LUMINESCENCE

ELSEVIER
Journal

JOURNAL

OF

of Luminescence 70 (1996)

129-143

Electron-phonon

interactions in semiconductor nanocrystals

T. Takagahara

NTT

Basic Research

Luhorutories,

3-1 Morinosuto

Wakutniya,

Atsuqi,

Kuncrgucva 243-01, Japan

Abstract

The electron-phonon interactions in semiconductor nanocrystals, especially concerning the acoustic phonon modes are derived and the size dependence of the coupling strength is clarified for typical coupling mechanisms. On the basis of these results, the commonly observed linearly temperature-dependent term of the excitonic dephasing rate and the proportionality of its magnitude to the inverse square of the nanocrystal size are attributed to the pure dephasing due to the deformation-potential coupling. The luminescence Stokes shift and the Huang-Rhys factor due to acoustic phonon modes in Si nanocrystals are discussed in conjunction with the origin of the recently observed luminescence onset energy.
Kr~~ords: Electron-phonon interaction; Semiconductors; Nanocrystals

1. Introduction Semiconductor nanocrystals of a size comparable to or smaller than the exciton Bohr radius in the bulk material are attracting much attention from the fundamental physics viewpoint and from the interest in the application to functional devices [l]. Especially, their novel optical properties due to the discrete electronic energy levels have been investigated extensively [2,3]. In semiconductor nanocrystals, not only the electronic energy levels but also the lattice vibrational modes become discrete due to the threedimensional confinement. The consequences of the latter feature, namely the phonon confinement, have been studied extensively. The longitudinal optical (LO) phonons in semiconductor nanocrystals were observed by the resonance Raman scattering [4-61 and the size dependence of the electron-LO phonon coupling strength was discussed [S, 7,8]. Also the size-quantized acoustic phonon modes were observed by the low-frequency Raman scattering [9]. In addition to these studies, the excitonic dephasing in various semiconductor nanocrystals has been mea0022-2313/961$15.00 0 1996 Elsevier Science B.V. All rights reserved PII SOO22-23 13(96)00050-6

sured in detail as a function of the nanocrystal size and the temperature. In CuCl nanocrystals, the homogeneous linewidth of the excitonic transition was measured from the luminescence line width under size-selective excitation [lo] and by the spectral hole burning [ 111. In nanocrystals of II-VI compound semiconductors, the excitonic dephasing constant was measured by the spectral hole burning [12-l 51 and by the four-wave mixing [ 16, 171. The commonly observed T (temperature)-linear behavior of the excitonic dephasing rate suggests the importance of the electron-phonon interaction with acoustic phonon modes, although the relevant temperature range is dependent on the nanocrystal size and the material. In the higher-temperature region, the temperature dependence of the excitonic dephasing rate deviates from the T-linear behavior, indicating the participation of LO phonons. Si nanocrystals are now attracting much attention because of their visible photoluminescence [ 1, 181. However, the mechanism of the photoluminescence is still controversial. In order to resolve the controversy, it is necessary to study finely the

130

T Tukayuhara/

Journd

of Lumineswncr

70 (1996~

129-143

excitonic energy spectra in Si nanocrystals by sizeselective spectroscopies [ 19, 201 and to compare them with theoretical calculations. For example, from the onset energy of photoluminescence and from the temperature dependence of the luminescence lifetime, the singlet-triplet exchange splitting of the excitonic states was claimed to be observed as a function of the nanocrystal size [ 191. The observed splitting is of the order of 10 meV and is much larger than the bulk value [21]. This suggests strong enhancement of the electron-hole exchange interaction in a Si nanocrystal. On the other hand, Martin et al. [22] reported a theoretical calculation which shows that the onset energy of photoluminescence cannot be explained in terms of the excitonic exchange splitting but may be partially ascribed to the Stokes shift of photoluminescence. At the same time, step-like structures due to optical phonons were observed in the luminescence spectra [19, 201, although the structures associated with acoustic phonons were not resolved well. In this paper, we derive the electron-phonon interactions with acoustic phonons in semiconductor nanocrystals and clarify the size dependence of the coupling strength for typical coupling mechanisms, i.e., the deformation potential coupling and the piezoelectric coupling. On the basis of these results, we identify the origin of the commonly observed T-linear term of the excitonic dephasing rate and of the proportionality of its magnitude to the inverse square of the nanocrystal size. We also formulate the electronphonon interaction in Si nanocrystals for acoustic phonon modes to calculate the luminescence Stokes shift and the Huang-Rhys factor. As a result, we find that the luminescence onset energy can be explained mainly in terms of the excitonic exchange splitting in contrast to the conclusion of Ref. [22], although the contribution from the luminescence Stokes shift is not negligible.

following, the shape of a nanocrystal is assumed to be spherical and the anisotropy of the elastic constants is neglected for simplicity of the arguments. Then the vibrations of an elastically isotropic sphere can be described by d2 PSU = (j. + p)graddivu + pD2u, (2.1)

where u is the lattice displacement vector, p the mass density and i and ,Uare the Lames constants [23]. The eigenmodes of the above equation under the stressfree boundary condition were studied by Lamb for the first time [24]. There are two kinds of eigenmodes, namely torsional modes and spheroidal modes. Under the spherical geometry, each mode can be expanded in terms of vector spherical harmonics [25]. These harmonics for an angular momentum 1 and its z-component m are defined as -&(kr) N,,(kr) 1 = -~Qi1(,,(k~), k 1 = -rot Ml,(kr) k M/,fkr)=rot(ry~Y,,(kr)), (2.2)

with Yl,(kr) = j,(kr)Y,,(Q), where j,(kr) is an Ithorder spherical Bessel function and Y,, is a spherical harmonics. ( 1) Torsional modes: The displacement vector apart from a normalization factor is written as u(r) = &@-) =j,(kr) 2 2 0 A-8% ( sin 02cp Y/,(Q), (2.3)

where 13 1 because M,,,, for 1 = 0 vanishes identically. This mode is purely transversal because div u = 0. The eigenfrequency w is related to k by pk2 = pw2. The stress-free boundary condition leads to an equation which determines the size-quantized k- values, namely, kRj;(kR) - jl(kR) = 0, (2.4)

