Home

Search

Collections

Journals

About

Contact us

My IOPscience

Impact ionisation rate and soft energy thresholds for anisotropic parabolic band structures

This content has been downloaded from IOPscience. Please scroll down to see the full text. 1988 Semicond. Sci. Technol. 3 48 (http://iopscience.iop.org/0268-1242/3/1/008) View the table of contents for this issue, or go to the journal homepage for more

Download details: IP Address: 193.194.76.5 This content was downloaded on 24/11/2013 at 19:39

Please note that terms and conditions apply.

Semicond. Sci. Technol. 3 (1988) 48-53. Printed in the UK

Alan R Beattie Department of Applied Mathematics, University College, PO Box 78, Cardiff CFI IXL, Wales, U K
Received 31 July 1987
Abstract. The effect of anisotropic parabolic energy bands and of indirect

parabolic energy bands (conductionandvalence band extremaat different points in wavevector space) on the transition probability per unit time for impact ionisation is investigated. The resulting direction-dependent threshold energy is described by introducing a threshold ellipsoid. It is found that for the lucky-drift model for impact ionisation in an electric field the transition probability rate which is applicableincreases from zero abovetheenergy threshold with a cubic power for anistropic or indirect parabolic bands, or, in special circumstances a 2 power compared with the more usual square law of the Keldysh result for direct isotropic parabolic band structures

soft, thein that probability transition increases from zero as the energy increases above threshold and luckyTo calculateimpactionisationcoefficients in semicon-drifttheorycan now take this into account (Ridley ductors precisely, requires a detailed knowledge of all 1987). To use soft-threshold theory one requires a simple scattering processes to be brought together with an to describe the impactionisation accuratedescription of the impactionisationtransitionanalyticalexpression as a function of electronenergy. Such an in a large transport computer program. However, it hasprobability expressionhasbeen known forsometimeand is the longbeenfelt thatmoreunderstanding of the physical famous Keldysh result (Keldysh 1960), namely processes can be obtained by formulating a simple model from which analytical results may be obtained. P p , ) X ( E ,- E$ (E,>&) (1.1) Notable success has been achieved in this dating from Shockleys lucky-electrontheory(Shockley 1961) to Ridleys morerecentandextremelyeffective modifica-where P ( ~ Iis ) the Probability Per unittime that an b, 1987, electron of wavevector k , and energy EI will impact tion, lucky-drift theory (see Ridley 1983~1, and M ~ and Burt K 1987). ~ I~ the ~ lucky-electron ~ ~ ~ionise and where ETis the threshold energy. Implicit in theorythescatteringprocessesaredescribed by a the Keldysh result is the assumption Of Constant matrix relaxation tirne andthe model envisages anelectronekmentS,non-degeneratematerialandisotropicparabolic bands. Additional softness in the threshold can be being accelerated by the field until it is scattered with a probability governed by this relaxation time, in which introduced through the matrix elements oia the Overlap caSe it loses all momentum and energy. Occasionallyan functions or through degenerate material oia the probaelectron will gain sufficientenergy so that it may cause bilities of OccuPancY and vacancy of the appropriate states (see Beattie 1985), but it is the object of this an impact ionising transition. I~ the qucky-drift theory paperto investigate solely theeffect of anisotropic the momentum relaxation time t , is separated from the parabolic bands, with also the possibility of indirect energy relaxation time tE and it is assumed that t,GtE bandst. With the development of new materials and so that the electron will gain energy as it drifts in the greater knowledge of band structure, it is becoming field before it is scattered energetically and, again, may increasingly important to beable to deal with nonoccasionally pick up sufficient energy to make an isotropic energy bands. In this paper ellipsoidal paraboimpact ionisation possible. This minimum energy is lic conduction and valence bands areassumed(the known as the threshold energy and initially hard principal axes of the conduction and valence bands may thresholds were used, in that the probability of impact ionisation was taken as zero below threshold and unity +rhc cond,lction h;,llci m~n~m,lmiIl,tl ,,illcncc hand milxinlu,n do above it. However, in practice, the thresholds are 1101 occur a t thc samc point in A-spacc.
0268-1242/88/010048+06 $02.50 @ 1988 IOP Publishing Ltd

