r Cas Series in Ato ee 6 . Berlin Hel berg etn cd iB.I. Shklovskii A.L. Efros Electronic Properties of Doped Semiconductors With 106 Figures Springer-Verlag Berlin Heidelberg New York Tokyo 1984Professor Dr. Boris J. Shklovskii Professor Dr. Alex L. Efros AE IOFFE Physico-Technical Institute, Academy of Sciences of the USSR, Politekhnicheskaja, Leningrad 194021, USSR Dianslater Dr. Serge Luyi Bell Laboratories, 600 Mountain Avenue, Murray Hill, NJ 07974, USA Series Editors: Professor Dr. Manuel Cardona Professor Dr. Peter Fulde Professor Dr, Hans-Soachim Queisser ‘Max-Planck-Institut ftir Festkérperforschung, Heisenbergstrasse 1 1D-7000 Stuttgart 80, Fed. Rep. of Germany Title of the original Russian edition: Elektronniye svoistva legirovannykht poluprovodnikoy © by "Nauka Publishing House, Moscow 1979 ISBN 3-540-12995-2 Springer-Verlag Berlin Heidelberg New York Tokyo ISBN 0-387-12995-2 Springer-Verlag New York Heidelberg Berlin Tokyo Libeary of Congress Cataloging in Publication Data, Shklovski, B. 1, (Boris lonovich), 1M. Electronic properties of doped semiconductors. Springerseriesin solid-state sciences 45), Translation of: Eiekironnye svoistva legicovannykh poluprovodnikov. Includes bibliographical references and index. 1. Doped semi- conductors, 2. Flectron-electron interactions. 3, Hopping conduction, 4, Materials at low temperatures. I. Efros, A-L. (Alex L-), 1938. IL Title, II, Series. QC6I.8.D6685513 1984 $37.622 84-5420 ‘This work is subject fo copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically tose of translation, reprinting, reuse of illustrations, broadcasting, reproduction by photocopying machine or similar means, and storage in data banks. Linder $4 ofthe German Copyright Law, where copies are made for other than private use, a fee is payable to “Verwertungsgeselischalt Wort”, Munich. © by Springer-Verlag Berlin Heidelberg 1984 Printed in Germany The use of registered names, trademarks, etc, in this publication does not imply, even in the absence of a specific statement, that such names are exerint from the relevant protective laws and regulations and therefore free for gencral use. ‘Offset printing: Beitz Offsetdruck, 6944 Hemsbach/Bergstr. Bookbinding: J, Schiffer OHG, 6718 Griinstad 2153/3130-543210Preface First-generation semiconductors could not be properly termed “doped” — they were simply very impure, Uncontrolled impurities hindered the discovery of physical laws, baffling researchers and evoking pessimism and derision in advocates of the burgeoning “pure” physical disciplines, The eventual banish- ment of the “dirt” heralded a new era in semiconductor physics, an era that had “purity” as its motto. It was this era that yielded the successes of the 1950s and brought about a new technology of “semiconductor electronics”. Experiments with pure crystals provided a powerful stimulus to the develop- ment of semiconductor theory. New methods and theories were developed and tested: the effective-mass method for complex bands, the theory of impurity states, and the theory of kinetic phenomena, These developments constitute what is now known as semiconductor phys- ics, In the last fifteen years, however, there has been a noticeable shift towards impure semiconductors — a shift which came about because it is precisely the impurities that are essential to a number of major semiconductor devices. Technology needs impure semiconductors, which unlike the first-generation items, are termed “doped” rather than “impure” to indicate that the impurity levels can now be controlled to a certain extent. New problems have arisen in the theory of the electronic states of doped semiconductors. They concern electrons located not in an ordered field of crystal atoms, but in the chaotic field of impurities, and the potential energy of the latter is by no means small. At low temperatures a doped semiconduc- tor crystal becomes a disordered system, which in its general characteristics resembles an amorphous system. This is true for both lightly and heavily doped semiconductors: the lighter the doping, the lower the temperature at which these characteristics are exhibited. The aim of this book is to present in logical fashion the theory of electron- ic states and conduction in doped semiconductors at low temperatures, that is, in the region where the properties of the