146 Supplement of the Progress of Theoretical Physics, No. 57, 1975 Metal-Nonmetal Transition in Doped Semiconductors Kazuo MORIGAKI and Fumiko YONEZAWA * Institute for Solz"d State Physics, University of Tokyo Roppongi, Tokyo 106 *Department of Physics, The City College of CUNY New York, N. Y. 1003Jtl (Received January 21, 1975) The metal-nonmetal (M·NM) transition observed in some doped semiconductors is discussed on the basis of the Matt-Hubbard-Anderson scheme. A classification of characteristic donor (acceptor) concentration regions is given. The M-NM transition concentration is defined as a concentration at which the states corresponding to the Fermi level become delocalized. Several experimental data near the transition concentration are analyzed and interpreted as lending support to the above-mentioned scheme. A possible picture concerning the position of the Fermi level relative to the mobility gap is proposed and some experimental results are described, which are explained in consistent with this picture. § 1. Introduction In some non-crystalline systems in which atomic configurations are disordered in some way or other, the metal-nonmetal (M-NM) transitions are observed when appropriate physical parameters such as densities, impurity concentrations, electric fields, temperatures, pressure are changed. Doped semiconductors which we shall discuss here are one of the most typical and best studied examples of those disordered systems that show the M-NM transitions. Other examples are metal-ammonia solutions, 1> metal-rare gas mixed solids, 2> supercritical metallic fluids, 3) high density excitons, 4 > etc. To a certain extent, electrons in these materials are treated on more or less the same ground in view of the essential feature, common to most of these disordered systems, that electrons are interacting with one another as well as with fluctuating potential due to randomly distributed atoms. So far, two major theoretical schemes have been put forward to explain the M-NM transitions and the related properties of electrons in these systems; that is, the Mott- HubbardAnderson scheme 5)-7) and the percolation theory for inhomogeneous conduction regions. 8)' 9) t> On leave from Department of Applied Physics, Tokyo Institute of Technology,. Meguro-ku, Tokyo 152. Metal-Nonmetal Transition in Doped Semiconductors 147 The purpose of this short article is to study electronic properties of doped semiconductors near the M-NM transition concentration Nc along the line of the Matt-Hubbard-Anderson scheme. Because of the limited space allowed for us, we specifically confine ourselves to the donor concentration region N D;SNc of n-type doped semiconductors without compensation although most of the arguments given here are, with appropriate modifications, equally applied top-type doped semiconductors and to the case where samples are compensated. It must be noted, however, that in actual samples a certain amount of compensation exists inevitably even when the purity and the doping of the impurity material are highly controlled, and this small amount of compensation enables the hopping conduction to take place in the low concentration region. In §2, we classify the donor concentration regions according to the behaviour of the states which provide theoretical basis for the Matt-Hubbard picture in disordered systems. In §3, experimental results in the concentration region below Nc, i.e., N D;SNc are studied in order to see how they are explained within the framework of the Mott-Hubbard-Ander.son scheme. In § 4, we suggest a possible position of the Fermi level with respect to the mobility gap near the transition concentration and some concluding remarks will be given in §5. §2. Characteristic concentration regions and M-NM transition As is clearly shown in the famous and almost classical paper by Fritzsche,lO) the donor concentration regions of doped semiconductors are classified into three major characteristic categories according to the types of impurity conduction; i.e., the low concentration region where the conduction is due to the hopping between donor centers; the intermediate concentration region where the conduction is observed mainly of an activation type with an activation energy which is termed e2; and the high concentration region where the conduction is nearly