2. Energy spectra of acoustic phonon modes in spherical nanocrystals In semiconductor nanocrystals, even the acoustic phonon modes become discrete due to the size quantization. As long as the nanocrystal size is not too small, its acoustic properties can be described in terms of the elastic vibration of a homogeneous particle. In the

where R is the radius of a spherical nanocrystal. It is interesting to note that this equation is independent of the elastic constants of the material and the k-value is determined only by the radius. (2) Spheroidal modes: The displacement vector is written as 12.5)

T Tukugaharal

Journal of Luminrscence

70 (1996)

129-143

131

where the eigenfrequency w is related to h and k by po2 = (2. + 2p)h2 = ,uk2. This mode has a mixed character of the longitudinal and transversal modes. The coefficients p,m and qlm have to satisfy a set of linear equations

The displacement

vector can be generally expanded as

(2.6)
with ~lrn = -02hRjr(hR) B/n, = IkRjl(W :jlnl = -a*hRjr(hR) 61, = (l+ 1)[2(1+ 2(1+ 2)jl+,(hR), - 2/(1+ 2)jl+r(W, (2.7) (2.8) (2.9) (2.10)

where al,,,j and b,, are the boson annihilation operators of the torsional and spheroidal phonon modes, respectively, and the factor of (- 1) appears due to the property, i.e., Yk = (- 1) Y,_,. The coefficients from the 5hJ and a/mj are now to be fixed. Starting Lagrangian density given as 6w = ids - ;(I. + 2p)(divu)* ~ ip(rot u)~ (2.15) and postulating the commutation relation between the displacement vector u and the canonically conjugate momentum vector pti, we find that r/mj and O/mj have to satisfy 2p0~~~r~,,,~ = fi and 2pw~,/0~,,,~ = k, (2.16)

+ 2(E - l)ji_l(hR), l)j,_I(kR) - kRj,(kR)],

where 0 = (i + 2p)/p and h and k are determined from ~l~81,,~- fllrn~lrn = 0. In the case of 1 = m = 0, the above equation should be replaced by - a*hRjo(hR) + 4jl (hR) = 0. (2.11)

Now, the typical phonon energy spectra of a CdSe nanocrystal with 2OA radius are shown in Fig. l(a), (b) for the torsional and spheroidal modes, respectively. The material constants employed are [26] i. + 2~ = cl, = 7.49 x IO(dyncm-*), p = ~44 = 1.315 x 10(dyncm-2) and p = 5.81(gcmP3). (2.12)

respectively, where m:,, (0 ) is the eigenfrequency of the torsional (sphezdal) phonon mode. The total phonon Hamiltonian and the secondquantized form of the acoustic phonon displacement can then be written as

These size-quantized acoustic phonon modes were observed by the low-frequency Raman scattering [9]. In the calculation of the electron-phonon interaction, the second-quantized form of the phonon displacement becomes necessary. Here the procedure of the second quantization is described. The torsional and spheroidal modes will be denoted by Tlmi(r) and Sl,j(r), respectively, where 1 and m are the angular momentum indices as above and j denotes the radial quantum number. They are normalized within a nanocrystal as d3r1Tf&r). = T/,(r) = 1. (2.13)

and

+C

hJ

h(b,mj d +
2Pw:mJ

(- 1)mbj_,,l, )Slmj(r).
(2.18)

3. Derivation of the electron-phonon s

interactions

d3r S,,!j(r) . Sl,j(r)

The electron-phonon interaction with acoustic phonon modes arises mainly through the deformationpotential coupling and the piezoelectric coupling.

132

T. Tukugalruru,i

Jountul of Luminexence

70 (1996) 129-143

Deform&on potential coupling: When the crystal lattice deforms, the electronic energy structure changes in proportion to the strain tensor and the proportionality constant is called a deformation potential. Although the detailed form of the deformation potential coupling is dependent on the crystal symmetry [27], the most dominant term can be described by Eddiv U, where Ed is a deformation potential constant. Hereafter only this term will be taken into account. Within this simplification, the torsional modes do not contribute to this coupling because of their transversal character. In the case of a spheroidal mode, only the longitudinal component contributes to the deformation potential coupling and we obtain Ed div SI,,~,= Ed divhlph,(~~-) + ~~~~.jNdkjf)) (3.1)

spherical coordinates are given in Appendix A. The electron-lattice interaction is given by the potential energy of the lattice polarization in the electric field induced by an electron and is represented as (3.5) where c is the intermediate-frequency dielectric constant. In the calculation of this integral, it is convenient to employ the expression

= ~Edh,P/nl,j/(h;r)Y/,,,(~).