electronic states differ most from those of Bloch waves. Depending on the doping, the electronic states of a semiconductor at zero temperature may be localized or delocalized, An important advance in the theory of disordered systems was the so-called Anderson theorem, which posits the existence of strictly localized states under certain conditions. A dis- cussion of this question (Chap. 2) begins our exposition of the theory of clec- tronic states, which differs from that for ideal crystals in that it must accountVI Preface for electron-electron interaction even at the lowest electron concentrations. To this end, a nonlinear screening theory was developed, based on the self- consistent field method (Sect. 3.4), This method does not, however, work in the vicinity of the Fermi level, where the density of states has interesting and peculiar features (Chap. 10). If the Fermi level is in the localized-state region, then conduction is due to electron hopping and is exponentially dependent on temperature and the im- purity concentration. The hopping conduction phenomenon was identified Jong ago, but several major advances have taken place in the last decade. A theory was developed which describes the temperature, concentration and magnetic field dependences quantitatively. This theory is based on a new mathematical discipline known as “percolation theory”. Today, the percola- tion method is as essential to the study of low-temperature conduction as the Kinetic equation method is for band conduction; the term “percolation level” is as frequent in the relevant literature as “relaxation time”. Good reviews of the percolation theory do exist — we cite many in Chap, 5 — but these were written relatively long ago and are inappropriate to the study of hopping conduction. For this reason we found it necessary to write a separate chapter detailing the main tenets of percolation theory (Chap. 5), replete with bibliog- raphy on the topic. In all chapters devoted to hopping conduction there is a thorough compa- rison of theory and experiment, a comparison that we find, on the whole, favorable. We have tried to point out discrepancies and theoretical problems that in our view remain unsolved. Although the book is devoted to crystallic semiconductors, many of the ideas and methods also apply to amorphous semiconductors, so much so that “amorphous digressions” are an integral part of the text. Occasionally (see Chap. 9) we use experimental data on amorphous semiconductors to support certain concepts. Our book is not intended solely as a specialists’ monograph, but also as an extension of an ordinary course in semiconductor theory that touches on a new range of problems. Chapter 1 and Sects. 4.1 and 11.1 serve to connect this book with standard courses in the theory of “pure” semiconductors. The book is aimed at a wide readership: theoretical and experimental phy graduate students, and engineers acquainted with the basics of solid-state physics. An easier version of our book can be obtained by omitting Sects. 1.3, 2.3-2.5, 5.4, 5.5, 8.3, 10.1, 10.2, 11.3, 12.3, 13.4. It is useful to keep in mind that as a rule all questions are discussed twice, first qualitatively and then quantitatively. For the reader not interested in mathematical detail the quali- tative explanation should suffice; sections which may be omitted are usually designated as such in the text. We substantially updated the book for the English edition, adding new re- sults in percolation theory and hopping conduction. During the five years since we had written our book, several significant developments had occurred in the physics of disordered systems. One of these was the creation in 1979 ofPreface vu the scaling theory of localization by Anderson with coworkers and Thouless. We have incorporated this theory, as we understand it, into Chapter 2 of the present edition, New ideas have also emerged in the understanding of electron-electron in- teraction in disordered systems. In 1975 we proposed the idea that a Coulomb gap may form in the vicinity of the Fermi level, which if correct would make it necessary to revise Mott’s law for variable-range hopping conduction. In the Russian edition we ‘idly devoted only one section to this question (in Chapter 10), but since then, a number of authors have made both theoretical and experimental