control the microstructure and impurity levels one can presume that at this early stage of development the efficiencies are still below the values which are reported for polycrystalline ingot material.
6. SINGLE CRYSTAL AND EPITAXIAL GALLIUM ARSENIDE
III-V semiconductors such as GaAs, InP and multinary compounds such as AIGaAs, InAsP or GalnAsP have been receiving considerable attention for high speed and opto-electronic applications. These materials have also exceptional characteristics for photovoltaic applications. Because of their high cost, solar cells utilizing these III-V semiconductors are usually considered for terrestrial concentrator cells or for space applications, in which the cost of the cell is less important because of the requirement for high performance or for a good resistance to increased radiation in space. The majority of solar cells operate with higher efficiencies under higher irradiance conditions (see Chap. 2.1) which requires, however, focusing lenses and direct sunlight. Concentrator cells thus do not utilize the diffusive component of the sunlight which accounts for about 20% on a sunny day. Therefore these systems can only compete with their flat panel counterparts in very sunny locations. Nonetheless, attempts are made to reduce the cost also for III-V semiconductors so that they may offer alternative solutions even for low cost cells in the future. Most of the III-V semiconductors have a direct bandgap and hence a high absorption coefficient so that only thin films of about 1 pm thickness are required to absorb the incident light. Gallium arsenide has a bandgap (Eg = 1.4 eV) that matches the solar spectrum almost perfectly therefore this semiconductor has attracted most of the interest. In addition, it has a very high electron mobility ( ~<8500 cm2/Vs) which makes it very suitable for the p-base of a solar cell. Since during recent years the deposition techniques to fabricate high quality GaAs layers have become available because of the parallel development for device applications the same film and device techniques are also employed for solar cell structures. High quality GaAs films are epitaxially grown on GaAs single crystal substrates. Their quality is important for the successful growth of good epitaxial layers. The study of single crystalline GaAs has also provided insight into the basic properties of the material and its defect structure. Therefore in the first part of the chapter the main results on growth of single crystal GaAs
SEMICONDUCTORS
367
and the structural and electronic properties shall be discussed. In the second part the properties of epitaxiaUy grown GaAs and related compounds such as A1GaAs are presented. Other III-V compounds are also explored for solar cell applications but their development is still at a rather early stage and will therefore not be included here.
6.1. Single Crystal Growth Techniques Most of the single crystal GaAs today is either used for opto-electronic applications or as a substrate material for integrated circuit (IC) devices. Because of the different requirements for the material basically two crystal growth techniques are employed which are most suitable for the particular purpose. Doped material for the opto-electronic application is usually grown by the horizontal Bridgman technique (HB). The Bridgman technique uses low temperature gradients which results in very little thermal stresses in the growing single crystal. Therefore one can keep the dislocation density low which is essential for the stability of the optical performance of laser diodes for instance. In this technique the GaAs crystal is grown in a boat or crucible and therefore some kind of contamination is usually present. Since in addition GaAs expands considerably upon freezing the crucible material has to withstand quite high stresses. The requirement limits the choice of materials and thus the possibility to avoid contamination by suitable crucible materials. For the IC technology undoped semi-insulating (SI) GaAs is required which is difficult to grow by the Bridgman technique because of the contamination problem. For this purpose the liquid encapsulated Czochralski technique (LEC) has been developed which does not require a boat and allows the growth of purer crystals. But this method on the other hand needs higher temperature gradients for the crystal growth which results in higher dislocation densities. Typically the dislocations densities in LEC material are an order of magnitude larger compared to HB gallium arsenide. Since dislocations are less important for the IC fabrication most of the commercial GaAs substrates are grown by these techniques. So far semi-insulating GaAs serves mainly as a substrate material for the following epitaxial growth of the GaAs layers which will be discussed in the next section. Nonetheless, the semi-insulating properties of undoped GaAs are decisive for the fabrication of IC's and thus have been studied intensely during recent years. The material related issues for GaAs and other III-V compounds are more complex compared to elemental semiconductors for several reasons. Besides the usual lattice defects such as impurities or dislocations compound semiconductors can also contain numerous intrinsic point defects such as vacancies on the different sublattices and antisites. These defects are mainly introduced when deviations from the stoichiometry occur. Since in many cases the growth of compound semiconductors is difficult because one of the constituents is rather volatile (for instance arsenic in GaAs) the control of the stoichiometry is a complex task. The requirements for GaAs which can be used for solar cells differ from those for device application. If one considers the conventional design of a p-n junction or heterojunction solar cell for instance in Fig. 23 (see Chap. 3) where the two contacts are placed on opposite sides of the structure and the photo-generated current passes through the entire cell it is evident that a good conductivity in all layers is required. Therefore only doped GaAs can be employed and the semi-insulating material itself has no importance for photovoltaic applications. The investigation of semi-insulating GaAs has provided, however, enormous insight into the structural and electronic properties of this material and of compound semiconductors in general, therefore some of the main results shall be discussed in the next section.
368
PROGRESS
IN MATERIALS
SCIENCE
The EL2 defect. The electronic properties of melt grown undoped GaAs are dominated by the presence of a defect, referred to as EL2, which introduces a donor level in the middle of the band gap. This defect occurs with a rather constant concentration of about 1 to 5 1016cm -3 in single crystals. The characteristics of the EL2 defect are two donor charge states at Ev + 0.52 eV and Ev + 0.75 eV which are responsible for the compensation of residual acceptor impurities and which can be observed by optical absorption or photocapacitance measurements (Fig. 94).0t5'266) Energy levels reported in the literature vary for instance for the upper level between Er = 0 . 7 6 - 0.82 eV below the conduction band edge. EL2 occurs in As-rich crystals which are grown by a variety of techniques such as CZ liquid encapsulated crystal (LEC) growth, horizontal (HB) or vertical Bridgrnan (VB) growth, or vapor phase epitaxy (VPE, MOVPE). EL2 appears to be an intrinsic defect of the semiconductor and in order to obtain a semi-insulating material it has to be compensated by acceptor defects or impurities. Its presence thus determines the electrical properties of semi-insulating GaAs considerably. Because of the importance to control the defect from the perspective of the production of high quality GaAs material, the nature and properties of the EL2 defect have been studied intensively and received greater attention in the last 10 years than probably any other aspect of semiconductor materials. Some of the conclusions which point towards the intrinsic origin of the EL2 defect are based on the observation that the concentration of the defect is linked to the stoichiometry of the compound. In particular it has been found that the EL2 concentrations depend on the ratio As = [As]/[Ga] and increases linearly with the excess of arsenic3267) The concentrations are also higher if the GaAs crystals are prepared at elevated temperatures necessary for melt growth which evidently promotes atomic disorder. On the contrary GaAs prepared by thin film deposition techniques at lower temperatures (see next section) contains lower concentrations. Table 6.1 shows typical concentrations for the different growth techniques. The concentration of the defect is also affected by the cooling conditions after solidification. Fast cooling can lead to the generation of dislocations either by excessive thermal stresses or the supersaturation of vacancies which coalesce into dislocations. In general, moving dislocations generate point defects (vacancies or self-interstitials) probably by climb processes and in GaAs it has been observed that dislocation motion and the formation of the EL2 and other midgap level defects are linked to each other. This becomes particularly evident in doped crystals where n-doping above about 8 x 1016cm -3 (the intrinsic carrier concentration at 1100 K) decreases the dislocation and EL2 concentration whereas for p-doping the
Light
T 0.$ eV
0/+ + / ++
0.3 eV
1.4[ V e
T
Ev ~
~ E v
FIG. 94. Schematic representation of the energy levels of the EL2 and ASGa defect in GaAs. (jl5,116)
FIG. 95. Structural model for the Asoa-Asi(EL2 ) defect in GaAs (a) and the proposed metastable configuration (split interstitial). As-atoms (open circles), Ga-atoms (black circles), Asc,a (atom 1) and As i (atom 2) on a tetrahedral interstitial site (Tdt), and after illumination in a split interstitial configuration3 "6)
SEMICONDUCTORS
FOR
SOLAR
CELL
APPLICATIONS
369
opposite effect is observed. This has been explained by the assumption that gallium vacancies are generated which are considered unstable and react with neighboring arsenic atoms by which they form an arsenic antisite defect according to the reaction
+ + VEa + ASAs Asca + VAs + 4e-
Under the assumption that these defects introduce trap levels in the bandgap and become charged, it is obvious that the formation of the antisite defect is suppressed with an increase of the electron concentration. In addition the mobility of dislocations can depend on the doping level as well as has been observed in silicon and germanium (Chap. 4.3). It is now commonly assumed that the EL2 defect is related to the arsenic antisite and the application of many experimental techniques such as electron spin (ESR) and electron nuclear double resonance (ENDOR), DLTS or optical absorption have determined many of the structural and electronic properties of the EL2 and the AsGa defect. The antisite defect which has two extra electrons which are not used for the covalent bond formation is a double donor with the energy levels indicated in Fig. 94 and the energy level ( o / + ) is close to the level observed by DLTS measurements on crystals containing the EL2 defect. Despite these and many other similarities there are experimental results which indicate that the EL2 defect is not completely identical to the arsenic antisite but rather a complex ASGa-X. Several suggestions for the unknown component X have been made and the most widely accepted assumption seems to be now that an arsenic interstitial (X = Asi) is bound to the antisite defect. One of the characteristic features of the EL2 defect is that it can be transformed into a metastable state by optical excitation (with hv ~ 1.1 eV) at low temperatures <130 K (persistent photoquenching effect). In this configuration the trap level disappears completely from the bandgap but can be reactivated by annealing above > 130 K. It was suggested that the metastable state is formed by a slight shift of the atomic positions of the arsenic in the ASGa-Asi complex towards a split interstitial configuration (Fig. 95). A key problem in the identification of the EL2 defect is the lack of direct experimental evidence confirming the structure of the metastable state. This is mostly due to the difficulties with ESR measurements which were so successful in the case of silicon. In III-V compounds the spectra usually consist of very broad lines mostly due to the multitude of the nuclear spin moments of the group III and V elements. Although there is thus still a controversy about the exact nature of the AsG~-X complex and the metastable configuration the main aspects seem to be understood
now.
Finally it should be noted that EL2 is not the only electrically active defect which appears after crystal growth in gallium arsenide. Other deep levels have been observed by DLTS but the concentrations of these defects is usually less than 10% of the EL2 concentration as can be seen from Table 13. In most cases the identification of these levels is not possible so far or is still under discussion. A large number of trap levels can also be introduced by electron irradiation as has been discussed in Chap. 4.2. These defects limit the performance of gallium arsenide solar cells under irradiation for instance in space and their properties are thus important for the radiation resistance of the material. Besides the intrinsic point defects there are also impurities which can become relevant for the electrical properties of GaAs. Of the shallow donor atoms that may be unintentionally present silicon poses the greatest problems. This contaminant may be introduced if the GaAs ingots are sealed in quartz ampoules or tubes. At the melting temperature of GaAs of about 1238C quartz decomposes rapidly enough to release silicon and contaminate the melt.
370
Table 13. Typical concentrations of the EL2 defect for GaAs crystals prepared by various growth techniques and under different conditions. For comparison the concentrations and energy levels of some other electrically detectable defect levels are also included t115'322)
Growth method Horizontal Bridgman (HB) Czrochalski (LEC)
Vapor Phase Epitaxy (VPE, MOVPE) Liquid Phase Epitaxy (LPE) MolecularBeamEpitaxy EL2 (MPE)
Defect EL2 EL6 EL3 EL2 EL6 EL5 EL2 EL16 EL2
Energy level Ec - Er [eV] 0.77 0.28 0.53 0.82 0.28 0.59 0.82 0.39
Concentration [cm-3] 5 x 1016 5 x 1015 5 x I014 1 1016 7 1015 1 1014 2 x 1014 1 x 1013 -< 1014
Growth conditions Composition slow cooling, As-rich low thermal gradients fast cooling, high As-rich thermal gradients low temperatures (~700C) low temperatures (~ 700C) low temperatures ( < 600C) [As]/[Ga] 10-L102 Ga-rich
Carbon on substitutional arsenic sites is the most common unintentional acceptor impurity. This contaminant is introduced from heated graphite parts in the LEC pullers. The presence of carbon is usually beneficial since it compensates some of the EL2 defects and whatever donors may be present. This raises the resistivity but at the expense of the carrier mobility which is lowered by high impurity concentrations (see Chap. 4.1). Thus it is generally desired to keep both the carbon and EL2 concentration lOW. (26s)
6.2. Epitaxial Thin Films of GaAs and Related Ternary Compounds The conventional doping techniques for silicon such as diffusion and ion implantation are less suitable for gallium arsenide. The main problem is the evaporation of arsenic during the necessary heat treatments either for the diffusion or the removal of the lattice damage after implantation. This phenomenon leads to change of the stoichiometry near the surface and the introduction of native defects as discussed in the previous section. Although several techniques such as capping of the surface with dielectric films (Si3N4, SiO2), rapid thermal annealing, or annealing in arsenic overpressure can be applied to suppress the out-diffusion or arsenic, the current gallium arsenide technology utilizes to a large extent epitaxial growth techniques for the fabrication of opto-electronic devices and integrated circuits. The most widely used techniques are the deposition from the gas phase by vapor phase epitaxy (VPE, MOVPE) or from the liquid phase (LPE), and by molecular beam epitaxy (MBE). The typical processing temperatures and conditions are shown in Table 13. These techniques not only allow to deposit very thin and doped layers ( < 10 nm) but also a variety o f compositions and thus the formation of heterostructures. The most widely used configuration involves the deposition of AIGaAs layers on GaAs. These heterostructure configurations are extensively used for devices where the confinement of charge carriers or light in a two dimensional layer is required. MBE grown layers have the advantage of a rather precisely controlled impurity profile particularly of the aluminium across the AIGaAs/GaAs interface. Nonetheless, MBE material still suffers from the drawback that it contains a higher density of so-called oval defects compared to MOVPE material which has prevented its commercial use so far3269) Generally, the epitaxially grown layers have, however, superior qualities compared to the single crystal bulk material. For instance the concentrations of EL2 and other deep trap centers are at least an order of magnitude lower as can be seen from
SEMICONDUCTORS
371
Table 13. This is partly due to the lower process temperatures or the growth under excess gallium concentrations (etc. for LPE) which also suppresses the EL2 formation. From the perspective of the photovoltaic application GaAs heterostructures are the natural choice for the fabrication of solar cells. Since the absorption of light requires only a thickness of a few micron solar cells can easily be fabricated entirely by a suitable sequence of epitaxial layers. A typical device concept is depicted schematically in Fig. 23 and shows that these structures also utilize the optical properties of ternary A1GaAs compounds. Since the replacement of gallium by aluminium atoms increases the bandgap proportional to the aluminium content A1GaAs layers are very suitable as windows for solar cells.
Material problems. Epitaxially grown layers can contain a variety of lattice defects such as microtwins, stacking faults, dislocations, or antiphase domain boundaries which limit their utilization for device applications. These defects arise during the growth process and originate at the substrate interface or the surface. Further processes which degrade the film properties are the interdiffusion of atoms at the interfaces for instance of aluminium across the A1GaAs/GaAs interface which is of particular importance for superlattice structures where the thickness of the epilayers is very low (< 100A) and atomically sharp transitions are required. The composition of epitaxial layers can also be changed by phase separation and long range ordering particularly in ternary or quaternary compounds. These phenomena have been observed for instance in InGaAs epilayers.(27) The interface between substrate and the epilayer plays an important role in the formation of the lattice defects. The structure of the substrate surface before the deposition determines to a large extent the quality of the deposited layer. The surface orientation, the roughness or the occupation with foreign atoms (absorption, chemisorption) are relevant for the quality of the deposited films. On the atomic scale the bond reconstruction, the atomic distances, the occurrence of small (atomic) steps or the intersection of lattice defects from the substrate (dislocations, grain boundaries) are factors which determine the occurrence of lattice defects in the growing layer. For a discussion of the various aspects of epitaxial growth the reader is referred to the vast literature on the subject. In order to illustrate some of the problems the following examples will be discussed briefly. A particular lattice defect which can occur during heteroepitaxial growth of compound semiconductors is the inversion-domain boundary (IDB) (also called anti-phase boundaries). It separates two neighboring domains which are related to each other by an inversion operation. This is equivalent to the exchange of atomic species occupying different sublattices. The IDB's in AB compounds are characterized by a high density of incorrect bonds of the type A-A or B-B. In a layer-by-layer growth process inversion-domain boundaries can occur at atomic steps on the substrate surface since the stacking sequence of the atomic layers in the film is shifted by the height of the steps (Fig. 96). Since for particular step heights the correct stacking sequence is maintained one can reduce or eliminate the IDBs by growing on slightly inclined surfaces which contain atomic steps of the correct height if the inclination angle is chosen correctly. It is evident that the substrate surface is important for the initial stages of growth. An analysis and study of the nucleation process is therefore necessary for the understanding of the structure and morphology of the growing layer. Dislocation and microtwin generation is of particular concern for heteroepitaxial growth of semiconductors with different lattice constants and thermal expansion coefficients. A brief description of the nucleation models which have been developed sheds some light on the process of dislocation generation.
372
FIG. 96. (011) projection of a structural model for an inversion domain boundary (IDB) in a binary AB compound with sphalerite structure formed at a step (e.g. during epitaxial growth). A-atoms (small circles), B-atoms (large dotted atoms), substrate atoms (large black and white circles). Light and dark atoms occupy positions in the paper plane and above at a distance a/x/2, respectively. (a, lattice constant). The theoretical treatments ~271)of the nucleation models show that they can be differentiated by the following factors, the bond energies between two atoms in the epitaxial film EAA and across the interface to the substrate EAS, respectively, and the relative (absolute) lattice mismatch between film and substrate Aa/a. Three different models of growth are usually considered and are illustrated in Fig. 97. The coherent two dimensional layer-by-layer growth is often referred to as the Frank-van der Merwe mode. It is favored when the lattice mismatch is very small Aa/a ~ 0 and EAs/EAA> 1. The Vollmer-Weber is characterized by the nucleation and growth o f three dimensional crystallites, and occurs for IAal/a>>O and EAs/EAA < 1. The Stranski-Kristanov mode is an intermediate case and is favored by the condition Aa/a ~ 0 and EAs/EAA < 1. The growth begins two dimensional and the lattice mismatch will be accommodated by elastic strain until a critical thickness is reached. The critical thickness depends on the lattice mismatch and is usually of the order of a few monolayers. According to van der Merwe ~272~a mismatch of less than 7/'0 can be accommodated by elastic strain below the critical thickness. When the total strain energy reaches the energy required to generate dislocations the growth switches either from laminar to three dimensional growth or misfit dislocations are introduced. Therefore the three dimensional growth mode is accompanied by the formation of lattice defects such as dislocations or twin boundaries. The experimental approaches to reduce the defect generation arising from lattice mismatch are the intermediate growth of lattice-matched buffer layers or of strained layer superlattices. These problems are less severe for lattice matched systems such as the growth of A1GaAs on GaAs. A further attempt to reduce the dislocation nucleation is the doping with isoelectronic elements which are electrically not relevant. For instance in G a A s it has been found that
(a)
(b)
(c)
FIG.97. Three different modes of epitaxial growth. (a) Coherent two dimensional layer-by-layergrowth (Frank-van der Merwe mode). (b) Stranski-Kristanov mode is an intermediate case where initially a strained layer forms before three dimensional nucleation begins. (c) Vollmer-Webermode is characterized by the nucleation and growth of three dimensional crystallites.
373
doping with indium (at concentrations of about 2 x 1019 cm -3) which is isoelectronic to gallium can efficiently reduce the dislocation density in LPE grown layers. These doping concentrations may be high enough so that the dislocation motion is impeded by elastic interaction. It has also been suggested that the reduction of the dislocation density is linked to modifications of the concentrations of native defects which is caused by the dopant impurity. It remains unclear, however, how the alteration of the intrinsic defect concentrations affects the dislocation density3273~ Besides the occurrence of extended lattice defects the quality of MBE or MOVPE grown epilayers is also degraded by the presence of deep level point defects. High temperature growth usually results in a lower density of traps but has major drawbacks as far as achieving abrupt interfaces on the atomic scale, in minimizing interdiffusion in the heterostructure and in maintaining narrow diffusion profiles. One growth scheme to achieve better quality at lower temperatures is also the doping with isoelectronic elements such as indium or antimony. This has been shown to reduce the density of deep traps and enhance the mobility of the charge carriers of the material (see Table 14). Because of the technological importance of epitaxial GaAs layers for device applications there is a strong driving force for the further investigation of the growth processes and the defect generation which is linked to it. From the perspective of the photovoltaic application of GaAs these developments can be directly incorporated into the solar cell technology and basically no additional efforts are necessary which may be only relevant for this technology. It is however evident considering the complexity of the problems which are encountered for GaAs that the utilization and investigation of other binary or multinary compounds as solar cell materials will be a formidable task for the future. 7. POLYCRYSTALLINE THIN FILMS The technical advantages of polycrystalline and amorphous thin film semiconductors for photovoltaic applications have been discussed previously. In this chapter the properties and material issues of the major polycrystalline thin film materials will be discussed. These materials include II-VI compounds which will be discussed in the first part and a wide range of multinary semiconductors mainly based on copper-ternary compounds with the chalcopyrite structure. Particularly the promising potential of the group of ternary semiconductors for solar cell application is by far not fully explored. Therefore it seems justified to discuss some basic aspects of these semiconductors in more detail here. 7.1. H - V I Compounds The major representatives of binary II-VI compounds are based on the elements Cd, Zn and Hg in combination with S, Se and Te. Some of the compounds and their relevant
Table 14. Mobility of isoelectronicallydoped and
undoped MBE grown GaAs and A1GaAs 24) Mobility Doping impurity [cm2/Vs] MBE-GaAs Sb (isoelectronic) 2880 (550C) Undoped 2020 MBE-AIGaAs Sb (isoelectronic) 2230 (620C) Undoped 1770
JPMS 35-3/4~L
374
PROGRESS
IN MATERIALS
SCIENCE
Table 15. S t r u c t u r a l a n d electronic p r o p e r t i e s o f s o m e m o n o - a n d p o l y c r y s t a l l i n e I I - V I