(3.6) where 7, = max(r,r,), r< = min(r, re ), PI, and BI,,, are components of the angular part of the vector spherical harmonics as given in Appendix B. Substituting Eq. (3.6) into Eq. (3.5) and carrying out the integration, we obtain the electron-phonon interaction Hamiltonian as

This interaction is of the form of a contact coupling between an electron and the lattice displacement and the relevant Hamiltonian is given by Eq. (3.1) only by substituting the electron coordinates. The exciton phonon interaction Hamiltonian is composed of the interaction terms of the conduction band electron and of the valence band hole. The deformation potentials in a nanocrystal might not be very much different from those in bulk materials as long as the nanocrystal is not in the regime of small clusters. Piezoelectric coupling: In polar semiconductors, the lattice strain produces the lattice polarization and this polarization interacts with an electron. The lattice polarization in the Cartesian coordinates is given as
PPZ(Y) = (e,5e=,~,e,5e?,=,e3,(e.~~ + e,?.) + e33e,)

(3.2) for the wurtzite structure and as &Z(Y) =

e14f(e,,z, ez~r,e,y)

(3.3) where ei, is

where d and dt represent the phonon operators of the torsional and spheroidal modes and the expression of ~~I,,?n&7l,, is too much complicated to obtain a physical insight. The important feature is the phonon mode selection rule, namely the allowed numbers of I and m for a fixed number of 1, and m,. These selection rules are summarized in the following. Zinc blende-type crystuls: IIILV compound semiconductors and copper halides are typical crystals of this symmetry. The allowed values of I and m are I=1,+2,1,,~1,~2~ and m=m,*2 (3.8)

for the zinc blende structure, respectively, the strain tensor defined by

for the case of torsional phonon modes and (3.4) l=l,t3,1,+1,...,)1,~3~ and by their cyclic permutations and e15, e31, e33 and e14 are the piezoelectric constants [28]. The explicit expressions of these tensor components in the and
m=m,+2 (3.9)

for the case of spheroidal phonon modes, respectively.

T. Tukaguhara / Journal qf Luminrswnce

70 (I 996 J 129-143

133

Wurtzite-type crystals: Most of II-VI compound semiconductors belong to this case. The allowed values of I and m are 1=1,+2,1,,~1,-21 and m=m, (3.10)

for the case of torsional phonon modes. It is to be noted that when m = 0, the coupling vanishes for any value of 1. On the other hand, for the case of spheroidal phonon modes, the allowed values of 1 and m are 1=1,+3, and m = m,. (3.11) I,+ l,..., min(l& ~ 31, 11, ~

piezoelectric coupling might be stronger than the deformation potential coupling. For the sake of reference, the size dependence of the coupling constant will be examined for the case of the Friihlich coupling with LO phonon modes. In a similar way to the case of the piezoelectric coupling, the interaction Hamiltonian is given by

fJd3rVr (A)

.P(r>,

(3.14)

11)
where P(r) is the lattice polarization vector induced by the LO phonon mode. The size dependence is estimated as (3.15)

These selection rules arise from the property of the coupling between the electronic polarization vector and the second-rank strain tensor. In the above, of course, the condition that 13 jml is implicitly assumed. It is very important to examine the size-dependence of these electron-phonon interactions in order to understand and identify the mechanisms of the excitonic dephasing. In the case of the deformation-potential coupling, the size dependence arises from div u and can be estimated as 111 divu x -~~ R@&YF 1 (3.12)

where the first factor comes from the volume integral, the second one from V,.( l/lr - r,i), the third one from the normalization of the phonon mode and the fourth factor comes from the quantization of the LO mode and is almost size-independent.

4. Linearly temperature-dependent of excitonic dephasing rate

term

where the first factor comes from the operation of div, the second one from the normalization of the phonon mode and the third one comes from the quantization of the phonon mode. The eigenfrequency (r) is given by (A + 2p)h2 = pco2 or pk2 = po2, where h and k are determined in Section 2 and are scaled as 1/R. Thus, the eigenfrequency w scales as 1/R and we obtain the dependence in Eq. (3.12). In the case of the piezoelectric coupling, the strain tensor has the same size dependence as div U, namely 1lR2 and we obtain

where the first factor comes from the volume integral, the second one from V,( I/lr - ~~1) and the third one comes from the piezoelectric polarization Ppz(r). From these size-dependencies, we see that in the small size region the deformation-potential coupling would be dominant, whereas in the large size region the

Now that the electron-phonon interactions in semiconductor nanocrystals are derived, we can calculate the excitonic dephasing constant. The excitonic dephasing rate is the decay rate of the excitonic polarization and in general consists of the pure (adiabatic) dephasing constant and a half of the longitudinal decay constant. The former part arises from the fluctuation of the excitonic energy due to the virtual emission and absorption of phonons without the change of the excitonic state, whereas the latter part comes from the phonon-assisted transitions of the relevant excitonic state to other excitonic states. In other words, the diagonal and off-diagonal matrix elements of the electron-phonon interaction with respect to the excitonic states are responsible to the former and latter parts of the dephasing rate, respectively. When specified to a two-level model consisting of the electronic ground state 1~) and the lowest excitonic state iex), the diagonal part of the relevant Hamiltonian in the

134

T. Tukaguharmi Journal qf Luminescence 70 (1996)

129-143

Franck-Condon

approximation

can be written as

terms, namely

(4.6) (4.1) where the subscript j is the phonon mode index, wJ the mode frequency and EO is the adiabatic electronic excitation energy in which the lattice relaxation energy is not included. The coupling constant 7j is the matrix element of the electron-phonon interaction Hamiltonian taken between the same exciton wave functions. The term proportional to yI in Eq. (4.1) represents the shift of the origin of the lattice vibration in the excitonic state relative to the ground state and can be interpreted also as representing the adiabatic fluctuation of the excitonic energy level. The homogeneous line width can be estimated from the line width of the absorption spectrum. The absorption spectrum of the above system is proportional to with z = (E - EO + ili~,,/2)/(JzD) and D2 = C ?l,?(2n(hWj) + 1).
i

(4.7)

(4.8)

[291
Re { LX dtexp [i(-iLR)i (4.2)

+-S+S+(t)+S_(t)]}
with
S*(t) = c

( 1
;i G J

efiwf(~(hCOj)

i 5

i),

(4.3)

s = S+(O) + S_(O),
and

where y,, is the longitudinal relaxation rate of the excitonic state, n(??o) the phonon occupation number and ELR is usually called the lattice relaxation energy. Expanding the exponential function in Eq. (4.2) in powers of S+(t) and S_(t), we obtain the multitude of phonon sidebands. If the low-energy phonon modes are distributed quasi-continuously, the envelope of the phonon sidebands gives the homogeneous broadening of the absorption spectrum. That envelope can be obtained approximately by expanding s+(t) up to t2