1. Introduction

Anisotropic parabolic band structures

be in different directions). Non-parabolicity could be dealt with by methods similar to thatof Beattie (1985). It is found that the thresholdenergy depends on the direction of the ionising electron and can be described by a threshold ellipsoid where the square of the distance from the origin to the surface in the appropriate direction gives the threshold energy. For direct bands ellipsoid is atthe origin andfor thecentre of the indirect bands is displaced from the origin. For luckyelectron theory the Keldysh results, equation (l.l),is applicable with the threshold being given by the direction of the field. For lucky drift, since the momentum relaxation randomises the direction to a large extent a more appropriate quantity is the impact ionisation probability rate averaged over direction. Thiscan introduce additional softness as it is possible for the electron to have sufficient energy to impact ionise in some directions and insufficient energy in others. When this averaging is done and if the principal thresholds of the ET2 and E T 3 then it is found threshold ellipsoid are E T I , that near threshold ( 9 P(kJ c (El - E d 3 (ii) P(kJ and (iii) P ( k , )cc ( E ,- ET,)* when E , 2 ET3 = En = ET,. As can be seen from result (iii) the Keldysh result is given by the spherical case, but even in the spherical case a cubic dependence can occur for indirect bands.
2. The impact ionisation transition rate and threshold ellipsoid

when ET3

Em > E , 3 ETlt

( E ,-E,,)*

when ET3 E , 3 ET*= E,,

Figure 1. The impact ionisation transition

It will be simpler to give, in the first instance, a method forderiving the Keldysh result(equation (1.1)) for isotropic parabolic bands before proceeding to the anisotropic case where the argument is exactly the same but the algebra is rather more tedious. The transition probability per unit time, P ( k , ) , that an electron of wavevector k , will impact ionise as in figure 1 may be written as (see Beattie 1985) 1-cosx dk, dk; (2.1)

be computed exactly has been shown by Beattie (1985) but it is not the purposeof this paper to obtain accurate numerical values butrather, as with the lucky-electron and lucky-drift models, to gain agreater physical understandingthrougha simplified analyticalresult. Examination of equation (2.1) shows that the softness of the threshold can beaffectedthrough IM1* by the overlap functions, through 8 if the material is degenerate and through energy and momentum conservation, embodied in the delta-function term (1 - cos x)/x2. It is in the last of these that we are currently interested and which yields the Keldysh result for spherical bands and so lM1* is taken as constant and 8 as one. For parabolic bands and spherical energy surfaces equation (2.2) may be written as

-2pkl, k2,+ 2p k , . k , 2pk2.. k , ]- E G ) (2.3) where p is m c l m h ,and m, and mh are the conduction and hole effectivemasses respectively. This is a quadratic form and can be reduced to a sum of squares by a linear transformation such as
. .

where C is composed of fundamental constants, M is the matric element, 8 is the probability of vacancies in state 1 and 2 and occupancy in state 2 and

- E*(kl+k ; - kt) - EG] (2.2)

with Ei(kj) being the kinetic energy of state i with vector ki and EG being the energy gap. How this integral may
T When this result is applied to a calculation of lifetime it agrees with ;I recent result given hy Cornhescot and Cornbescot (19x7).

S = (1 2p)l* k J ( l +p)12-pk,/(l +py2(1 + 2p)2

(2.4)

so that

49

A R Beattie

The integral of equation (2.1) may then be performed over R-S spaceand when six-dimensional spherical polar coordinates are used this yields

evaluatedasbeforeandthe following quantitiesare useful in giving the result so obtained

P.I = ap
P, = bq
,U, = cr

By defining a reduced wavevector K , by

K

lrd

1-cosx dx 7

(2.6) six-

where r'= R'+ S', dQ2, is the incremental dimensional solid angle and

where The integral of (2.6) is non-zero if x = 0 lies within the range of integration, a condition satisfied by

1+ 2,u

E , - EG>O

p,,: = (1 + 2p:)ptV- 2,u;,41:,

Y = (1 +2P.r)(1+ 2 ~ v ) ( 1 + 2 ~ :4~.ry(1+ ) 2,~:)