c o m p o u n d s c274) Single crystals ZnS CdS ZnSe CdSe HgSe ZnTe CdTe HgTe Poly-crystals ZnSe ZnTe CdSe CdTe CdTe CdTe Hgl _ xCdxTe x =0.10 x =0.15 x = 0.20 a, c [.~,] E~ [e~7] 3.8 2.3 (5) 2.7 1.8 --0.15 2.26 1.60 -0.14 /~n [cm2/Vs]
~/p
[cm2/Vs]
Carrier concentration [cm -3 ] n n, p n n p n, p n,p
5.669 4.299 7.015 6.084 6.104 6.481 6.461
0.53 x 103 0.65 x lO 3 22.00 0.34 1.05 32.00 x x x x lO3 103 103 103
28 50 -110 80 -----------
(0.3-4.7) x 1016 (1.0-9.6) x 1015 (0.8-3.0 (1.0-6.1) (0.7-4.2) (2.2-6.9) x x x 1017 1016 lO 15 1017
0.01-0.1 0.I-I0 20-380 20-400 200-400 30-40 20 x 103 12 x 103 8 x 103
109-1012 101-102 1011-1016 lOt2-1014 102-1021 1.5 X 1022 2 X 1017 -1 1017
i n p n n n
electrical properties are summarized in Table 15. Compounds with smaller band gaps have attracted considerable interest for optical applications such as photodetectors and lasers. For photovoltaic applications only the Cd-compounds are directly suitable. Since they are direct bandgap semiconductors they have high absorption coefficients and therefore can be used as thin films materials. CdS was among the first semiconductors which have been used for solar cell applications, and several attempts have been made to develop a commercial CdS/Cu2S solar cell. Although frequently described as the "CdS-cell" the optically active layer is Cu2 S whereas CdS with a rather large band gap of about 2.3-2.5 eV serves mainly as a substrate material for the deposition of the Cu2 S layer. Because of its wide band gap and the ease with which it can be fabricated it is still frequently used as a window material particularly in a variety of thin film solar cells (see also Section 7.2). The most suitable semiconductor for solar cells is CdTe with a band gap of about 1.6 eV which matches the solar spectrum very well. This semiconductor has been receiving variable attention as a candidate for photovoltaic applications since the early 1960s. Despite its potential the utilization of this semiconductor has been hampered by the difficulty to control the material properties. Much of the pioneering and early work with this semiconductor was accomplished on single crystals. As-grown single crystals tend to be semi-insulating and doping is therefore required. Metallic dopants such as copper that do lower the resistivity have very high diffusion coefficients which may be one reason for the frequently observed instability of the material and the solar cells. It is also difficult to make electrical contacts to CdTe--especially to the p-type material. The interest in CdTe increased enormously when it was reported that efficiencies of greater than 10% were achieved for sintered polycrystalline thin film cellsfl) This observation stimulated the further research on polycrystalline thin films and several techniques were explored to produce larger area cells and higher efficiencies. In fact it turns out that a variety of methods, such as electrodeposition, (closed space-) vapor deposition etc., exist for making
375
good device quality thin films which makes this material particularly attractive from the commercial point of view.
7.1.1. Basic properties

Most of the II-VI compounds crystallize in the cubic zincblende structure, only CdSe prefers the wurtzite structure at room temperature. There seem to be, however, only small energy differences in the lattice structures of the cubic and hexagonal phase of CdTe and ZnSe, since the wurtzite structure can easily occur during thin film growth. Since the electronic properties are presumably different this causes problems concerning the homogeneity of the films. The tetrahedral covalent bonding between next neighbors is the same in both structures and the next neighbor distance varies only slightly between 2.46 (ZnSe) and 2.80 (HgTe). Because of the difference in the electronegativity for these elements the covalent bonding has a strong ionic contribution. With the exception of the mercury compounds the semiconductors have a direct band gap which varies considerably within this group of compounds. HgTe and HgSe are already semimetals indicated by their negative bandgap energy. HgTe can easily form a ternary compound with CdTe which offers the possibility to vary the bandgap continuously by changing the composition. Experimental results for Hg I _xCdxTe show that the bandgap changes linearly with the composition x according to the relation Eg--EHgTe + X (EcoTe- EHgTe). For small x the bandgap width is close to zero which can be used to develop a semiconductor with an adjustable small bandgap for infrared lasers. For photovoltaic applications larger bandgaps are required and Hg~ _xCdxTe with small additions of mercury (x > 0.85) have been used because the ternary compound offers a lower resistivity and easier contacting. The wide band gap semiconductors have absorption coefficient ~ which usually exceed 104cm -1 above the absorption edge so that typically films with a thickness of about 1 #m are sufficient to absorb the light. 7.1.2. Thin film preparation Many of the crystal growth and thin film deposition techniques are carried out at high temperatures. It is therefore important for the determination of the stoichiometry of the compounds to control the vapor pressures of the components at these temperatures. This becomes difficult or even impossible if one component of the compound is very volatile like for instance arsenic in GaAs or antimony in InSb. The vapor pressures of the components of the cadmium and zinc tellurides and selenides in equilibrium with the compound differ only slightly. Therefore one can generate stable vapor pressure conditions and obtain rather easily crystals with a homogeneous stoichiometry. In the case of compounds containing mercury which is a rather volatile element some deposition techniques cannot be employed because of this difficulty. In the field of vacuum technology methods with one, two or three (for ternary compounds) sources for the thermal evaporation are used. These techniques are described in many publications and the reader should refer to the references for more information. A review of their application to II-VI compounds is given. (274)None of these techniques seems to have a distinct advantage for the preparation and quality of the II-VI compounds and their application is thus mainly determined by other (practical) considerations. The microstructure of the films is mainly determined by the substrate temperature and to some degree by the lattice match of the compound. In general substrate temperatures above about 100C are necessary for the zinc and cadmium compounds to obtain a polycrystalline film, whereas lower temperatures can be used for the mercury compounds. Selenides have a greater tendency to form the amorphous phase probably because of the occurrence of Sen
376
chains (n = 2-8) in the gas phase which makes it more difficult to form the compound during deposition. The grain sizes of the polycrystalline films are typically in the range below about 2 #m. For orientated substrates and a lattice structure which matches the lattice of the compound to some extent epitaxial growth can be achieved in a particular temperature range. The deposition of CdTe can result in the formation of the cubic fl- and the hexagonal ct-CdTe modification. For instance thin films prepared by the MBE technique with neutral beams mainly consisting of ~-CdTe when deposited at low substrate temperatures (T ,~ 100C) whereas the fl-modification dominates at temperatures above 300C. It has also been observed that silver which is an acceptor dopant also promotes film recrystallization, and under optimal conditions grain sizes of about 1-2 cm can be obtained. In addition, the hexagonal phase forms predominantly.
7.1.3. Electrical properties

Some of the fundamental electronic properties of the II-VI compounds are also listed in Table 15. In general the mobilities of electrons are considerably larger compared to the hole mobility. Undoped crystals have a conductivity type which is determined by the stoichiometry and changing the composition usually converts the conductivity type. In most cases it is not clear which intrinsic defects determine the conductivity in undoped crystals. For instance for CdSe it is assumed that the n-type conductivity is caused by either the cadmium interstitials or the selenium vacancies. The carrier concentrations of as-grown crystals are in the range of 1017 to 10~Scm and particular precautions have to be taken to obtain a stoichiometry of -3 the crystal with a lower carrier concentration. The situation is even less clear for CdTe which is also n-type in the as grown state with carrier concentrations of about 1016 c m -3. Annealing in Cd vapor leads to an increase of the conductivity similar to CdSe although it is not clear which intrinsic defects cause it. Correspondingly the conductivity of p-type material heated in Te vapor increases for high resistivity material. A shallow level acceptor with a level at 0.05 eV above the valence band edge is observed and has been related to the cadmium vacancy which may occur in excess in Te-rich CdTe. For low resistivity CdTe, however, the opposite effect is observed which is due to the occurrence of a number of defect levels in the range 0.1 to 0.28 eV above the valence which compensate the semiconductor. The nature of these defects has not been identified yet. The situation is also complicated by the fact that the crystals and particularly the thin film are not homogeneous and may contain impurities, precipitations or dislocations in an uncontrolled way which makes the comparison and interpretation of the results of different authors difficult. Generally the identification of the defect structure and the corresponding trap levels is certainly still in an infant state for the II-VI compounds compared to the elemental and some of the III-V compounds. The electrical properties of the polycrystalline thin films in general differ from the properties of the single crystals. This is due to the high density of the grain boundaries and their influence on the carrier concentration and impurity distribution which has been discussed in general in Chap 4.6. The main effects are the compensation of the doping elements due to the grain boundary states and the change of the mobility because of the grain boundary barriers. The experimental data for the zinc and cadmium compounds given in Table 15 show indeed the much lower carrier concentrations compared to the single crystals.(274)Although there are no systematic data on the doping dependence available, the results for CdTe show that for high doping levels both for indium doped n-type CdTe and silver doped p-type CdTe carrier concentrations can be reached which are only slightly smaller than the single crystal values for the same doping level. This is in general agreement with the
SEMICONDUCTORS
377
model for grain boundary controlled conductivity in polycrystalline materials which shows that in a narrow range of doping levels (usually about 10 TM cm -3 for this grain size) the conductivity increases over several orders of magnitude and saturates almost at the single crystal values. Another way to reduce the resistivity at lower doping levels is the formation of Hg I xCdxTe films. As can be seen from the data in Table 15 the addition of smaller amounts of mercury (x > 0.85) still provides a large bandgap but reduces the resistivity. The electrical properties of CdSe and CdTe are influenced by the absorption of impurities at the surface, particularly oxygen and water vapor. This causes instabilities of the material which is not desired for solar cell applications. The doping and stability problems and to some extent the difficulty to fabricate good ohmic contacts on CdTe are partly overcome by the development of suitable cell designs. Most of the cell concepts are based on heterostructure designs which employ a variety of materials. High efficiencies have been reported for CdS/CdTe (9.4%) and SnO2/CdTe (10.5%) solar cells. Other concepts use the ternary compound HgxCd I _ xTe which provides the lower resistivity and easier contacting. Efficiencies of about 10% have been achieved for thin film CdS/Hgl _xCdxTe solar cells. The full potential of polycrystalline CdTe based solar cells has certainly not been fully explored yet, but this semiconductor is one of the promising materials for thin film solar cells if the material problems are better understood and can be improved.
7.2.
Multinary Compounds
A large number of ternary and quaternary semiconductors is currently investigated for their potential as high-performance inexpensive solar cells which can serve as an alternative to the single- and polycrystalline silicon technology. Several groups of ternary semiconductors with the composition A I W 1IX vl or A xl B Iv X v (chalcopyrites), and AIB" C v Nowotny-Juza compounds) have attracted interest because of their diverse optical, electrical and structural properties. ~275'276)These compounds offer a broad range of optical band gaps and carrier mobilities, as well as the ability to accommodate different dopants, which has led to their emergence as technologically significant device materials, including applications in photovoltaic solar cells, light emitting diodes and various nonlinear optical devices. Quaternary compounds--e.g. A B XC 1_xX2--can be formed by substituting isoelectronic atoms and offer the possibility to change the band gap and other electronic properties continuously. Other quaternary compounds with chalcopyrite structure which can be derived from ternary A ~B 111X compounds are of the type A 1B HI C Iv X4 In this structure vacant lattice sites on vl vl. the cation--sublattice are included. These compounds seem to have band gaps similar to the corresponding ternary compound, (277) but their potential as photovoltaic materials has not been systematically explored so far. A first selection of materials for photovoltaic conversion is based on a number of requirements which have been discussed in a previous Chap. 2.5. They include the following conditions for a solar cell semiconductor. (a) a direct band gap with nearly optimum values for either homojunction or heterojunction devices; (b) a high optical absorption coefficient which minimizes the requirement for high minority carrier lengths; (c) the possibility to produce n- and p-type material, so that the formation of homojunction as well as heterojunction devices is feasible. Generally, p-type material is preferred because electrons in many cases have a higher mobility and the materials therefore
378
Table 16. Structural and electronic properties of ternary ABX2 compounds with chalcopyrite structure, (a) for A = Ag, Cd, Zn and (b)A = Cu. u is the displacemeht of the cations given in Fig'. 36, 2C electron affinity, Eg the band the gap energy, and/*, the mobility018'279)
Ternary AglnS2 AgGaSe2 AglnS% AgGaTe2 AglnTe2 CdGeP2 CdSnP2 ZnSnP2 CdSiAs2 ZnGeAs2 CuGaS 2 CulnS 2 CuGaSe2 CulnS% CuA1Te2 CuGaT% CulnTe2 a (/~) 5.87 5.985 6.09 6.283 6.406 5.74 5.901 5.651 5.885 5.672 5.35 5.52 5.596 5.782 5.964 5.994 6.167
c/a
1.910 1.793 1.916 1.897 1.962 1.8772 1.951 2.00 1.849 1.966 1.959 2.016 1.966 2.0097 1.975 1.987 2.00
u
0.25 0.278 0.25 0.26 0.25 0.283 0.265 0.239 0.298 0.250 0.272 0.2295 0.243 0.235 0.25 0.25 0.225
Eg (ev) 1.87 1.83 1.24 1.2* 1.0" 1.72 1.17 1.66 1.55 1.15 2.43 1.53 1.68 1.04 2.06 1.23 1.04'
z (eV) -----3.98 4.41 4.25 3.85 4.20 ---4.58 ----
/~ (cm2/Vs) --600 (n) 100 (n) 2000 (n) 55 (n) 500 (n) 23 (n) -300 (n) 20 (p) 900 (n) -40 (p) 200 (n)
exhibit a higher m i n o r i t y c a r r i e r length. A n o t h e r r e a s o n is t h a t m o s t suitable w i n d o w m a t e r i a l s h a v e n - t y p e c h a r a c t e r a n d a p - t y p e a b s o r b e r is n e e d e d in a h e t e r o j u n c t i o n device; (d) a g o o d lattice a n d e l e c t r o n affinity m a t c h with w i n d o w m a t e r i a l s such as C d S o r Z n O so t h a t h e t e r o j u n c t i o n s with low interface state densities can be f o r m e d a n d deleterious c o n d u c t i o n b a n d spikes c a n be a v o i d e d . S o m e p r o p e r t i e s o f c h a l c o p y r i t e c o m p o u n d s w h o s e b a n d gaps lie in the r a n g e 1.0 t o 2.5 eV are s u m m a r i z e d in T a b l e 16. T h e t e r n a r y c o m p o u n d s which have been m o s t t h o r o u g h l y i n v e s t i g a t e d so far a r e the c h a l c o p y r i t e s with the c o m p o s i t i o n A I B ~I X w a n d b a s e d on the t r a n s i t i o n elements A = C u o r Ag, B = A1, G a , In, T1, a n d X = S, Se, Te. H i s t o r i c a l l y the C u - t e r n a r y a p p e a r e d to be the m o s t a t t r a c t i v e c a n d i d a t e s b e c a u s e o f the p r e v i o u s d e v e l o p m e n t s o f a Cu2 S - C d S s o l a r cell a n d the research focused o n three o f the c o p p e r t e r n a r y c o m p o u n d s - - C u l n S 2 , C u l n S % , a n d C u l n T e 2 - - 0 f which CulnSe2 seems to be the m o s t p r o m i s i n g s e m i c o n d u c t o r so far. T h e extensive p r o g r e s s t h a t has been m a d e in e x p e r i m e n t a l studies o f the f u n d a m e n t a l electrical a n d s t r u c t u r a l p r o p e r t i e s o f t e r n a r y a n d q u a t e r n a r y C u - c h a l c o p y r i t e s will be p r e s e n t e d first in the next section. Since CulnSe2 is the l e a d i n g c a n d i d a t e for p h o t o v o l t a i c a p p l i c a t i o n m o s t o f the e x p e r i m e n t a l d a t a have been a c c u m u l a t e d for this material. The e m p h a s i s will be on the critical discussion o f the defect c h e m i s t r y f r o m the perspective o f the p h o t o v o l t a i c p r o p e r t i e s . I n the following p a r a g r a p h v a r i o u s processing techniques for b u l k a n d thin film m a t e r i a l s will be p r e s e n t e d a n d the p r o b l e m s t h a t are c o n n e c t e d with the design o f a s o l a r cell. T h e m o r e c o m p l e x c h a l c o p y r i t e crystal structure c o m p a r e d to e l e m e n t a l a n d b i n a r y s e m i c o n d u c t o r s which crystallize in the d i a m o n d cubic, sphalerite o r wurtzite structure a n d the l a r g e r n u m b e r o f a t o m i c species leads to a g r e a t e r v a r i e t y o f intrinsic p o i n t defects a n d p o s s i b l e a t o m i c structures o f lattice defects like g r a i n b o u n d a r i e s a n d dislocations. It has been e s t a b l i s h e d t h a t the electrical a n d o p t i c a l p r o p e r t i e s are d o m i n a t e d by the presence o f
SEMICONDUCTORS
379
structural defects, and their control and investigation is, besides the technical design of the solar cells, the greatest challenge for the further development of these materials. The entire spectrum of novel multinary semiconductors is certainly not completely explored from the perspective of photovoltaic applications. A group of ternary compounds which is currently investigated are A ~Bn C v (Nowotny-Juza) compounds. (275'27s) They are mainly Li-based semiconductors with B = Zn, Cd and C = P, As and crystallize in a so-called filled zinc-blende structure. In this structure the B and C atoms take the same positions as the cations and anions in a III-V sphalerite unit cell, whereas the lithium (A) atom occupies one of the tetrahedral interstitial sites at (a/2) (111) or (3a/4) (111) (Fig. 35). Some of these compounds such as LiZnAs, LiCdP and LiZnP have band gaps in the range 1.1-2.1 eV. There are also investigations of novel binary compounds with the composition A~ Bw H which crystallize in a layered structure with an orthorhombic unit cell. Band gaps of 1.44 eV and 1.2 eV have been reported for Bi2$3 and Sb2 Se3, respectively. These developments are still in a very early stage and it would go beyond the scope of this review to discuss these preliminary results. The further development of all new semiconductors depends critically on the ability to fabricate low-resistivity material and to control the doping concentrations of p- and n-type material. Since the complexity of multinary compounds offers such a wide range of electro-optical properties certainly more suitable photovoltaic semiconductors can be expected in the future. 7.3.
Copper-ternary Compounds
The chalcopyrite lattice structure has been described in Chap. 4.1.1 and is characterized by the lattice constant, a, the tetragonal distortion of the unit cell, r / = c/2a, and the anion displacement u. Table 16 gives a compilation of experimental data for a, r/, u, and the band gap energies for some ternary chalcopyrites3279'28) The closed packed planes in the chalcopyrite lattice are the {112} planes (corresponding to the {111} planes in the zinc-blende structure) which consist of a double layer of anions (X) and cations (A, B). The usual growth direction of thin films is perpendicular to these planes along the (221) direction. Together with the {100} planes they are also the common cleavage planes of single crystals. The ternary phase diagrams for most of the compounds have not been completely determined yet. For the Cu-In-Se and Cu-In-S system the most complete data on the crystal structure and composition are available. The binary phase relations that border the ternary composition field(~5'~st) show a variety of intermetallic compounds (Fig. 98) and several pseudobinary relations between some of these compound have been established. The best known pseudobinary phase diagrams~282) are published for the phases Cu2S-In2S 3 and Cu2 Se-In2Se3 one of which is shown in Fig. 99. Several other pseudobinary relationships have been determined, for instance the CulnS2 phase can also coexist with InS and In. It appears to be a common feature of many Cu-ternaries that a phase transition from the chalcopyrite (7) to the cubic sphalerite structure (6) occurs at elevated temperatures due to the lowering of the free energy by the entropy contribution of the random distribution of A and B atoms in the cation sublattice. (Some ternaries exist in the cubic phase already at room temperature.) The transition temperatures for the three compounds are 810C (CulnSe2), 975C (CulnS2), and 672C (CulnTe2). It appears to be that the chalcopyrite structure can be maintained over a rather wide range of compositions. Results for polycrystalline CulnSe2 thin films show a composition window of about 10% for all three components. The deviations from the composition and valence
380
PROGRESS IN MATERIALS SCIENCE Se
C u S e 2 ~ ~" Cu3In~$eg--~ CuS~

