Está en la página 1de 2



Lamb's letterA describes a ' new procedure' that, in fact, is
merely a straightforward Fourier analysis obtained by a
lengthy method. A statement is made in the second paragraph
of the letter that implies that the response of a system to a
signal a sin cot contains harmonics. If the system is linear,
which is assumed when frequency responses are plotted, and if
measurements are made after any initial transient has died
away (otherwise the measurements would be meaningless),
the output from the system can only be of the form
Ka sin (co + <f>), and there are no harmonics in the output.
Standard Fourier analysis shows that the mean product over
an intergral number of cycles (the crosscorrelation) of

However, Conwell1 has shown that this simple equation

leads to an overestimate of n\ an average of relaxation time
over all electron velocities is needed, and the relationship
between the three mobilities is not readily expressible in
analytic form. A further complication arises for heavily
doped semiconductors, particularly for electron mobilities in
direct-gap compounds. Here the effective mass increases
considerably with impurity concentration, so that the
constant' A in the relation for impurity scattering itself
becomes a function of N.
There are many advantages in retaining a simple reciprocal
relationship for combining mobilities, and we may write this
M =


where Hi is an impurity scattering mobility to allow for simple

combination. This expression in turn reduces to

Ka sin {cot + 4>) a n d a sin co{t + x) is


(cos cox cos <j> sin cox sin <j>)

The method described is therefore not new. It suffers from

the obvious disadvantage of requiring each point on the
correlation curve to be determined separately. Fig. 2 seems
to indicate some 80 computations to plot the curve.
Determination of phase and magnitude from such a trace
seems likely to introduce considerable random errors.
Although correlation has noise rejection properties, it is
doubtful whether the system proposed, 80 measurements for
each frequency taken, can compare with the use of simple
Fourier t.f.a.s and averaging the results of a few separate
30th May 1974


Department of Electronics & Electrical Engineering

University of Surrey
Guildford, Surrey, England
A LAMB, J. D.: 'Determination of system frequency response direct from
crosscorrelation functions of sinusoidal signals'. Electron. Lett.
1974, 10. pp. 176-177

Here No is a normalising impurity concentration introduced to

make the expression dimensionless. The function / can be
made as complex as necessary to match experimental results
over a wide range of N, and one would expect the form to
differ for elemental and compound semiconductors. The
simplest form for / is a power relationship, with f(N/N0) =
(N/No)x, and the Brooks-Herring theory of impurity scattering
shows that x will be less than unity. No is now the impurity
concentration at which the mobility in the semiconductor is
half the mobility in pure material. If we concentrate on the
case of maximum device interest, electron mobilities at
300 K, we note that No is 1017 cm" 3 for a number of important materials, including Si, Ge, GaAs, InP and InSb. We
may further deduce the value of x for those materials where
there is sufficient available data by extracting values of Hi,
i.e. MLKMLIM-1), from the literature, and plotting ni as a
function of ND, the donor concentration. All of the data are
not relevant. Many samples are compensated, and the
presence of acceptors causes a reduction in mobility. For a


Indexing terms: Carrier mobility, Impurity electron states and
effects, Semiconductors
Analysis of experimental data shows that a simple relationship
connects the electron mobility with the donor concentration
ND. At room temperature, the mobility is inversely proportional to 1 + V ( N D / 1 0 1 7 ) for a number of important semiconductors.

In the design of semiconductor devices, to maximise performance it is often necessary to strike a balance between
doping heavily to reduce series resistance and reducing
impurity concentration to raise the breakdown voltage. The
calculations needed for an optimum design are complicated,
because there is no simple relationship between the carrier
concentration n and the conductivity nepi, and the mobility n
is itself a complex function of n. The mobility in any doped
semiconductor is limited by two scattering processes: lattice
scattering and impurity scattering. The mobility limit due to
lattice scattering HL has been calculated for most semiconductors of device interest, and can be assumed to be a
fixed parameter. Various formulas have been suggested for
impurity scattering, but most reduce to a relationship of the
where N is the impurity
form Mi = Aj{N(B-CxInN},
concentration. Since the resultant mobility is due to the
combination of the two relaxation times, it is tempting to
combine the scattering mechanisms by the equation