In nanocrystals, the energy spectra of acoustic phonon modes are discrete as shown in Fig. 1. However, various combination of such discrete modes can form a continuous distribution of phonon sidebands unless the electron-phonon coupling is concentrated to a special phonon mode. Thus, the above envelope function approximation can be used even for the case of nanocrystals. Concerning this approximation, it is worth mentioning on the optical phonon modes. The optical phonon modes which have an almost sizeindependent energy even in nanocrystals contribute to the definite phonon sidebands in the absorption spectrum rather than causing the energy fluctuation of the excitonic level and the homogeneous broadening of each sideband is determined by the lowenergy acoustic phonons. Thus, the summation over j in Eq. (4.8) is carried out over the low-energy acoustic phonons but not over the optical phonon modes. The homogeneous line width I-h estimated from the half-width at half-maximum (HWHM) of Eq. (4.6) is plotted in Fig. 2 as a function of D scaled by fiy,,. It is seen that the proportionality of rh to D2 holds in the range of D < hy,, In the high-temperature region such that keT > hu,~, D2 is proportional to the temperature and this leads to a T-linear term in rh under a condition that D < fiy,,. The T-linear term of D2 is proportional to Cj $/fic~~. The size dependence of this coefficient will be examined. In the case of the deformation-potential coupling, since yj x 1/R2,wj cc l/R and the phonon mode density scales as R, we have c, ~~?/?Lo, (x l/R2. Precisely speaking, ,, the size dependence of the coupling constant ;i is determined not only by the electron-phonon interaction

T Tukayuhara/ .Iourtzd of Lunzinescence 70 /1996) 129-143

135

0 0

CdSe R=ZOA

0
b)

2
Angular

4
Momentum

I
for (a)

Fig. 1. The acoustic phonon energies of a CdSe nanocrysral torsional modes and (b) spheroidal modes

with 20A radius are plotted as a function of the angular momentum

Hamiltonian but also by the exciton wave function. However, the gross features can be grasped by the above argument since the size dependence arising from the latter factor is not very strong. On the other hand, in the case of the piezoelectric coupling, a similar argument shows that Ci $/u, is independent of the size because yj 3: 1/R, Cuj 0: l/R and the phonon mode density scales as R. These are the key arguments to clarify the underlying mechanisms of the excitonic dephasing in semiconductor nanocrystals. The excitonic dephasing rate in a CdSe nanocrystal with 11 8, radius was measured as a function of the temperature and a typical T-linear dependence was observed over a wide range of temperature [ 161, namely rh = TO + AT + . . . The theoretically estimated pure dephasing rate of the lowest excitonic state for this nanocrystal is shown in Fig. 3 as a function of the temperature with the experimental results. The calculation has been carried out without any adjustable parameter employing the material parameters such as Ed,ql,e33 and ers of bulk CdSe [26] and the exciton wave function calculated by the method

of Ref. [30]. The overall good agreement is obtained between the theory and the experiment except for a constant background (-3 meV), whose origin will be discussed later briefly. The coefficient A of the T-linear term of I-h which is theoretically estimated around 100 K is plotted by a solid line in Fig. 3 as a function of 1/R2. The linearity to l/R2 is clearly seen indicating that the deformation potential coupling is dominantly determining the T-linear term. The experimental value of 0.136 meV/K for an 11 A radius nanocrystal is reproduced well by the theory. For the sake of reference, the experimental values of A for CdSeo,ssSo.r2 nanocrystals [17] are plotted in Fig. 4 by open circles. Since the material parameters of CdSeo.ssSa.t2 are not much different from those of CdSe in the virtual crystal approximation, we can confirm the characteristic l/R2 dependence of A due to the deformation potential coupling. Recently, Mittleman et al. extended the measurement to several CdSe nanocrystals with radii from 10.5 to 20 A [31]. The estimated coefficient As of the T-linear term are plotted in Fig. 4 by solid dots.

136

T. Takaqahara J Journal of Lun~inescrncr

70 (1996)

129-143

R(nm)
643 I 2 1.5

)-

2-

0.2

0.4 0.6 1/R2(nni2)

0.0

l,

3 1

Fig. 4. The coefficient of the linearly temperature-dependent term of the pure dephasing rate in CdSe nanocrysrals is plotted by a solid line as a function of the inverse square of the radius. The solid circles represent the experimental values for CdSe nanocrystals (Refs. [ l6,3 I]), while the open circles depict the experimental values for CdSeo 88So 12 nanocrystals (Ref. [ 171).

2 DI Y,,h

Fig. 2. The homogeneous linewidth Fh estimated for HWHM of Eq. (4.6) is plotted as a function of D. Both Fh and D are scaled by fL;.,, The dashed line represents d%?D as an asymptote of the solid line

4o
t

Gs?
R.111

2 30
E c

2o

50

100 T(K)

150

200

Fig. 3. The pure (adiabatic) dephasing rate of the lowest exciton state in a CdSe nanocrystal with I I ..& radius is plotted as a function of the temperature. The experimental data of Ref. [16] are also plotted with error bars.