~~uZnSe2
11'Im~ In2$e3
.]l~Se
n 4'
FIG. 98. Schematic ternary phase diagram for CuInsc2, derived from crystal structure considerations.(28~)
stoichiometry of a compound ABX 2 are usually characterized by the molecularity Am and the valence stoichiometry As, quantities which have been discussed in Chap. 4.1.1 [eq. (118)]. The chalcopyrite structure within the composition range can be maintained by the incorporation of intrinsic point defects such as vacancies, interstitial atoms, and antisite defects. These
IO00
_~ - - ' o o
800
600 u
Y / I I
E 400 I-
/I
(b) o
i
40 50
200 -030 CuzSe
I I
!I
I I I I I
l
Mole % of In=Se3
6O
70
In2Se3
FIG. 99. Pseudobinary phase diagram for CulnSe2-based on Cu2Se-ln2Se3.(2s2)Open circles represent experimentally (DTA) determined transition points.
SEMICONDUCTORS
381
defects are considered the main electrically active defects and have been related to the electrical properties of the material as will be discussed in Section 7.2.3. 7.3.1. Film and crystal growth techniques The first successful developments of Cu-ternary solar cells with efficiencies of about 12% were based on single crystalline CulnSe2 and encouraged the further exploration of these materials. Several principal techniques have been used to grow single crystals of chalcopyrite I-III-VI2 semiconductors325) (a) The first experiences with the solidification of stoichiometric melts of CuInSe2, CuInS2 and CuInTe2 uncovered a problem that has hindered the production of large crystals. The chalcopyrite-sphalerite transition and the anisotropic thermal expansion of the crystals along the a- and c-axes leads to a high density of cracks in the material, and promotes the tendency to form polycrystals. The best quality and largest crystals have been produced by a high pressure Bridgman process. The successful growth of n- and p-type crystals has been reported. (b) Direct growth from the melt is also possible from binary constituents (e.g. Cu2 Se-In2 Sea or CuSe-InESe3 for CuInSe2). Lower pressures are required here which minimizes the danger of explosion during growth. (c) Large platelets of CuInSe2 single crystals have also been grown by chemical vapor transport, and thin layers ( ~ 6#m) on ZnSe substrates using liquid-phase epitaxy (LPE). In the latter case the solute was prepared from a stoichiometric melt and Bi was used as a solvent. The Cu-ternaries are primarily considered for thin film solar cells and the application and development of film growth techniques is therefore of greater importance. A variety of standard thin film techniques, such as sputtering, physical vapor deposition (PVD), electrodeposition, or spray pyrolysis has been employed so far but none of which is considered as the definitive approach. The highest efficiencies have been obtained for films which are deposited by PVD or selenization of metal films (for CuInSe2): (a) Flash evaporation and (resistive heated) single boat deposition using single phase powder has the difficulty to control the loss of either one of the components if their vapor pressure is too different. This is the case for CuInSe2 where deficiencies of Se can result and the stoichiometry is difficult to control. Therefore two and three source schemes have been developed in which either CuInSe2 and Se are evaporated from separate sources or the films are deposited directly from the elemental constituents. The n- and p-type character of the films can be controlled by varying the different partial pressures. (b) A technologically quite simple process is the selenization of metal films. Metals layers of the different components are deposited on a substrate with a thickness according to the desired stoichiometry of the compound. The reaction of the films with selenium at evaporated temperatures occurs when they are exposed to a Se or H2Se atmosphere, or by depositing Se directly on the metal layers.~283)Relatively large grains (> 1 #m) develop under these conditions. Polycrystalline films are usually deposited on a metal layer (e.g. molybdenium for CuInSe2) which forms a good ohmic contact. The deposition techniques produce films which can have grain sizes of about 1 #m or less depending on the deposition conditions. It appears to be that for optimal performance grain sizes > 1 #m are required in some cases. 025)
382
PROGRESS
IN
MATERIALS
SCIENCE
T1 +T:~
N1
i!% ....
Cu3d etU In-Se Bond In - Se Bond Cu3d
-10
CulnSe2 Ceperley 3orrelatior
-12 Se 41
f
-14
Se 4s
FIG. 100.Calculatedband structureof CulnSe2using(a) Ceperleycorrelationand (b) Slaterexchange method.(118)Bothapproximationsyielda band gap whichis smallercomparedto experimentalresults.
7.3.2. Band structure and optical properties Several attempts have been made to calculate the band structure of a variety of Cu-ternaries. The most comprehensive and systematic study is an ab initio calculation by Zunger et al. 0~8) for CUBS2 and CuBSe2 chalcopyrites (B = A1, Ga, In). The applied method is a local-density-functional formalism where the electron problem is solved self-consistently and the interelectron interactions are based on the Ceperley's correlation or Slater exchange methodologies. The calculated local density of states compare well with XPS and optical data and yield a direct band gap for these semiconductors in accordance with experimental results (Fig. 100). A disadvantage of the applied local density theory is the failure to predict the exchange coefficient to the experimentally observed optical band gap. The band gap energies (Table 16) in comparison with the isoelectronic binary compounds given in Table 15 are considerably lower which makes ternary compounds therefore generally more suitable for photovoltaic applications. An essential feature of this class of semiconductors is the significant ionic contribution to the covalent bonding of the atoms. This is shown in the total electron density maps (Fig. 101) calculated for instance for CulnSe2. The A-X bond generally appears to be more polarized towards the anion site compared to the B--X bond. In fact in many cases the interpretation of experimental results is based on the assumption that the chalcopyrites are more ionic than covalent materials. A further important result of the electronic structure calculations is the significant contribution of the Cu d-states in the Cu-X bond (Cu--d-X-p interaction) to the
SEMICONDUCTORS
FOR
SOLAR
CELL
APPLICATIONS
383
;lrf .
;1
o 2 3 4 5 6 7
ou,o o,
8 9 I'o
FIG. 101. Calculated total valence charge density for next neighboratoms in CuInSe2. Logarithmically spaced contour plot with shading highlighting the formation of a covalent lobe around the Cu-Se contact.OTM upper valence band whereas the Cu s-states contribute to the conduction band. ]8'284) It has been pointed out in this study that the anomalous reduction in the band gaps relative to the isoelectronic binary semiconductors is correlated with the existence of the d-bonding and the anion displacement, u. The theoretical calculations allow to predict chemical trends for the changes in band gap energies and the structural parameters (a, c and u) for various chalcopyrite semiconductors.
2000
1600
1200
O eg "O m m
800
v-
.<
400
1.0
1.5
2.0
2.5
hv ( e V )
FIG. 1 0 2 .
Absorption spectrum of CuInSe 2 compared with other photovoltaic semiconductors. For comparison the solar emission spectrum (for AM1.5) is included (dashed line).
384
/
12
e~
0.96
100
104
1.08 hv (eV)
1.12
I. f6
FIG. 103. Plot o f the absorption parameter (=hv) 2 vs photon energy hv for various CulnSe 2 samples showing the direct band gap (see eq. (3)) and the variation of the band gap energy: (11) single crystal; (O) polycrystalline thin film; (A) annealed thin film. (2s5)
The optical properties of ternary semiconductors have been studied by a variety of techniques which have been reviewed~7'285)and partly summarized by Kazmerski et al. ~25'286~ The investigation of the fundamental absorption is usually used to determine the band gap of the material. Results for single- and polycrystalline CulnSe2 in comparison with other semiconductors show the enormously high absorption for this material (Fig. 102). Assuming an allowed direct band gap transition, the absorption coefficient as a function of the photon energy is given by eq. (3). For hv > 0.95 eV the absorption spectrum can be described by this equation which confirms the direct band gap character of the semiconductors. Specimens prepared by different techniques and subjected to various annealing treatments show, however, that the empirical band gaps vary between 0.92-1.04 eV (Fig. 103). It has been suggested that band gap narrowing due to high concentrations of impurities is responsible for the observed scatter of the band gaps. In Chap. 2.2 the band gap narrowing induced by high concentrations of shallow doping elements, N, has been described by eq. (7) where the band gap is inversely proportional to the Debye screening length L D = (e 2N/eeo KT) 1/2. An analysis of the band gap energies as a function of the concentration of ionized impurities and free carriers in a variety of samples, indeed confirms the 1/LD dependence. (284)The extrapolation to 1/LD = 0 gives the band gap energy of the undoped crystal Eg --- 1.01 eV which is close to the currently accepted value Eg = 1.04eV (Fig. 104). This value is also typical for polycrystalline thin films with a comparable free electron or hole concentration which indicates that the concentration of scattering impurities is lower compared to doped single crystals. The difference between electrically active and scattering impurities can be explained
SEMICONDUCTORS
1.02

, l
385
o.gg
g
0.9E
0.93 0
3
L-I
6 (in 10%m-i)
FIG. 104. Plot of the band gap width Eg vs the reciprocal o f the screening length for n-type CuInSe 2 single crystals.
by the assumption that single crystals are highly compensated whereas in polycrystalline films the concentration of compensating impurities is lower. In the low-absorption-coefficient region (hv < 0.95 eV) of single- and polycrystalline CulnS2 and CulnSe2 deviations (tails) from the absorption behavior for a direct band gap material are observed. These tails have been attributed to defect (Cu-vacancies, grain boundaries) related transitions from defect states into the conduction band or phononassisted absorption with the participation of long-wavelength optical phonons, respectively.~287~The analysis of an acceptor-to-conduction band transition at low temperatures has been used to determine the effective masses of electrons and holes by fitting eq. (7) to the absorption spectrum of CulnSe2. Calculation of the effective densities of states of the conduction and valence band for the T = 300 K yielded Arc= 1.1 x 10TM cm -3 (mo = 0.09 mo) and Nv = 2.6 x 1019 c m -3 (mh = 0.73 mo).m4,28s'29) The deviations in the low energy region which are frequently observed may be the reason for some of the ambiguous results about the direct band gap character of CulnTe2. Generally, the measurements support the direct band gap prediction for Cu- and Ag-ternaries, with some further explanation of the additional absorption near the band edges. 7.3.3. Lattice defects and electrical properties Thermoelectric power, Hall effect and conductivity measurements as a function of temperature are usually carried out to determine the conductivity type, the carrier concentrations, mobilities and the impurity ionization energies. Many of these measurements are performed on single crystals with the purpose to study and understand the fundamental properties and defect chemistry of the ternary semiconductors. The technological interest however is directed towards polycrystalline thin film cells although single crystal solar cells have also been fabricated successfully3289) In previous chapters it has been discussed that electrical properties of single and polycrystalline semiconductors can differ considerably because of the presence of grain boundaries. Since there is little information available about the electrical properties of grain boundaries in ternary semiconductors their influence can only be considered in a very general manner so far. 7.3.3.1. Single crystals. The general observations about the electrical properties of CulnSe2, CulnS2 and CulnTe2 are the following: CulnSe2 and CulnS2 can be made n- or p-type by changing the stoichiometry of the crystals, whereas CulnTe2 can only be prepared as a p-type
386
material so far which prevents the fabrication of homojunction devices. Extrinsic doping by shallow impurities has been reported (25) but is usually not applied for device applications. Typical values for the carrier concentrations in n- and p-type single crystal CuInSe2 are in the range 1016 to 1017cm-3. The mobilities of n-type samples can be as high as /z, = 900 cm2/Vs and are generally lower for p-type material:/zp ,~ 30 cm2/Vs. Measured Hall mobilities as a function of temperature (29) have been analyzed assuming scattering by acoustic, polar-optical and non-polar phonons and by ionized impurities using eqs (39-42). The results yielded ionized impurity concentrations of about Na = 8.4 x 10 is c m -3 (p-type) and Nd = 7.9 x 10]7cm -3 (n-type) which was considerably higher compared to the free carrier concentrations. These results can be explained by a high degree of compensation and seems to be characteristic for Cu-ternary semiconductors as has been mentioned above. An extensive study of the relationship between conductivity type and stoichiometry has been carried out for CulnSe2 by Neumann e t al. (29n)An excess (As > 1) or deficiency (As < 1) of selenium can be achieved by crystal growth or annealing under high or low vapor pressures of S or Se. As is determined from eq. (118) with sl = 2, s: = 1 and s3 = 3 and the following results shown in Fig. 105 are obtained: (1) For small deviations from the ideal stoichiometry (IAml < 0.08 and IAsl < 0.06) p-type specimens are obtained for As > 0 and Am > 0. Results from previous studies also indicate p-type behavior for Am < 0. (292)n-Type conductivity is observed for As < 0 and Am both positive and negative. The excess (As > 0) or deficiency (As < 0) of selenium determines the p- or n-type character, respectively. (2) For larger deviations from the ideal stoichiometry (I Am I > 0.08 and IAs I > 0.06) a more complex behavior occurs. The region with n- and p-type character appears to be separated along a line which does not correspond to As = 0 but can be defined by a critical As corresponding to each value of Am: p-type conductivity occurs if As > Asc
112flu 0.2 In 0.6 Se 0.24 In 0.24flu 0.52 Se o~0o
0.3Cu 0.2 In
o 2so,, o.~,c,, 0.241n /0~-,.~[ \ 0251n 0.51 S e ~ 0 5 1 S e
Oo
'.
~--~ '
/
.~"2se3
o . ~ . " "0.2 flu
0.3 In
o.s s , / ~ . % " :
/__\gTd
/ \
"zs'3
Cu2Se
O / ~
,..r
~ flu
~. 0.4 In
o.2cu
\
0.27Cu 0.24 In O.&9Se 0.26Cu 0.251n 0.49Se 0,25Cu 0.26 In o!,g Se 0.24Cu O.Z'/In O/,gSe 0.5Cu 0~2In (~3Se OJ,flu 0.3In 0.3Se 0.3Cu Off,in 0.3Se 0.2 Cu 0.5 In O.3Se
(a)
(b)
FIG. 105. Schematic representation of the composition of n-type (closed circles) and p-type (open circles) CulnSc2 single crystals in the ternary phase field near the ideal stoichiometry. (a) For small deviations of the stoichiometry (lAx I < 0.06, lAy I< 0.08, and (b) for larger deviations (lAx [ > 0.06, [Ay l > 0.08.(290
387
and n-type behavior for As < Asc. Corresponding to the previous case p-type conductivity is still observed for As > 0 and Am > 0, whereas n-type behavior occurs for Am < 0 and As both positive and negative. The observed conductivity properties for the group with small deviations have been consistently explained on the basis of a point defect model, which assumes that intrinsic point defects compensate the deviation from the ideal stoichiometry and introduce shallow acceptor and donor levels in the band gap. This model which will be discussed next summarizes the main results of a number of electrical and optical measurements, but fails so far to describe the electrical behavior of specimens with larger deviations from the ideal stoichiometry. This may be due to the fact that a higher density of point defects leads to the formation of complexes which increases the number of possible donor and acceptor species and increases the difficulty to predict the conductivity type. The intrinsic point defects which have to be considered are vacancies, interstitials and antisite defects as has been discussed in Chap. 4.4.1. In a compound with the general formula ABX 2 12 different native defects can occur which are compiled for CuInSe2 in Table 17. Their concentrations are determined by the formation enthalpies AH = A E - T A S and the stoichiometry of the crystal. The formation energies AE have been calculated previously (119'12'293) and are included in the table. Though the values differ the trends in the relative magnitudes are similar. Assuming that the formation entropies are not too different the results show that the defects with the highest concentrations in an ideal single crystal are the antisites Incu and Cui,, and the vacancies V~n, Vcu and Vs. The formation energies of the other defects such as the interstitials are so high that they are less likely to occur. In a non-stoichiometric crystal the concentrations are determined by the deviations from molecularity Am and stoichiometry As. The majority defect pairs which can occur for a given composition have been discussed in Chap. 4.1.1. Taking into account the formation energies of these defects from Table 17 one obtains the defect pairs with the lowest formation energies given in Table 18. (The defects with the lower formation energies are always listed first). Some of these defects can only occur in one of their charged states for a given composition which has to be taken into account in their formation energies. Since their trap levels are not known this contribution has generally been neglected in previous discussions, however, for shallow level impurities the additional energy may be small compared to the high formation energies.
Table 17. Calculated bulk formation energies H F of all intrinsic defects in CuInSe2. HFB are calculated formation energies for the same defects formed at incorrect bond sites at grain boundaries (e.g. in a polycrystallinematerial). The donor or acceptor character is determined assuming an ionic compound. For the selenium vacancy a donor character has also been suggested assuming a covalent model
Vacancy/Interstitial He [eV] [119] Hv [eV] [120] Hvs [eV] [1201 Acceptor/Donor Anti-site defects Hv [eV] [119] HF [eV] [120] HrB [eV] [120] Acceptor/Donor
Vcu 2.6 3.2 1.6 A CUln 1.5 1.9 1.6 A
V~n Vse 2.4 2.8 2.4 2.6 1.2 1.3 A A/D Incu 1.4 1.6 1.I D Cus~ 7.5 5.4 2.4 A
Cui 4.4
-
Ini 9.1
-
Se~ 22.4
-
D Inse 5.0 5.0 2.3 A
D Secu 7.1 6.0 2.8 D
A S% 5.5 5.2 2.7 D
388
P R O G R E S S IN M A T E R I A L S S C I E N C E Table 18. Majority defect pairs with the lowest formation energies (from Table 17) in single crystalline CulnSe 2 for a given stoichiometry Am and As. The donor/ acceptor character of the defects and the experimentally determined conductivity type o f single crystals are also included 02,292~ Am >0 <0 >0 <0 As >0 >0 <0 <0 Small Large p p* n n p n n, p n Culn-h A Incu-Cuin D-A Inc~-Vln D-A Incu-e D Bulk defects pairs Vin-h A Incu-Vcu D-A CuI~-Vs~ A-D Vcu-h A Incu-Cul. D-A Vcu-Vs~ A-D -IncuWcu D-A Culn-Vs~ A-D --Vcu-Vs~ A-D
In the absence of additional doping elements of comparable concentrations the electrical properties of the crystals are determined by the energy levels which are introduced by the intrinsic point defects. Though several attempts have been made to estimate their electrical character, the properties of none of the defects has been determined directly and unambiguously so far. In most cases the acceptor or donor character of the defects has been estimated from the perspective of the ionic character of the compounds as indicated in Table 17. Several authors pointed out, however, that in a covalent model the selenium vacancy would be an acceptor, 2~) so that for this defect a consistent description has not emerged. Generally it has been assumed that any of these defects can posses a shallow level and contribute to the doping of the material. Since in most cases charge compensating defects (acceptors and donors) occur in comparable concentrations the materials are likely to be compensated as it is actually observed. The conductivity type is thus determined by a small excess of either one of the possibe defect pairs. In general, one has to calculate the relative concentrations of all defect pairs for a given composition to determine the carrier (electron or hole) concentrations correctly. Qualitatively, the results from Table 18 suggest however that for small deviations from stoichiometry the observed conductivity character of the compounds is almost correctly predicted from this point defect model as has been discussed by several authors. For p-type material (As > 0, Am > 0) the majority defects are the Cu~n antisite and the indium vacancy VI, which are both acceptor impurities. For indium-rich material (Am < 0) the p-type character is still maintained for small deviations though the majority defect appears to be the Incu antisite which is a donor. It may be overcompensated by the acceptors CUln and Vcu. For larger deviations in this case the conductivity of the material finally is converted into n-type character. The n-type behavior of selenium deficient material (As < 0) is thus consistent with the occurrence of the Incu antisite. For larger deviations no conclusive results can be obtained in the case Am > 0, As > 0, as has been mentioned above, and a more refined model may be needed to understand the behavior. The interpretation of conductivity and Hall effect measurements remains still rather speculative and it is desirable to support the results by other experimental techniques which allow a more direct verification of the nature of the defects. Most of the additional results have been obtained from photoluminescence measurements on single crystals with various compositions and after annealing treatments in air or in excess of selenium. Typical PL spectra show two groups of lines (A1-3, B1-5) whose relative intensities vary with the composition of the specimens (Fig. 106). From a systematic study of these lines it is concluded that the A-lines correspond to optical transitions between the valence band and either the conduction band (A1), an exciton level (A2) or a donor level (A3). The five B-lines belong to transitions between an acceptor level and either the conduction band (B1-2) or a donor
SEMICONDUCTORS
FOR SOLAR
CELL
APPLICATIONS
389
I~
,v
III
I L
-g
.... "71
I I
Ii'.' <" 1 ' IOlll

