Professional Documents
Culture Documents
Jan Erik Ramstad Institute for Informatics University of Oslo janera@fys.uio.no 27th March 2006
Contents
1 Essay background 2 Processes 2.1 CMOS-MEMS process steps . . . . . . . . . . . . . 2.2 CMOS Process-sequence . . . . . . . . . . . . . . . 2.3 MEMS Process . . . . . . . . . . . . . . . . . . . . . 3 Materials 4 CMOS-MEMS parameters 4.1 CMOS-MEMS compability . . 4.2 Pre-CMOS Micromachining . 4.3 Intra-CMOS Micromachining 4.4 Post-CMOS Micromachining 5 Existing CMOS-MEMS products 6 Circuit and system integration 7 Discussion and analysis 8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 2 2 2 3 4 4 5 5 6 7 8 8 8
2 Processes
MEMS evolved from CMOS processes in the late 90s, and MEMS has since then been behind CMOS in process technology. This development is thoroughly described in [2]. Because of the technology trend, CMOS has larger wafer sizes than MEMS. Some manufacturers have used laser equipment to cut the CMOS wafer to match MEMS wafer sizes. This enables the possibility to use MEMS process technologies to make MEMS structures in the CMOS wafer. This is known as PostCMOS Micromachining. However, the cutting of wafer size led to an inefcient economy. This is because number of dies per wafer was reduced dramatically, and the yield was reduced. CMOS compabilitiy as discussed in [2], indicates that CMOS is the dominant semiconductor technology. This means that MEMS has to adapt to CMOS and not the other way around. Still, this leaves three different choices of producing CMOSMEMS systems: 1. Pre-CMOS Micromachining 2. Intra-CMOS Micromachining 3. Post-CMOS Micromachining These three micromachining processes are hereby dened as modules. A set of fabrication steps of CMOS or MEMS is also dened as a module. For intra-CMOS, a MEMS module is made rst, then a CMOS module is made, and nally a nishing MEMS module is processed. Pre-CMOS indicates a MEMS module processed before the CMOS is made, and vice versa for post-CMOS. Pre-, intra- and post-CMOS all have in common that the production of the MEMS is adapted to the CMOS processes. This means that the CMOS process may have to modify the original CMOS process in order to perform MEMS process steps. It is also possible to make the CMOS and MEMS parts seperately in foundries specialized on CMOS or MEMS. Seperate manufacturing of MEMS and CMOS requires proper connection between the CMOS and MEMS parts, as well as minimal inuence on the CMOS part if the MEMS module is fabricated after the CMOS module. To efciently design and implement MEMS systems, reference [2] suggests a CMOS compatible path. This is due to the importance of using IPs (Intellectual Properties) and embedded software to set up the whole system. Reference [2] suggests to use Canvendish Kinetics to produce a monolithic IC. This manufacturing is based on a post-CMOS process to create a System on a Chip (SoC). Design IPs for the MEMS elements and an embedded MEMS process module are used. There are design IPs for the CMOS infrastructure as well. Because the CMOS technology is the contraint in this type of process, not all MEMS elements can be implemented.
Essay background
The intention of this essay is to investigate the possibilities of integrating MEMS and CMOS together in one or two seperate processes. There are a variety of factors which are to be taken into account when designing a CMOS-MEMS integrated circuit. Some reasons for designing CMOS-MEMS [1] are shown below: No discrete components Less parts assembly and packaging High quality factor (Q) High operational frequency Possibility for mixing and ltering to be performed simultaneously This essay will try to take a look at available processes, materials for the design, parameteres for both CMOS and MEMS, the possibilities for full circuit integration as well as reviewing existing CMOS-MEMS product. The essay will have its focus towards RF MEMS related processes.
