Managing Editor:
M. HAZEWINKEL
Centre/or Mathematics and Computer Science, Amsterdam, The Netherlands
Editorial Board:
Volume 66
Mathematical Models in
Electrical Circuits:
Theory and Applications
by
C. A. Marinov and
P. Neittaanmliki
Department of Mathematics.
University of Jyvăsky/ă.
Jyvăsky/ă. Fin/and
ISBN 9789401055215
'Et moi, .... si favait su comment en revenir, One service mathematics has rendered the
je n'y seTais point alle.' human race. It has put common sense back
Jules Verne where it belongs. on the topmost shelf next
to the dusty canister labelled 'discarded non
The series is divergent; therefore we may be sense',
able to do something with it. Eric T. Bell
O. Heaviside
Mathematics is a tool for thought. A highly necessary tool in a world where both feedback and non
linearities abound. Similarly, all kinds of parts of mathematics serve as tools for other parts and for
other sciences.
Applying a simple rewriting rule to the quote on the right above one finds such statements as:
'One service topology has rendered mathematical physics .. .'; 'One scrvice logic has rendered com
puter science .. .'; 'One service category theory has rendcred mathematics .. .'. All arguably true. And
all statements obtainable this way form part of the raison d'e"tre of this scries.
This series, Mathematics and Its Applications, started in 1977. Now that over one hundred
volumes have appeared it seems opportune to reexamine its scope. At the time I wrote
"Growing specialization and diversification have brought a host of monographs and
textbooks on increasingly specialized topics. However, the 'tree' of knowledge of
mathematics and related fields does not grow only by putting forth new branches. It
also happens, quite often in fact, that branches which were thought to be completely
disparate are suddenly seen to be related. Further, the kind and level of sophistication
of mathematics applied in various sciences has changed drastically in recent years:
measure theory is used (nontrivially) in regional and theoretical economics; algebraic
geometry interacts with physics; the Minkowsky lemma, coding theory and the structure
of water meet one another in packing and covering theory; quantum fields, crystal
defects and mathematical programming profit from homotopy theory; Lie algebras are
relevant to filtering; and prediction and electrical engineering can usc Stein spaces. And
in addition to this there are such new emerging subdisciplines as 'experimental
mathematics', 'CFD', 'completely integrable systems', 'chaos, synergetics and largescale
order', which are almost impossible to fit into the existing classification schemes. They
draw upon widely dif%ferent sections of mathematics."
By and large, all this still applies today. It is still true that at first sight mathematics seems rather
fragmented and that to find, see, and exploit the deeper underlying interrelations more ef%fort is
needed and so are books that can help mathematicians and scientists do so. Accordingly MIA will
continue to try to make such books available.
If anything, the description I gave in 1977 is now an understatement. To the examples of
interaction areas one should add string theory where Riemann surfaces, algebraic geometry, modu
lar functions, knots, quantum field theory, KacMoody algebras, monstrous moonshine (and more)
all come together. And to the examples of things which can be usefully applied let me add the topic
'finite geometry'; a combination of words which sounds like it might not even exist, let alone be
applicable. And yet it is being applied: to statistics via designs, to radar/ sonar detection arrays (via
finite projective planes), and to bus connections of VLSI chips (via dif%ference sets). There seems
to be no part of (so<:alled pure) mathematics that is not in immediate danger of being applied.
And, accordingly, the applied mathematician needs to be aware of much morc. Besides analysis and
numerics, the traditional workhorses, he may need all kinds of combinatorics, algebra, probability,
and so on.
In addition, the applied scientist needs to cope increasingly with the nonlinear world and the
vi SERIES EDITOR'S PREFACE
extra mathematical sophistication that this requires. For that is where the rewards are. Linear
models are honest and a bit sad and depressing: proportional ef%forts and results. It is in the non
linear world that infinitesimal inputs may result in macroscopic outputs (or vice versa). To apprcci
ate what I am hinting at: if electronics were linear we would have no fun with transistors and com
puters; we would have no TV; in fact you would not be reading these lines.
There is also no safety in ignoring such outlandish things as nonstandard analysis, superspace
and anticommuting integration, padic and ultrametric space. All three have applications in both
electrical engineering and physics. Once, complex numbers were equally outlandish, but they fre
quently proved the shortest path between 'real' results. Similarly, the first two topics named have
already provided a number of 'wormhole' paths. There is no telling where all this is leading 
fortunately.
Thus the original scope of the series, which for various (sound) reasons now comprises five sub
series: white (Japan), yellow (China), red (USSR), blue (Eastern Europe), and green (everything
else), still applies. It has been enlarged a bit to include books treating of the tools from one subdis
cipline which are used in others. Thus the series still aims at books dealing with:
a central concept which plays an important role in several dif%ferent mathematical andlor
scientific specialization areas;
new applications of the results and ideas from one area of scientific endeavour into another;
inDuences which the results, problems and concepts of one field of enquiry have, and have had,
on the development of another.