Table 19. Summary of experimental results for the energy levels of shallow acceptors and donors and their assignment to particular intrinsic defects in CuInSe22s) Character Defect Energy levels [meV] Donors Incu 10-35 Vs~ 60-80 Acceptors Cu~n 30-40 Vln 35-90 Vcu 30-50
. i. o.,,:jAii--.
.
0.800
1.000 Photon energy (eV)
1.200
FIG. 106. Typical photoluminescence emission spectra of nand p-type CulnSe2 single crystals with different compensation ratios. Most of these transitions are detected at a wider range of energies which vary with the composition.(292)
level (B3-5). These results have been used to develop the positions of various shallow donor and acceptor levels in the band gap. (292) Similar to the conductivity measurements the assignment to a particular defect is deduced then from assumptions about the corresponding defect chemistry. Some of the results for the five antisite and vacancy defects, which have been derived by various methods and show some agreement, are summarized in Table 19. Other authors have also discussed the occurrence of defects such as interstitials and cations on a selenium site and have assigned shallow levels to one of these defects. Considering that the higher formation energies of Table 17 are correct such defects are less likely to occur, however. The interpretation so far is based on the assumption that the antisite and vacancy defects predicted from the stoichiometry in the ionic model are related to the observed shallow level impurities. This m a y be questionable considering the fact that in binary compounds (GaAs for instance) antisite defects (AsGa) with midgap levels are observed. A drawback of the experimental techniques which have been applied so far is that none of these defects can be observed directly. Therefore studies with more structure-sensitive methods such as electron paramagnetic resonance (EPR) in combination with D L T S for instance are certainly necessary to understand the complex defect chemistry in ternary compounds. Some recent D L T S measurements have shown that deep level defects are present in single-crystal CulnSe2 but a systematic investigation which would allow to identify the nature of the defects has not been carried out yet. In summary, a consistent picture for the defect chemistry and the energy levels of the various defects has thus not emerged yet. 7.3.3.2. Polycrystalline thinfilms. A wide range of electrical properties of the polycrystalline Cu-ternaries has been reported. It appears to be that they are critically dependent on the method of preparation and annealing treatments after fabrication. At present a simple evolution of the measurements is therefore not possible. As an example electrical measurements on CulnSe: are presented in Fig. 107 and shows the conductivity, mobility, and the carrier concentration for a p- and n-type films grown by the dual source technique. (25) The parameters are exponentially dependent on the inverse temperature and increase with
JPMS 35-3/4--M
390
l0 ]
t0 -z
f0~7
~01
10-4
i015
5 FIG. 107. Mobilities/xp (O), #. (O),
4 103/T (K -i)
5 n (A) upon
inverse
conductivity cr. (ll), and carrier concentration
temperature for n- and p-type polycrystalline CulnSe 2 thin films. 25)
increasing temperature. On the contrary recent measurements on flash evaporated p- and n-type films showed a maximum in the conductivity at about 144 K and a decrease for higher temperatures. (294) A systematic study of the resistivity and carrier concentration as a function of the composition of the films has been carried out by Noufi e t al. (295)The composition of the films during the evaporation process has been determined by changing the ratio Am = [Cu]/[In], the molecularity, while maintaining the selenium concentration. A typical result is given in Fig. 108 which shows that in a narrow range of compositions the carrier concentration drops
n/p
[cm -3]
Poly-CulnSe
P [cm~] 10 19 1020 10 16 10 17 1016 1015

10 14
Poly-CulnSe2
10 20 10 19 1018 1017 1016 1015

10 TM
/L~11
m ~ '~ Bill
JB
I I I I I
=- 0.02
Am =- 0.04
I I I I I I I
0.7
0.9 1.1 [Cu]/ [In]
1.3
0.7
0.9
1.1
1.3
lSe] / ( [Cu] + [In] )
FIG. 108. Carrier concentration of p- and n-type CulnSe 2

single crystals as a function of the composition Am = [Cu]/[In]. Black symbols n-type, white symbols p-type .(29s)
FIG. 109. Majority carrier (hole) concentration o f CulnSe z single crystals as a function of the ratio [Se]/(Cu + In] for two molecularity ratios Am = [Cu]/[In]/295)
S E M I C O N D U C T O R S FOR S O L A R C E L L A P P L I C A T I O N S
Table 20. Majority defect pairs with the lowest formation energies (from Table 17) in polycrystalline CulnSe2 for a given stoichiometry Am and As. Defects which are formed at grain boundary sites are indicated by subscript b. The donor/acceptor character of the defects and the experimentally determined conductivity type of polyerystalline thin films are included(t2o,29t)
391
Am >0 <0 >0 <0
As >0 >0 <0 <0
p, small Am p, large Am low, (p) med, (p) low, (p) high, (p) low, (p) high, (n) low, (p) low, (n)
Grain boundary and bulk defect pairs Vl.b-h Cul,b-h A A Incub--CUlnb Incub--Vc,b D-A D-A Inc~b--Vtnb Vs,~:2Uln b D-A D-A Incub--e Vcub-h D A Cutn-h A Vs~b-Vcu b D-A Inc.-Vln D-A Incu D Vln-h A Incu-CUln D-A
Vseb--Vcub
D-A
by several orders of magnitude and the conductivity changes from n- to p-type behavior. A corresponding X-ray analysis has shown that for larger deviations, Am < -0.2, a second phase, Cu2_ xSe, appears which m a y explain why the electron concentration for very copper poor material is apparently not correlated to the composition any more. The effect of the selenium/metal (Cu + In) ratio on the carrier concentration for a p-type film in the single phase region is illustrated in Fig. 109 for two fixed Cu/In ratios (Am = --0.02 and -0.04). The material is p-type for selenium deficient films, or As < 0, and increases with increasing selenium concentration by several orders of magnitude. A summary of the experimental data which shows qualitatively the dependence of the conductivity on the composition is given in Table 20. If the composition of the films is near the ideal stoichiometry the material is always p-type, whereas a large deficiency of copper converts the conductivity into n-type behavior. The conversion is evidently also correlated with a strong decrease in the carrier concentration and conductivity. It should also be pointed out that the correlation between conductivity and composition for selenium deficient polycrystalline films (As < 0) is different from single crystals which can be seen from a comparison with Table 18. For solar cell applications the conductivity of the polycrystalline thin film is difficult to control because of the sensitivity to the composition. In order to achieve high efficiency in a CdS-CuInSe2 heterojunction cell a p-type absorber with an optimal carrier concentration of about 1016-10~7cm -3 is required which lies in the concentration range where slight changes in the composition change the carrier concentration considerably. It is evident from the discussion in Chap. 4.6 that polycrystalline thin film semiconductors can have electrical properties which differ significantly from the single crystal behavior. This is due to the influence of the grain boundaries which become increasingly important when the grain sizes are in the micron range or below, typical for polycrystalline thin films. The experimental results on silicon and germanium and theoretical considerations have shown that a characteristic behavior of the conductivity in thin films can be a strong variation over several orders of magnitudes in a narrow range of doping levels. This occurs when grain boundary traps are filled and the free charge carrier concentration is rapidly reduced. It has been shown that the important parameters which determine the electrical behavior are the microstructure such as the grain size and composition, the grain boundary trap distribution in the forbidden band gap, and the doping level. The experimental results on the doping dependence of the conductivity and carrier concentration for polycrystalline CuInSe2 also indicate a strong influence of grain boundary states, however, no attempt has been made yet to take this aspect into account quantitatively.Unfortunately, important parameters which determine the electrical conductivity are
392
frequently not reported for many published experimental results. In ternary compounds a complication also arises because the doping is usually related to the composition. The previous results on CuInSe2 single crystals have shown that the relationship between the composition and the concentration of the intrinsic point defects that determine the doping level has not been established yet. Therefore an independent determination of the concentration of the doping species in ternary polycrystalline films is not possible, contrary for instance to the situation for elemental semiconductors where the doping level is determined by extrinsically introduced impurities. An understanding of the electrical behavior of polycrystalline ternaries is therefore even more complex than in single crystals since the parameters which finally determine the majority carrier concentration and conductivity--the doping concentration, the concentrations of the compensating point defect and the grain boundaries states--are basically not known. A comparison of electrical and structural measurements between single and polycrystals is further complicated by the fact that the defect chemistry in semiconductors with a high density of grain boundaries can be considerably different from single crystals. A recent analysis of grain boundary structures in the chalcopyrite lattice has shown that incorrect anion-anion or cation-cation sites are part of the atomic structure for large angle grain boundaries in compound semiconductors. 2) It has been pointed out and discussed in Chap. 4.4 that the formation energies for intrinsic point defects are lower if they are formed at these incorrect bond sites. A comparison of the atomic structure of an antisite defect and a vacancy in the bulk and at an incorrect bond site is shown in Fig. 110. The different next neighbor environments compared to the same defects in the bulk indicates that the trap levels of these defects are likely to be different, too, if they are formed at the boundary. The deviations from the stoichiometry are therefore compensated then by point defects which are either formed at the grain boundaries and or by intrinsic bulk defects. An indication that native point defects are preferentially formed at grain boundaries may be an experimental observation which has been mentioned earlier. The band gap of polycrystalline thin films measured by optical absorption remains rather constant (about 1.0 eV) independent of the deposition technique. This indicates a reduced number of bulk defects compared to single crystals where band gap narrowing due to high concentrations of compensating intrinsic defect is frequently observed (see 7.3.2). This result can be understood if intrinsic defects are preferentially formed at grain boundaries and due not to contribute to band gap narrowing defects because they are essentially part of the already disturbed grain boundary structure. An approximate calculation of the formation energies of intrinsic point defects formed at particular grain boundary sites has been given in Ref. (120) and the results are included
In
In In
Cu
Cu
(a)
(b)
FI. 110. Structural atomic models of a vacancy Vs~b (a) and an antisite defect |ns~b (b) at an incorrect Se-Se bond. Incorrect bonds occur at grain boundaries in the CuInSe2 lattice. (12)
SEMICONDUCTORS
393
in Table 17. It shows that the formation energies of the major intrinsic point defects in single and polycrystalline ternaries are different. The defects with the lowest formation energies are now Incub (1.1 eV), V~nb(1.2 eV), and Vs~b (1.3 eV) (using the abbreviated notation for grain boundary defects, see Chap 4.4) compared to the bulk defects Incu (1.6 eV), Cu~n(1.9 eV), and VIn (2.4 eV). Based on these results the majority defect pairs with the lowest formation energies for a given stoichiometry which can form at grain boundaries and in the bulk are included in Table 20. It shows that a variety of acceptor and donor defects occurs, however, unless their concentrations and trap levels, and their influence on the grain boundary states are unknown a prediction of carrier concentrations and conductivity type is not possible. Nonetheless a number of authors have tried to analyze their results by making assumptions about the majority defect, which seems in view of the complexity of the problem still rather speculative. Qualitatively one can see, however, that changes in the stoichiometry can shift the balance between donor and acceptor defects and lead to strong changes in the occupation of the grain boundary states and the free carrier concentration which mainly determines the conductivity. 7.3.3.3. Quaternary compounds. The relatively small band gap of CulnSe2 below the optimal value for photovoltaic applications of 1.5 eV limits the utilization of this material for single-junction devices. Nonetheless, most of the device structures have been based so far on the heterostructure concept (e.g. CdS-CulnSe2 cells) since the highest efficiencies have been obtained here for any thin film material [for a compilation of data see Ref. (25)]. The research efforts were mainly directed towards the optimization of the device parameters. A further improvement can be expected if multi-junction designs are employed in combination with a larger band gap material (e.g. a-Si on CulnSe2). Solar cells based on this concept have recently reached efficiencies of 14.6/o38) Another approach is to change the band gap of the semiconductor by alloying. Since the band gap increases for the compound series CulnSe 2 (1.04eV), CuGaSe 2 (1.68 eV), and CuAISe2 (2.67eV) the electro-optical properties of the quaternary compounds Culn~_xGaxSe2 and Culn~ xAlxSe2 compounds have attracted interest recently. The band gap of Culn~ _xGaxSe2 as a function of the composition, x, at room temperature increases according to the empirical relationship E ( x ) = 1.018+0.575x +0.108x2. (283) The most important property is the high absorption coefficient exceeding 2 x 104 cm-i near the optical absorption edge. Other than CulnSe2 which can be made p- and n-type Culn~_ x Gax Se2 films are always p-type which is assumed to be due to the lowered formation energy of shallow acceptor defects such as Cu~n and Vcu in the presence of gallium. The conductivity as a function of the composition, x, at room temperature shows a minimum at x = 0.5 and a strong increase towards the Cu-rich side most likely due to the formation of a secondary phase Cu2Se (Fig. 111). The full potential of these and other quaternary compounds cannot be estimated at present. It is evident that the complexity of the material such as composition, defect chemistry, and their relation to physical properties increases. Currently the main emphasis is still on the fabrication of thin films and the characterization of the electro-optical properties and it is unclear in many cases if the material problems can always be solved. 7.3.4. Interface properties and oxidation behavior It has been discussed in Chap. 3 that many of the device structures which are appropriate for polycrystalline compound semiconductors such as heterostructure, multi-junction or Schottky barrier designs contain a variety of interfaces betweeen different materials. The
394
......
I,o,o
,o"
~o-~
1(~' ' ~
;1~s
.:\"..
'r ,, 3 0 0 x
............
10 4
o. 7
0:,9 o:so o'.s,
is2
X-'P"
FIO. 1 1l. Conductivity of Cu(ln, Ga)Se 2 thin films vs ratio o f [Ga]/[ln]. (x and y define different compositions o f the compound [Cu2Se]l-x [(GayInl-y)2Se3]x .(283)
structural, chemical and compositional characteristics of the interfaces are critical to the performance and operational lifetime of such solar cells. This has also been demonstrated for Cu-ternary solar cells and the correlation between interface properties and the device parameters has been studied intensely. The greatest improvements in the performance of Cu-ternary solar cells have been made for the CuInSe2-based cells compared for instance to CuInS2- and CuInTe2- ceils. The most efficient device structure so far is the Cd(Zn)S-CuInSeE-heterojunction cell and two typical designs are shown in Fig. 29. Frontwall configurations, i.e. light incident on CuInSe2, have also been tried but have generally shown lower efficiencies. There are several reasons for this which seem to be connected with the CdS-CuInSe2 interface. First, since the deposition temperatures for CuInSe2 are higher (T ~ 350-400C) compared to CdS (T ,~ 150-200C), the interdiffusion of S and Se at the interface during fabrication especially along grain boundaries is minimized if the CdS deposition at the lower temperature is the second process. Second, the CdS or Cd(Zn)S seems to protect the light absorbing CuInSe2 layer from surface-environment interactions which degrade the efficiency. Third, it appears that an important annealing step in air which is necessary for the improvement of the performance (discussed below) affects the properties of the cells only if CdS is on top. The design of the cells has been further modified because of the observation that for low resistivity material on either side of the junction, which is necessary to reduce the series resistance (see Chap, 2.5), copper precipitates form at the interface which protrude into the CdS layer and destroy the junction properties. (25) Therefore thin layers of high resistivity (Cu-poor CuInSe2 and undoped CdS) are deposited on either side of the junction which seems to prevent the formation of the copper phase. Another solution to the problem is the deposition of Cd(Zn)S with a concentration gradient towards the interface. Cd(Zn)S 05-30 at% Zn) has also a wider band gap which improves the efficiency slightly. The obvious importance of the interface for the device characteristics (296) has stimulated a number of investigations with techniques such as XPS, EELS or EBIC. For instance results on CdS-CuInSe2-interfaces have shown the presence of either Cu2 S or Cu2 Se and hence indicate the possibility of chemical reactions at the interface. A complete analysis of the structure and chemistry of this interface has however not emerged yet.
SEMICONDUCTORS
395
The defect structures of the interface, the grain boundaries and the bulk is especially important to understand the effect of oxidation on the performance of CdS-CulnSe2-solar cells. A major and critical step in the fabrication of high efficiency heterostructure CdS and Cd(Zn)S-CulnSe2-solar cells is an annealing treatment (at 200C) in air or oxygen after deposition. This step improves the open-circuit voltage and the short-circuit current. It has also been observed that this effect is to some extent reversible if the films are annealed in vacuum or soaked in a reducing solution. (297)Results from a variety of electrical and optical measurements show changes of the electrical activity of the interface and the bulk. Since oxygen diffusion in single crystals is slow but may be enhanced along grain boundaries, it is likely that grain boundary states are mainly responsible for the observed effect. An experimental result that supports this view is a dependence of the required annealing time on the grain size. The major effect seems to be an increase in the net acceptor density due to reduction of the donor density in the compensated thin films. Capacitance and Hall effect measurements on CulnSe: films have shown an increase in the acceptor concentration from about 1014 to 1016cm -3. Infrared (IR) measurements provide evidence for the formation of In-O bonds and scanning tunnelling microscope (STM) investigations indicate the accumulation of oxygen at grain boundaries. The annealing effect is less pronounced in selenium rich films (As > 0) which points to the participation of the selenium vacancy which has a low concentration in these films (Table 20). These and other observations have led to a defect model that postulates the neutralization of selenium vacancies (Vseb) which are formed at grain boundaries and are treated as donor defects. One can see from Table 20 that the selenium vacancy at the grain boundary is one of the possible donor defects because of the low formation energy. A schematic illustration of the atomic structure of a selenium vacancy in Fig. 110 shows that an I n 4 ) bond can be formed if the oxygen atom occupies the vacant selenium site. This bond formation is supposed to neutralize the Vseb donor and increase the net acceptor concentration in the material. Though a process like the proposed one seems feasible a better understanding of the defect chemistry in single and polycrystalline CulnSe2 films is certainly necessary to confirm the model.
8. AMORPHOUSTHIN FILMS The microstructure of thin films which can be deposited on a substrate by a variety of deposition techniques such as evaporation, sputtering or chemical vapor deposition etc., depends on the substrate temperature. For instance for silicon a polycrystalline material (grain size ~ 600-1000/~,) is obtained for a substrate temperature above 580C, and a microcrystalline (~50-80 A) or an amorphous structure below this temperature, usually between 300-250C. These low deposition temperatures for amorphous films offer the possibility to deposit amorphous silicon on many different substrates for instance on polymers. Besides the fundamental scientific interest in the investigation of the amorphous state there also exist favorable electro-optical properties which have stimulated the investigation of this material. Although the potential of amorphous silicon (a-Si) was recognized almost 20 years ago, the electrical and optical properties of evaporated or sputtered films were not encouraging enough at the beginning to utilize the films for technical applications. The conductivity could not be influenced by doping nor did these films show any measurable photoconductivity. It soon became evident that the films had a high density of defect states which prevented a significant and variable conductivity.
396
PROGRESS
IN MATERIALS
SCIENCE
This situation changed when it was discovered that amorphous silicon which was deposited by glow discharge decomposition of silane (Sill4) exhibited a high photoconductivity and that the electrical conductivity was sensitive to small additions of PH 3 or B2H6 during deposition. It turned out that these enhanced electrical properties were related to hydrogen which is released during the decomposition process and incorporated in the amorphous structure. This breakthrough was an important discovery for electronic and photovoltaic applications. Since then, many structural and electrical properties of amorphous silicon have been studied which are summarized in many books and review articles. (37'298-3)At the same time the technical development has produced solar cells for low power applications such as watches and calculators. Amorphous solar cells now account for about one-third of the world's photovoltaic market. 8. l. Structure and Electro-Optical Properties of Amorphous Silicon 8.1.1. Band gap states Several deposition processes such as chemical vapor deposition (CVD), reactive sputtering, sputtered assisted plasma CVD and others can be used now to produce amorphous silicon for solar cells, although the (glow discharge) plasma deposition remains the major film deposition method. Hydrogenated films contain about 4--40 at.% of hydrogen and are generally denoted by a-Si: H. The importance of incorporated hydrogen can be demonstrated for instance by annealing hydrogenated films up to about 600C. Figure 112 shows that above about 500C when the hydrogen is released from the film the conductivity decreases abruptly by several orders of magnitude. Although it seems evident that hydrogen passivates defect levels in the amorphous films great efforts have been taken to study the nature of these defects and their interaction with hydrogen. An essential observation was that the loss of hydrogen and the decrease in photoconductivity are accompanied by an increase in the density of electron spins which can be measured by electron spin resonance (ESR). (3j) The ESR signal measures the existence and concentration of unpaired spins. Therefore the results indicate that hydrogen is trapped at defects with dangling bonds which are electrically active centers and if unsaturated reduce the lifetime of free charge carriers. In a perfect crystalline structure all four covalent bonds of an atom are saturated by next neighbors and no unpaired spins occur. Structural models of an amorphous covalently bonded crystal show that the average number of next neighbors is lower than four so that a certain fraction of atoms are only coordinated with three or two nearest neighbors. The
1/Ns [cm 3] 10 -14 10 -15 10 -Is lO -17 lO -18 lO -is
o(T)
[] P (RT) 1.0 10 -1 10 -2 10 -3 lO -4 lO -s
2OO
400
6OO
T [ C]
F]G. 112. Inverse of the spin density N s and normalized photoconductivity a (300K) o f undoped a-Si: H after different annealing steps. ) The corresponding dependence for 1/Ns indicates the relationship between the spin density and the lifetime o f electrons Ns = const.
SEMICONDUCTORS
397
threefold coordinated atoms, T~, possess an unpaired (dangling) bond which produces an electron spin signal in an ESR experiment. In order to describe the defect the notation, T q, will be adopted here, where T stands for the atom, q for the charge state, and z describes the coordination number. Twofold coordinated atoms, T~, can also occur and release the strain twice as effectively as a threefold coordinated atom. (3]) The neutral T~ center has only paired spins and thus does not produce an ESR signal. Though not directly observed, it is assumed that these defects also occur. More complicated defects have also been postulated such as pairs of two and threefold coordinated atoms, Tf-T~-, complexes of several defects, or microvoids. Hydrogen free amorphous silicon has spin densities of up to 102ocm -3 depending on the deposition conditions and it is assumed that they are related to uncoordinated covalent bonds mostly of the T~ atom. A hydrogen atom with a single electron can easily supply an electron and complete an unsaturated covalent. This bond formation will remove the dangling bond level from the forbidden band gap (passivation) and trap the hydrogen as discussed already in Chap. 4.1.4. For instance in a good (glow discharge) plasma deposited material the spin density can be reduced down to 10Is cm -3 after passivation with hydrogen. In order to obtain more insight into the electronic properties of amorphous silicon it is necessary to determine the distribution of the trap states in the band gap. The application of a variety of experimental techniques has provided information about the density of states. Figure 113 shows a result from field-effect measurement in n-doped a-Si : H which depicts the main results. (33) Since the majority of atoms are still fourfold coordinated the essential features of the crystalline band structure, the existence of extended valence and conduction band states, are still maintained. The amorphous state lacks however the long range order which gives rise to a gradual (probably exponential) decrease of the density of states (band tails) given by Ec - E E - - Ev NTC = Wooexp ~'TTc ; NTV = N,,o exp KTT------~
1021
(204)
lO 2o
/
i
,? E o
10t9 uJ h(n
Eq
1017 1.5
. . . . . .
~
. . ,,
E~
1.0 E N E R G Y BELOW E c
0.5 (eV)
FIG. 113. Density of states for n-dopes a-Si:H determined from field-effect measurements. (33)Position of the Fermi level E v indicated by arrow.
398
PROGRESS
IN MATERIALS
SCIENCE
with about Nco = Nvo = 1021-1022 eV-' cm -3, KTTc and KTTv are the characteristic widths of the conduction and valence band tails, respectively. Experimental values are for instance KTTc = 26 meV and Taw = 42 meVo4) which means that the valence band tail extends deeper into the gap. These tail-states near the band edges are assumed to be associated with the distorted bonds of fourfold coordinated neutral silicon atoms, T~, in the amorphous structure. Because of the gradual transition of the density of the extended states the band width is not clearly defined. Nonetheless, particular experimental techniques allow to determine parameters which can be used to attribute a certain value for the band width. For instance the optical absorption which will be discussed in the next section can be analyzed analogous to a crystalline semiconductor and is usually used to determine the "optical band width". A fraction of tail states also originates from silicon atoms with three silicon neighbors and one hydrogen atom in tetrahedral configuration. These additional states from the silicon-hydrogen bond changes the optical band width depending on the hydrogen content. The optical gap increases from about 1.5 eV for pure a-Si to about 2.0 eV when 30% H is present335) All states in the valence bandtail, T~-/, are donor like, namely positive when empty and neutral when full. The states in the conduction bandtail, T~/-, are acceptor like, they are neutral when empty and negative when occupied. Grossly distorted bonds form states deeper inside the gap and there exists a critical limit forwhich the states have to be considered as localized states. These critical energies are called the mobility edges (Ec, Ev) and are located below the optical band gap edges. The localized states inside the mobility edge essentially determine the Fermi energy and the electrical transport properties. Superimposed on the band tail states are two deep level bands in the lower and upper half of the band gap (at about E ~ - 0.4 eV, Ev + 0.6 eV) which are thought to be related to dangling bonds of mainly threefold coordinated silicon atoms (see Fig. 114). They are considered as the defect bands of two different charge states of the neutral dangling bond, T~. The separation of the two levels is due to the electrostatic correlation energy U of the two electrons. Since the local environment of the dangling bonds varies in the disordered structure one can assume that the positions of their trap levels in the gap varies and leads to the observed distribution of states. It is mostly assumed that the upper level E]/b corresponds to the negatively charged defect T3 = T~ + e, when an electron is trapped, and the lower level EJ-b/ to the positively charged defect T~ = T~ + h. In order to understand the effect of the Fermi energy on the occupation of the dangling bond levels one can use the eqs (115) from Chap. 4.1.1. The numbers of dangling bonds in the different charge states NJ-b and N& are given by N~- ---- N~b exp [ b ~
with the total number given by Ndb = Ndbo+ N~-b+ N~-b. It is evident from this equation tot o that a shift in the Fermi level changes the concentration of positive and negatively charged states as indicated in Fig. 114. In an undoped material the bonds are neutral and occupied by one electron. In a p-type material the lower levels (E~b/) begin to empty and the positively charged dangling bond T~ dominates which carries no electrons. In n-type material the upper levels (E~/b) become partly (or totally) occupied and the negatively charged dangling bond T f with two electrons dominates. Therefore the ESR signal which measures only the neutral bond with an unpaired spin should decrease when the Fermi level shifts towards the band edges. This reduction of dangling bond T~ centers is actually observed for low doping levels.(3o6)
(205)
SEMICONDUCTORS
399
Ec
-E F
D
Ev
N (El
N (E)
N (E)
(a)
, j
(b)
(c)
EF
N (E) (d) (e)
N IE)
FIG. 114. Simplified density of states model for dangling bond states in undoped and doped a-Si:H. E c and E v denote the conduction and valence band mobility edges. (a) undoped, (b) moderately phosphorus-doped, (c) heavily P-doped, (d) moderately boron-doped, (e) heavily B-doped. The defect states are labeled by their state when filled if they are below E r and empty if they are located above E F .
The capture cross sections for electrons an and holes O'p for the donor T~-/ and acceptor T~/- differ considerably (an+/>>a p+/oand a o/-<< a g/-, respectively), whereas the capture cross sections for electrons and holes for the neutral state are about equal a+n/ ~ ap/+ (see Fig. 114). These dangling bond states are effective recombination centers especially in undoped a-Si-H where they can capture electrons and holes. A consequence of the different charge states is therefore that the lifetime in undoped a-Si:H is rather low and depends very sensitively on the position of the Fermi energy. Amorphous silicon usually also contains impurities which contribute to the density of states in the band gap. The most common impurities are oxygen, carbon and nitrogen with concentrations typically in the range of about 10~8-102cm -3. These impurities have been shown to affect the diffusion length and thus the efficiency of solar cells, but shall not be discussed further here [see for instance Ref. (307)]. 8.1.2. Doping in a-Si:H The problem of doping in amorphous silicon becomes evident if one considers that atoms in the disordered structure are to some extent rather threefold (or less) coordinated compared to the fourfold coordination in crystalline silicon. Doping atoms in the crystalline structure supply holes and elecrons due to an excess or deficiency of electrons compared to the number of electrons which are necessary to form four covalent bonds (formulated as the 8-N rule). It has been argued that a doping atom on a substitutional threefold coordinated site will therefore not be able to supply holes or electrons according to this basic argument.
400