3 Materials
To be able to integrate MEMS into CMOS (CMOS-MEMS), it is important that the materials offered from the IC foundry are capable MEMS materials [3]. CMOS has always focused on electrical conduction and insulation, while MEMS materials rely heavily on mechanical, chemical and thermal characteristics. A set of test structures must be implemented in order to characterize the CMOS materials and to see if they are good enough as MEMS materials. A test structure is a diagnostic device which contains necessary functions in order to extract material properties. There are four main test structure tests: Simple geometry test, robustness test (mechanical and fracture), minimal micromechanical structuring (minimize undesired complication of the process when using MEMS process steps) and simple measurement method (electrical and optical testing). The following list shows the most important materials used in the processes: Silicon, Si Most common substrate material. Can be doped to change electrical properties. Single-crystal silicon has excellent
mechanical properties. The amount of doping in the silicon depends on the microfabrication to be done. For example, a membrane of a pressure sensor requires low p-doped silicon to be able to etch fast enough with KOH solutions. Polysilicon Used as gate material for MOSFETs. Is also used as electrode, thermoelectric material and as resistor materials for piezorestitive sensing. Metal layers Used as interconnects, electrodes, thermistors. Aluminum, titaniun tungsten are most used and are deposited by sputtering. Copper is used in state of the art CMOS technology because of lower resistance and higher electromigration resistance. Electromigration is the transport of ions in a conductor due to momentum transfer between conducting electrons and diffusing metal atoms. A high electromigration will allow high current densities. Insulator layers and masks: Silicon dioxide, SiO2 - Etch resistant Silicon nitride, SiNx - More resistant to chemical etch than SiO2 Polymers - Used as photoresist, dielectric material or passivation layer. Electrical restitivity, thermal conductivity and thermal capacity are important parameters for MEMS. These parameters can be tested by making test structures where the parameters can be extracted out from the behaviour. For example, a micromachined van der Pauw structure can be created to measure in-plane thermal conductivities of thin lms. The structure consists of a four armed greek cross structure suspended over a micromachined cavity. Another example is to extract the thermal response coefcient by applying a range of frequencies [5]. An important part for MEMS sub-systems is to extract mechanical material properties. A set of test models are applied in order to determine the mechanical properties of the materials. Parameters for load-deection, strain deformation, residual stress, Poissons ratio can be achieved from Finite Element Method analysis or from test structures. Examples of test structures can be square diaphragms or long diaphragms. MIT [6] or Sandia National Laboratories extracts Youngs modulus and pull-in voltage for a set of structure by performing electrostatic tests on beams [7]. For RF MEMS, thermoelastic friction may cause irreversible heat generation. This may limit the quality factor of micromechanical resonators. Still, there are many factors in CMOS-MEMS which limits the optimal functionality of the system when considering material behaviour. The CMOS technology has excellent geometric and electrical characteristics. The additional mechanical and thermal material characterization of MEMS is less mature than CMOS, thus making it a challenge to successfully implement optimal microstructures with given material parameters [3].
modied in a way which would inuence characteristics and yield of the circuits. There are a few CMOS process steps which may affect the following MEMS steps in an undesired way. Because of this, some CMOS steps may have to be modied. Such process modications must be thoroughly qualied. Because of this, it is important to introduce minor process modications to the foundry. Strongly doped p-type substrate for the CMOS circuit is important in order to avoid latch-up. This will affect anisotropic etch rates and etch directions. Reduced pdoped epitaxial layers will reduce this problem. Excess interstitial oxygen concentration is present in order to reduce metal impurities in CMOS processing. This excess oxygen concentration will affect anisotropic etch quality, leaving rough edges or cavities. The problem can be solved by either using an etch-stop ring around the membrane or using an electroplated metal ring. Reducing excess oxygen concentration is possible, but external gettering must be employed to reduce crystal defects on the back of the wafer. A getter is a small circular trough or channel lled with metals that oxidize quickly. It is important not to employ high temperature MEMS process steps if the CMOS circuitry is made rst. Use LPCVD to deposit polysilicon and the annealing step afterwards must be within the so called thermal budget. For interconnects, CMOS often use aluminium. Because aluminum only withstands 450 C, there are a few choices: The rst one is to deposit aluminium after MEMS circuitry (if possible). The other choice is to have a thermal budget of 450 C. The third choice is to use another material for interconnects other than aluminum. At the end of the CMOS process, a passivation layer is deposited. This passivation layer may be modied in order to match residual stress with layers from the MEMS process. It is important that the stress proles match in order to not introduce compressive or tensile stress. Also, stress in the passivation layer may inuence the underlying metal interconnects. The design of a fully integrated system can be done using CMOS processes and CMOS design programs. In order to be able to create MEMS structures, certain rules in the program must be violated because the design software does not understand the process modication. For example, the design program will not allow making a via without having metal below and top of it. It is possible to avoid these rule violations by expanding the standard design rule set with the new MEMS modications, but this will be time consuming. For complex MEMS designs, this is preferable. CMOS-MEMS consists of electronic circuits with one or more micromechanical transducers. IC designers use electronic design automation (EDA), while MEMS designers typically use nite element modeling software. In order to simulate CMOS with transducers, behavioral models for the transducer models must be made. Because transducers contain several signal domains, lumped element circuit models of the transducer must be developed manually. An example of a software designed to handle these types of systems is INTEGRATOR from Coventor. This software will perform mechanical simulations and then export these simulations to standard circuit simulators. For layout-versus-schematics (LVS), these libraries must also be extended in order to include the MEMS elements [3].