To quote the authors: 'In any mathematical approach to a real world problem, the first step is to
establish an adequate and useful mathematical model'. Here 'adequate' and 'useful' means that once
these problems are understood (more or less completely) the solutions and techniques are indeed of
value in practice. That is indeed the case in the present instance: the modelling of electronic and
electrieal circuits, notably MOS circuits. Fortunately for the mathematicians, the mathematics
involved is nontrivial: and, fortunately for the engineers, techniques are far enough advanced to be
able to calculate and design in concrete situations.
Electrical and electronic engineering has an enviable reputation as an area in which the intui
tions of engineers and the preciseness and exactness (and sometimes surprising insights) of
mathematics reinforce each other rather than fight each other. This book will do much to enhance
that reputation.
The shortest path between two truths in the Never lend books, for no one ever returns
real domain passes through the complex them; the only books I have in my library
domain. are books that other folk have lent me.
1. Hadamard Anatole France
La physique ne nous donne pas seulement The function of an expert is n01 10 be more
roceasion de resoudre des problemcs ... elle right than other people, but to be wrong for
nous fait pressentir la solution. more sophisticated reasons.
H. Poincare David Butler
vii
V111 Contents
ix
x Preface
demic" (not realistic) problems on which the mathematicians work. Our book
proposes an armistice. On the one hand it uses efficient tools developed in func
tional analysis, differential equations and numerical analysis to solve problems of
circuit theory. Conversly, to applied mathematicians, the book offers new models
derived from the engineering of integrated circuits, a field in full and fast progress.
We would like to express our graditude to Dr. V. Hara and Dr. A. Lehtonen for
their stimulating collaboration and to JP. Santanen Ph.lic. for his help in prepa
ration of this book. We wish to thank P. Kemppainen M.Sc. and A. Roikonen
M.Sc. for their skilful typing of the manuscript in 'lEX. We are also indebted to the
Academy of Finland, Research Council for Technology for their financial support.
1.0. Introduction
give much better information about the behaviour of solutions than the Lipschitz
type estimates
If(t,x)  f(t,y)1 ~ Llx  yl ,
because w, unlike L, may be negative. This fact was extended to the case when,
for all t, the operator f(t,·) has the domain and range in a Hilbert space satisfying
(f(t,x)  f(t,y),x  y) ~ 0.
Such an operator has been called di~~ipative since the energy of the corresponding
system does not increase (Phillips [1957,1959]). However, there are several problems
in partial differential equations where it is more natural to consider Banach spaces
rather than Hilbert spaces. To simulate the inner product techniques on Banach
spaces the concept "semiinner product" has been introduced and one has studied
operators which are dissipative with respect to such a semiinner product (Lumer
and Phillips [1961]).
During the last thirty years the theory of dissipative operators in connection
with the theory of semigroups has been proved to be a fertile mathematical field.
Bn\zis [1973,1987], Goldstein [1970], Barbu [1976], Martin [1976], Pazy [1983] are
few of the more wellknown textbooks of the domain. Let us note that many authors
prefer to call A monotone or accretive if A is dissipative. Various applications
2 I Dissipative operators
Let X be a Banach space over the field K (R or C) and let II . II be the norm
on X. Strong neighbourhoods of a point x E X are the open balls B(x,r) = {y E
X ; Ily  xii < r} for any r > o. The boundary and the closure of B(x,r) are
S(x,r) = 8B(x,r) = {y EX; Ily  xii = r} and B(x,r) = B(x,r) U S(x,r)
respectively.
The space X* of all linear continuous functionals on X is called the dual space
of X. The space X* is also Banach with respect to the norm
and
Proposition 1.1. If X is a Banach space and its dual X' is strictly convex, then
for all x I 0, G( x) consists of a single elemcnt.
Proof. Suppose, for contradiction, that there are h, fz E G( x), h I fz. By
convexity, (1  t)h + tfz E G(x) c S(O,l). But S(O,l) does not contain line
segments. 0
Let us consider now thc sub differential of the norm at the point z E X defined
by
811 zl1 = {f E X' ; liz + xII
IIzll + Ref(x)
~ for all x EX}.
We have the following characterizations of 811zll:
Lemma 1.2.
811zl1 = {f E X'; (z,x)_ ::; Ref(x)::; (z,x)+ for all x E X}
= {f E X' ; Ref(x) ::; (z,x)+ for all x E X}.
Consequently,
(z,x)_ ::; Ref(x)::; (z,x)+.