n [cm'3] 1019 10 18 1017 10 16 1015 1014 1013 1012 1011 1010 I 4 I 10 is I I 1017 I
f
/
I 10 l I
N [cm -3 ]
FIG. 115. Free carrier concentration n vs phosphorus concentration N d for crystalline and amorphous silicon. (1) monocrystalline, (2) microcrystalline, (3) polycrystalline, (4) CVD-deposited amorphous, (5) plasma-deposited amorphous. )
Nonetheless one can observe a doping effect in a-Si:H, however, only a fraction of the incorporated impurity atoms contributes to the free carrier concentration as compared to the doping situation in crystalline silicon. This behavior is shown in Fig. 115 where the free carrier concentration as a function of the impurity (phosphorous) concentration is presented for crystalline, micro- and polycrystalline and amorphous silicon. The polycrystalline silicon (curve 3) shows the characteristic decrease of the carrier concentration below about 10~8cm -3 (for a grain size of about 0.1 #m) due to a trapping of charge carriers in the grain boundary states, which has been discussed in Chap. 4.6. The hydrogen free amorphous silicon deposited by a CVD-process (curve 4)~3)has a trap density of about ~ 1019 cm -3 in the gap which is indicated by the rapid decrease in free carrier concentration below this value, when free carriers become trapped in these states. The low carrier concentration could however also be due to an insufficient activation of the doping atoms when it is incorporated on a site where it is electrically inactive. The behavior of the hydrogenated a-Si: H (curve 5) shows that both effects are important, namely trapping of carriers in bandgap states and low activation of the impurity atoms. Below about 1018cm -3 which correspond to the trap density in this case the free carrier concentration decreases rapidly. However, even above about 1019 c m -3 the free carrier concentrations remains several orders of magnitude below the impurity concentrations which shows that only a fraction of the doping atoms is activated. Phosphorous is the most common n-type dopant in a-Si:H. Normal structural bonding for phosphorous is threefold and one might expect that all P-atoms are preferentially incorporated in the ~ configuration where it is electrically inactive and merely a network relaxer and defect compensator. Experimental results show however that P-atoms are also incorporated in the fourfold configuration where they act as a donor P4 +/. A possible explanation may be that phosphorous atoms in the fourfold configuration form a complex with another defect which is energetically comparable with the ~ configuration. For instance the P~--T 3 pair has been proposed as a possible complex configuration.(32) Although the configuration of the fourfold P-atom may still be controversial it is clear that a fraction of the incorporated atoms are electrically active and contribute impurity states to the band tails
401
or just below. One can define the doping efficiency r / = [P4]/[Ptot] of the electrically active component with respect to the total concentration of doping impurities. From experimental data for phosphorous and arsenic the following estimations have been obtained: r / ~ 0.3 - 0.4 (P) and r / ~ 0.2 (As). (37) Other authors 8) also report much lower values of r/< 0.06 for P which indicates some difficulties to determine the doping efficiency. Despite these problems it is of practical interest that one can vary the Fermi energy over a wide range up to about 0.1 eV below the mobility edge for high concentrations of phosphorous. Boron is the usual p-type dopant in a-Si:H. Because of its complex chemistry it is also difficult to predict the atomic configurations in the network where it becomes electrically active. Experimental data indicate that the vast majority of the boron atoms are in the neutral threefold coordinated configuration B~. (32) Similar to phosphorus it has therefore been suggested that a defect pair must form to stabilize the electrically active fourfold configuration e.g. B4-T~. Doping efficiencies of about r / < 0.06 have been reported which indicates the low activation of boron. ~39)For high boron concentrations the Fermi energy can also be shifted to about 0.1 eV above the mobility edge. The doping behavior of phosphorous and boron in amorphous silicon differs also in another way from crystalline silicon. ESR measurements show that for slight doping levels the concentration of dangling bond centers decreases and finally completely disappears. This is in accordance with the conversion of the dangling bond center T~ into the spin-paired centers T~- or T~- when the Fermi energy shifts as discussed in the previous section. With the introduction of more impurity atoms a new spin signal appears, however. The density of the dangling bond states Ndb has been found to increase as Ndb '~ (N~t) I/2 where N~t is the total concentration of the doping element.(3~)The increase in the density of states for a doped film can be seen for instance in Fig. 116 if one compares the increased defect concentration near midgap with the density of about l016 c m - 3 of undoped a-Si: H. Additionally, the distribution of the defect levels is also changed and varies with the deposition conditions. For heavily doped films the ESR signal strength decreases again. (3H)These doping induced deep levels can also be observed in the optical absorption as an increase of the subband gap absorption (see next section). Several explanations have been proposed for this behavior. For instance the observed increase of the number of dangling bond states in the case of n-doping has been explained by the assumption of a chemical equilibrium between the inactive impurity (po) and the electrically active (P~-) center via the simple reaction(39~ p0,__~p~- + D3 (206)
where D 3 is a dangling bond center that is formed by the incorporation of phosphorous. Applying the law of mass action yields the observed dependence [D3] "~ [p0]~/2. The nature of the dangling bond center remains unclear in this model. Another explanation for the formation of new spin centers is that with the shift of the Fermi energy towards the mobility edges other trap centers become occupied which possess an odd number of electrons so that they produce an ESR signal. For instance it has been suggested that a twofold coordinated silicon atom T~ may be activated as T{ or T~- depending on the type of doping. The quenching of the signal for heavy doping is then easily explained by the formation of T 2- or T 2+ centers. The increase in the total number of defect centers and the empirical relationship for the spin density however cannot be explained so easily in this model. Despite the controversial explanation so far the important result is that the doping in amorphous silicon has two effects. First, it changes the Fermi energy and thus increases the concentration of free carriers. Second, it introduces new deep level centers which partly
402
10 t9
_.-,.
>
i
1018
cJ
E
oD 1017
>Ira L=J 3
B2H6 PH 3
(ppm) (ppm}
I0~6
oooooooo
ooooooo
0 4 9 17 37 44
290 260 340 300 340 280
I01'~'{ "
Y 2.0 4
1.6
1.2 0.8 ENERGY BELOW E (eV)
0.4 E
FXG. l l 6 . Density o f states derived from D L T S measurements for phosphorous doped a-Si:H containing various amounts o f boron counterdoping. The gas-phase PH3 and B2H 6 concentrations are indicated.O 3)
compensate the doping impurity. The density of states of doped a-Si: H is thus increased and changed compared to the intrinsic material. It is evident that the new defects also have a significant influence on the electrical conductivity and the lifetime in a-Si:H.
8.1.3. Electrical transport
The character and distribution of the band gap states determine the mobility and the lifetime in a-Si: H in different ways. While the shallow tail states essentially limit the mobility of charge carriers, the midgap states dominate the recombination. These midgap states change in concentration and distribution depending on the position of the Fermi level as discussed in the previous section. In undoped material the neutral dangling bond dominates which is an especially active recombination center, therefore their density has to be kept low to achieve a reasonable lifetime. In doped material with the Fermi energy shifted towards the mobility edges most of the deep trap centers are occupied and therefore are rather trapping than recombination centers. However, since their density usually increases by several orders of magnitude (>1017cm -3) they can also have a detrimental effect on the lifetime. Also important for the lifetime are band tail states which can become recombination centers under illumination as the quasi-Fermi level approaches the band edges. This loss mechanism is not present in crystalline semiconductors, thus a Si:H has a somewhat lower theoretical conversion efficiency than an ideal semiconductor with the same band gap. (312) Generally the
SEMICONDUCTORS
403
lifetime in doped a-Si : H is very short and most of the carrier collection in a solar cell has to be in undoped layers. Therefore the p-i-n configuration is usually used for amorphous solar cells where the intrinsic region is the most active zone. Lifetimes have been estimated for undoped material with values ranging between 3-30 #S. (313) The band tail states have a strong influence on the mobility in amorphous silicon. Undoped a-Si: H is almost an insulator (a < 10-10 ~ - ~cm- 1) and even doped amorphous silicon still has a very low conductivity of about tr ~ 10-1 f~-~ cm-l. This is due to a small mobility with typical values below 0.1 cm2/Vs at room temperature. This indicates that the mobility is determined by mechanisms other than scattering of charge cartier in the extended states above the mobility edges. Because of the high density of localized band tail states a significant fraction of carriers is trapped in these states. Conduction in the localized states is only possible by thermally activated hopping or tunneling between these states which in both cases results in a very low mobility (below 0.01 cm2/Vs) and a strong temperature dependence. Charge carriers can however also be thermally activated into the extended states above the mobility edge. The mobility in amorphous silicon may thus be in general a combination of carriers moving just above the mobility edge in the extended states and carries hopping in the localized states. The mobility in localized states can be measured by the time-of-flight (TOF) method which is most often performed on insulators. In such an experiment a sheet of carriers is generated in a sample for instance at one side of a p-i-n-diode by a short light pulse and drawn through the sample by an applied field. The time is measured to reach the collecting electrode at the opposite side. After the flash of light carriers drop back to the conduction band edge and can be trapped then by the localized states and re-emitted. For strong applied fields the measurements mainly determine the mobility in the extended states, whereas for lower fields the drift mobility ~td due to capture and re-emission in the localized states is obtained. A typical example for the drift mobility #d of electrons as a function of the temperature is shown in Fig. 117. It shows that the drift mobilities in the localized states are thermally activated with activation energies of 0.15 eV for electrons and 0.35 eV for holes. (314)Experimental values for the mobilities of electrons and holes in the extended state are #, = 13 cm2/Vs and #p = 0.5 cm2/Vs, respectively.4) The carrier mobility ~H which is determined by Hall effect measurements, the usual method for conducting crystalline materials, is another method to study the mobility in amorphous semiconductors. The Hall mobility measures the cartier transport by diffusion and may be more related to the mobility which determines the conductivity and diffusion length. Hall mobility and drift mobility differ in general in an amorphous material. In solar cells, both #H and /~d may play an important role, depending on whether there are field-free regions in the structure, in that case, carrier transport is by diffusion and directly related to the magnitude of #H. In the presence of strong electric fields for instance in the junction region the mobility may be more determined by #dIt seems to be a rather general observation in amorphous semiconductors that the sign of the Hall voltage is opposite to that of the dominant charge carrier. This reflects the anomaly of the transport mechanism in disordered structures. The Hall mobility # . as a function of the temperature is shown in Fig. 118 for a number of p- and n-type specimens. 15~ The temperature dependence is exponential for boron, but shows a more complicated behavior for phosphorous. The experimental data for phosphorous have been analyzed by assuming a two path model. The results suggest that above 250K conduction is mainly through extended states whereas below this temperature it takes place by hopping between localized states which in doped material would be mainly impurity levels. For high doping levels above 5 1018cm -3 impurity conduction becomes the dominant transport mechanism even at room
404 10
ie% i =
'o
"%%
020
I I I
10.1 "~o,,
--
'~
olo
----'~" - .'~,,,.
AA
{--
O,O4
10"; 0.02 0.01 2

I 3 I I 4 5 IOa/T (*K")
1(~
o~
IO00/T (*K")
FIG. 117. Temperature dependence of the electron drift mobility for an undoped a-Si:H thin film with an applied field 104Vcm -~ (solid circles) and 2 x 104 V cm -~ (open circles). Solid lines are theoretical fits.(3~4)
FIG. 118. Temperature dependence of the Hall mobility for doped a-Si: H. For n-type samples the volume gas ratios PHflSiH4 are (1) 98 ppm and (2) 3 x 104 ppm. For the p-type samples the volume gas ratios BEH6/SiH4 are ( i , A ) 5 104 p p m and ( 0 ) 2.3 x 104 ppm. t315)
temperature. m6)Other groups however have obtained different results so that a controversy about the interpretation of the results remains. The Hall mobility in undoped a-Si:H (with a slight excess of electrons) is given in Fig. 119 and shows an exponential increase above 360K with an activation energy of about 0.13 eV. This corresponds to the activation energy for the drift mobility gd of electrons from TOF measurements and indicates that the mobility occurs through localized states above room temperature in this case. The mobility at low temperatures is assumed to occur through tunneling. The data have been analyzed using eq. (37) in Chap. 2.4.1 assuming the two different modes of transport. The important parameters for the performance of the solar cell, mobility and lifetime, enter most of the equations which describe the performance of a solar cell as a product g~. Therefore experimental results are frequently presented for the mobility-lifetime product gz which can be determined from photoconductivity and diffusion length measurements (see eq. (38) in Chap. 2.4.1). A high performance solar cell must efficiently collect all light-generated carriers in the active region of the device. This condition can only be met if the diffusion length L is greater than the thickness of the active region of the cell. The absorption behavior of amorphous silicon requires a thickness of about 0.5 gm to absorb most of the light. Using the expression L = (KTgT/e) 1/2 this translates into a mobility-lifetime product gz > 10 -7 c m 2 / V . Considering the mobility values for electrons in undoped a-Si :H at room temperature from Fig. 119 (gn ~ 0.1 cm2/Vs) and the lifetime values mentioned above (10ps) one obtains gT ,,~ 10-6cm2/V which is sufficient for the operation of the cell. The mobility-lifetime product in undoped a-Si: H depends primarily on the density of deep trap states. Experimentally it has been observed that both for electrons and holes the following relationship is valid for the product gz Ndb = const. (Fig. 112).~4> Since the dangling bond
SEMICONDUCTORS
T(K) 300
I~T t i
405
500
I
400
}
250
I
200
I
,o
-3 0.1-lIE
~[ 3:
~ ~
E:O.13eV"~"
~m2N] 10 -s 10 -7 10 -s
10 -9
a - Si:H
J
N o Electron e le
O.O1
2.0
5.0 4.0 I/TxlO 3(K'l)
5.0
1.0 2.0 3.0 B-doping level [ppm] FIG. 120. Mobility-lifetime product/~z as a function o f
FIG. 119. Temperature dependence o f the Hall mobility
for electrons in undoped a-Si:H. (1) dark and (O) illumination with 400 Wm-2. 016) The solid line is a fit of eq. (37) to the experimental points.
the boron concentration for a-Si:H. l)
trap density Ndb varies with the deposition technique the diffusion length can be optimized by the deposition processes. Further examples for the doping dependence of the product/~z for electrons and holes are given in Fig. 120('') for boron doped a-Si:H films. The low mobility values for the electrons (minority carriers here) at higher doping levels (#z < l0 -8 cm2/V) show that the collection efficiency in a p-doped region of a solar cell will be rather poor. It is therefore important to minimize the thickness of the doped regions usually below l0 nm (see for instance the typical configuration for a p-i-n solar cell in Chap. 3.6, Fig. 24). It should also be noted that for undoped films #z is considerably smaller for holes than for electrons, thus the diffusion length is mainly determined by electron transport. This difference suggests for the design of p-i-n solar cells to minimize the average travel path for the holes rather than for electrons. In fact the more efficient cells are produced with a p+-layer as the window for the incident light. The conductivity of doped a-Si:H is determined by the mobility and concentration of minority carriers. The previous results have demonstrated that both parameters depend in a complex way on the doping level and the temperature. The interpretation of the results from the perspective of the band structure and the possible transport mechanisms is thus difficult. Nonetheless, the measurement of the conductivity is important because it directly characterizes the electrical transport in a-Si: H, and can also give valuable insight into the electrical transport and doping behavior of the material. Experimental results for the room temperature conductivity for p- and n-type films as a function of the doping level and various doping elements are shown in Fig. 121.(317)Rather high doping levels are required to obtain sufficient conductivity necessary for solar cell applications. A numerical analysis of this and other
JPMS 35-3/~-N
406
/
'~ I to)
i
j, ibi
I
i
J
So I
:l,o.
,d
,0'
,o" i"
n I (cm "31
FIG. 121. Room temperature conductivity a0 as a function of implanted impurity concentrations n x for elements of group III and V. (a) p-type: B (O), A1 (O), Ga (A), In ( ~ ) T1 (). (b) n-type: P (O), As (O), Sb (A), Bi (V). 17)
results by various authors has shown that by the incorporation of the doping impurities new defects states are produced near the midgap region in accordance with the results of other measurements discussed earlier. 8.1.4.
Optical absorption
Crystalline silicon as an indirect semiconductor has a rather low absorption in the visible spectrum range since only a few of the electronic states near the band edge can take part in the absorption process due to the k-selection rule. This situation is different for amorphous silicon, since most of the electronic states near the band edges (tail states) are available for nonradiative optical transitions. Correspondingly, the optical absorption coefficient is much higher compared to crystalline silicon as can be seen in Fig. 122. Over most of the visible light region (1.9 eV <hv < 4.0 eV) the absorption is greater than 104 cm -l. For high photon energies (hv > 1.6 eV) the absorption can be described by the energy dependence of eq. (3) for direct band transitions. The band width Eg which is obtained as a fitting parameter here is called the "optical band width" in order to distinguish it from the mobility band width which is inferred from transport measurements. For hydrogen free a-Si:H the optical gap is about 1.7 eV and for photon energies above the gap the absorption coefficient is about ten times higher compared to c-Si. This has the consequence that a film thickness of about 0.5/~ m is already sufficient to absorb most of the light. It is interesting to note that it is the reduced optical penetration which compensates for the low carrier diffusion length and the poor conductivity in a-Si. For photon energies below the optical gap the absorption drops off slower compared to the energy dependence extrapolated from higher energies. This increased subband absorption is similar to the absorption in heavily doped crystalline semiconductors and can be described by the Urbach law [eq. (8) in Chap 2.2].
ao (v) = C(v) exp{ EgEohv'~j