CMOS-MEMS parameters
As a summary, the pre-CMOS fabrication module does not interrupt the CMOS process, nor does it place any constraints on the thermal budget of the CMOS process. There are many choices of CMOS technologies, but this approach requires much development and preparation of the micromachined wafers. The MEMS wafers are the start material and must have proper contact to the following CMOS circuit. In addition to this, there will most probably be a requirement of post-CMOS processing in order to release devices [3].
Technical University of Aachen and the Fraunhofer Institute IMS creates a pressure sensor during an intra-CMOS module. Deposition of SiNx , sacricial oxide layer and polysilicon structure layer is performed after implantation of source and drain areas. Then the polysilicon membrane is released by an HF etch. This technique allows the construction of an array of polysilicon membranes in parallell for measuring different pressure ranges [11]. Bosch and Fraunhofer Institute ISiT proposes a way to implement thick polysilicon structures without inuencing the source/drain doping characteristics. A surface micromachined accelerometer is made by depositing a sandwich of a sacricial oxide and a LPCVD polysilicon layer. An epitaxial layer forms the circuit areas, and then an epi-poly layer is grown with the LPCVD polysilicon layer, creating a 10 m thick structure with excellent mechanical properties [12]. IBMs T.J. Watson Research Center avoids high temperature steps by creating dielectric lms using PECVD at about 400 C. Microstructures are removed by removing an organical sacricial layer using oxygen plasma [13]. In summary, this fabrication module brings many constraints to the CMOS technology, thus making it harder to proceed to more advanced CMOS technologies. In addition, it is preferable to have fabrication and micromachining at the same foundry. The critical factor is the high temperature steps required for fabrication of microstructures, thus violating the thermal budget of the CMOS process. This fabrication module is not favorable [3].
1. Add layers on top of completed CMOS wafer: This type of fabrication consists of using surface micromachining techniques. It can be performed with low temperature modules (100-150 C) with PVD, electroplating or CVD processes over 300 C. It is important to have a smooth surface to deposit on and good electrical contact with the CMOS circuit. The CMOS may be surrounding the MEMS or be located underneath the microstructures. By using metal-silicon contacts and tungsten interconnect in CMOS, it it possible to perform high temperature MEMS processes. A passivation layer planarizes the wafer and protects the circuit below. The passivation layer may consist of phosphosilicate glass (PSG) and low-stress LPCVD SiN . Rapid thermal annealing may be performed, but still there might be a change in the doping proles of the transistors. [3] The high temperature process occurs when using polysilicon as a structural layer. This may be solved by using an alloy: Polycrystalline silicon-germanium which can reduce the temperature processes to 450C , which is below the thermal budget. Using PECVD to deposit poly-SiGe has increased deposition rates. Investigation of Poly-SiGe is described in [14]. Poly-SiGe might be a future structural material for post-CMOS as long as deposition stress and annealing parameters are improved. The top part of gure 6 is a post-CMOS microstructure made on top of a CMOS circuit based on this Poly-SiGe process. By using PVD, the temperature can be further reduced to 350 C. The silicon structural layer with TiW enables a MEMS structure with low electrical resititivity where the structural layer is released by dry etching. Another possibility is to use electroplating to form a microstructure by selectively electroplating in open mold areas. Electroplated processes are performed in room temperatures, however the control of thin lm stress and stress gradients are very important. Instead of
adding layers on top of the CMOS substrate, it is possible to manufacture CMOS and MEMS wafers seperately, and then solder bond the MEMS wafer on the CMOS wafer [3]. 2. Modify CMOS layers to make microstructures The bottom part of gure 6 on the preceding page shows how microstructures can be made with CMOS layers [1]. This requires less process steps both for release of structure and deposition steps. This process resembles bulk etching, and many commercial devices using this technique are piezorestitive pressure sensors. The CMOS process may be slightly modied to improve the doping prole for piezoresistors. It is popular to back etch the silicon substrate to make a cavity for membranes or cantilevers. Anisotropic KOH etch or etch stop can be used to make the cavity. Using SOI from CMOS offers the possibility to use a buried oxied layer as an etch stop. Using p++ etch-stop techniques requires an intra-CMOS module to diffuse highly doped p-regions. A special wafer holder can be used to protect the front of the wafer which contains the CMOS circuit. Then the back of the wafer is wet etched. Anisotropic wet etch with electrochemical etch stop