contributions to this subject, and we felt compelled to write a whole new chapter for the present edition (Chapter 14). It describes com- puter modelling of the Coulomb gap, the impurity-band structure, and hop- ping conduction. In 1979-1982 Altshuler and Aronov published a series of remarkable papers devoted to the role of electron-electron interaction in disordered sys- tems with delocalized states. We were not able to consider these concepts in detail — otherwise we would have had to write a new book. One reason not to do so was furnished by APishuler and Aronov themselves, who discuss these topics in a chapter of the volume entitled Evectron-Electron Interaction in Dis- ordered Systems ed, by A, L. Efros and M. Pollak (North-Holland, Amster- dam 1984). Other contributions to that volume also substantially complement our book, It is a pleasure to us that Springer-Verlag has undertaken to publish the English translation of our book. We thank Professor Mike Pollak, our long- time friend, and Dr. Serge Luryi for their hard and selfless labor in preparing this edition. We are grateful to our foreign colleagues B. N. Butcher, J. Chroboczek, E. Guyon, J. J. Hauser, S. Kirkpatrick, Y. Imry, P. A. Lee, R. Mansfield, N. F. Mott, G. A. Thomas, H. B. Stanley, J. P. Straley, and many others who regularly send us reprints of their works, thus keeping us abreast with the latest developments. Leningrad, December 1983 B, I, Shklovskii - A. L. EfrosContents Part1 Lightly Doped Semiconductors » > w . The Structure of Isolated Impurity States . . Localization of Electronic States . . . Percolation Theory 1.1 Shallow Impurities .. 1.2 Impurity Levels Near a Nondegenerate Band 1.2.1 Extremum at the Center of the Brillouin Zone . 1.2.2 Several Equivalent Extrema .. 1.3 Impurity Levels Near a Point of Band Degeneracy . vee 1.4 Asymptotic Behavior of the Impurity-State Wave Functions... 2.1 Narrow Bands and the Mott Transition . 2.2 Anderson Transition . 2.3 Examples of Transition from Localized to Delocalized States. Conductivity Near the Transition Point . 2.4 The Scaling Theory of Localization . 2.5 Localization in the Lifshitz Model The Structure of the Imp Doped Semiconductors . 3.1 General Remarks . 3.2 The Impurity Band at Low Degrees of Compensatior 3.3 Long-Range Potential at Low Degrees of Compensation . 3.4 The Impurity Band at High Degrees of Compensation . ity Band for Lightly A General Description of Hopping Conduction in Lightly Doped Semiconductors ... 4,1 Basic Experimental Facts 4,2. The Resistor Network Model ‘roposed by Miller and Abrahams 5.1 Lattice Problems .... 5.2 Continuum Problems . 5.3 Random Site Problems Shouanun 141x Contents 5.4 Theory of Critical Exponents .... + 5.5 Electric Conductivity of Random Networks of Conducting Bonds. Infinite Cluster Topology . . $.5.1 Dead Ends ..........05 5.5.2 The Nodes and Links Model . $.5.3 The Scaling Hypothesis and Calculation of the Conductivity a(x) . 5.6 Percolation Theory and the Electti Conductivity of Strongly Inhomogeneous Media ..........0602..000 05 . 6. Dependence of Hopping Conduction on the Impurity Concentration and Strain in the Crystal . 6A Resistivity 3 for Semiconductors with Isotropic Impurity | Wave Functions . 6.2 Resistivity p; for Semiconductors with Anisotropic Impurity Wave Functions .......seeeee ees bene e eee ener n nes 7. Hopping Conduction in a Magnetic Field 7A The Elementary Resistance Rin a Magnetic Field 7.2 Evaluation of the Magnetoresistance and Discussion of Experimental Data ... 8. Activation Energy for Hopping Conduction 8.1 Activation Energy e; at Low Degree of Compensation . 8.2. Activation Energies e, and ¢, at High Compensation 8.3 The Perturbation Method in Percolation Theory. General Theory of the Activation Energy & .-. 9. Variable-Range Hopping Conduction ... 9.1 Mott's Law... 9.2 Magnetoresistance in the Region of Vatiable- Hopping Conduction 9.3 The Dependence of Hopping Conduction in Amorphous Films on Film Thickness 9.4 The Preexponential Factor in Hopping Conductivity . 10. Correlation Effects on the Density of States and Hopping Conduction .. 10.1 Coulomb Gap in the Density of States 10.1.1 The Theory of the Coulomb Gap, 10.1.2 Possible Manifestations of the Coulomb Gap 10.2 Many-Body Correlations in Hopping Conduction . 10.2.1 Hubbard Current Correlations . 