Conversely, suppose (z,x)_ ::; Ref(x) ::; (z,x)+ for all x E X. Taking h = 1 in
(z,x)+ ::; h 1 (liz + hxlllIzl!), it follows that
liz + xlllIzll ~ (z,x)+ ~ Ref(x) for all x E X,
and hence f E 811zll. This establishes the first equality. If Ref(x) ::; (z,x)+ for all
x E X, then by i) of Lemma 1.1,
Ref(x) = Ref(x) ~ (z,x)+ = (z,x)_
and the final set equality is seen to be valid. o
Proposition 1.2. For every lEX' the following properties are equivalent:
i) Ref(x)::; IIxll for all x E X
ii) I/(x)l::; IIxll for all x E X.
o
Now we are able to prove the following important result:
1.1. Duality type functionals 5
Proof. Suppose first that f E 811zll. By Lemma 1.2 and by parts i), vi) and ix) of
Lemma 1.1 it follows that
and
liz + xillizil :::: ReJ(z + x) llzll = ReJ(z) ;
i.e. J E811z11. o
The relation between the set G(z) and the numbers (z,y)± is of fundamental
importance here. We have
Theorem 1.1. Ifz,y E X and.\ is a real number such that (z,y)_ :::;.\:::; (z,y)+,
then there is J E G(z) such that ReJ(y) =.\. In particular,
Proof. Taking X as a real linear space, let us consider the linear subspace of X
spanned by z and y, that is,
D={az+/1y; a,/1ER}.
We assume y =1= az for any a E R, since otherwise the assertion is immediate. Let
us define the linear function 9 : D ...... Rby g(az + /1Y) = allzll + /1.\. If /1:::: 0 then
also by Lemma 1.1. Consequently, j3).. ::; (z, j3y)+ for all j3 E R and we obtain by
Lemma 1.1 that
This fact, together with the first part of the theorem give the required set equality.
o
and
Hence,
(1.2)
JJxJJ = g(x) = Reg(x  hy) + Reg(hy) ::; Reg(x  hy) ::; Jg(x  hy)J ::;
::; IIx  hYII for all h > 0,
such that v) holds.
Now, if vi) is true we can write
and letting ,\ ..... 0+ one obtains iii). Finally, if ii) is valid we have
Lemma 1.5. If w E R and x, y EX, then the following assertions are equivalent:
i) For all f E F(x), Ref(y) ::; wJJxJJ2
ii) For all 9 E G(x), Reg(y) ::; wJJxJJ
iii) (x, y)+ ::; wllxJJ
iv) (x, y  wx)+ ::; O.
Proof. The equivalence of i) and ii) is clear. Also, the equivalence of iii) and iv) is
derived from Lemma 1.1 x).
To prove that ii) is equivalent with iv), means to prove that Re g(y  wx) ::; 0
for all 9 E G(x) is equivalent with iv). But this follows directly from Theorem 1.1.
D
Let us consider now the following problem: when does (x,y)_ = (x,y)+ for all
x, y E X with x i= 0 ? Proposition 1.1 combined with Theorem 1.1 shows that this
fact holds if X' is strictly convex.
8 I Dissipative operators
If we consider now a real Banach space with a strictly convex dual, by Theorem
1.1 we have a unique I E G(x) such that (x,y)_ = (x,y)+ = fey) for all X,y E X
and x =f O. But this means exactly the Gateaux differentiability of the norm at
the point x =f O. By Lemma 1.3, the Gateaux derivative f of the norm is just its
sub differential. Also, we observe that the norm is never Gateaux differentiable at
x = 0 because
(O,y)_ =f (O,y)+ for all y EX, Y =f O.
Now we shall compute the duality functionals for some spaces appearing in our
applications.
To begin with, if Y is a compact space, let us consider the Banach space
IIfll = l~'~n
max max Ifi(x)l·
xEY
where
M(f)={(P,x);PEl,n, ,xEY, IIp(x)I=IIIII}·
Proof. Let {Ed k with Ek + 0+ for k + 00. For every pair (k, i) of indices we
choose Xk,i E Y such that
1.1. Duality type functionals 9
Consequently,
IIf + Ek911 = l:::;.:::;n
max Ifi(xk,i) + Ek9i(xk,i)1 =
(1.3)
= Ijp(xk,p) + Ek9p(xk,p)l,
\Ve choose from {xk,ph a convergent subsequence (keeping the same notation)
{xk,ph, such that Xk,p + Xp, for k + 00. By (1.3), (p,xp) E M(f). On the other
hand, it is easy to see that there exists an index N such that for k > N we have
sgn [Jp(Xk,p) + Ek9p(Xk,p)] = sgn fp(Xk,p) = sgn fp(xp). Hence, taking also into
account the relation (1.3) we obtain for k > N:
In order to prove the converse inequality, we take (p,xp) E M(f) and we observe
that
[I + E9p (X p) [ 1
Ilf + E911  Ilfll ?: If (x )1 fp(xp)
E P P E
. . ~(~)
Due to the fact that the right hand Side tends to Ifp(xp)1 fp(xp) when E tends to
0+, the inequality follows. D
The following two lemmas are slight extensions for complex valued functions of
the results of Sato [1968].