metallic. Among several possible mechanisms which may be responsible for the activation energy e2, the most probable candidate is the activation across the energy gap between the donor band and the D- band, the latter of which is the band grown from the energy level of an electron added to an isolated neutral donor atom. This idea of the D- band1°) is essentially the same as that of electron correlations proposed for an explanation of the so-called Mott transition, 11 > formulated later by Hubbard 6> for a simplified model system. These two bands are now generally referred to as the lower and upper Hubbard bands. Besides the effect of electron correlations, there is another mechanism, possible only for disordered systems, which is expected to hinder electrons from moving around freely. This is generally known as Anderson's localization-delocalization mechanism 7> due to potential fluctuation. In the Matt-Hubbard-Anderson scheme, these two mechanisms of electron 148 K. Morigaki and F. Yonezawa correlations and Anderson's localization are regarded to be the most essential cause for the M-NM transitions in disordered systems in general, although detailed interplays of these mechanisms are of course different for different disordered materials. In doped semiconductors, the dominant factor is considered to be the electron correlations while the actual transition is explained to be the Anderson transition.5),1 2) Both for theoretical and experimental analysis' sake, it is convenient to subclassify the intermediate and high concentration regions as follows: (i) Intermediate concentratz"on region I; in which there exists a real gap between the upper and lower Hubbard bands as shown in Fig. 1 (a). The shadowed regions in the figures denote that the states in these energy regions are localized in Anderson's sense. (ii) Intermediate concentration region II; in which there is no longer a real energy gap, but a mobility gap still remains (see Fig. 1 (b)). In both cases (i) and (ii), s2 is defined as IEF- Ec I, where Ec is the upper edge of the mobility gap. (iii) High concentratz"on region I; in which the states at the Fermi level is delocalized but still in such an energy region as is characteristic of the impurity bands (see Fig. 1 (c)). (iv) High concentration region II; where the tail from the conduction band covers well over the energy region to which the Fermi level belongs (see Fig. 1 (d)). (a) (b) (c) (d) E, Fig. 1. Schematic diagram of the density of states for the impurity band and the conduction band versus energy. Figures (a)-(d) approximately correspond to the situations for the concentration regions (i) -(iv) respectively. Metal-Nonmetal Transition z'n .Doped Semiconductors 149 The predicted behaviour of the density of states as shown in Fig. l is theoretically confirmed by calculating the spectra for these systems in the "so-called first principle" manner. Our numerical results are shown in Figs. 2 to 4_13)-15) We have used the Hubbard model for electron correlations. The spectra in Fig. 2 are the conduction band and the two Hubbard bands relevant to the impurity conduction of a system in which host atoms (A) on regular lattice sites are substituted for by donor atoms (B). They have originally been calculated for a disordered binary system Al_zBx (x=0.4), 13> in which the effect of disorder is treated by the coherent potential approximation.16) In actual doped semiconductors, the effective Bohr radius of donor electrons is large compared with the atomic spacing of the host lattice, so that one should regard the distribution of donor atoms to be structurally disordered. However, in an approximate sense, the variation of the spectra with the donor concentration (or the average distance between donors) can be seen in Fig. 2 by changing the value of Ll/ U, where Ll and U denote the bandwidth and the repulsive Coulomb interaction of two electrons with opposite spins on the same site respectively. The large value of Ll/ U corresponds to the situation of the higher donor concentration. Figure 3 has been obtained by regarding the distribution of donor atoms to be structurally disordered, but not taking into account the conduction band. A single-site theory1 7) proposed for this kind of disorder has been used. In view of the fact that the effective Bohr radius of a donor atom in doped semiconductors is comparable to the mean distance between donor atoms, this assumption of totally disordered atomic distribution E/U Fig. 2. Calculated density of states of the two Hubbard bands and tbe conduction band for X= 0.4 and various values of LJ / U. E A and E B designate the energies of tbe neutral donor level and tbe center of the gravity of the conduction band respectively. Small triangles in the figures indicate the position of the Fermi level. (After Ref. 13).) 