10.2.2 Coulomb Correlations 118 123 125 126 129 130 137 137 144 155 156 163 180 180 186 191 202 202 210 216 222 228 228 228 238 244 246 250Contents XI Part II Heavily Doped Semiconductors 11. Electronic States in Heavily Doped Semiconductors 255 11.1 Linear Screening Theory . 1 255 11.2 Density of States Near the Bottom of the Conduction Band 259 11.3 Derivation of the Quasiclassical Formula for the Density of States 264 12. The Density-of-States Tail and Interband Light Absorption ...... 268 12.1 The Optimum Fluctuation Method . 268 12.2 The Uniformly Charged Sphere ‘Approximation. ‘The Spectrum of the Majority Carriers 270 12.3 Exact Distribution of Impurities in Optimum. Fluctuations . 277 12.3.1. Derivation of Basic Equations 277 12.3.2 Derivation of (12.2.12) .. 279 12.3.3 Density-of-States Asymptotics in the Absence of Impurity Correlations . 281 12.4 The Spectrum of the Minority Carriers 283 12.4.1 Classical Case (y> E,) . 284 12.4.2 Quantum Case (y a7. For the typical semiconductors, a7! is of order 10°? of the reciprocal lattice constant, which justifies our approximations. The full wave function of an impurity center is determined by (1.2.6) in which we can let By: = d’B,. Hence we have VO) = DB, Wea , £ By, (k) = (1.2.19) - WF DA Wanged™ (1.2.20) Neglecting, as above, the dependence of x, on k we find 1 vO = Wye uno) F(x) . (1.2.21) o(t) DB, (k)e** K Thus, the wave function of an impurity state is a Bloch wave function at the bottom of the band, modulated by a large-scale hydrogenlike function. Let us conclude our consideration of the case of a single extremum at the zone center by presenting Table 1.1 [1.2], in which the theoretical and experimental values of donor levels are compared. Discrepancies are due to the fact that the effective-mass method poorly describes the potential at short distances from the center. Table 1.1, Dielectric permittivities, effective masses, and theoretical and experimental values of the ground-state energy for selected semiconductors nn Material =x _ Ey (theor) Ey, (expt) 0 [meV] [mev} GaAs 12.5 0.066 5.67 Ge: 6.1; Si: 5.8 Se: 5.9; S:6.1 Sn: 5.9 InP 12.6 0.08 6.8 7.28 CdTe 10 O1 13 13.88 1, The Structure of Isolated Impurity States 1.2.2. Several Equivalent Extrema The conduction band of such semiconductors as germanium and silicon possesses several extrema located at equivalent points of the Brillouin zone. The conduction band minimum in silicon is located at k° = 0.85 (2x/aq) in the [100] direction, ao being the lattice constant. The constant energy surfaces represent ellipsoids of revolution around [100]. The cubic symmetry of the crystal requires the existence of identical ellipsoids in each equivalent direction. Thus, there must be six equivalent minima in silicon. Choosing the z axis along [100], we find the following expression for the energy near a 1100] extremum: ae yr ag a E = > (k,-kp)? + <— (k? + ky) (1.2.22) 2my Im, The longitudinal (m,) and the transverse (m,) effective masses in silicon have. the following values (1.2): me = 0.916 mo and m, = 0.190 sq, where ig is the free-electron mass. The extrema of the conduction band of germanium lie exactly on the Brillouin zone boundary in the direction of the cube's body diagonals. Bach extremum is therefore shared by two zones, so that the number of extrema equals half the number of the equivalent directions, i.c., equals 4. The constant energy surfaces in k space are ellipsoids of revolution with their axes along the body diagonals. The effective masses are [1.2]; my = 1.58 mo and mh = 0.08 mo, The effective-mass method can easily be generalized to the case where several extrema are present. In complete analogy with (1.2.21) the solution of (1.2.5) near an extremum at k = k; is of the form x = F/G, {0) , (1.2.23) where @; is a Bloch function with k = ky normalized to the unit cell, and F is a smooth envelope function satisfying the equation faveuw - Elnw -ERW, 2.24) where the function E,(k) is the conduction-band energy in the vicinity of kj. ‘The energy eigenvalues # are, of course, the same for all equivalent extrema. Because of this, the impurity levels in the effective-mass approximation have an additional N-fold degeneracy (N being the number of equivalent extrema), In germanium and silicon this degeneracy is partially lifted because of corrections to the effective-mass method. These corrections