They show substantial deviation from the theoretical line, indicating the fortuitous agreement at R = 11 A. At present, the origin of this discrepancy is not well understood, although some possible origins can be suggested. In the very small size regime, some dephasing processes associated with surface defects or trap states may be contributing much more than the pure dephasing process and the coefficient A is affected by those contributions having different size dependence. In the theoretical analysis in Ref. [32], rh is identified with HWHM of the Gaussian envelope function, namely rh = mD with D in Eq. (4.8) without taking into account the longitudinal decay rate ;,,. Also the deformation potential constants and the piezoelectric constants are fixed at the bulk values. In general, the theoretical fitting of the temperature dependence of rh has to be done by varying ;,, and D based on Eq. (4.6). Then we can obtain information coupling constants about ;,, and the electronPphonon in nanocrystals compared with the bulk values. Also there may be some uncertainty on the experimental side, especially the large error bars in Fig. 4 of Ref. [ 161 and the inevitable size inhomogeneity in actual samples. Anyway, a more detailed analysis is left for future study. In order to see the relative ratio between contributions from the deformation-potential coupling and

T. Takagahova I Journal qf Luminescence

70 f 1996) 129-143 Rhm)

137

CdSe T=80 K 101 r$DF)

865 CUCI

0.1 l/R2h62)

0.2

Fig. 6. The coefficient of the linearly temperature-dependent term of the pure dephasing rate in CuCl nanocrystals is plotted as a funtion of the inverse square of the radius. The open circles represent the experimental values of Ref. [IO].

5
Rhd

10

15

Fig. 5. The squared pure dephasing rate is decomposed into two components rz(DF) and rt(PZ) arising from the deformation potential coupling and the piezoelectric coupling, respectively. These components in CdSe nanocrystals at 80 K are plotted as a function of the radius.

the piezoelectric coupling, we decompose (4.8) as rl(ac.) = ri(DF) + ri(PZ), where the first (second) term is the contribution from the deformationpotential (piezoelectric) coupling. In Fig. 5 these contributions are plotted as a function of the radius for CdSe nanocrystals at 80 K. As expected before, in the small size region Tz(DF) is dominant, whereas in the large size region ri(PZ) is larger than Tt(DF). The crossover between the two components occurs around R = 70 A. In CuCl nanocrystals embedded in NaCl matrices, the excitonic dephasing rate was measured systematically as a function of the temperature and the nanocrystal size [lo]. The size dependence of the coefficient A of the T-linear term of ri, was found to be well described by 1/R2. Since the size of experimentally studied nanocrystals is several times the exciton Bohr radius in bulk CuCI ( 7 A), the exciton wave function can be approximated well by that in the center-of-mass confinement regime. The energy scale of acoustic phonon modes of a CuCl

nanocrystal is similar to that for a CdSe nanocrystal of the same size because the elastic constants of both materials are similar to each other. The coefficient A is calculated employing the material parameters of bulk CuCl [26] and is shown in Fig. 6 as a function of l/R2 with the experimental data. In this case also the characteristic dependence of A on l/R2 due to the deformation potential coupling can be confirmed both theoretically and experimentally. Finally, we will discuss briefly the contribution of the longitudinal relaxation processes to the dephasing constant. As such processes, we can consider phonon-assisted transitions to other excitonic states or to trapped states associated with the nanocrystal surface and defects. In the case of CdSe, as long as the wurtzite symmetry can be assumed even in nanocrystals [33], the energy splitting between the A and B excitons may be very close to an LO phonon energy (-26 meV) [26] and the LO phonon-assisted transition to the B exciton is expected to contribute significantly to the dephasing rate of the A exciton at high temperatures. In fact, the deviation of the experimental dephasing rate from the T-linear behavior at high temperatures (> 200 K) may be explained by this mechanism. The localized excitonic states associated with the surface or defects were found more than several tens of meV below the lowest excitonic state [34]. Thus, the transitions to those states occur through multi-phonon emission and contribute an almost temperature-independent term to the excitonic

dephasing rate giving rise to a constant which persists even at 7 = 0 K.

background

action in bulk Si was derived by Laude et al. [36]. At the valence band top at the r-point, the interaction can be written as

5. Luminescence factor

Stokes shift and Huang-Rhys -3bl [(L - iL)e,, - VQ, [(LJ,, -3h2[(L,cry ~ vQZ[(LXI + c.p.1 + c.p.] + c. p.]

We have so far been discussing the excitonic dephasing due to the electron-acoustic phonon interaction. There is another intriguing phenomenon caused by this interaction, namely, the intrinsic Stokes shift between edges of the absorption and luminescence spectra. The term intrinsic is meant to exclude the luminescence associated with defects and impurities. As mentioned in Section 1, the luminescence Stokes shift and the excitonic exchange spIitting in Si nanocrystals were discussed in conjunction with the origin of the luminescence onset energy. Martin et al. [22] reported that the onset energy of photoluminescence cannot be explained in terms of the excitonic exchange splitting but may be partially ascribed to the Stokes shift due to acoustic phonon modes. However, they dealt with the dielectric screening erroneously in the exchange integral and introduced some phenomenological assumptions about the exchange integrals and the phonon modes in nanocrystals whose validity is not quite obvious. On the other hand, in Ref. [19], the expression of the excitonic exchange splitting for the directgap material was used for Si nanocrystals. However, the justification of this simple-minded extension to the case of indirect-gap materials with a multi-valley structure is not obvious and an accurate estimation of the exchange splitting cannot be expected. In order to clarify these features and to search for a quantitative interpretation of the observed onset energy, we calculated the excitonic energy spectra in Si nanocrystals focusing on the strong enhancement of the exchange splitting, employing the multi-band effective mass theory and assuming a spherical shape of nanocrystal to simplify the analysis but to retain the essential physics [35]. Here, we formulate the exciton-phonon interaction in Si nanocrystals for acoustic phonon modes to calculate the Stokes shift. Since Si is a non-polar material, the piezoelectric coupling is absent and only the deformation potential coupling is effective. The explicit form of the electron-acoustic phonon inter-

+ L$Jf&,,. - ;L . o)e,,

+ &P%)~.Y~~ C.P.1, +

(5.1)

where Li (i = x, Y, z) are the angular momentum operators in the manifold of L = 1, oi (i = x, y, z) the spin operators, aI, b 1, bz, d 1, d2, deformation potential constants, c.p. means the cyclic permutation and the last two terms describe the stress-dependent spin-orbit interaction. On the other hand, at the conduction band valley bottom on the d-line, the interaction Hamiltonian is written as H~~,,i = n [Ql(t?,.~-+ e,.,. + e;,)l - $(err + e!., + ez,)l}] n, (5.21