Let Y be a normed space on which a aalgebra and a measure m are defined
and let LP be the set of measurable complex functions h defined on Y for which
Proof. We have
= 1 Igi
Yo
dm + lim
.0+ VYo
1 If I
11+cB.ll
c
f dm.
But,
If(x)1
11 + c RXl 1 1 :::; Ig(x)1
g(x)
c
and
. 1+ c ffitl
hm
1 1
= Re
g(x)
f( )' for every x E Y  Yo .
• 0+ c X
r If + cglPdm
h
= cP r IglP dm + l~[ IflPdm + l~r If + cglPdm.
l~
But
[ If + cgl
ly,
Pdm = r IfIP (1 + 2cFr + c 1F12); dm
ly,
2 ,
it holds
r If + cgl
lY2
P dm = r Ifl [1 + ~2 (2cFr + c21F12) + o(c)] dm =
lY2
P
It follows that
+ cgll lIfll) =
r.
c 1 (llf
[l Ifl dm + cp l (l Ifl dm
1 1
= c 1 P
UY
2 Frlfl P dm + o(c))'  c 1 P
Let us consider A: 'D(A) ....... X, 'D(A) C X, with its "range", R(A) = {Ax; x E
'D(A)}.
If w E R, A is said to be wdi33ipative iffor all x, y E 'D(A), any of the following
equivalent properties (see Lemma 1.4) are valid:
Dl. There is f E F(x  y) such that
Proposition 1.3. If A is dissipative then, for all a > 0, (I  aA)l is well defined
and it is a contraction on nCI  aA).
Proof. Actually we have to prove that n(I  A) = X implies n(I  aA) = X for
all a > 0 with condition
(see (1.4».
Take an arbitrary vEX. We must show that there is U E V(A) such that
v = u  aAu, or equivalently
1.2. Dissipative operators 13
1
R(I  an A) = X, for all n E N and a > 2'
from which we easily derive (see Oharu [1966, p. 1150]) the desired result. 0
Proof. Assume by contradiction that there exists Xo rt. "D(A) and yo E X such that
(1.7)
By Lemma 1.1 ix) we find Xo = Xl E "D(A) and from (1.7), Yo = AXl = Axo, that
is a contradiction. 0
and the domain V(A) of A is the set of all x E X for which the above limit exists.
Lemma 1.10. Let Al and A2 generate contraction semigroups Sl and S2, re
spectively, such that Sl(t)S2(S) = S2(S)SI(t) for all s,t E [0,00[. Then, for each
x E V(AJ) n V(A2)
Proof. As
Lemma 1.11. Let S be a closed and densely defined linear operator on X, and
let 11 E peS). Then). E peS) if and only if I  (11  ).)(IlI  S)I has a bounded
inverse; in this case
1.3. Semigroups of linear operators 15
and similarly,
(J.lI  S)ITI(>"I  S) = Ilv(s).
Thus>.. E p( S) and the desired equality holds.
The proof of the converse is equally easy.
Now, we are ready to prove the main result about linear semigroups.
Proof. A. Necessity.
For each x E D(A),
d+
dt S(t)x = AS(t)x = S(t)Ax , x E D(A). (1.8)
d
djS(t)x = AS(t)x = S(t)Ax. (1.9)
Let us remark in passing that, (1.8) and (1.9) show that u(·) = SOx solves the
initial value problem
du
{ dt = Au(t) , t~0, (1.10)
u(O) = x
when A is the generator of S and x E D(A).
16 I Dissipative operators
t d t
S(t)x  x = Jo dsS(s)xds = Jo AS(s)xds = Jo S(s)Axds.
t (1.11 )
Now let t > 00; the closedness of A and the Dominated Convergence theorem imply
foco e>"S(s)xds E D(A) and
x = (AI  A) 1 00
eA'S(s)xds, xEX
x= 1 00
e>"S(s)(AI  A)xds, x E D(A).
1.3. Semi groups of linear operators 17
Moreover,
II(AI  A)lyll ::; [YO e>'sIIS(s)11 ·llyll ds ::; Ilyil/'\.
This completes the proof of the necessity.
B. Sufficiency.
For ,\ > 0, set
A>. = '\A(AI  A)l = ,\2 (AI  A)l  AI.