are due to the fact that for small r the potential cannot be described by an expression of the form (1.1.1) since the macroscopic dielectric constant is meaningless at short distances, One has to introduce an additional short- range potential whose matrix elements between two plane waves do not decay1.2 Impurity Levels Near a Nondegenerate Band, 9 with & as in (1.2.12) but remain finite at large k. These matrix elements connect wave functions belonging to different extrema. This leads to a splitting of the impurity states which were degenerate within the effective- mass approximation, as well as to a shift of energy levels which is usually called the chemical shift. The form of such splitting can be predicted on the basis of symmetry alone, without a detailed knowledge of the short-range potential. We shall discuss only the ground-state splitting. The correct wave functions are linear combinations of the form x 09 Dox, PLN. (1.2.25) i The form of the splitting, as well as the coefficients af? for germanium and silicon, are determined from the symmetry properties with respect to transformations of the tetrahedral group Ty. It turns out that in silicon the sixfold degenerate ground state is split into one singlet, one doublet, and one triplet. The lowest energy state is the singlet. Its wave function is of the form 6 1 Pex (1.2.26) A v6 In germanium the fourfold degenerate ground state is split into one singlet and one triplet, The ground state is singlet, and its wave function is of the form o aed = Diy (1.2.27) dl 2 In the effective-mass approximation, the energy of an impurity state is determined by solving (1.2.24), Substituting into this equation the ellipsoidal spectrum (1.2.22) we find (1.2.28) This equation contains one dimensionless parameter, which is usually chosen to be the ratio of the transverse and longitudinal masses: 7 = m,/me. Let us recast (1.2.28) in dimensionless form, using atomic units corresponding to the transverse mass, viz. r'= r/a,, E'= E/E, a, = Refe?m,, and E, = #/2m,a2. We obtain 2, # Sot fy az? * ox 2 tee Fe ay? or (1.2.29) This equation can be solved exactly in two limiting cases: y= 1 and y = 0. The former case corresponds to the ordinary hydrogen atom problem, the10 1, The Structure of fsolated Impurity States latter to the two-dimensional Coulomb problem. Naturally, one would like to know the ground-state energy as a function of the parameter y in (1.2.23). This problem was solved by Kohn and Luttinger [1.4] by a variational method which proved to be very successful. The trial function was chosen in the form 2.4.9,2) 2 F = Gra? a Mevol- — +S | : (1.2.30) 4 ail and the two parameters a4 and ay were found by minimizing the energy. For 77 = | the variational procedure yields the exact result: a, = ay, and the wave function is identical to that of the hydrogen atom ground state. A less trivial fact is that even for -y = 0 the variational energy differs from the exact one by less than 8%, The dependence E'(y) obtained by the variational method is shown in Fig. 1.2. Using the above values for the effective masses and the dielectric constants of 11.40 for silicon and 15.36 for germanium, we find that the effective-mass approximation of the ground-state energy is 31.27 meV for silicon and 9.81 meV for germanium [1.2] oe 4 Exact value 3 2 1 si ~~ Exact Fig. 1.2. Ground-state energy in value —| units of E, calculated with the trial function (1.2.30) [1-4] and 0 ae oF rm 78 7a yt plotted as a function of y!/? The excited-state energies are calculated in a similar way. Equation (1.2.29) is axially symmetric and hence the good quantum numbers are the magnetic quantum number m and the parity. Levels corresponding to the same magnitude of m but different signs are degenerate. It is customary to label the excited states of (1.2.29) by the corresponding hydrogenic states at y= 1. However, for arbitrary -y one has additional level splittings. For example, the 2p state of hydrogen is replaced by one state 2p9 with m =0 and two degenerate states 2p. with m = +1. We conclude this section by presenting Table 1.2 [1.2], which shows both theoretical and experimental values of energy for the ground and excited states in silicon and germanium. As is clear from the table, the effective-mass description of the ground state is far from ideal. The main error is introduced1.3 Impurity Levels Near a Point of Band Degeneracy u ‘Table 1.2, Donor ionization energies in Ge and Si. Theoretical values were calculated in the effec tive-mass approximation. Experimental values correspond to the impurities indicated in paren- theses, Energies of all s-state levels, split due to the chemical shift, are listed separately Material jy, meV] Egy (meV] Si (theor) 31.27 ist Si (P) 45.5 33.9 32.6 11.45 Si (As) 53.7 326 © 31.2 11.49 Si Gb) 427 32.9 30.6 11.52 Ge (theor) 9.81 44 Ge (P) 129 9.9 475 Ge (As) 1417 100. 