+62{~

where 8, and 82 are deformation potential constants, E the strain tensor, .n = k,~/lk~ 1 and kd is the wave vector at the valley bottom. Although these interactions are derived for a spatially homogeneous strain field, they can be extended to the case of nanocrystals as long as the wave vector dependence of deformation potentials is reasonably weak around the valence band top and the conduction valley bottom. Then the matrix element of the exciton-phonon interaction can be calculated by integrating Eqs. (5.1) and (5.2) multiplied by exciton envelope functions in nanocrystals. In the following calculation, the inter-valley electron-phonon interaction is not taken into account for simplicity. The acoustic phonon modes in Si nanocrystals can be classified as torsional modes and spheroidal modes as in Section 2 under assumptions of the stress-free boundary condition and isotropic elastic constants. The mode spectra for a Si nanocrystal with 20 A radius are shown in Fig. 7. The material constants employed are [37] i + 2,~ = cl1 = 16.772 x IO(dyn ~1= ~43 = 8.036 x lO(dyn cm-) cnP2),

T. Tukayuharu!

Journal

of Lwninrscwm

70 (1996)

129--/U

139

Si

F
=I

R-20;

t
2 Angular 4 Momentum 6 8

2 Angular

4 Momentum

I
with 20A

b)
radius

I
momentum for

Fig. 7. The acoustic phonon energies (a) torsional modes and (b) spheroidal

of a Si nanocrystal modes

are plotted

as a function

of the angular

and p = 2.329(g cmP3). (5.3)

The mode spectra are rather sparse as compared with those in Fig. 1 for a CdSe nanocrystal with the same radius. This is due to the larger elastic constants and the smaller mass density for Si compared with CdSe. The relevant exciton-phonon interaction Hamiltonian can be written in the form of Eq. (4.1). Then the luminescence Stokes shift A which is twice the lattice relaxation energy in Eq. (4.5) and the dimensionless coupling strength 3 which is a sum of the Huang-Rhys factors are given in the Franck-Condon approximation as
A = 2 C y;/hOj
i

and

i = C ;f/(tiOj)2,
J

(5.4)

where the summation is taken over relevant acoustic phonon modes. The above 2 is equal to S in Eq. (4.4) at T = 0 and will be called simply the Huang-Rhys factor. The contribution from the torsional modes turns out to be two or three orders of magnitude

smaller than that from the spheroidal modes. Through a similar analysis to that in Section 3, we see that A(j) is inversely proportional to R3(R2) in the small size regime. The maximum value of the Huang-Rhys factor within the manifold of the S3;2 exciton levels varies from 0.7 to 0.3 for the nanocrystal radius from 14 to 20 A, indicating the weak coupling regime of the exciton-acoustic phonon interaction. However, the luminescence Stokes shift is not negligible and is of the order of several meV in the above size range. The Stokes shift is dependent on the excitonic levels. In the estimation of the onset energy of photoluminescence, the photon-absorbing state is identified with the excitonic state in the S3:2 manifold having the largest oscillator strength, while the luminescent state is identified with that having the lowest energy when the Stokes shift is included. The onset energy estimated in this way is plotted in Fig. 8 by a solid line as a function of the nanocrystal radius (upper abscissa) or as a function of the exciton absorption energy (lower abscissa). The calculated excitonic exchange splitting [35] is plotted by a dashed line. Thus,

l2p

17

Radius(*)
l?

,,:r::-:---: __-I
o.

deformation potential coupling. For Si nanocrystals, the Huang-Rhys factor and the Stokes shift of photoluminescence due to acoustic phonon modes are calculated and their size dependencies are clarified. It is found that the observed onset energy of photoluminescence from Si nanocrystals can be interpreted mainly in terms of the excitonic exchange splitting, although the contribution from the luminescence Stokes shift is not negligible.

ooB __--_--

___O___---

__---

Appendix A: Strain tensor components orthogonal curvilinear coordinates

2.2 Energy (eV1 2.4

in general

Fig. 8. The theoretical excitonic exchange splitting and the result including the Stokes shift are plotted by a dashed line and a solid line, respectively, as a function of the excitonic transition energy. The experimental data of Ref. [19] are exhibited by circles. On the upper abscissa, the nanocrystal radius is shown corresponding to the exciton energy in the lower abscissa.

the exchange splitting is contributing to the onset energy more than the Stokes shift. The quantitative agreement between the experiment (open dots) and the theory (solid line) can be regarded as satisfactory, although the fluctuation of the exciton energies due to the actually present inhomogeneity in the nanocrystal size and shape is not taken into account. An important conclusion is that the observed onset energy of photoluminescence can be explained mainly in terms of the excitonic exchange splitting in contrast to the conclusion of Ref. [22], although the contribution from the Stokes shift is not negligible.