This bounded operator is called Yo.~idaappToximation of A. This is because,
lim A>.x = Ax for all x E D(A).
>.~oo
Clearly IIS(t)1I ::; 1, and the above equation holds for all x E X. Moreover
S(t)S(s) = Set + s), S(O) = I. Next, for x E D(A)
S(t)x  x = lim etA,x  x =
>'~oo
= lim
>.+00
t esA'A>.xds = 10t S(s)Axds
10
(1.14)
by the bounded convergence theorem. Thus S(·)x is continuous on [0, oo[ for each
x E D(A) and hence for each x E X. Thus S is a Cocontraction semigroup.
Let B denote its generator. Then (1.14) implies B ::> A i.e. D(B) ::> D(A) and
Blv(A) = A. By the necessity part of the theorem, 1 E pCB); also 1 E peA). Hence
(IB)l = (IA)l since (IB)l ::> (IA)l and both are bounded operators.
If follows that B = A. This completes the proof. 0
Proof.
a) By Theorem 1.2, i) and iii) hold. Let x E V(A). If 9 E G(x),
Reg (S(t)x  x) = Reg (S(t)x) lIxll ~
~ Ig(S(t)x)IlIxll ~ IIS(t)xlllIxll ~ a.
If we divide by t and let t > 0, ii) follows.
Proof.
d+
d S(t)x
t
= hO+
lim h I [S(h)S(t)x  S(t)x] = AS(t)x
104. Linear differential equations 19
and
d S(t)x = lim S(Tlt  h)h 1 [S(2 1 t + h)x  S(Tlt)X] =
dt hO+
= lim S(2 1 t  h)h 1 [S(h)S(2 1 t)x  S(2 1 t)x] =
hO+
= S(Tlt)AS(Tlt)x = AS(t)x.
Moreover, if t ;::: 8 > a then S(8)x E V(A), AS(t)x = Set  8)AS(8)x and it
follows that the map t f+ AS( t)x is continuous from ]0, oo[ into X. 0
Let us consider the Banach space X over the field K and A : V(A) C X f+ X,
a linear operator on X.
\Ve consider in this section the existence and uniqueness of the solution for the
Cauchy problem in X:
A function u E C1(]O, 00[; X) verifying u(t) E V(A) for all t > 0 such that (1.15)
holds, is said to be a classical solution of (1.15) or, simply, a solution.
A function u E C(]O, 00[; X) is called a strong solution of (1.15) if it is absolutely
continuous on every compact of ]0,00[, u(t) E V(A) and satisfies (1.15) a.e. on
]0,00[.
20 I Dissipative operators
Theorem 1.4. Let D(A) be a subspace of X such that D(A) = X and A : D(A) t+
X be a linear I'm "dissipative operator. Let also J E CI([O, 00[; X) and Uo E D(A).
Then, the Cauchy problem (1.15) has a unique classical solution.
Proof. Taking into account Proposition 1.3 we see that the "m"dissipativityof
A implies )O,oo[e peA). Then by Theorem 1.3 b), A generates a Cocontraction
semigroup denoted by Set). If u is a solution, then
d
ds (S(t  s)u(s)) = Set  s)J(s)
and we obtain
u(t) = S(t)uo+ 1'S(tS)J(S)dS. (1.16)
Clearly u E CI([O, 00[; X) iff v E CI([O, 00[; X), in which case u'(t) = AS(t)uo +
v'et). This means that (1.15) holds iff
v'et) = Av(t) + J(t). (1.17)
Then
d
v'et) = dt 10t S(a)J(t  a)da =
Otherwise, by (1.18)
dw
{ dt = Aw(t) , t 2:: 0;
w(O) = 0 .
d
ds Set  s)w(s) = Set  s)Aw(s)  Set  s)Aw(s) = 0
It is not difficult to show, essentially with the same proof as above, that we have:
Also, starting with the "mild" solution (1.16) it is easy to prove the socalled
"smoothing effect on initial data". This means that the solution may be of the
classical type even in the case Uo <t D(A). Precisely, we have the following result
(Martin [1976, Proposition 4.2, Ch. 7]):
Moreover, this property holds for f being a Holder continuous function. For
v EjO,lj, C"(O, T ; X) will denote the space of functions for which there exists
M > 0 such that Ilf(td  f(t2)11 ::; Mltl  t21" for all tI, t2 EjO, T[.