(2.17)

giving the well known threshold condition -4pt2(1+2p.r) - 4 ~ : r ( 1 + 2 ~ v ) + 1@.r>P>:p:r. (2.18) which is the Keldysh result with ET given by (2.8). For anisotropic parabolic bands the conduction kinetic energy is taken as (2.10) where p = 2rn,,/h2 etc, and the hole kinetic energy is of the form The condition for non-zero P ( k , ) is that the term in the square bracket of (2.15) is greater than zero. Thus the energyat which impactionisationbecomespossible will, in general, depend on the direction of the ionising electron. Hence from (2.15) one can define a 'threshold ellipsoid' by the equation
-( a , . ,K~,,+/3,,zKI,KIj +cyclicterms)

YEG

= 1.(2.19)

which allows for the principal axes of the conduction andhole ellipsoids to benotparallel.Initially,the bands will be taken to be direct but the result is extended to indirect bands in 84. When these energies are substituted into (2.2) this again results in a quadratic form of the nine components of the vectors k , , k?. and k , . This is reduced to a sum of squares by a transformation of the form.

Thesquare of the distance from the origin tothe surface of this ellipsoid in any direction then gives the threshold energy appropriate to this direction. For isotropic parabolic bands
P , =p!. = p = :p

and

px, =p,.: = p 2 r = 0

so that
c l r x z = (1 + p ) ( l

+ 2/42

P.r>:=

y=(1+2p)3

and hence the thresholdellipsoid becomes a sphere and (2.15) reverts to the Keldysh result. When the conduction and hole bands have the same principal axes, i.e. px,, = p j z=p,, = 0, then the threshold ellipsoid has also the same principal axes and (2.19) becomes

where wherethe coefficients ai, determined by themethod described in Jeffreysand Jeffreys(1962, p 137) are given in Appendix 1. The integral of (2.1) may then be
50

Anisotropic parabolic band structures quadratic form of the energy term to a sum of squares may then be followed as beforewith the transformation

3. Soft thresholds for 'lucky electron' and 'lucky drift'

Since, as has been shown in 02, the minimum energy required by an electron to cause an impact ionisation depends on the direction of its momentum the thresholds applicable to the 'lucky-electron' and 'lucky-drift' models require to be examined. For the 'luckyelectron' model, since the electron's accelerated flight has the direction of the electric field then it is clear that the threshold appropriate to this direction should be used and the Keldysh result may be applied. However, for the 'lucky-drift' model the situation is not so clearcut.Inthismodelthemomentumrelaxationtime is considered to be much shorter than the energy relaxation time so that there are many momentum relaxing collisions for each energy relaxation. Thus the direction of the momentum is continually changing and, although there must be some small net momentum in the direction of the field, it would seem appropriate that with of thisrandomisationthetransitionprobabilityrate (2.15) should be averaged over direction. In terms of the reduced wavevector K , , defined in (2.14),the surfaces of constant energy are spheres and thethreshold ellipsoid of (2.19) canberegardedas and ET?,say, having three principal thresholds E T , , ETZ such that E T 3 2 E T 2 2 E T , . When the principal thresholds are all different there will be no impact ionisation when E < E T 1 as the energy sphere is wholly contained within the threshold ellipsoid, but as E is increased so that E T ? > E > E T , some of the energy sphere will lie outside the thresholdellipsoid in directions around that of the principal axis corresponding with ET1 and impact ionisation may take place only in these directions. This contrasts with thespherical casewhereanincrease above threshold allowsimpactionisation in all directions. Hence it can be seen that when the directionaveraging process is applied an additional softness in the threshold is introduced for the elliposidal case, This averaging is done in Appendix 2. For the case where E T 3 2 ET?> ETI,then the transition probability rate goes as a cubic power near threshold i.e. P ( k , ) cc ( E ,- ET,)?as seen in case 1 of Appendix 2 while the behaviour becomes modified as E , increases above ,ET2 and E T 3 . For the case where there is an axis of symmetry such that E T , > ET?= ET, there is a 4 power law near threshold i.e. P ( k , ) cc ( E ,- ET,)"' as seen in case 2(a) of Appendix 2, and for the spherically symmetric Keldysh case with ET37 E T ' = ET, it reducestothe resultnamely P(kl) ( E ,- ET,)',