where E o is the Urbach energy which relates directly to the characteristic distribution width of the band tail states. An example is given in Fig. 123 where the characteristic energies
SEMICONDUCTORS
FOR SOLAR
CELL APPLICATIONS
105 i i t
I L
I
407
104
Wavelength
1.0
0.8
0.7
(pro) 0.6
=-. 103 I
0.5
/
60 meV - ~ / / /
0.4
m u
102
_'~ 10 s p,
o J
< 10
--~
a-Si f
j.
/
./'/
/ / - - ~ 48
meV .e.~ I
103
102 , 1.0 FIG. 122.
10"1
0.8
1.4
Optical
2.2 Photon Energy (eV)

absorption coefficient
1.8
2.6
3.0
1.0
1.2
1.6 1:8 Photon Energy (oV)
11.4
21.0
2.2
of mono-
FIG. t23. Optical absorption coefficient of a-Si:H as a
crystalline silicon, hydrogenated and non-hydrogenated amorphous silicon as a function of the photon energy.(3)
function of the photon energy. Theroretical fits of the Urbach law (eq. (8)) to the data indicate exponential band tails with characteristic widths of Eo = 48 and
60 meV. (318)
Eo ~ 60 meV and ~48 meV have been obtained. The lower value corresponds well to the characteristic width of the valence band tail states obtained from TOF measurements on material prepared under similar conditions. (31g)This suggests and it is generally accepted now that the Urbach edge is primarily due to the absorption from the band tail states. An important consequence of the tail states is that nonradiative recombination can occur which limits the lifetime and is thus inescapable for amorphous semiconductors. This is important for the photovoltaic applications since a limitation of the lifetime also limits the open circuit voltage in solar cells which can be seen from eq. (96) in Chap. 3.1 where the reduction of the lifetime increases the saturation current and thus reduces the open circuit voltage. It has been estimated that the band tail states inferred from the transport measurements described above limit the open circuit voltage of an amorphous silicon solar cell with an optical gap of 1.7 eV to about 1.0 eV. (312) For comparison a crystalline semiconductor with the same band gap but only radiative recombination would have a maximum output voltage of 1.4 eV. The deeper energy levels within the band gap have a strong influence on the absorption for lower photon energies. Figure 121 shows for instance that hydrogen free a-Si has a significantly larger absorption below the mobility gap of about 1.7 eV for lower photon energies which is due to the much higher density of trapping states in the gap. Since the
408
PROGRESS
IN MATERIALS
SCIENCE
density of these band gap states depends strongly on the deposition conditions the absorption strength varies accordingly. The passivation of these states in a-Si: H decreases the subband gap absorption significantly. The absorption in this range can be used to estimate the density of midgap states and to study their behavior. At the same time the incorporation of hydrogen also changes the optical band width which varies between about 1.6 and 1.8 eV. Doping has a similar effect, it generally increases the absorption in the entire visible range but particularly at lower photon energies. This has important consequences for the device design since doped regions also have high recombination losses and as much as 2 0 o of the incident light can be lost by absorption for instance in the boron doped p-layer ( g 10 nm thick) of a p-i-n cell. (37) 8.1.5. Amorphous silicon alloys The development of multi-junction cells seems to be the most promising device technology for improved efficiencies. Since multi-junction concepts require materials with different bandgaps stacked upon each other to utilize the incident light more efficiently (see Chap. 3.6) it was necessary to develop a-Si films with different band gap. This can be achieved by amorphous silicon alloys where the band gap can be varied over a wide range, so that the designer has substantial flexibility in constructing optimized devices. So far amorphous silicon-germanium alloys appear to be the most promising semiconductors for the narrow bandgap portion of double- or triple junctions devices. Since amorphous germanium has a lower bandgap it reduces the bandgap of the silicon-germanium alloy with increasing germanium concentration. In general, increasing the germanium content decreases the mobility-lifetime product and hence the photoconductivity. Since at the same time no increase of the midgap density of states, nor an increase in the tail density of states is observed, it has been suggested that the poor transport properties are an intrinsic property of the amorphous structure of the alloys. In particular it is proposed that a preferential bonding between Ge-Ge occurs which reduces #z. (35)The situation can be improved by the addition of hydrogen which seems to promote the Si-Ge bonding as has been observed by Raman spectroscopy. In addition, alloys prepared with hydrogen also show less light degradation compared to the hydrogen free amorphous alloys. Amorphous silicon alloys with a wider bandgap can be obtained by the addition of carbon and the formation of a-Si 1_xCx :H alloys. However, many of the parameters relevant for the photovoltaic performance are not yet available and the development of these alloys is still at the exploration level. It shall also be mentioned that microcrystalline films of a-Si:H and a-SiC: H which can be obtained at slightly higher deposition temperatures are currently investigated from the perspective of their utilization in multi-junction solar cells. 8.2. Stability and Light lnduced Degradation Hydrogenated amorphous silicon presents an important problem which is less desirable for photovoltaic applications. It was first reported in 1977 that the photoconductivity of a-Si:H degrades when exposed to light. This phenomenon which mainly affects undoped films is known as the Staebler-Wronski (S-W) effect named after those who reported it first. (34'35) The main features of the effect are shown in Fig. 124. First, it is observed that the photoconductivity of the films decreases during illumination, and secondly, that the dark conductivity decreases below its starting value. After heating the film in the dark at above 150C the photo-induced effect can be annealed and the conductivity returned to its original
S E M I C O N D U C T O R S FOR S O L A R CELL A P P L I C A T I O N S
409
i(~ 4
I LL U M I N A TIO N ' - - - - - - - - - . - - I
7
to-g
I
i 0" ? I I e~ o
I io-a
\
u6 9
50
I00 TIME (MIN)
150
ZOO
250
FIG. 124. Staebler-Wronski effect for undoped a-Si:H films showing the decrease of the photoconductivity (solid line) and dark conductivity (dashed line) during illumination with 2000 Wm -2 filtered tungsten fight.~)
state. The effect is perfectly reversible and stable at room temperature. It can involve enormous changes in the conductivity which has of course strong implications for the technical applications. Whereas the original state has a clearly defined conductivity, the measurable state can have a range of conductivities depending on the light exposure. The reversible S-W effect has been observed for a-Si:H films fabricated under a wide range of deposition conditions and also for a-Si-Ge alloys. The continuous transition from the stable to the metastable state depends, however, on the original properties of a-Si film and the level of illumination3319) At room temperature and illumination of about 1 sun (typical for photovoltaic applications) it may take several hours to reach the metastable state. Later EPR measurements have shown that during illumination the dangling bond density increases from a low starting value of about 1016cm -3 up to values of about 1017cm -3. The photo-induced increase in the dangling bond density Nab increases the density of mid-gap states which can then capture free carriers and thus reduce the dark conductivity. If one expresses the dark conductivity by = aoeXp{-K~ t (207)
where Ea = E c - EF is the activation energy for mobile carriers (here electrons) then experimental results show that for undoped films for instance Ea can change from about 0.55 to 0.9 eV and for phosphorous-doped films between 0.3 and 0.55 eV corresponding to the shift of the Fermi energy towards the band center. The direct correlation of the change in conductivity and illumination with the corresponding movement of the Fermi level is however
410
complicated by the fact that the prefactor tro changes simultaneously corresponding to the experimentally determined expression a o = C exp(DEa) (208)
where C and D are constants. This exponential dependence of the prefactor on the activation energy is known as the Meyer-Nedel rule (32)and accounts for the observed saturation of the conductivity decrease. However, there is a large range of values for the constant C and D which have been reported for the different material and deposition conditions. The creation of dangling bonds is thus a self-limiting effect which reaches a saturation value after several hours. The Staebler-Wronski effect is mainly observed for undoped films but also occurs both in n- and p-type doped films. It has also been reported that for slight n-doping the dark conductivity can increase rather than decrease upon exposure of light.(319) For higher doping levels above 1017 c m -3 the Fermi level is essentially pinned by the acceptor or donor states and not affected any more by the formation of deep trap levels at a concentration between 1016-1017cm -3. Although there is general agreement that the formation of metastable compensating defects are responsible for the Fermi level shift, there is no satisfying explanation yet for the mechanisms which can cause the observed large changes in the prefactor. The transition from the metastable state to the original state by annealing is a thermally activated process which has a unique activation energy of about 1.5 eV independent of the material and the conditions under which the metastable state has been reached. It has been observed that this value is close to the activation energy for the diffusion of hydrogen in amorphous silicon which may suggest that hydrogen can play a role in the formation and/or passivation of the metastable defects. The degradation effect can also be observed in the dark when large currents are injected under forward bias of a p-n junction. (321~This important observation indicates that it is the trapping and/or recombination of the light-generated carriers rather than light absorption itself which causes the photo-induced degradation effect. The light induced degradation problem has become one of the most investigated phenomena in the photovoltaic research field obviously because of its impact on this promising technology. Almost every conceivable technique has been applied to elucidate the problem but a generally accepted model has not emerged yet to explain the degradation. A general model has to explain not only the increase in dangling bond states but also several other features which can be observed and additionally characterize the photo-induced degradation. It is evident now that the carrier recombination and/or capture induces the formation of metastable defects which return into their initial state by annealing. An important question for the explanation of the effect is whether the degradation is due to impurities or an intrinsic property of the disordered structure. Experiments which have been performed at very pure a-Si:H films seem to rule out the participation of common impurities such as oxygen or nitrogen. Most models have therefore focused on processes which are related to the amorphous structure itself, particularly the weak (strongly distorted) silicon-silicon bonds in the bandtails, T4. It has been proposed that an elementary process can be the breaking of a weak bond generating two adjacent dangling bonds which are stabilized by a relaxation of the network or the switching of the bonds (Fig. 125). A large number of other models have been suggested and one should consult the references for more details. (3'319~None of these models has emerged yet as a generally accepted theory therefore it appears that the problem of the Staebler-Wronski effect is not completely solved yet.
411
FIG. 125. Schematic representation of the bond-breaking model in a-Si:H to account for the observation of metastable, light induced dangling bonds as proposed in Ref. (331)
The degradation phenomena have important consequences for the operation and the design of the solar cells. In general, most of the degradation under illumination occurs in the first 48 hr of operation, and is accompanied by a drop in the efficiency of the solar cells by 15-20/o. (307) During illumination both the photoconductivity and the dark conductivity decrease as discussed above and shown in Fig. 124. In view of the previous results the dynamics of the process and its impact on the efficiency are rather complex and cannot be evaluated easily. It is useful to consider the parameters which determine the photoconductivity (eq. 28a,b) for instance for undoped material aph = a + eG #n z, (209)
where G is the generation rate of electron-hole pairs (per unit volume and time). Since/~. % is higher for electrons in undoped material (see Fig. 120) the hole contribution has been neglected here. During illumination the dark conductivity tr (eq. 207) decreases as discussed above. The effect of the photo-induced defects on the second term is less evident from the previous results. The illumination can in principle effect the generation rate G, the mobility #., and the lifetime z.. The generation rate G can be effected if the spectral response which is mainly determined by the band structure near the band edges is changed by light-generated defects. Similarly one would expect a change in the mobility if the band tail states are greatly affected. Experimental results show, however, that both generation efficiency and mobility are hardly changed which indicates that the light generated defect levels lie mainly in the center of the gap. Consequently the light-induced changes in the photoconductivity are primarily due to the decrease of the lifetime resulting from the introduction of the metastable defects. The impact on the efficiency is determined by the change of the short-circuit current, I=, the open-circuit voltage, Vo, and the fill factor FF (Chap. 2.3). From the previous discussion it is evident that I~ is less affected whereas the reduction of the minority carrier lifetime mainly increase the saturation current [eq. (96)] and thus decreases the open circuit voltage. This is in qualitative agreement with experimental results obtained for amorphous solar cells. The simultaneous decrease of the dark conductivity increases the series resistance of the solar cell and therefore mainly affects the fill factor (FF) (see Chap. 2.3). Both factors reduce the efficiency of a solar with the duration of the illumination until the saturation of the defect formation has occurred. Conventional solar cell designs for amorphous films use the p-i-n structure (see Chap. 3.6). Since the degradation is primarily caused by changes in the properties of undoped a-Si:H layers it is the intrinsic layer in a p-i-n cell which is mostly affected. Slight doping of the intrinsic layer generally improves the stability but at the expense of the efficiency. Other technical solutions utilize the fact that thinner (undoped) layers are generally less susceptible to degradation effects. Better stabilities are also reported for multi-junction cells and the use
412
of amorphous silicon alloys. The best results (12.4%) have been reported for a-Si:H with a tandem p-i-n layer structure which consists of two p-n junctions (see Fig. 30 in Chap. 3). This result is still well below the theoretical limit for a single-junction cell of about 20% which follows from the upper limits for I~ from the optical absorption data, for the open circuit voltage of about 1.05 eV and for the fill factor FF = 0.86. (312)
REFERENCES 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. W. SMITH, Nature 20, Feb (1973). C. E. FRITTS, Lumiere Electr. 15, 226 (1985). L. O. ARNODAL and P, H. GEIGER, Trans AIEE 4 6 , 357 (1927). W. SCHOTTKYand W. DEUTSCHMANN,Phys. Z. 30, 839 (1928). W. SCHOTTKY, Z. Phys. l l , 460 (1930). n . i . CHAPIN, C. S. FULLER and G. L. PEARSON,J. appl. Phys. 25, 676 (1954). D. C. RAYNOLDS, G. LEmS, L. L. ANTES and R. E. MARBURGER,Phys. Rev. 96, 533 (1954). L. L. KAZMERSKI,Intern. Mat. Rev. 34, 185 (1989). R. B. GODFREY and M. GREEN, Appl. Phys. Letters 34, 790 (1979). A. HELLER, B. A. PARKINSONand B. MILLER, Conference Record of the 13th IEEE Photovoltaic Specialists Conf., p. 1253, IEEE, New York (1988). W. SHOCKLEYand W. T. READ, d. appL Phys. 32, 510 (1961). A. W. BLAKERS,A. WANG,A. M. MILNE, J. ZHAO and M. A. GREEN, Appl. Phys. Letters 55, 1363 (1989). M.A. GREEN, A. W. BLAKERSand S. R. WENHAM,Proc. 9th E.C. Photovoltaic Solar Energy Conference, p. 301, Kluwer, Dordrecht, (1989). T. SUREK and B. CHALMERS,J. Cryst. Growth 29, 1 (1975). K.V. RAvl, R. C. GONSIORAWSKIand A. R. CHAUDHURI,Proc. of the 18th IEEE Photovoltaic Specialists Conf., p. 1222, IEEE, New York (1985). T. F. OSZEK and J. L. HURD, Proc. Symp. on Electr. and Opt. Properties of Polycrystalline or Impure Semiconductors and Novel Silicon Growth Methods (edited by K. V. Ravi and B. O'Mara), p. 213, Electrochemical Soc., Princeton (1980). F. V. WALD, Crystals 5, 147, Springer, Berlin (1981). D. L. MEIER,J. S. SPITZNAGELand R. B. CAMPBELL,Proc. of the 20th IEEE Photovoltaic Specialists Conf., p. 415, IEEE, New York (1988). K. KURmAYASm, H. MATSUMOTO,H. UDA, Y. KOMATSU,NAKANOAL and S. IKEGAMI,J. appl. Phys. 22, 1828 (1983). Y. S. TYAN, S. VAZANand T. S. BARGE,Proc. 17th Photovoltaic Specialists Conf., p. 840, IEEE New York (1984). M. B. PRINCE, Proc. 8th E.C. Photovoltaic Solar Energy Conf., p. 1632, Kluwer Academic Publ., Dordrecht (1988). H.F. MACMILLAN,N. R. KAMINAR,M. S. KURYLA, M. J. LADLE, D. D. LIU and G. G. VmSHUP, Proc. of the 20th IEEE Photovoltaic Specialists Conf., p. 462, IEEE, New York (1988). I. WEINBERG,C. K. SWARTZ,R. E. HART and T. J. CouTrs, Proc. of the 20th IEEEPhotovoltaic Specialists Conf., p. 893, IEEE, New York (1988). T. J. COUTTSand M. YAMAGUCm, Current Topics in Photovoltaics 3 (edited by T. J. COUTTSand J. D. MEAKIN), p. 79, Academic Press, New York (1988). L. L. KAZMERSKIand S. WAGNER, Current Topics in Photovoltaics (edited by T. J. COUTTSand J. D. MEAKIN), p. 41, Academic Press, New York (1985). D. MOREL, Solar Cells 24, 157 (1988). H. J. HOVEL, Semiconductors and Semimetals 11, (edited by R. K. WILLARDSONand A. C. BEER),Academic Press, New York (1975), M. A. GREEN, Solar Cells, Prentice Hall, Englewood Cliffs (1982). M. A. GREEN, High E~ciency Silicon Solar Cells, Trans Tech Publ., Aedermannsdorf, Switzerland (1987). D. L. PULFREY, Photovoltaic Power Generation, Van Nostrand Reinhold, New York 1978. W. D. JOHNSTON, Solar Voltaic Cells, Dekker, New York (1980). K. J. BACHMANN, Current Topics in Material Science 3, (edited by E. KALDIS), North Holland, Amsterdam (1979). Y. HAMAKAWA,Current Topics in Photovoltaics, (edited by T. J. COUTTSand J. D. MEAKIN), p. I 1 I. Academic Press, New York (1985). C. WRONSKI and D. STAEBLER,J. appl. Phys. 51, 3262 (1980). C. WRONSKI and D. STAEBLER,AppL Phys. Lett. 31, 292 (1977). E. S. SABISKY,J. Non-Cryst. Solids 87, 43 (1986). R. K. WILLARDSONand A. C. BEER, (eds.), Semiconductors and Semimetals, 21A-D, Academic Press, New York (1984).
S E M I C O N D U C T O R S FOR SOLAR CELL A P P L I C A T I O N S 38. 39. 40. 41. 42. 43. 44. 45. 46. 47. 48. 49. 50. 51. 52. 53. 54. 55. 56. 57. 58. 59. 60. 61. 62. 63. 64. 65. 66. 67. 68. 69. 70. 71. 72. 73. 74. 75. 76. 77. 78. 79. 80. 81. 82. 83. 84. 85. 86.
413
S. M. SZE, Physics of Semiconductor Devices, p. 790, Wiley, New York (1981). M. P. THEKACKARA,NASA Technical Report No. R-351 (1970). M. P. THEKACKARA,Suppl. Proc. 20th Annu. Meet. Inst. Environ. Sci. 21, (1974). C. E. BACKUS,ed., Solar Cell, IEEE Press, New York (1980). C. H. HENRY, J. appl. Phys. 51, 4494 (1980). Terrestrial Photovoltaic Measurement Procedures, ERDA/NASA Report 1022-77/16 (1977). J. I. PANKOVE, Optical Processes in Semiconductors, Prentice Hall, Englewood Cliffs, N.J. (1971). V.L. BONC-BRUEVICand S. G. KALASNIKOV,Halbleiterphysik, VEB Deutscher Verlag der Wissenschaften, Berlin (1982). H. P. D. LANYONand R. A. TUFT, IEEE Techn. Dig., Int. Electron Device Meet., p. 316 (1978). J. P. MCKELVEY, Solid State and Semiconductor Physics, p. 422, Harper & Row, New York (1966). J. DRESNER, Semiconductors and Semimetals, 11C, (edited by R. K. WmLARDSON and A. C. BEER), p. 193, Academic Press, New York (1984). H. BROOKSand J. HERRING, Adv. Electr. Electr. Phys. 7, 85 (1955). D. E. BURK and V. DE LA TORRE, IEEE Electron Device Lett. 5, 231 (1984). M. G. BUEHLERand W. R. THURBER, IEEE Trans. Electron Device 23, 968 (1974). L. A. VERHOEFand W. C. StNKE, IEEE Trans. Electron Devices 37, 210 (1990). L.A. VERHOEF,A. ZONDERVAN,F. A. LINDHOLM,M. B. SPITZERand C. J. KEAVNEV,Y. appl. Phys. 63, 4683 (1988). R. LABUSCHand W. SCHR6TER, Dislocations in Solids (edited by F. R. N. NABARRO),p. 129, North Holland, New York (1980). F. DUSTER and R. LABUSCH,Phys. Stat. Sol. (b) 60, 161 (1973). R. PODatELSKI and H. J. M6LLER, Poly-microcrystalline and Amorphous Semiconductors (eds P. PINARD and S. KALBITZER),p. 365, Les edition de Physique, Les Ulis (1984). J. Y. W, SETO, Y. Appl. Phys. 46(12), 5247 (1975). J. DZIEWIOR and W. SCHMID, Appl. Phys. Left. 31, 346 (1977). M. LANNOOand J. BOURGOIN,Point Defects in Semiconductors I, Solid State Sciences 22, p. 158, Springer, Berlin (1981). F. GREUTER and G. BLATTER, Semicond. Sci. Technol. 8, 111 (1990). P. PANAYOTATOS,E. S. YANG and W. HWANG, Solid State Electronics 25, 417 (1982). A. ROMANOWSKIand A. BUCZKOWSKI, Solid State Electronics 28, 645 (1985). D. P. Josm, Solid State Electronics 29, 19 (1986). G. BLATTERand F. GREUTER, Phys. Rev. B33, 3952 (1986). G. BLATTERand F. GREUTER, Phys. Rev. B34, 8555 (1986). W.T. READ, Phil. Mag. 45, 775 and 1119 (1954), W. SCHROTERand R. LABUSCH,Phys. Stat. Sol. 36, 359 (1969). H. EL GHITANI and S. MARTINUZZI, Mater. Res. Soc. Syrup. Proc. 106, 225 (1988). S. PIZZINL A. SANDR1NELLI,M. BEGH1,D. NARDUCCI,F. ALLEGRETTIand F. PORCHIO, J. Electrochem, 135, 155 (1988). A. BRON1ATOWSKI, Phys. Rev. 1336, 5895 (1987). M. AUCOUTURIER,A. BRONIATOWSKI,A. CHARI and J. L. MAURICE, Polycrystalline Semiconductors (edited by H. J. M()LLER, H. STRUNK and J. WERNER), Springer Proceedings in Physics 35, p. 64, Berlin (1989). J. WERNER,Polycrystalline Semiconductors (edited by G. HARBEIO~),Springer Series in Solid State Science 57, p. 95, Springer, Berlin (1985). G. PETERMANN and P. HAASEN, Polycrystalline Semiconductors (edited by H. J. MOLLER, H. STRUNK and J. WERNER),Springer Proc. Physics 35, p. 332, Berlin (1989). W. SZK1ELKOand G. PETERMANN,Poly-microcrystalline and Amorphous Semiconductors (edited by P. PINARD and S. KALBITZER), p. 379, Les edition de Physique, Paris (1985). M. ROSSINELLI, G. BLATTERand F. GREUTER, Br. Ceram. Proc. 41, 177 (1989). F. GREUTER, G. BLATTER, M. ROSSINELLIand F. SCHM~KLE, Mater. Sci. Forum 10, 235 (1986). C. DONOLATO, J. appl. Phys. 54, 1314 (1983). A. COHN and G. CALEDONIA,J. appl. Phys. 41, 3767 (1970). N. TABET,C. MONTYand Y. NARFAING,Polycrystalline Semiconductors (edited by H. J. MOLLER,H. STRUNKand J. WERNER),Springer Proceedings in Physics 35, p. 89, Berlin (1989). Principal conclusions of the American Physical Society Study Group on Solar Photovoltaic Energy Conversion, American Physical Society, New York (1979). L. A. VERHOEF,PhD Thesis, University of Utrecht, Holland (1990). A. OURMAZDand W. SCHROTER, Appl. Phys. Lett. 45, 781 (1984). J. LINDMAYERand J. F. ALHSON, Conference Record of the llth Photovoltaic Conference, p. 40, IEEE, New York (1972). H. B. SERREZE, Conference Record of the 13th Photovoltaic Conference, p. 609, IEEE, New York (1978). M. G. COLEMANN, Conference Record of the 12th Photovoltaic Conference, p. 313, IEEE, New York (1976). M. A. GREEN, A. W. BLAKERS,S. R. WENHAM,S. S. NARAYANAN, R. WILLISON,M. TAOUKand T. SZPITALAK, M. Proc. 18th Photovoltaic Specialists Conf., p. 39, IEEE, New York (1985).