475 Ge (Sb) 10.32 10.0 474 by an incorrect treatment of the short-range part of the potential, As discussed above, including this correction gives rise to a splitting of the ground level and a substantial shift which increases the ground-state binding energy. The effective-mass method works much better for p states. The reason is that wave functions of these states vanish at the origin, so that their energy is not sensitive to the exact form of the potential near the impurity center, 1,3 Impurity Levels Near a Point of Band Degeneracy Let us suppose that eigenstates of the Hamiltonian (1.2.1) are degenerate at a certain point in k space, ‘This means that there exist t > 1 Bloch functions 2 satisfying the equation Hobha = En (Weix. f= 12, a3 In what follows we shall consider the case when the degenerate point coincides with the center of the Brillouin zone, k = 0, Such is the case for the valence band of typical semiconductors. This can be simply viewed as a consequence of splitting the atomic p orbital. At k = 0 the band is sixfold degenerate, spin included. For finite & the degeneracy is partially lifted (Fig. 1.3). ‘The effective-mass method is casily generalized to include this case [1.5]. It is natural to seek a solution to the Schrédinger equation in the form of a superposition of Bloch functions with their wave vector corresponding to the degeneracy point: * v= > Fy Md)0 W. (1.3.2) fel ‘The coefficients F,(r) are slowly varying functions of the coordinates and they2 1, The Structure of Isolated Impurity States Fig, 1.38, b, Band diagram of typicat semiconductors. (a) Spin-orbit cou- feife pling is absent and the valence band is sixfold degenerate at k=0. (b) Spin-orbit coupling splits a subband of total angular momentum J = 1/2 Pye fala a) » satisfy an equation analogous to (1.2.15), viz. [3 LD) D AP bbs t+ UMS, | Fp = BF; F [eo b=—iav. (1.3.3) The difference is that here the envelope function constitutes a ¢-component column vector, so that the Hf? form a matrix not only with respect to the Cartesian indices ,6 but also with respect to the ¢ components of Fj. This result provides a quite natural generalization of the effective-mass method discussed in the preceding section. For a detailed derivation of (1.3.3) the reader is advised to consult the original paper by Luttinger and Kohn [1.5]. In that work the matrix H is expressed in terms of the matrix elements of momentum, taken between Bloch functions, Another possible approach to the problem is simply to write down the matrix H from symmetry considerations. This approach is chosen below. So far we have not been taking into account effects due to spin-orbit coupling. To include these effects one must add to the ideal-lattice electron Hamiltonian Ha term of the form 1 - Hg me lex VVIp, (1.3.4) which represents a relativistic correction, as follows from the Dirac equation. Here o is the spin operator (Pauli matrices), V(r) is the potential energy of an electron in a periodic field which appears in (1.2.1), and ¢ is the speed of light, The valence-band structure is pronouncedly affected by the spin-orbit interaction. In the presence of such interaction the spin and the orbital angular momenta are no longer conserved individually, the conserved quantity being only the total angular momentum. Consider the sixfold degenerate p state of an atom. Each of the three coordinate wave functions is multiplied by a two-component spinor. Under the spin-orbit interaction these six states1,3 Impurity Levels Near a Point of Band Degeneracy 3 break into two groups of different energy. One group contains four states and corresponds to the total momentum 3/2, while the other contains two states of momentum 1/2. A similar splitting occurs in the crystal. When spin-orbit coupling is taken into account, the valence