22 I Dissipative operators
Proof. Let Set) be the semigroup generated by A. Since the map t I> S(t)uo is
continuously differentiable on [0,00[, it suffices to show that
VI(t) = [S(tS)[J(S)J(t)]dS
and
V2(t) = [S(t  s)J(t)ds for all t ~°
and note v = VI + V2. Moreover, it is easy to see that V2(t) E V(A) and AV2(t) =
S(t)J(t)  J(t) is continuous on [0,00[. Since S is analytic, let M be large enough
so that IIS(t)1I :s; M and IIAS(t)1I :s; t I M for all t E]O, T]. For each positive integer
n define ri: = ktn I for k = 0,1, ... ,n  1. If
nI
nI
Since
as t > 0+, we have that t ~ AVl(t) is continuous at t = O. Now let t > 0 and
G > o. Choosing (j E]O, t/2[ so that
for r E [t/2,2t] and noting that 1f!o(r) :=: J;o AS(r 8 )[J(8)  J(r)] d8 is continuous
at r = t, we see that
Since this holds for each <: > 0 we have that t 14 AVI(t) is continuous. o
This section deals with the existence and uniqueness of classical and strong
solutions for the Cauchy problem on the space X,
du
{ d = A(t)u(t)
(1.20)
utO) = Uo E D(A(t» ,
Proof. Since Reg(u(t»::; Ig(u(t))1 ::; lIu(t)1I and Reg(u(s» = lIu(s)1I we have
Dividing both sides by t  s and letting t + S from above, we obtain Re g( u:"( s)) ::;
(d/ds)lIu(s)lI. Letting t + S from below we obtain the reverse inequality. 0
The usefulness of the duality type functionals depends mainly on the following
lemma (Coppel [1965)):
Lemma 1.15. Let I be a real interval and f : I ....... X such that d f(t)/dt exists
for tEl. Then, dlIf(t)II/dt exists and
Illf(t)IIlIf(t
h
 h)1I _ Ilf(t)  hd f(t)/dtlllIf(t)111
h
= I"f(t  h)IIlIf~tl hd f(t)/dtlll
::; I f(t)  {(t  h) _ d:?) 11+ 0 as h + 0+.
o
The following result of this section refers to the Cauchy problem (1.20) with an
everywhere defined and continuous operator A. This theorem was independently
obtained by Lovelady and Martin [1972] and Pavel [1972 a,b]. See also Pavel [1984,
p.65].
Theorem 1.8. Suppose that the function (t,x) ....... A(t)x, [O,+oo[xX ....... X, is
continuous and that there is a continuous function c : [0, +00[ ....... R such that for
each t 2: 0, A(t)  c(t)I is dissipative. Then, for each Uo E X, the problem (1.20)
has a unique classical solution. Furthermore,
1.5. Nonlinear differential equations 25
and
IIUI(t)  u2(t)11 ~ lIulo  u2011 exp ( [ e(s) dS) for alIt :::: 0,
v'(t) = A2(t)V(t) ,
{
v(O) = Uo.
It follows that the function U: [0, +00[ ...... X defined by
(Bu)(t) = u'(t),
26 I Dissipative operators
with V(B) = {u E C ; u' E C , u(o) = o} where u' denotes the strong derivative
of u. It is known that B is "m"dissipative.
According to a theorem of Webb [1972]' U +B is "m"dissipative too, i.e. for
°
each ,\ > there is u,\ E V(B) such that
{
'\u,\(t) + u~(t) = A(t)u,\(t) ,
u,\(o) = °. (1.21 )
d
di11u,\(t)11 S IIA(t)OII a.e. on [0, Tj.
This implies Ilu,\(t)11 S TM for all t E [0, Tj and ,\ > 0, where M = sup{IIA(t)OIl;
t E [0, T]}. Returning to (1.21), we see that
u~(t)  u~(t) = A(t)u,\(t)  A(t)ul'(t) + f.1,ul'(t)  '\u,\(t) ,
d
Ytllu(t)  uoll = (u(t)  uo,A(t)u(t»)_ S
S (u(t)  uo,A(t)u(t)  A(t)uo)_ + IIA(t)uoll S
S c(t)llu(t)  uoll + IIA(t)uoll·
1.5. Nonlinear differential equations 27
Solving this differential inequality we obtain the desired result. The second inequal
ity of the theorem can be proved in a similar way. 0
Let us consider now the operator A : 'D( A) >+ X and a real number c such
that A  cT is "m"dissipative. Following Proposition 1.3 we can define, for each
positive integer n, the resolvent of A  cT, namely
Also, let us consider the "Yosida approximation" of (A  cT) i.e. the everywhere
defined operator
An = (A  cT)Jn = neT  I n).