414
87. M. A. GREEN, J. ZHAO, A. WANG,C. M. CrtONG, S. S. NARAYANAN, W. BLAKERSand S. R. WENHAM, A. Proc. 8th E.C. Photovoltaic Solar Energy Conf., p. 167, Kluwer Academic Publ., Dordrecht (1988). 88. R. M. SWANSON,Proc. 8th E.C. Photovoltaic Solar Energy Conf., p. 156, Kluwer Academic Publ., Dordrecht (1988). 89. K. W. MITCHEL, C. EBERSBACHER,J. ERMER, D. PIER and A. MILLA, Proc. 8th E.C. Photovoltaic Solar Energy Conf., p. 167, Kluwer Academic Publ., Dordrecht (1988). 90. L. M. FRAAS,Current Topics in Photovoltaies (edited by T. J. COUTTSand J. D. MEAKIN),p. 169, Academic Press, New York (1985). 91. I. WEINBERG,Solar Cells 29, 225 (1990). 92. C.J. KEAVNEY,V. E. HAVENand S. M. VERNON, Proc. 2nd Int. Conf. on InP and Related Materials, IEEE-LEOS, Piscataway, NJ (1990), 93. P. V. MEYERS, Solar Cells 27, 91 (1988). 94. W. D. JOHNSTONand W. M. CALLAHAN, Appl. Physics Lett. 28, 150 (1976). 95. S. M. SZE, Physics of Semiconductor Devices, p. 273, Wiley, New York (1981). 96. S. M. SZE, Physics of Semiconductor Devices, p. 825, Wiley, New York (1981). 97. R. P. GALE, R. W. McCLELLAND, B. D. KING and J. C. C. FAN, Solar Cells 27, 99 (1988). 98. J. A. BRAGAGNOLO,IEEE Transactions on Electron Devices 27, 645 (1980). 99. H.J. HOVEL, Semiconductors and Semimetals I1, (edited by R. K. WILLARDSONand A. C. BEER), Academic Press, New York (1975). 100. L. L. KAZMERSKI,Current Topics in Photovoltaics (edited by T. J. COUTTSand J. D. MEAKIN), p. 41, Academic Press, New York (1985). 101. A. CATALANO,R. R. ARYA, M. BENNETT, L. YANG, J. MORRIS, B. GOLDSTEIN, B. FIESELMANN,J. NEWTON and S. Wn;DEMAlqN,Solar Cells 27, 25 (1988). 102. S. M. SZE, Physics of Semiconductor Devices, p. 94, Wiley, New York (1981). 103. E. C. BOES,Proc 14th IEEE Photovoltaic Specialists Conf., p. 994, IEEE, New York (1980). 104. R. SAHAI,D. D. EDWALLand J. S. HARRIS, Proc. 13th IEEE Photovoltaic Specialists Conf., p. 946, IEEE, New York (1978). 105. J. M. GEE and G. F. VmsmJp, Proc. 20th IEEE Photovoltaic Specialists Conf., p. 754, IEEE, New York (1989). 106. D. A. ANXONIADlSand I. MOSKOWITZ,J. Appl. Phys. 53, 6788 (1982). 107. W. FRANK, U. GO~ELE, H. MEHRERand A. S~EOER, Diffusion in Crystalline Solids II, p. 63, Academic Press, New York (1984). 108. A. SEEGERand K. P. CmK, Phys. Stat. Sol. 29, 455 (1968). 109. F. MORErmAD, Mater. Res. Soc. Symp. Proc. 104, 99 (1988). I10. G. D. WATK[NS and J. R. TROXELL, Phys. Rev. Left. 44, 593 (1980). 111. B. LERCY, J. appl. Phys. 50, 7996 (1978). 112. T. Y. TAN and U. GOESELE,J. appl. Phys. 53, 4767 (1982). 113. A. F. W. WILLOUGHBY,Impurity Doping Processes in Silicon, p. 1, North-Holland, New York (1982). 114. F. A. KR6GER, The Chemistry of Imperfect Crystals, North Holland, Amsterdam (1973). 115. H. C. GATOSand J. LAGOWSKI,Mater. Res. Soc. Symp. Proc. 46, 155 (1985). 116. VON, H. J. BARDELEBENand D. STXEVENARD,Mater. Res. Soc. Symp. Proc. 104, 353 (1988). 117. Proc. 4th Int. Conf. on Ternary and Multinary Compounds, Jpn. J. appl. Phys. 19-3, (1980). 118. J. E. JAFrE and A. ZUNGF~, Phys. Rev. B28, 5822 (1983). 119. H. NEUMANN, Verbindungshalbleiter, p. 392, Akadem. Verlag, Leibzig (1986). 120. H. J. M6LLER, Solar Cells, 31, 77 (1991). 121. J. A. THORNTON, T. C. LOMMASSON,A. F. BURNETTand L. H. CHOU, Progress Report, SERI Contract No. XL-5-04131-1, Golden (1986). 122. J. A. GROENINK and P. H. JANSE, Z. Phys. Chem. Neue Folge 110, 17 (1978). 123. B. O. KOLB~EN and A. M~rlLaAtr~R, Solid State Electronics 25, 759 (1982). 124. U. GOESELEand T. Y. TAN, .4ppl. Phys. A28, 79 (1982). 125. J. R. PATEL, Semiconductor Silicon, p. 189, Electrochem. Soc., Pennington (1981). 126. J. C. MIKKELSEN, Mater. Res. Soc. Symp. Proc. 59, 26 (1986). 127. R. NEWMAN,Mater. Res. Soc. Syrup. Proc. 104, 25 (1988). 128. A. OURMAZD,A. BOUm~ETand W. SCm~6TER, J. Appl. Phys. 56, 1670 (1984). 129. L. C. K]MERLING, Mater. Res. Soc. Symp. Proc. 59, 83 (1986). 130. D. MATmOT, Mater. Res. Soc. Syrup. Proc., 104, 189 (1988). 131. A. BOURRET,J. 2~dmAULT-Dr~SEAUXand D. N. SE]DMAN,J. appl. Phys. 55, 825 (1984). 132. A. BOURRET, Microscopy of Semiconducting Materials, Inst. Phys. Conf. Ser. 87, 197 (1987). 133. T. Y. TAN, Mater. Res. Soc. Syrup. Proc. 59, 269 (1986). 134. K. TEMPLEHOE, R. GLEICHMANN,F. SPmGELBERGand D. WRUCH, Phys. Stat. Sol. A56, 213 (1979). 135. F. HAM, J. Phys. Chem. Solids 6, 335 (1958). 136. J. LEROUE1LLE,Phys. Stat. Sol. (a) 67, 177 (1981). 137. V. CASZARAand P. ZUNINO, J. appl. Phys. 51, 4206 (1980). 138. M. TAJIMA,T. MASUhT. ABE and T. hzur.A, Semiconductor Silicon, p. 72, Electrochem. Soc., Pennington (1981). 139. A. R. BEANand R. C. NEWMAN, J. Phys. Chem. Solids 32, 1211 (1971).
S E M I C O N D U C T O R S FOR SOLAR C E L L A P P L I C A T I O N S 140. 141. 142. 143. 144. 145. 146. 147. 148. 149. 150. 151. 152. 153. 154. 155. 156. 157. 158. 159. 160. 161. 162. 163. 164. 165. 166. 167. 168. 169. 170. 171. 172. 173. 174. 175. 176. 177. 178. 179. 180. 181. 182. 183. 184. 185. 186. 187. 188.
415
W. ZULEHNER, Aggregation Phenomena of Point Defects in Silicon, p. 89, Electrochem. Soc., Pennington (1983). J. L. BENTON, M. T. ASOM, R. SALTERand L. KIMERLING, Mater. Res. Soc. Syrup. Proc. 104, 85 (1988). R. C. NEWMANand J. WAKEFIELD, Phys. Chem. Solids 19, 230 (1961). J. J. P. KAtJUS, L. A. LADD and U. GOESELE,Appl. Phys. Lett. 45, 269 (1984). U. GOESELEand T. Y. TAN, Aggregation Phenomena of Point Defects in Silicon, p. 17, Electrochem. Sot., Pennington (1983). H. FOLL, U. GOESELEand B. O. KOLBESEN,J. Crystal Growth 52, 907 (1981). U. GOESELE,Mater. Res. Soc. Symp. Proc. 59, 419 (1986). D. HELMREICH, Proc. Syrup. Electrical and Optical Properties of Polycrystalline or Impure Semiconductors and Novel Silicon Growth Methods (edited by K. V. RAVI and B. O'MARA), p. 184, Electrochemical Soc., Princeton (1977). J. DIETL, D. HELMRE1CHand E. SIRTL, Crystals 5, p. 43, Springer, Berlin (1981). S. MOLLERand H. J. M()LLER, Proc. 91h Photovoltaic Solar Energy Conf., p. 439, Kluwer Publ. Dordrecht (1989). J. P. KALEJS and B. CHALMERS,J. Crystal Growth 79, 487 (1986). H. J. MOLLER, Mater. Res. Soc. Syrup. Proc. 205, 271 (1991). E. WEBER, Appl. Phys. A30, I (1983). S. FISCHLER, J. appl. Phys. 33, 1615 0962). M. SEIBT and K. GRAFI~,J. appl. Phys. 63, 4444 (1988). M. SE1BTand K. GRAFP, Mater. Res. Soc. Syrup. 104, 215 (1988). J. R. DAVIS, A. ROHATGI, R. HOPKINS, P. BLAIS, P. RAI-CHOUDHURY, J. McCORMICK and H. C. MOLLENKOPF, IEEE Trans. Electron. Devices 27(4), 677 (1980). S. T. PICREAUX,F. L. VOOK and H. J. STEIN, Def. and Rad. Effects in Semiconductors, Inst. Phys. Conf. Ser. 46, 31 (1979). J. I. PANKOVE, Cryst. Latt. Def. Amorph. Mat. 11, 203 (1985). S. J. PEARTON, J. W. CORBETTand T. S. SHI, Appl. Phys. A43, 153 (1987). S. J. PEARTON, 13th Int. Conf. Defects in Semiconductors, Metall. Soc. AIME, Wavendale, p. 737 (1985). J. W. CORBETT, D. PEAK, S. J. PEARTONand A. SGANGA, Hydrogen in Disordered and Amorphous Sol., p. 61, Plenum, New York (1986). J. I. PANKOVE,D. J. ZANZUCCHI, W. MAGEEand G. LUCOVSKI, Appl. Phys. Lett. 46, 421 (1985). C. A. VAN WIERINGEN and N. WARMHOLTZ,Physica 22, 849 (1956). T. ICHIMXYAand A. FURUICm, Int. J. appl. Rad. Isotopes 19, 573 0968). J. W. CORBE'rr and J. C. BOURGOIN,Point Defects in Solids, p. 1, Plenum Press, New York (1975). H.Y. TADA,J. R. CARTER,B. E. ANSPAUGHand R. G. DOWNING, Solar Cell Radiation Handbook, JPI Publication 82 (1982). D. J. FLOOD and H. W. BRANDHORST,Current Topics in Photovoltaics, Vol. 2, p. 163, Academic Press, London (1987). C. GORADIAand I. WEINBERG,J. appl. Phys. 57, 4752 (1985). D. FLOODand C. Z. SWARTZ,Space Photovoltaic Research and Technol. Conf. p. 239, NASA Conf. Pub; 2169 (1980). I. WEINBERG,C. K. SWARTZ, R. E. HART and R. L. STATLER, Proc. of the 19th IEEE Photovoltaic Specialists Conf., p. 548, IEEE, New York (1987). J. J. KREBS and G. G. STAUSS,Phys. Rev. B 16, 971 (1977). D. V. LANG and L. C. KIMERHNG, Phys. Rev. Lett. 489 (1974). J. C. BOURGOINand J. W. CORBETT, Phys. Rev. A38, 135 (1972). G. D. WATKINS, J. Phys. Soc. Jap. 18 (Suppl. 2) 22 (1961). G. D. WATKINS, Lattice Defects in Semiconductors, p. I, Inst. Phys. Conf. Publ. 23 (1974). I. WEINBERG, Current Topics in Photovoltaics, p. 87, Academic Press, London (1990). G. D. WATKINS and K. L. BROWER, Phys. Rev. Lett. 36, 1329 (1976). J. M. TROMBETTIand G. D. WATKINS, Appl. Phys. Lett. 51, 1103 (1987). I. WEINBERG,C. GORADIA, J. W. STUPICAand C. K. SWARTZ, J. Appl. Phys. 60, 2179 (1986). H. ALEXANDERand P. HAASEN,Solid State Physics 22, 27, Academic Press, New York (1968). H. ALEXANDER,Dislocations in Solids (edited by F. R. N. NABARRO), p. 114, Elsevier Science Publ., London (1986). H.J. LEAMY,G. E. PIKEand C. H. SEAGER(eds), Grain Boundaries in Semiconductors, Mat. Res. Soc. Syrup. Proc. 5, North Holland, Amsterdam (1982). A. ROCriER (ed.), Polycrystalline Semiconductors, J. Physique 43---C1 (1982). G. HARBEKE(ed.), Polycrystalline Semiconductors, Springer Series Solid State Science 57, Springer, Berlin (1985). H.J. MOLLER, H. STRUNKand H. WERNER(eds), Polycrystalline Semiconductors, Springer Proceedings in Physics 35, Springer, Berlin (1989). W. BOLLMANN, Crystal Defects and Crystalline Interfaces, Springer, Berlin (1970). R. C. POND, Polycrystalline Semiconductors, Springer Series Solid State Science 57 (edited by G. HARBEKE), p. 27, Springer, Berlin (1985). W. SKROTZK1,H. WENDT,C. B. CARTER and D. L. KOHLSTEDT, Phil. Mag. A57, 383 (1988).
416
189. 190. 191. 192. 193. 194. 195. 196. 197. 198. 199. 200. 201. 202. 203. 204. 205. 206. 207. 208. 209. 210. 211. 212. 213. 214. 215. 216. 217. 218. 219. 220. 221. 222. 223. 224. 225. 226. 227. 228. 229. 230. 231. 232. 233. 234. 235. 236. 237. 238.
P R O G R E S S IN M A T E R I A L S S C I E N C E A. PAPON, M. PETIT and J. J. BACMANN,Phil. Mag. A49 (1984) 573. M. D. VAUDIN,B. CUNNINGHAMand D. G. AST, Scripta Met. 17, 191 (1983). A. BOURRETand J. J. BACMANN,Revue Phys. Appl. 22 (1987) 563. A. BOURRETand J. L. ROUVIERE,Polycrystalline Semiconductors, Springer Proceedings in Physics 35 (edited by H. J. MOLLER,H. STRUNKand H. Wm~NER),p. 8, Springer, Berlin (1989). H. J. MOLLER,Phil. Mag. A43 (1981) 1045. M. PAPON and M. PETIT, Scripta Met. 19 (1985) 391. H.J. M()LLERand H. H. SINGER,Polycrystalline Semiconductors, Springer Series Solid State Science 57, Springer, Berlin (I 985); Grain Boundary Structure and Related Phenomena, Suppl. to Transactions of the Japan Institute of Metals. 27, 987 (1986). J. T. WETZEL,A. A. LEVIand D. A. SMITH,Mater. Res. Soc. Syrup. Proc. 63, 157 (1986). S. PHILLPOTand D. WOLFF,Mater. Res. Soc. Syrup. Proc. 122, 103 (1988). A. T. PAXTONand A. P. SUTTON,Acta Metall. 37, 1693 (1989). M. C. PAYNE,P. D. BRISTOWEand J. D. JOANNOPOULOS,Phys. Rev. Lett. 58, 1348 (1987). N. M. JOHNSON,D. K. BIEGELSENand M. D. MOYER,Appl. Phys. Left. 40, 882 (1982). G. E. PIKE, P. L. GOURLEYand S. R. KURTZ, Appl. Phys. Lett. 43, 939 (1983). M. G. SPENCER,W. J. SCHAVFand D. K. WAGNER,Appl. Phys. 54, 1429 (1983). E. ZIEGLER,W. SmGEL,H. BLUMTRITTand O. BREITENSTEIN, Phys. Stat. Sol. (a) 72, 593 (1982). R. HERMANN,G. NACHTWEIand W. KR~K, Phys. Stat. Sol. (a) 83, K207 (1984). Y. TAKEDA,Ceram. Bull. 67, 1961 (1988). D. B. HOLT, J. Phys. Chem. Solids 25, 1385 (1985). H. F. MATARE,J. appl. Phys. 56, 2605 (1984). G. BLATTERand F. GREUTER,Polycrystalline Semiconductors, Springer Series Solid State Science 57 (edited by G. HARBEKE),p. 118, Springer, Berlin (1985). X. J. Wu, V. SZKmLKOand P. HAASEN,J. Physique 43-C1, 135 (1982). A. BRONIATOWSKI,Polycrystalline Semiconductors, Springer Series Solid State Science 57 (edited by G. HARBEKE),p. 95, Springer, Berlin (1985). J. J. YANG, P. D. DAPKUS,R. D. OuPUlS and R. D. YINGLING,J. appl. Phys. 51, 3794 (1980). A. J. MADENACHand J. WERNER, Phys. Rev. B38, 1310 (1988). J. WERNER,Polycrystalline Semiconductors, Springer Series Solid State Science 57 (edited by G. HARBEKE),p. 76, Springer, Berlin (1985). G. BLATTERand F. GREtrrER, Phys. Rev. !i33, 3952 (1986); Phys. Rev. B34, 8555 (1986). J. WERNERand M. PEISL,Mat. Res. Soc. Symp. Proe. 46, 575 (1985). H. J. MrLLER, Defects in Crystals (edited by E. MIZERA),p. 318, World Scientific Publishing Co., New Jersey (1988). J. WERNERand H. STRUNK,J. de Physique 43-C1, 99 (1982). W. E. SPICER,I. LINDAU,P. SKEATHand C. Y. Su, J. Vac Sci. TechnoL 17, 1019 (1980). M. GUTTMANN,Surface Science 53, 213 (1975); Interracial Segregation, p. 251, ASM Seminars, Metal Park (1979). E. D. HONDROS,J. Physique 36-C4, 117 (1975). M. AUCOUTURIER,Polyerystalline Semiconductors, Springer Series Solid State Science 57 (edited by G. HARBEKE),p. 47, Springer, Berlin (1985). A. BOUR~T, J. THIBAULTand D. M. SEIDMAN,J. appL Phys. 55, 825 (1984). D. MCLEAN, Grain Boundaries in Metals, Clarendon Press, Oxford (1957). W. R. TYSON,Acta Met. 26, 1471 (1978). I. M. LIFSCmTZand V. V. SLVOZOV,J. Phys. Chem. Solids 19, 35 (1961). C. WAGNER,Z. Elektrochemie 65, 581 (1961). J. S. LANGERand A. J. SCHWARTZ,Phys. Rev. A21, 948 (1980). H. J. MOLLER,J. Crystal Growth, in press (1991). L. L. KAZMERSKIand P. E. RUSSELL,J. Physique 43-CI, 171 (1982). L. L. KAZMERSKI,A. J. NELSONand R. G. DX-mRE,J. Vac. Soc. Sci. Techn. AS, 1994 (1987); Proc. 19th IEEE Photovoltaic Specialist Conf., p. 944, IEEE New York (1987). C. R. M. GROWNOR, P. E. BATSON,D. A. SMITHand C. WONG, Phil. Mag. AS0, 409 (1984). M. M. MANDURAH,J. appL Phys. 51, 5755 (1980). L. L. KAZMERSKI,Polycrystalline Semiconductors, Springer Proceedings in Physics 35 (edited by H. J. Mt3LLER, H. STRUNKand H. WEANER),p. 96, Springer, Berlin (1989). T. TOTKEN,W. SCt-mrTERand H. J. Mt3LLER, Mater. Res. Soc. Symp. Proc. 122, 109 (1988); Polycrystalline Semiconductors, Springer Proceedings in Physics 35 (edited by H. J. MrLLER, H. STRUNKand H. WERNER), p. 108, Springer, Berlin (1989). H. J. MOLLER,U. JENDRICH,L. HUANGand A. FOITZIK,Mater. Res. Soc. Symp. Proc. 209, 411 (1991). K. MATSUDA-JINDO, Polycrystalline Semiconductors, Springer Proceedings in Physics 35 (edited by H. J. MrLLER, H. STRUNKand H. WERNER),p, 52, Springer, Berlin (1989). J. C. FISHER,J. appl. Phys. 22, 74 (1951). R. T. P. WHIPPLE, Phil. Mag. 45, 1225 (1954).
S E M I C O N D U C T O R S FOR SOLAR CELL A P P L I C A T I O N S 239. 240. 241. 242. 243. 244. 245. 246. 247. 248. 249. 250. 251. 252. 253. 254. 255. 256. 257. 258. 259. 260. 261. 262. 263. 264. 265. 266. 267. 268. 269. 270. 271. 272. 273. 274. 275. 276. 277. 278. 279. 280. 281. 282. 283. 284. 285. 286. 287.
417
T. SUZUOKA,J. Phys. Soc. Jap. 19, 839 (1974). H. BAUMGART,H. S. LEAMY,G. K. CELLER and L. E. TRIMBLE,J. Phys. 43-C1, 363 (1982). J. L. LIOTARD, R. BIBERIANand J. CABA~, J. Phys. 43-C1, 213 (1982). A. GAT, L. GERZnmtG and J. F. GIBBONS, Appl. Phys. Lett. 33(8), 775 (1978). H. SCHABER,D. CUTTER, J. BINDERand E. OaERMEIER, J. appl. Phys. 54(4), 4633 (1983). T. I. KAMINS, J. Electrochem. Soc.: Solid State Science and Technol. 127, 686 (1980). R. PODBIELSKIand H. J. MOLLER, Proc. 13th Int. Conference on Defects in Semiconductors, Metallurgical Society of AIME (edited by L. C. KIMERLINGand J. M. PARsEr) 435 (1985). H. J. MOLLER and V. SCHUCHTING, Polycrystalline Semiconductors, Springer Proceedings in Physics in Physics 35 (edited by H. J. M6LLER, H. STRUNK and H. WERNER), p. 326, Springer, Berlin (1989). T. H. WANG, T. F. CISZEK and T. SCHtrVLER, Solar Cells 24, 136 (1986). P. T. LANDSBERGand G. S. KOUSlK, J. Crystal Growth 56, 1696 (1984). P. J. ROKSNOERand M. M. B. VAN DEN BOOM,J. Crystal Growth 53, 563 (1981). A. J. R. DEKOCK, Appl. Phys. Lett. 16, 100 (1970). P. M. PETROFFand A. J. R. DEKOCK, J. Crystal Growth 30, 117 (1975). W. DIETZE, W. KELLER and A. MfrHLBAUER, Crystals 5, 2, Springer, Berlin (1981). D. HELMREICH,Silicon Processing for Photovoltaics H (edited by C. P. KHAa-rAKand K. V. RAVI), p. 97 Elsevier Science Publ, London (1987). D. HELMREICH, Proc. Syrup. on Electrical and Optical Properties of Polycrystalline or Impure Semiconductors and Novel Silicon Growth Methods (edited by K. V. RAVl and B. O'MARA), p. 184, Electrochemical Soc., Princton (1977). F. SCHMID and D. VIECHNICKI,J. Am. Ceram. Soc. 53, 528 (1970). S. MOLLER and H. J. M(')LLER, Proc. of the 9th E.C. Photovoltaic Solar Energy Conference, p. 439, Kluwer, Dordrecht (1989). L. L. KAZMERSKI,P. J. IRELANDand T. F. CISZEK, Appl. Phys. Lett. 36, 323 (1980). U. JENDRICH, and H. J. MOLLER, Colloque de Physique 51-C1, 197 (1990); Mater. Res. Soc. Syrup. Proc. 163, 579 (1990). T. SUREK, Proc. Symp. on Electrical and Optical Properties of Polycrystalline or Impure Semiconductors and Novel Silicon Growth Methods, p. 173, Electrochemical Soc., Princeton (1977). A. GOETZBERGERand A. R_KUaER, Techn. Digest of Photovoltaic Solar Energy Conference 4 (1989). Proe. of 1st Int. Conf. on Shaped Crystal Growth, J. Crystal Growth 87 (1987). A. EYER, A. R~UBER and J. G. GRABMEIER,Proc. of the 9th E.C. Photovoltaic Solar Energy Conference, p. 17 Kluwer, Dordrecht (1989). J. C. SWARTZ, T. SUSEKand B. CHALMERS,J. Electron. Mat. 4, 255 (1975). T. F. CISZEK, Crystals 5, Springer, Berlin (1981). H. LANGEand I. SCHWIRTLICH, J. Crystal Growth 104, 108 (1990). B. K. MEYER, D. M. HOFFMANNand J. M. SalTer, Proc. 14th Int. Conf. Defects in Semiconductors, p. 311 Transtech, Switzerland (1986). L. SAMUELSON,P. OMLING, H. TITZE and H. G. GmMMElSS, J. Cryst. Growth 55, 164 (1981). R. E. KREMER, D. FRANCOMANO,G. H. BECKHART,K. M. BURKEand T. MILLER, Mater. Res. Soc. Syrup. Proc. 144, 15 (1989). S. MATTSONand H. D. SI-nH, Appl. Phys. Lett. 48, 47 (1986). S. MAHAJANand M. A. SHAH/D, Mater. Res. Soc. Syrup. Proc. 144, 169 (1989). M. M. AL-JASSIM,J. P. GORAL, P. SHELDONand K. M. JONES, Mater. Res. Soc. Syrup. Proc. 144, 183 (1989). J. H. VAN DER MERWE and N. G. VAN DER BERG, Surface Sci. 32, 1 (1972). G. M. BLOM and J. M. WDODALL, J. Electron Mater. 17, 391 (1988). S. IGNATOWICZand A. KOBENDZA (eds), Semiconducting Thin Films of II-VI Compounds, Ellis Horwood, Chichester (1990). K. KURIYAMAand F. NAKAMURA,Phys. gev. ii36, 4449 (1987). P. GmART, L. GOLDSTEINand J. L. DORMANN, in Ref [117], p. 184. J. C. WOOLLY, R. G. GOODCHILD, O. H. HUGHES, S. A. LOPEZ-RIVERAand B. R. PAMPLIN, Proc. 4th Int. Conf. on Ternary and Multinary Compounds, Jpn. J. appl. Phys. 19-3, 146 (1980). R. BACEWICZand T. F. CISZEK, Phys. Left. 52, 1150 (1988). J. E. JAFFEand A. ZUNGER, Phys. Rev. B29(4), 1882 (1984). N. ROMEO, Proc. 4th Int. Conf. on Ternary and Multinary Compounds, Jpn. J. appl. Phys. 19-3, 5 (1980). S. WAGNER, Proc. Syrup. Mater. New Process. Techn. Photovolt., Electrochem. Soc. 83-11, 410 (1983). L. S. PALATNIKand E. J. ROGACHEVA,Soy. Phys.-Dokl. (Engl. Transl.) 12, 503 (1967). H. W. SCHOCK, Polycrystalline Semiconductors, Springer Proceedings in Physics 35 (edited by H. J. M611er, H. Strunk and H. Werner), p. 246, Springer, Berlin (1989). H. NEUMANN, Solar Cells 16, 317 (1986). J. L. SHAH and J. H. WERNICK, Ternary Chalcopyrite Semiconductors, Pergamon, New York (1975). L. L. KAZMERSKI, Ternary Compounds (edited by E. D. HOLAH), Inst. Phys. Conf. Set., 35 217 (1977). C. RINCON, J. GONZALEZ,G. SANCHEZ-PEREZand C. BELLABARA,Nuovo Cimento Soc. Ital. Fis. 2D, 1895 (~983).
418
288. 289. 290. 291. 292. 293. 294. 295. 296. 297. 298. 299. 300. 301. 302. 303. 304. 305. 306. 307. 308. 309. 310. 311. 312. 313. 314. 315. 316. 317. 318. 319. 320. 321. 322. 323. 324. 325. 326. 327. 328. 329. 330. 331. 332.
P R O G R E S S IN M A T E R I A L S S C I E N C E H. NEUMANN, H. SOBOTTA,W. K1SSINGER,V. RIEDE and G. KOHN, Phys. Stat. Sol. B108, 483 (1981). S. N. DERMATISand J. W, FAUST IEEE Trans. Commun. Electron. 82, 94 (1963). T. IRIE,, S. ENDO and S. KIMURA, Jpn. J. appl. Phys. 18, 1303 (1979). H. NEUMANNand R. D. TOMLINSON, Solar Cells 28, 301 (1990). F. ABou-ELEOTOUH, D. J. DUNLAVYand T. J. COUTTS, Solar Cells 27, 237 (1989). G. KOHI~ and H. NEUMANN,Z. Chem. 27, 197 (1987). R. D. KRISTENSEN, S. N. SAHU and D. HANEMAN,Appl. Surface Science 33, 1285 (1988). R. NOUFL R. AXTON, C. HERRINGTON and S. K. DEB, Appl. Phys. Lett. 45(6), 668 (1984). J. M. POATE, K. N. TU and J. W. MAYER, Thin Films: Interdiffusion and Reactions, Wiley, New York (1978). R. J. MATSON, R. NOUFI, K. BACHMANNand D. CAHEN, Appl. Phys. Lett. 50) 158 (1987). J. D. JOANNOPOULOSand G. LUCOVSKY(eds), Hydrogenated Amorphous Silicon 1 and II, Springer, Berlin, New York (1984). D. ADLER, M. SILVER, M. P. SHAW and V. CANNELLA, Mater. Res. Soc. Syrup. Proc. 70, 119 (1986). G. WINTERLING and G. Mf3LLER, Physica Scripta TI3, 45 (1986). D. J. WOLFORD, J. A. REIMERand B. A. SCOTT, Appl. Phys. Lett. 42, 369 (1983). D. ADLER, Semiconductors and Semimetals 21A (edited by R. K. Willardson and A. C. Beer), p. 291, Academic Press, New York (1984). W. E. SPEAR,P. E. LECOMBERand A. J. SNELL, Phil. Mag, 33, 935 (1976). T. T1EDJE,Semiconductors and Semimetals 21C (edited by R. K. WILLARDSONand A. C. BEER), p. 207, Academic Press, New York (1984). G. D. CODY, C. R. WRONSKI, B. ABELES, R. B. STEPHENSand B. BROOKS, Solar Cells 2, 227 (1980). J. MARGARINO, D. KAPLAN, A. FRIEDRICH and A. DENEUVILLE,Phil. Mag. 1345, 285 (1982). D. E. CARLSON, Semiconductors and Semimetals 21D (edited by R. K. WILLAROSONand A. C. BEER), p. 7, Academic Press, New York (1984). P. E. LECOMBERand W. E. SPEAR,Phys. Rev. Lett. 25, 509 (1970). R. A. STREET, Phys. Rev. Lett. 49, 1187 (1982). G. M~ILLER, H. MANNSBERGERand S. KALBITZER, Phil. Mag. B53, 257 (1986). R. A. STREET, D. K. BmGELSENand J. C. KNIGHTS, Phys. Rev. B24, 969 (1981). T. TIEDJE, Appl. Phys. Lett. 40, 627 (1982). A. J. SNELL, W. E. SPEAR and P. E. LECOMBER,Phil. Mag. 43, 407 (1981). T. TmDJE, J. M. CEBULKA,D. L. MOREL and B. ABELES, Phys. Rev. Lett. 46, 1425 (1980). P. E. LECOMBER,D. I. JONES and W. E. SPEAR, Phil. Mag. 35, 1173 (1977). J. DRESNER, Appl. Phys. Lett. 37, 742 (1980). S. KALBITZER,G. MOLLER, P. E. LECOMBERand W. E. SPEAR, Phil. Mag. 1341, 439 (1980). C. B. ROXLO, B. ABELES,C. R. WRONSKI, G. n . CODY and T. TmDJE, Solid State Commun. 47, 985 (1983). C. WRONSKI, Semiconductors and Semimetals 21C (edited by R. K. WILLARDSON and A. C. BEER), p. 347, Academic Press, New York (1984). W. MEYER and H. NEDEL, Z. Techn. Phys. 12, 588 (1937). D. L. STAEBLER,R. S. CRADALLand R. WILLIAMS, Appl. Phys. Lett. 39, 733 (1981). G. MARRAKAKCHI,G. CHAUSSEMV,A. LAUGER and G. GUILLOT, Mater. Res. Soc. Symp. Proc. 144, 27 0989). J. W. CORBETT,J. C. CORELLI, DESNICAand L. SNYDER, Mater. Res. Soc. Syrup. Proc. 46, 243 (1986). U. A. BENSAOULA,K. D. JAMISON, H. C. CHEN and A. IGNATIEV,Mater. Res. Soc. Syrup. Proc. 144, 85 (1989). C. RINCON, C. BELLABARA,J. GONZALEZand G. SANCHEZPEREZ, Solar Cells 16, 335 (1986). G. OTTAVIANI, Thin Solid Films 140, 3 (1986). M. W. WANLASS,K. A. EMERYT. A. GESSERT, G. S. HORNER, C. R. OSTERWALDand T. J. COUTTS, Solar Cells 27, 191 (1989). A. BENNETTand L. C. OLSEN, Conf. Rec. 13th IEEE Photovoltaic Specialists Conf., p. 868 IEEE, New York (1988). S. HOCINE and D. MATHIOT, Appl. Phys. Lett. 53, 1269 (1988). P. CULLEN, P. HAP,BISON, D. V. LANG and D. ADLE.R, J. Non-Cryst. Solids 50/60, 261 (1983). J. H. DERSCH, M. STUKEand J. BEICHLER,Appl. Phys. Lett. 38, 456 (1981). A. G. MILNES, Deep Impurities in Semiconductors, John Wiley & Sons, New York (1973).