Fig. 1

Vol. 10 No. 13


Dependence of mobility on donor concentration

T= 300 K
GaAs;/* L = 10 000cm
InP; fiL = 5700 cm
S\;HL = 1400 cm /Vs
Ge; ML = 4200 cm 2 /Vs


given impurity concentration, we select and plot data from

those samples that show relatively high mobilities. Some data
chosen on this basis appear questionable, in that the mobilities
are appreciably higher than those observed in a number of
samples of lower impurity concentration. It may be that
these samples were inhomogenous, and as a result the data
are in error. Nevertheless they are retained in the plots for
The rather surprising result of this analysis is that x is
close to one-half for all the materials analysed, so that the
simple expression


adequately express the dependence of electron mobility at

300 K on donor concentration. This relationship is plotted in
Fig. 1 as a continuous line, together with the experimental
data. The data 1 ' 2 for Si and Ge are shown as broken lines.
Those for the compounds InP and GaAs are less extensive3' 4
and partly unpublished, so individual data points are given.
The square-root expression certainly gives at least as good an
agreement with experiment as any previously suggested, but
it does lead to an underestimate approaching 20% at the
very high impurity concentrations. Nevertheless, it provides
an extremely simple way of allowing for mobility reduction
with impurity concentration when predictions of device
performance need to be calculated.
Acknowledgments: I am grateful to C. Clark and L. Taylor
for the provision of unpublished data on InP, and to D. Ashen,
G. Gibbons and B. Mullin for similar data on GaAs. This
letter is published by permission of the Director of the RRE.

link are identified so that these can be divided out. From

Kron's link-at-a-time formula,3 the new nodal impedance
matrix resulting from the addition of a voltage-controlled
current source is easily shown to be
N' _ dN +

where the previous nodal impedance matrix N/D was

perturbed by a mutual conductance of n/d connected from
node / to ; and controlled by the voltage from node k to /.
All quantities are implicitly assumed to be functions of the
Laplace variable s. The following abbreviations are used
JV.f = z'th column of N, the matrix of numerator polynomials
A'fc = ArthrowofiV
The above result used by Barham is useful because the
polynomial division by D in the numerator is exact and
prevents unnecessary common factors occurring.*
Suppose a scalar multiple of the same link is again added.
The new nodal impedance matrix has a denominator
D" = d'D' + n'N'IJki


iVV. = dNiJkl +


= dNlJkl


the term in braces being identically zero. Thus

1 CONWELL, E. M.: 'Properties of silicon and germanium', Proc. Inst.
Radio Eng., 1952, 40, pp. 1327-1337
2 MOLL, J. L.: 'Physics of semiconductors' (McGraw-Hill, 1964),
pp. 192-193
3 RODE, D. L.: 'Electron transport in GaAs', Phys. Rev., 1971, 3B, pp.
4 RODE, D. L.: 'Electron transport in InSb, InAs and InP', ibid., 1971,
3B, pp. 327-3299


where n'/d' is the added link. However, expanding N'lJki by


30th May 1974

Royal Radar Establishment

St. Andrews Road
Great Malvern, Worcs. WR14 3PS, England


D" = d'dD + (dn' + d' n)NiJkl


and, if d' = d, but n' = bn, b being a scalar multiplier,

D" = d{dD-n(l+b)Nukl}


The (p, q)th element of the new numerator is

N"pq = d'N'pq +

. . .


From eqn. 1, however,

N'pt = dNpi +
= dNpl +




Indexing terms: Linear network analysis, Lumped-parameter
networks, Perturbation techniques
The application of Kron's link-at-a-time formula to s plane
analysis methods is investigated to facilitate optimisation
and variational studies of linear lumped networks characterised in the s domain.

The purpose of this letter is to demonstrate that effective

perturbation methods can be applied to 5 plane analysis
methods. As a result, optimisation in the s plane can be
performed more efficiently. In the modified pivotal reduction
method of Downs,1 the only apparent method of effecting
element perturbations economically is to order the network
equations so that the initial reduction stages eliminate all
nodes except those on variable elements and input/output
ports, and to update and solve this smaller system of equations
for each element perturbation. However, the iterativesolution matrix-construction technique proposed by Barham2
lends itself well to the task of adding perturbing links.
Perturbing links can be added using Kron's link-at-a-time
formula,3 thereby effecting considerable savings over the above
approach. The possible common factors that can result from
adding a scalar multiple of a link in parallel with an existing

N'pJ = dN

so that

= d(Nkq-Nlq)

and, substituting these, together with eqn. 3, into eqn. 6,

N"M = d'N'pq +


and, ifd' = d and n' = bn, as above,

X"PQ = d[K'pq + n'/D'{N'p9Nljkl-d(Npl-Npj)(Nkq-N,qy}]


Thus both the numerator and denominator have a common

factor d. Further, the denominator expressed in the form of
eqn. 5 will be in its most primitive form if the common
factor d is cancelled. It can be shown, by substitution into
eqn. 7, that the numerator also takes on the expected
primitive form when d is divided out:
N"pq =


* BARHAM, R. A.: Private communication

ELECTRONICS LETTERS 27th June 1974 Vol. 10 No. 13