Finally, let Bn : X >+ X be the operator
Lemma 1.17. Let A  cT be "m"dissipative and suppose that for each sequence
{xn}n in'D(A) such that Xn + x and that IIAx n 11 are bounded, it follows that
x E 'D(A) and AX n ~ Ax. Then:
i) If {Yn}n is a sequence in X such that Yn + Y and that IIAnYnll are
bounded, then Y E 'DCA), AnYn ~ (A  cI)y and BnYn ~ Ay.
ii) If z is in 'D(A) then Anz ~ (A  cT)z and Bnz ~ Az.
28 I Dissipative operators
i.e. the IIAxnll are bounded. This means that Y E D(A) and AX n ~ Ay i.e.
AnYn = Axn + eXn
~ (A  eI)y. Consequently BnYn ~ Ay. Thus i) is true.
Part ii) follows from i) with Yn = z and Lemma 1.16 vi). 0
Let us consider now a family of operators with the same domain D(A(t» =
D, {A( t) ; A( t) : D >+ X , t 2: o} having one of the properties:
HI. There is a continuously differentiable function e : [0,00[>+ R such that
A(t)  e(t)I is "m"dissipative for all t 2: 0.
H2. There is a continuous function d : ([0, OO[)3 >+ [0, oo[ such that
H3. If t 2: ° °
for all t, s 2: and all x in D.
and {xn}n is a sequence in D such that Xn ~ x and IIA(t)xnll
are bounded for n 2: 1, then xED and A(t)x n ~ A(t)x.
Lemma 1.18. IfH2 is valid for each bounded subset QeD and there is a 8>
°
and an M > such that if x is in Q and t, s E [0, TJ with It  sl 8, then s
°
IIA(t)x  A(s)xll S It  slM (1 + 2I1A(s)xID·
Proof. Take M = 2sup{d(t,s, IlxlD ; x E Q, t,s E [0, Tn and 8 = 11M. For x E Q
and It  sl s 8 we obtain from H2:
IIA(t)x  A(s)xll sit  slM (1 + IIA(t)x  A(s)xll + 21IA(s)xll) 12 s
s IIA(t)x  A(s)xII/2 + It  slM(l + 21IA(s)xll)/2
and the assertion of the lemma follows. o
Lemma 1.19. If HI and H2 are fulfilled and Q is a bounded set of X, then there
is a constant J{ such that IlJn(t)xll s
J{ for all (t,x) in [0, TJ x Q and all n 2: 1.
Proof. Let M be such that Ilxll s M for all x E Q, let zED, and take J{ =
M + sup{IIA(t)z  c(t)zll ; t E [0, Tn + 211zl1 (see Lemma 1.18). By part i) of
Lemma 1.16,
Lemma 1.20. If HI and H2 hold, then (t,x) f+ Bn(t)x is continuous from
[0, oo[xX into X.
IIlnCt)x  In(s)xll
= jjJn(t) [I  n 1 (A(s)  c(s)I)] In(s)x
Thus,
IIBn(t)x  Bn(s)xll
= II [n + e(t)] [JnCt)x  In(s)x] + [c(t)  c(s)] In(S)XII
:::; 11 + n1c(t)IIIA(t)Jn(s)x  A(s)Jn(s)xll
+ 11 + n1c(t)llc(t)  c(s)llIln(s)xll .
If t E [0, T] and x is in a bounded set Q, from Lemmas 1.18 and 1.19, it follows
that there is a (j > 0 and constants M' and K such that if It  sl :::; (j, then
IIBn(t)x  BnCs)xll
:::; 11 + n1c(t)llt  slM' (1 + 21IA(s)Jn(s)xll) + 11 + n1c(t)1 icCt)  c(s)1J<.
Taking into account the continuous differentiability of c we conclude that for each
bounded set Q, there exists a (j > 0 and an M > 0 such that
(1.22)
for all n 2: 1, x in Q and t, s E [0, T] with It  sl :::; (j. Moreover, Lemma 1.16 iv)
gives
Theorem 1.9. Let us consider the family of operators {A(t) j A(t) : 1> 1+ X,
t ~ o} restricted by assumptions HI, H2, H3 and let Uo be in 1>. Then, there is
a unique function U : [0,00[1+ 1> with the following properties
i) U is Lipschitz continuous on bounded subintervals of [0,00[.
ii) u(O) = Uo, the weak derivative U;" of U exists, is weakly continuous, and
satisfies u:"(t) = A(t)u(t) for all t ~ 0.
°
iii) The derivative du/dt ofu exists almost everywhere on [0, oo[ and du/dt =
A(t)u(t) for almost all t ~ (i.e. u is a strong solution of (1.20)).
iv) If Ul and U2 are two strong solutions of (1.20) corresponding to initial
conditions UlO and U20 respectively, then
dUn/dt = Bn(t)un(t)
{ (1.23)
un(o) = Uo E 1>.