band, originally sixfold degenerate at k = 0, splits into two bands, which are fourfold and twofold degenerate (Fig. 1.3b). The twofold degenerate band is shifted down by the amount A, which is the energy of the spin-orbit interaction (values of A for Ge and Si are given in Table 1.3) ‘Table 1.3. Parameters of the Luttinger Hamiltonian, spin-orbit spliting energy A, and dielectric constanis « for Si and Ge [1.2] Material v1 v2 Ya A K Si 4.22 0.39 144 004 114 Ge 13.35 4.25 $69 0.29 15.4 ‘The general form of the matrix H in the presence of spin-orbit coupling is rather complicated and we shall not write it down. Considerable simplification is achieved in two limiting cases. The first case corresponds to a weak interaction when the spin and coordinate variables are decoupled and the matrix H“®#,.Bq is, in fact, a 3 x 3 matrix. The second case is when the interaction is strong, and the band of total momentum 1/2 is far removed in energy, so that its influence can be ignored in the vicinity of the valence-band edge. In this case only the four top bands need be considered in (1.3.2), and the matrix Hf,py reduces toa 4 x 4 matrix. In relation to the structure of an impurity state, we can use the first approximation when the spin-orbit splitting A is small compared to the impurity ionization energy, and the second approximation in the opposite case. It will be seen below that the strong spin-orbit coupling approximation is valid for germanium and for a number of other semiconductors. Let us begin with this approximation. We have to construct the matrix H°f,f, in the neighborhood of a fourfold degenerate band. The Hamiltonian must satisfy the requirements of crystal symmetry under rotations of the coordinate system. The wave function is represented by a column of four coordinate functions. Quantum- mechanically this implies that an equation for a spin-3/2 pasticle is needed In order to write such an equation, one usually introduces a pseudovector J which represents the spin momentum operator. Its components are fourth- rank matrices whose meaning is analogous to the Pauli matrices for spin-1/2. An explicit form of the J matrices can be found in [Ref. 1.3, p. 335].4 1. The Structure of Isolated Impurity States There exist only two invariants which are quadratic in p, namely ? and (@-D?, under arbitrary rotations of the coordinate axis. A rotationally invariant Hamiltonian would then contain only two independent parameters, and the matrix H could be written in the form -([é@ ae nt By =1:9°]. (3.8) However, cubic symmetry is less restrictive than full spherical symmetry. In @ cubic crystal there is one more invariant which is quadratic in 6, namely p2 J} + By J? + Be J2. Therefore the full Hamiltonian can be written in the form et 34] 2 y@ent a= [[: +7 »| 77 BG D + (yoy) (G2 J? + BP Ip + 2 | (1.3.6) This Hamiltonian is named after Luttinger, who was the first to derive it (1.6), The parameters y, -¥, and y3 are well known for many semiconductors, Their values for Ge and Si are given in Table 1.3. It should be noted that (1.3.6) represents the complete Hamiltonian only for crystal lattices possessing a center of inversion. For crystals without inversion symmetry, the Hamiltonian contains not only quadratic but also linear terms in p. For example, in crystals with a zine-blende lattice (InSb, GaAs, etc.) the Hamiltonian may include the following term: By Ux UG} — JD) + By Uy, U2 - JD) + BA, U2 - ID), 3.7) where the braces stand for an anticommutator, (a,b) = ab + ba. However, the linear terms are very small in all extensively studied semiconductors, and can usually be neglected. For this reason we shall not consider them further. We have obtained a system of four second-order equations (1.3.3) which determine the energy and the wave functions of an impurity center. Before embarking on its solution let us consider the system (1.3,3) for free particles by setting U(r) = 0. The Hamiltonian (1.3.6) describes motion of a spin-3/2 particle. Its first term represents the kinetic energy operator, and the remaining terms describe a peculiar spin-orbit interaction. Inasmuch as the Hamiltonian does not explicitly depend on the coordinates, one should seck a solution to the Schrédinger equation in the form: Fy= A; ike Tid er (1.3.8) where the A, are components of a four-dimensional eigenvector of the matrix