150 K. Morigaki and F. Yonezawa ~ 0~-A~llW~L-~~~~J--llllW~--~ ,:: 0 0 EF u=0.3 0 EF Fig. 3. Calculated density of states of two Hubbard bands for various values of u which denotes U in units of the binding energy of donor electrons. The shaded r.egion designates the region with a localized character. (After Ref. 14).) ll./U=0.4 Un(E) 0.6 .6 E/U E/U Fig. 4. Calculated density of states of two Hubbard bands with each different bandwidths. The model corresponds to the case where the width of the .D- band is three times larger than that of the donor band. (After Ref. 15).) is considered to be almost adequate. Figure 4 shows the result of the calculation in which the original Hubbard model is extended so as to take into account the difference of the band widths of the donor band (the lower Hubbard band) and the D- band (the upper Hubbard band). Since the effective Bohr radius Metal-Nonmetal Transi#on in Doped Semiconductors 151 for the D- band is nearly 2 to 3 times larger than that for the donor band, a proper care of this difference must be taken so that the theoretical discussion can be more reliable. A characteristic feature is an asymmetry of the two Hubbard bands, the explanation of which we intend to give in a forthcoming paper. § 3. Experimental evidence for the Mott-HubbardJlnderson scheme near ]Vc If the behaviour of the density of states accompanying the change of the donor densities is as described by Fig. 1, experimental data are expected to give some indications for the existence of the mobility gap at N D~Nc. Let us see, in the following, whether this is the case. 3.1. Existence of the gap According to our prediction that the Hubbard bands (upper and lower) are separated from each other either by a real gap in the intermediate concentration region I or by a mobility gap in the intermediate concentration region II, the existence of the gap must be observed. The experimental fact for the activation type conduction through e2 in this concentration region is nothing but the most direct and certain evidence for the gap. Besides this almost complete evidence, we can give a few other experimental results indicating the existence of two kinds of electron systems, which are compatible with the existence of the gap. That is, for a finite temperature, electrons are partially excited from the lower Hubbard band to the upper Hubbard band across the gap. Since the states in the lower Hubbard band have a more localized nature than the states in the upper Hubbard band, we have consequently two kinds of electrons, one being conductive and the other being rather localized. In the intermediate concentration regions I and II, ESR enhanced conductivity 1 s>~ 21 > is measured as shown in Fig. 5, which is explained as follows :1 9),*) If we assume to have a donor spin system which is responsible for ESR and a system of mobile electrons which contributes to the conduction, energy from the microwave applied to the system is first absorbed by localized spins at resonance and thereafter transferred to mobile electrons via interaction between these two kinds of electrons, which brings about the increase of the kinetic energy of mobile electrons and accordingly the increase of the mobility. Another experimental support for the two kinds of electrons is given by the Raman spectra 2 3) for Ge: As on both the sides of Nc. The *> Recently, Mott and Kamimura5),12),22l have suggested that the result of ESR enhanced conductivity provides an evidence for the existence of the mobility gap (the pseudogap) in which the variable range hopping conduction may be enhanced by absorption of microwave energy from excited spins at resonance. 152 K. Morigaki and F. Y onezawa J f 0 I '\ :~'~! 