\(kl-kc)-(kc-kb)/ where the matrix A is as in (2.12). The transition probability rate may be given in terms of the reduced wavevector defined by

where again K ; = E , and the vector

in which case
P(k,)a(2Kl
1

'L-L'S-[u,,.:(K,.,-L,)'

+/3xJKlx- Lx)(KIyL,)

+ cyclic terms] - E G ) .
J

(4.4)

This results in a threshold ellipsoid whose centre is displaced from the origin, details of which are given in Appendix 3. The direction averaging of (4.4) then becomes quite complicated in the general case and it is probably better left to individual situations. However, it is interesting to note thatin the case of spherical bands an additional softness is introduced as a result of being indirect. For spherical bands (4.4) becomes

so that the threshold ellipsoid is a sphere with centre at -pL/( 1+ p ) and radius

(+
By writing

1 2p

(& +

L2 E.)

Ii2.

the direction averaging of (4.5) gives

4. Indirect energy bands

When the minimum of the conduction band is at wavevector kc and the top of the valence band is at k, the kinetic energies in the conduction and valence bands are respectively given by replacing k by k - k c in (2.10) and by k- k, in (2.11). The procedure of reducing the

(4.7) for E T + > E ,> E T - . Hencethedisplacement of the bandedgeshasintroduceda cubic dependence near threshold.
51

A R Beattie

5. Conclusion

The effect of anisotropic parabolic bands on the impact ionisation transition probability rate has been investigatedand it is shown how thedirection-dependent threshold energy can be described by a threshold ellipsoid. Simplepower-lawexpressions for the transition rates applicable to the lucky-electron and lucky-drift models have been obtained and the results extended to indirect bands.

Acknowledgment

The authoris grateful to ProfessorB K Ridley of Essex University for drawinghis attention to this problem at a meeting arranged by British Telecom their at Martlesham Heath laboratories.

Appendix 1. The coefficients of the transformation

With the quantities defined in (2.13) and the following quantities, namely

Appendix 2. Lucky-drift probabilities for direct bands

Since the surfaces of constant energy are spheres in K , space (see (2.14)) there is no loss of generality in evaluating the direction average of the transition probability rate of (2.15) by choosingaxes which are the principal axes of the threshold ellipsoid. Thus we may write and

d , = 4d - 2 ~ , ,, 2 ~ , , 1+ p z
52

where the X,are linear combinations of K,,, KI, and K,:.

Anisotropic parabolic band structures

Case 1. ET3 2 ET?> E , > ET1 With spherical polar coordinates z-direction is that of principal axis 1 Case 3. E , > ET3 3 ET22 ET1 is done as in Case 2 except that the limits for cos 6 are +- 1, hence

in which the This

Note that when ETI

= ET?= ET3 = ET

this reduces to

P ( k , )X ( E , / E T - 1)

the Keldysh result.

Appendix 3. Threshold ellipsoid for indirect bands

With the notation of 44 and from (4.4) the centre of the threshold ellipsoid is found to be at K , = a where
a.l = Ll

-D* J[L.,(4a,Ilu,l, -P;,,>

-SETI(ET? +ET~)]/(ET ETI)(ET~ ~ - ET,)? and P,,(x) is a Legendre polynomial. Case 2. E T 3 > E , > ET?> ET1 In this case the z axis is taken in thedirection of principal axis 3 so that
X = [ETEp

+ cyclic terms] - 1 = 0
where
1 E,,= L - 2L.a - - [u,,;(u, - L,) +p.r,2(~.l - L,)(u,. - L,,) Y

+ cyclic terms].
Note that for spherical bands ~,~,.,=(l+p)(1+2p) so that

PliZ=O

and y=(1+2p)

Case 2(a). ET,?>E , > ET?= ET, Again the z axis is taken in the direction of principal axis 3, giving

References

Beattie A R 1985 J . Phys. C: Solid State Phys. 18 6501 Combescot M and Combescot R 1987 Solid State Cotnmun. 61 821 Keldysh L V 1960 Sou. Phys.-JETP 37 509 McKenzie S and Burt M G 1987 Semicond. Sci. Technol. 2 275 Ridley B K 1983a J . Phys. C: Solid State Phys. 16 3373 1983b J. Phys. C: Solid State Phys. 16 4733 -1987 Semicond. Sci. Technol. 2 116 Shockley W 1961 Solid State Electron. 2 35

53