In this Appendix, we derive the strain tensor components in a general orthogonal curvilinear coordinates system and the transformation formulas for the strain tensor components between different coordinates systems. First of all, it is instructive to see the situation in the usual Cartesian coordinates. Let us consider two points denoted by P(x, Y, z) and Q(.x + Ir, Y + mr, z + nr), where Y is the distance PQ and 1, m and n are the direction cosines of the vector P& and assume that by an elastic deformation the point P moves to a new point P(x + u, y + v,z + IV). Then the point Q moves to a new point Q(X, r, Z) given by

X=x+lr+u+lr~+mr~+nrl:ll. (A.11 ay az
Y=.v+mr++h~+mr~+r2r~

(7y

SZ'

(A.21

Z=z+IZP+U+I,~+,r~+I?T(?W (A.3)
ax
dY
iiZ

and the distance PQ is calculated 6. Summary The electron-phonon interactions in semiconductor nanocrystals, especially concerning the acoustic phonon modes are derived and the size dependence of the coupling strength is clarified for typical coupling mechanisms. On the basis of these results, the commonly observed linearly temperature-dependent term of the excitonic dephasing rate and the proportionality of its magnitude to the inverse square of the nanocrystal size are attributed to the pure dephasing due to the

as

pe 41 + 12ex,m2eJ, +
+ mne,, + hwxz] with
exx
dU = -,

+ n2e_ +

/me,,.
(A.4)

ax

ex, =

au
v

ac
+ z

(A.51

and their cyclic permutations. Now we extend the argument to a general orthogonal curvilinear coordinates denoted by ([i, (2, &). In

T. Takayahara! Journal of Lumineswncr

70 (1996)

129-143

141

this system, the orthogonal e;=tt(g,J?$,Z$

unit vectors are defined as

(A.13)

(A. 14) with (A.15)

(A.61
and the distance P(~I,&,&) given by between two points +X2,53 denoted +&) by is and Q(<I +@I,&

(A.16) where the displacement vector ui at the point P can be identified with hiA;(i = 1,2,3) as long as Ai is very small. When specified to the spherical coordinates, namely ri=v, 52=0, &=cp, hr=l,

m = dh@%,)2

+ h;(6<2)2 + h:(653)2,

(A.7)

where, of course, &(i = 1,2,3) is assumed to be very small. Let us assume that the point P moves to a new point P( 5 I + d 1, (2 + AZ, (3 + 43 ). Accordingly, the point Q moves to a new point Q(X,,X2,Xj) given as

hI=r,
(A.17)

h3 = r sin 8,
we have 2% err = -,

SAj t&t2

dAj + (r36t3

&

(A.18) (A. 19)

(i = ~2~3).

(A.8) 1 8% +~+~cotH, -_- r sin 0 &p 8 !!!!

The distance PQ can be calculated by Eq. (A.7) using h,(i = 1,2,3) at the point P, namely

(A.20)

1 au, e,o=--+rr,

rdd

or ( r >

(A.21) (A.22)

(A.9) Retaining the terms up to the first order in A,(i =

e0, = 1 du, ++. &p

1,2,3), we have PQ g PQ [ 1 + /Tel 1 + lie22 + lie33

+ lll2ei2 + 1213823+ 13lle311,

eqr = ~

r sin8

(A.23)

(A. 10)

where li is the direction cosine of the vector PB in the system of (ei, e2, e3) at the point P and is defined by li = h&;/PQ. From the analogy to Eq. (A.4) we can define the strain tensor components in this curvilinear coordinates as (A.1 1) 1 c?al2 e22=---+x1+U3ah? ah h2 42 hihz 851

In order to derive the transformation formulas for the strain tensor components between different orthogonal curvilinear coordinates systems, we need only to note that the distance PQ in Eq. (A.lO) is the same for all the coordinates systems, namely l:eii + l&22 + lie33 + 11l2el2 + 1213e23+ 13lle31

= 1i2e{, + lk2ei2 + li2ei3 + l{lie{, + lolled, +l:l~e~,, (A.24)

h2hs X3

(A. 12)

where the primed quantities on the right-hand side are associated with a coordinates system different from

142

T Tukuyuharu;

Journal of Luminescence

70 (1996)

129-143

that on the left-hand side. When the relation between the direction cosines in two coordinates systems is substituted, we obtain the desired transformation formulas for the strain tensor components. For example, the transformation formulas between the Cartesian coordinates and the spherical coordinates are given as e1.Y sin 0 C0s2 cpe, + cos* 19 cos2 cpeoo+ sin2 pe,, = +i sin 2H cos2 qera - i cos (3 sin 2qeocp -+ sin 8 sin 2qe,,, (A.25)

are mutually orthogonal harmonic bases [25]. In the spherical coordinates, they are defined by

P/,(.Q) =(l>o,o)yl,(~), BdQ2) =

(J3.2)
_( -B ?;

d&l

( si; )
0, g>

Yh?l(~;2),
(B.3) Y/,(Q), (B.4)

eY). = sin 0 sin2 cpe, + ~0s 0 sin2 *eov + ~0s qe,, +i sin 28 sin2 (per0 + i cos 8 sin 2qeocp +i sin H sin 2qe,,, e, = cos2 eerr + sin 8eo,j - i sin 2Oe,,j, eX,, = sin* 0 sin 2qe, - sin 2qe,, + cos2 0 sin 2cpeoo (A.26) (A.27)

where Pi, is defined for all 1 (>O), whereas Bl, and Cl,,, are defined for 13 1. They are orthonormalized concerning the angular integral, namely

+ f sin 28 sin 2cpe,n (A.28)

+ cos 6, cos 2qeu, + sin f3 cos 2qe,,, e _VZ sin 20 sin (perr - sin 28 sin qefjfj = + cos 28 sin qero - sin 0 cos cpeo, + cos 8 cos cperVp, e,, = sin 20 cos 9erY - sin 28 cos qne(j() + cos 2H cos cpe,fj + sin 0 sin cpeoq - cos 0 sin (perV.

(A.29)

(A.30)

In terms of these angular harmonics, L, M and N fields can be written as

Appendix B: Vector spherical harmonics The vector Helmholtz [02 + @]Y(Y) = 0 equation given by (B.1) Nl,(kr) l(l+ 1) = ~ kr Z/(~)Ph(Q)

has three linearly independent solutions. They are usually denoted by Ll,(kr), M/,(/V) and N,,(kr) as given in Eq. (2.2), although the radial part can generally include spherical Hankel functions as well as spherical Bessel functions. These L,M and N fields are generally not orthogonal to each other. On the other hand, concerning the angular part, there

where z{(b)

is a radial function.