1 -12dB \ lQ-4 ~ \ lQ-5 1:- I lQ-6 1Xl0 18 No (cm- 3 ) Fig. 5. Resistivity change due to ESR plotted as a function of ND at 1.5 K and at the microwave power level of -12 dB for Si: P. (After Ref. 21).) spectra are analyzed to be the superposition of the Raman scattering line related to the valley-orbit splitting and the background spectra of singleparticle excitation. The result can be understood by regarding the former due to localized electrons and the latter due to conductive electrons. 3.2. The mobility gap If the transition is of Anderson's type, the gap observed on the nonmetallic side of Nc must be the mobility gap rather than the real density of states gap. This is guaranteed if we have a nonzero density of states at the Fermi level before the transition; i.e., n(EF)~O for Nn-:5Nc. One evidence for this is the fact that a variable range hopping 24> is observed in this region for uncompensated Si 25 > and Ge,26) and also for compensated Ge 27),28) with the T-114 dependence of the logarithmic resistivity at low temperatures, which possibly occurs only when n(EF) ~0. Another evidence is that the Pauli spin susceptibility29 >· 30> has no discontinuous change across the transition concentration Nc. This is consistent with the argument that, when a transition is the Anderson transition, the density of states at EF on the nonmetallic side is nonzero and undergoes no discontinuity as N n is increased onto the metallic concentration. 5), 12> Metal-Nonmetal Trans£tion in Doped Semz'conductors §4. 153 The Fermi level and the mobility gap near Nc The Anderson transition from nonmetal to metal is defined by the statement that the states at the Fermi level become delocalized. We notice that there are three possible ways in which this delocalization of the Fermi level occurs. These are schematically described in Fig. 6. The first possibility as shown in (a) is the case where the Fermi level remains in the mobility gap until the mobility gap itself disappears, which is quite unlikely since the spectra are far from symmetric especially where the impurity levels are shallow and the influence of the conduction band is important. The second possibility as given in (b) is the case where EF escapes upwards from the mobility gap through the top mobility edge while, in the third possibility (c), EF gets out of the mobility gap downwards through the bottom mobility edge. In what follows, we give a couple of experimental results which are consistently explained on the assumption of the third case. (1) Anomalous microwave hot electron eifect21) When the microwave is applied to a sample of an intermediate donor concentration under the non-resonant condition, the resistivity decreases due to an increase of the mobility of mobile electrons with the absorption of the microwave. On the other hand, when the donor concentration in Si : P is increased into the high concentration region I across Nc, a sharp increase of resistivity as shown in Fig. 7 is observed just above Nc accompanying the application of the microwave to the sample. When N D is increased further, the enhanced portion of resistivity .r::Jp/p becomes smaller monotonically and disappears around N D::::::::: 2 X 1019 cm-3, which is nearly the critical value at which the system transforms from the high concentration region I to II. 31 > This result is explained by assuming the third possibility of Fig. 6 (c). The sharp increase of the resistivity just above Nc is related to the microwaveassisted excitation of electrons from the region around the Fermi level into that of the mobility gap where the states are localized. This apparently EF (a) ~t EF (b) EF (c) Fig. 6. Schematic diagram showing the relative relation between the positions of E F and the mobility gap. Ec and Ec• designate the top and bottom mobility edges respectively. 154 K. Morigaki and F. Yonezawa 1.5K 0 OdB H=O w-3 - Q: ~ w-4 "" J0-5 5 6 7 8 9 10 19 No (cm- 3 ) 1. 