requires a contact potential. In either case, the wafers must be manually placed in the mechanical holder, which is a slow production process. This is done because this is the most reliable process to protect the front side of the wafer. DRIE etching is an alternative, providing excellent aspect ratio. The equipment to do this, however, is expensive. It is also possible to wet etch and dry etch the substrate from the front side, for example by using electrochemical etch-stop [3]. An interesting bulk micromachining technology from Carnegie Mellon University utilizes the top metal interconnect as an etch mask for the microstructure denition. By doing this, the CMOS technology masks will dene the length and width of beam, thus creating smaller structures than normal. A sandwich of dielectrics with polysilicon and metal layers is made and the structure is released by two dry etching steps. Anisotropic etch will etch areas not protected by the metal mask, while isotropic etch will underetch the oxide and thus releasing the structures. This technique, however, is exposed to large residual stress and stress gradients in between the dielectric layer sandwich. By using DRIE etch at the front, and anisotropic etch on the back of the wafer, this problem is solved. Anisotropic etch at the front removes SiO2 not protected by the top metal layer. The resulting thickness of the structure is 10-100 m [15]. A surface-micromachining technique developed in Zurich is used to remove thin lms of the CMOS process selectively. The process is known as SALE, sacricial aluminum etching. It consists of removing metal layers from the CMOS process to release microstructures. The microstructures consists of intermetal dielectric, the upper metallization layer and the passivation. It is also possible to make a nanocantilever by using laser beam litography on an aluminum etch mask. The polysilicon cantilever is released by removing eld oxide layer beneath the cantilever [16]. Conclusion of Post-CMOS micromachining: Post-CMOS microfabrication is the most exible fabrication module, especially because the CMOS and MEMS circuits may be made seperately. The constraints are: Bulk-CMOS micromachining requires a certain substrate material. New substrate materials must be thoroughly qualied.
The planarity of area for the MEMS device is important. Increasing amount of metallization layers may increase topographic variations. Because of this it is important to have proper planarization steps. The CMOS process introduces a thermal budget which the following MEMS fabrication module must keep below. This thermal budget is set from metallization layers and doping proles of the transistors. The advantages of post-CMOS are: The MEMS structure may be placed on top or beside the CMOS circuitry The CMOS process can follow the CMOS development track in order to reduce area of the microelectronic circuit. It is possible to use two different foundries for the CMOS and MEMS circuits. No expensive re-qualication of the CMOS process
The package of a CMOS-MEMS die requires special attention due the MEMS structure which needs external measurement equipment. RF MEMS requires hermetic sealing for high-Q resonators to avoid air damping. Packaging and testing consists of a little more than one third of the total costs [3].
The digital circuitry follows the CMOS development track. The analog circuitry has transistor sizes which are larger because of transconductance, noise and matching. RF design has an advantage in the steady progress of the CMOS development track due to an increase in transit frequency and reduction in parasitic capacitance. This has drastically increased the frequency range for RF MEMS. The fabrication of masks places a huge constrain on scaling between MEMS and CMOS, and it also increases the complexity of the mask development cycle. The mask development must be determined from simulations and experiments due to the lack of software integration tools. The interface between the CMOS-MEMS chip and the off-chip data processing unit can be done in three ways. Analog output places a big challenge in the conditioning of the signal off-chip without noise and perfomance degration. Digital output requires a microcontroller or FPGA for acquiring the signal. High-volume producers of CMOS and MEMS use complete systems where the sensed signal is thorougly processed. A standard digital interface display and analog read-out circuitry is normal. There is a trend going towards this system-on-chip module. Fully differential designs are preferred because the perfomance of integrated microsystems can be improved. The signal-to-noise ratio is improved by including lters. Digitalto-analog converters may also be used to tune the circuit (for example actuation), this enables the possibility to calibrate the integrated system. Using ADCs and DACs places a demand in accurate voltage references. Using a diode voltage and a voltage that is proportional to absolute temperature creates a temperature independent voltage reference approximately close to the bandgap voltage of silicon. Impedance matching to the wire bonds is an important factor in packaging.