HGHB

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

HGHB

Uploaded by

Copyright:

Available Formats

Progress in Materials Science Vol. 35, pp.

0079-6425/91 $0.00 + .50 1991 Pergamon Press plc

3.1. 3.2. 3.3. 3.4. 3.5. 3.6.

4. STRUCTURAL AND PHYSICALPROPERTIES OF LATTICE DEFECTS IN SEMICONDUCTORS

6. SINGLE CRYSTAL AND EPITAXlAL GALLIUM ARSENIDE

Copper Ternary Compounds

1.1. Historical Background of the Solar Cell Development

FOR SOLAR CELL APPLICATIONS

1.2. Technical Concepts for Photovoltaic Solar Energy Conversion

FOR SOLAR CELL APPLICATIONS

FOR SOLAR CELL APPLICATIONS

PROGRESS IN MATERIALS SCIENCE

FOR SOLAR CELL APPLICATIONS

FOR SOLAR CELL APPLICATIONS

215 ~c where ~ is the

(hv - Eg + hvphI )2 (hv - Eg - hVph2)2 Ctin(V) = ~l exp(hvphJKT) - 1 + ~2 1 - exp(hvph2/KT )

AE~/2 f ~ 4 (Eg-hv~3/2 ~ tFK(V)=v(Eg-hv)exp(-3\ er / J

S E M I C O N D U C T O R S FOR SOLAR CELL A P P L I C A T I O N S

3e 2 (e2N)~/2 3e 2 16zte ~ ~'~e~J = 16rCELD

Recombination at back surface

FIG. 3. Schematicdiagram of a p-n junction solar cell, defining basic parameters.

G(x, v) = o~(v)So(v ) (1 - R )exp(-o~(v )x ).

FOR SOLAR CELL APPLICATIONS

The carrier concentrations for a non-degenerate semiconductor are given by

/iEdivp +DAp +G---

where the ambipolar diffusion coefficient D and mobility/i are given by

PROGRESS IN MATERIALS SCIENCE

02p = c32n 6n Dp-~xZ -G+rP;zp D. ~ x 2 = -G+--.Zn

Fp2 are given by

Wn + sinh Lp A/L Dp sinh -Lp + cosh Lp j -

FOR SOLAR CELL APPLICATIONS

J. = eSo (1 - R) (~Ln)2 __ 1 e x p ( - ~Xp) {F~l -- F~2 exp(-- ~ Wp)}

FS.L. coshW,+sinhT. j / l-b-~-. sinh -w, + cosh wa. wa/r o, , L. Ln L. d

Jdp = eSo (1 -- R)exp(--~x,) (1 -- exp(--~W))

SR(hv) = Sp (hv) + J. (by) + Ja, (by) eSo(hv) (1 - R(hv))

0.2 2.0 5,0 hu(ev) 4.0 50

f DnFp DpFn ~ / Es Is = eANcNv_~L--~ + L - ~ ; exp ~--~-~).

FOR SOLAR CELL APPLICATIONS

VmpImp VmpImp FFV~I~ FF= Vo~I~ r/ = Pin Pin

PROGRESS IN MATERIALS SCIENCE

SEMICONDUCTORS FOR SOLAR CELL APPLICATIONS

Cor~p 37~7/2(2m*)5/2K3/2~ oE2o

Cop = 3(M1 + M2)ZEe37r7/E(2m*)3/2K 1/2

PROGRESS IN MATERIALS SCIENCE

FOR SOLAR CELL APPLICATIONS

= f s ( E , E ' ) N1 (E) f(E) N2(E' ) (1 - f ( E ' )) dE dE'

1 1 1 - = Spno + Sapl n~ + Sap2nop o = -- + - "IS Tp

p-Si a 77 K O 300K 6 400K

Doping concentration,NO (cm -3)

Doping concentration, N O (cm -3) (b)

SEMICONDUCTORS FOR SOLAR CELL A P P L I C A T I O N S

%o (no + nTl ) + Zno(Po +PTI) no + Po

k~ _ exp EF, -- E___ ___~ gr, KT '

FOR SOLAR CELL APPLICATIONS

gi exp{(iEF -- Ei)/KT} gj exp{ (jEF - EJ)/KT}

FOR SOLAR CELL APPLICATIONS

Ns (E)f~ (E + Vo) dE - eNs..

PROGRESS IN MATERIALS SCIENCE

nsPs--nlpl }dE. s.(n~+nl)+sp(p~+p~)

FOR SOLAR CELL APPLICATIONS

with the occupation function fs(E) for the surface states

f~(E) = 1 + exp{(E + V o - E v , ) / K T } "

S n = trpvpNso exp K ~ '

k~ _ exp EF, -- E_ _~ gr, KT '