5 2 Fig. 7. Resistivity increase due to microwave irradiation plotted as a function of N D in zero magnetic field at 1.5 K and at the microwave power level of 0 dB (the microwave power at 0 dB is about 30mW) for Si: P. (After Ref. 21).) works to increase the net resistivity. (2) Dependence of e2 on donor concentration Mott 12> argues that e2=! Ec-EF I approaches zero linearly as N D approaches Nc from below. If the M-NM transition occurs in the way as shown in Fig. 6 (b), this conjecture of Mott sounds reasonable. But, if the situation is as expressed by Fig. 6 (c), s 2 changes from a finite value to zero discontinuously when the Fermi level escapes from the mobility gap downwards and the transition from nonmetal to metal takes place. It is difficult to see whether e2 goes continuously zero or undergoes a discontinuous change at N 0 , partly because of the experimental difficulties and partly because the variable range hopping becomes important near N 0 • However, our careful analysis of experimental data1 0),18),19),32)-35) shows that they seem not to exclude the possibility of a discontinuous drop of e2 at Nc. § 5. Discussion We have studied experimental data of doped semiconductors near the M-NM transition concentration Nc in order to see how these experimental results fit in the Mott-Hubbard-Anderson scheme. By analyzing some observed properties of electrons carefully, we have also proposed a possible picture for a detailed behaviour of the Fermi level near the transition point. So long as existing experimental results, known to the authors, in this concentration region are concerned, the Mott-Anderson theories successfully work Metal-Nonmetal Transition in Doped Semiconductors 155 to give theoretical explanations of the transition mechanisms. A full discussion of all concentration regions will be given elsewhere, which also seems to indicate that the M-NM transition in doped semiconductors is satisfactorily understood by the Mott-Hubbard-Anderson scheme. References 1) M. H. Cohen and J. C. Thompson, Adv. in Phys. 17 (1968), 857. 2) H. Endo, A. I. Eatah, J. G. Wright and N. E. Cusack, J. Phys. Soc. Japan34 (1973), 666. 3) F. Yonezaw.a, M. Watabe and H. Endo, Butsuri 29 (1974), 665 (in Japanese); F. Yonezawa and M. Watabe, Butsuri 29 (1974), 1002 (in Japanese). 4) T. M. Rice, Proceedz"ngs of the International Conference on the Physics of Semiconductors, Stuttgart (1974), p. 23. 5) N. F. Mott, Metal-Insulator Trans#ions (Taylor & Francis Ltd., London, 1974). 6) J. Hubbard, Proc. Roy. Soc. A277 (1964), 237; A281 (1964), 401. 7) P. W. Anderson, Phys. Rev. 109 (1958), 1492. 8) M. H. Cohen and J. Jortner, Phys. Rev. Letters 30 (1973), 699. 9) N. Mikoshiba, Rev. Mod. Phys. 40 (1968), 833. 10) H. Fritzsche, J. Phys. Chern. Solids 6 (1958), 69. 11) N. F. Mott, Proc. Phys. Soc. A62 (1949), 416; Can. J. Phys. 34 (1956), 1356. 12) N. F. Mott, Adv. in Phys. 21 (1972), 785. 13) F. Y onezawa, to be published. 14) F. Yonezawa, M. Watabe, M. Nakamura andY. Ishida, Phys. Rev. B10 (1974), 2322. 15) F. Yonezawa, to be published. 16) For instance, see a review article by F. Yonezawa and K. Morigaki, Prog. Theor. Phys. Suppl. No. 53 (1973), 1. 17) Y. Ishida and F. Yonezawa, Prog. Theor. Phys. 49 (1973), 731. 18) K. Morigaki and M. Onda, J. Phys. Soc. Japan 33 (1972), 1031 and earlier papers quoted therein. 19) K. Morigaki and M. Onda, J. Phys. Soc. Japan 36 (1974), 1049. 20) S. Toyotomi, J. Phys. Soc. Japan37 (1974), 130. 21) N. Kishimoto, K. Morigaki and M. Onda, Proceedz"ngs of the International Conference on the Physics of Semiconductors, Stuttgart (1974), p. 756. 22) H. Kamimura and N. F. Mott, preprint. 23) J. Doehler, P. J. Colwell and S. A. Solin, Proceedings of the International Conference on the Physics of Semiconductors, Stuttgart (1974), p. 478. 24) N. F. Mott, Phil. Mag. 19 (1969), 835; N. F. Mott and E. A. Davis, Electronic Processes in NonCrystalline Materz"als (Clarendon Press, Oxford, 1971), p. 39. 25) K. Morigaki and M. Kishimoto, to be published. 26) R. H. Wallis, to be published. 27) F. R. Allen and C. J. Adkins, Phil. Mag. 26 (1972), 1027. 28) F. R. Allen, R. H. Wallis and C. J. Adkins, Proceedz"ngs of the International Conference on Amorphous and Liquz"d Semiconductors, Garmisch-Partenkirchen (1973), p. 895. 29) H. Ue and S. Maekawa, Phys. Rev. B3 (1971), 4232. 30) J.D. Quirt and J. R. Marko, Phys. Rev. Letters 26 (1971), 318; Phys. Rev. B7 (1973), 3842. 31) M. N. Alexander and D. F. Holcomb, Rev. Mod. Phys. 40 (1968), 815. 32) H. Fritzsche, Phys. Rev. 125 (1962), 1552. 33) E. A. Davis and W. D. Compton, Phys. Rev. 140 (1965), A2183. 34) C. Yamanouchi, J. Phys. Soc. Japan 20 (1965), 1029; Researches on the Electrotechnz"cal Labora· tory;, No. 656 (1965), (in Japanese). 35) C. Yamanouchi, K. Mizuguchi and W. Sasaki, J. Phys. Soc. Japan 22 (1967), 859.
© Copyright 2026 Paperzz