T Tukqahara/ Jaurnul of Luminescencr 70 (1996) 129-143

143
Collins, M.L.W. Thewalt and G.

References
[I] R.W. Collins, CC. Tsai, M. Hirose, F. Koch and L. Bnts, eds., Microcrystalline and Nanocrystalline Semiconductors, M. R. S. Symposium Proc., Vol. 358 (Materials Research Society, Pittsburgh, 1995). [2] L.E. Brus, Appl. Phys. A 53 (1991) 465; M.G. Bawendi, M.L. Steigerwald and L.E. Brus, Ann. Rev. Phys. Chem. 41 ( 1990) 477. [3] C. Flytzanis, F. Hache, M.C. Klein, D. Ricard and P. Roussignol, in: Progress in Optics, Vol. XXIX, ed. E. Wolf (Elsevier, Amsterdam, I99 I ) p, 32 I [4] A.P. Alivisatos, T.D. Harris, P.J. Carroll, M.L. Steigerwald and L.E. Brus, J. Chem. Phys. 90 (1989) 3463. [5] M.C. Klein, F. Hache, D. Ricard and C. Flytzanis, Phys. Rev. B 42 (1990) I 1123. [6] A. Tanaka, S. Onari and T. Arai, Phys. Rev. B 45 (1992) 6587. [7] T. Takagahara, Ultrafast Phenomena VI, eds. T. Yajima, K. Yoshihara, C.B. Harris and S. Shionoya (Springer, Berlin, 1988) p. 337. [8] S. Nomura and T. Kobayashi, Phys. Rev. B 45 (1992) 1305. [9] A. Tanaka, S. Onari and T. Arai, Phys. Rev. B 47 (1993) 1237. [IO] T. Itoh and M. Furumiya, J. Lumin. 48 and 49 (1991) 704. [l I] T. Wamura, Y. Masumoto and T. Kawamura, Appl. Phys. Lett. 59 (1991) 1758. [12] N. Peyghambarian, B. Fluegel, D. Hulin, A. Migus, M. Joffre, A. Antonetti, S.W. Koch and M. Lindberg, IEEE J. Quantum Electron. 25 (1989) 2516. [I31 P. Roussignol, D. Ricard, C. Flytzanis and N. Neuroth, Phys. Rev. Lett. 62 (1989) 312. [I41 M.G. Bawendi, W.L. Wilson, L. Rothberg, P.J. Carroll, T.M. Jedju, M.L. Steigerwald and L.E. Brus, Phys. Rev. Lett. 65 ( 1990) 1623. [I51 U. Woggon, S. Gaponenko, W. Langbein, A. Uhrig and C. Klingshirn, Phys. Rev. B 47 (1993) 3684. [I61 R.W. Schoenlein, D.M. Mittleman, J.J. Shiang, A.P. Alivisatos and C.V. Shank, Phys. Rev. Lett. 70 (1993) 1014. [I71 H. Shinojima, J. Yumoto, T. Takagahara and N. Uesugi, Quantum Electronics and Laser Science Conference (Baltimore, 1993), QWH26.

[I81 M.A.

Tischler,

R.T.

[19] [20] [2l] [22] [23] [24] [25] [26]

[27] [28] [29] [30] [31]

[32] [33] [34] [35] [36] [37]

Abstreiter, eds., Silicon-Based Optoelectronic Materials, M. R. S. Symposium Proc., Vol. 298 (The Materials Research Society, Pittsburgh, 1993). P.D.J. Calcott, K.J. Nash, L.T. Canham, M.J. Kane and D. Brumhead, J. Phys.: Condens. Matter 5 (1993) L9l. T. Suemoto, K. Tanaka, A. Nakajima and T. Itakura, Phys. Rev. Lett. 70 (1993) 3659. J.C. Merle, M. Capizzi, P. Fiorini and A. Frova, Phys. Rev. B I7 (1978) 4821. E. Martin, C. Delerue, G. Allan and M. Lannoo, Phys. Rev. B 50 (1994) 18258. A.E.H. Love, A Treatise on the Mathematical Theory of Elasticity (Dover, New York, 1944). H. Lamb, Proc. Math. Sot. London 13 (1882) 187. P.M. Morse and H. Feshbach, Methods of Theoretical Physics (McGraw-Hill, New York, 1953). 0. Madelung, M. Schulz and H. Weiss, eds., Physics of II-VI and I~VII Compounds, Vol. 17b of Landolt-Bornstein (Springer, Berlin, 1982). G.L. Bir and G.E. Pikus, Symmetry and Strain-Induced Effects in Semiconductors (Wiley, New York, 1974). W.G. Cady, Piezoelectricity (McGraw-Hill, New York, 1946). S. Nakajima, Y. Toyozawa and R. Abe, The Physics of Elementary Excitations (Springer, Berlin, 1980). T. Takagahara, Phys. Rev. B 47 (1993) 4569. D.M. Mittleman, R.W. Schoenlein, J.J. Shiang, V.L. Calvin, A.P. Alivisatos and C.V. Shank, Phys. Rev. B 49 (1994) 14435. T. Takagahara, Phys. Rev. Lett. 71 (1993) 3577. S.H. Tolbert, A.B. Herhold, C.S. Johnson and A.P. Alivisatos, Phys. Rev. Lett. 73 (1994) 3266. F. Hache, M.C. Klein, D. Ricard and C. Flytzanis, J. Opt. Sot. Am. B 8 (1991) 1802. T. Takagahara and K. Takeda, Phys. Rev. B 53 (1996) R4205. L.D. Laude, F.H. Pollak and M. Cardona, Phys. Rev. B 3 (1971) 2623. 0. Madelung, M. Schulz and H. Weiss, eds., Physics of Group IV Elements and IIl~V Compounds, Vol. 17a of LandoltBornstein (Springer, Berlin, 1982).