8 Conclusion
The integration of CMOS and MEMS together has been a eld of investigation and research for a long time. The trend has been to make CMOS and MEMS seperately and then wire bonding the two dies together in one package. The reason for this is because the MEMS processes are behind the CMOS development, as well as the costs for integrating them together. Therefore the trend has been to make CMOS and MEMS seperately, where only Analog Devices and Texas Instruments has been the only companies to volume produce CMOS-MEMS products. Making CMOS and MEMS on the same die requires preparation and modication of an existing CMOS process. It is both challenging and expensive to do this. Depending on the product to be made, the result may outweigh the effort to integrate CMOS and MEMS. Excluding discrete components, less packaging, less requirement for signal matching between CMOS and MEMS are all parameters that indicates that integrating CMOS and MEMS is a benet. RF MEMS in particular benets from the fact that less wire bonding increases the frequency range for the system because of the CMOS development as well as reduction of parasitic capacitances. Pre- and Post-CMOS are the most promising modules to integrate the two parts. Pre-CMOS has its advantages by placing no constraints on the CMOS part, and it has no thermal budget to stay below during the MEMS module. Post-CMOS is more exible because the MEMS structure can be placed on top or beside the CMOS, the CMOS may follow the CMOS development track, and it is also possible to manufacture the two modules at respective foundries. Modication of
the CMOS processes, mask development, system integration software tools, foundry processing and costs are the most constraining factors of integrating CMOS and MEMS. The making of a RF MEMS circuit with integrated microelectronics can be performed with the following candidates: INTEGRAM from QinetiQ, ASIMPS from MEMSCAP or Dalsas intelligent MEMS process.
References
[1] F. Chen, U. et al, CMOS-MEMS Resonant RF Mixer-lters , Micro Electro Mechanical Systems, 2005. MEMS2005. 18th IEEE International Conference on 30 Jan.-3 Feb. 2005 Page(s):24-27 [2] M. Beunder, Choosing a CMOS Compatible MEMS Manufacturing Approach , Cavendish Kinetics B.V.,CEO The Netherlands.,COMS 2004 [3] O. Brand, G.K. Fedder, CMOS-MEMS , Advanced Micro & nanosystems, Volume 2, WILEY-VCH Verlag GmbH & Co. KGaA, 2005, chapter 1,2,5 and 11. rmer, et al, Method of anisotropically etching silicon , [4] F. L a German Patent DE 4241045, US Patent 5501893, 1994 [5] L.J. Van der Pauw Philips Res. Rep. , 13, 1-9, 1958. [6] P.M. Osterberg, S.D. Senturia, J. Microelectromech. Syst. , 11, 512-520, 1997. [7] B.D. Jensen, et al, Proc. MSM 99, San Juan, Puerto Rico , pp. 206-209, 1999. [8] L. Parameswaran, et al, A merged MEMS-CMOS process using silicon wafer bonding , Proc. IEEE IEDM 95, pp. 613616, 1995. [9] T.A. Core, et al, Fabrication technology for an integrated surface micromachined sensor , Solid State Technol., 39-47, 1993. [10] T. Scheiter, et al, Full integration of a pressure-sensor system into a standard BiCMOS process , Sens. Actuators A, 67, 211214, 1998. [11] H. Dudaicevs, et al, Surface micromachined pressure sensors with integrated CMOS read-out electronics , Sens. Actuators A, 45, 157-163, 1994. [12] M. Offenberg, et al, Novel process for a monolothic integrated accelerometer , Proc. Transducers 95, pp. 589-592, 1995. [13] C. V. Jahnes, et al, Simultaneous fabrication of RF MEMS switches and resonators using copper based CMOS interconnect manufacturing methods , Proc. IEEE Microelectromechanical Systems Conference (MEMS 2004), pp. 789-792, 2004 [14] A E. Franke Polycrystalline silicon-germanium lms for integrated microstructures , J. Microelectromechan. Syst. 2003, 12, 160-171, 2003. [15] G.K. Fedder., et al, Laminated high-aspect-ratio microstructures in a conventional CMOS process , Sens. Actuation A, 57, 103-110, 1996. [16] O. Paul, et al, Sacricial aluminum etching for CMOS microstructures , Proc. IEEE MEMS 97, pp. 523-528, 1997. ller, et al, Solid-State Sensors, actuators and microsys[17] T. Mu tems , 2005. Digest of Technical Papers, Volume 2, 5-9 June 2005 Page(s):2074-2077 Vol.2 [18] J.A. Geen, et al, Single-chip surface micromachined integrated gyroscope with 50/h Allan deviation , IEEE J. Solid-State Circuits, 37, 1860-1866, 2002 [19] J. Chae, et al, A hybrid silicon-on-glass (SOG) lateral microaccelerometer with CMOS readout circuitry , Proceedings of the 15th IEEE international Conference on Micro Electro Mechanical Systems (MEMS 2002), Las Vegas, NV, pp. 623626, 2002. [20] C-C. Lo, et al, Assessment of silicon wafter material for the fabrication of integrated circuit sensors , J. Electrochem. Soc., 147, 1604-1611, 2000