Taking into account Lemma 1.20, Lemma 1.16 v) and Theorem 1.8 we deduce
the existence of the unique continuously differentiable function Un : [0,00[1+ 1>
satisfying (1.23) for all t ~ 0, and
(1.24)
Due to Lemma 1.16 vi) we have the boundedness of {IIBn(s)uol/}n and (1.24)
implies the existence of a constant K such that
°
Now let Q be a bounded subset of X which contains un(t) for all t E [0, T] and
n ~ 1. Moreover, let 0> 0 and M > be such that (1.22) is valid.
Taking 0< h :::; 0 and t E [0, T] and by using Lemma 1.15, Lemma 1.1 parts iv)
and vi), Lemma 1.16 v) and the inequality (1.22) we successively obtain:
d
dt lIun(t + h)  un(t)1I
= (un(t + h)  un(t), Bn(t + h)un(t + h)  Bn(t)un(t))_
:::; (un(t + h)  Un(t), Bn(t + h)un(t + h)  Bn(t + h)un(t))_
+ IIBn(t + h)un(t)  Bn(t)un(t)1I
:::; Ic(t)llIun(t + h)  un(t)1I + hM(l + 21IBn(t)un(t)ID.
1.5. Nonlinear differential equations 31
Consequently,
Since IIBn(O)uoll is bounded by part vi) of Lemma 1.16, it follows from Gronwall's
inequality (see e.g. Coppel [1965, p. 19]) that there is a constant K such that
Let Kl be a constant such that le(t)1 $ Kl for all t E [0, TJ, Let also c be a positive
number and c' = K l c/2(exp(K1 T) 1).
Following again Lemma 1.15 and also Lemma 1.1 part xi), there is h(c) > 0 such
that
d
di11um(t)  un(t)1I
d
dillum(t)  un(t)1I $ KIlIJm(t)Um(t)  In(t)un(t)il+
+ lIum(t)  Jm(t)um(t)II' h(c) + lIun(t)  In(t)un(t)1I . h(c) + c'
~ (Kl + h(c» (1Ium(t)  Jm(t)um(t)11 + lIun(t)  In(t)un(t)lI) +
+ Killum(t)  un(t)1I + c',
32 I Dissipative operators
But,
Ilun(s)  In(s)un(s)11 = n11IAn(s)Un(s)1I
::; nlIIBn(s)un(s)1I + nlllc(s)Jn(S)un(s)1I ::; n l K2
as (1.26), (1.25) and Lemma 1.16 i) give.
Then,
d
dillum(t)  un(t)1I
::; (Kl + h(c:» K 2 (n 1 + m l ) + KllIum(t)  un(t)11 + c:'
::; KIlIum(t)  un(t)1I + K]c:/(expKlT 1)
for all m,n ~ no = 4K2 [K] + h(c:)](exp(I{lT)  l)/Kl c:. Hence, the differential
inequality implies Ilum(t)  un(t)11 ::; c:, whenever m, n ~ no and t E [0, T]. Conse
quently, the sequence {u,,} n is uniformly Cauchy and since X is complete, there is
a continuous function U : [0, T] >> X such that u n(t) + u(t) uniformly on [0, T].
As IIdun/dtll are bounded for t E [0, T] and n 2: 1, it follows that u is Lipschitz
continuous on [0, T]. On the other hand, since un(t) + u(t) and IIBn(t)un(t)1I ::; K,
we obtain by Lemma 1.17 and the assumption H3 : IIAn(t)un(t)11 are bounded and
u(t) is in 1), Bn(t)u,,(t) ~ A(t)u(t) and IIA(t)u(t)1I ::; K.
Now let Q be the bounded set Q = {u(t); t E [O,T]} eX. By Lemma 1.18, let
{j and M such that
So we have proved parts i) and ii) of the theorem. As regards part iii) it is
sufficient to prove that t It A(t)u(t) is Bochner integrable on [0, T] and u(t) =
J;
uo + A( s )u( s) ds. For the proof, the reader can see Kato [1967, Lemma 4.6]. For
part iv) of the theorem we simply apply Lemma 1.15 and take into account the
hypothesis HI. 0
Chapter II
Lumped parameter approach of
nonlinear networks with transistors
2.0. Introduction
34
2.1. Mathematical model 35
Before describing the circuit under study, some prerequisites are needed. The
RN norm used below will be a weighted £1 one, namely IIxlid = I:~1 d;lxil where
db d 2 , ••• ,dN are strictly positive constants. If we denote by 11I·llId the matrix norm
induced by II· lid, and consider "the measure" of the N x N matrix M, namely
(2.1)
The utility of this concept for the dissipativity on RN, follows from the following
(almost straightforward) inequalities:
where a ERN.
Also, if f: [0,1] >+ RN is continuous, then
11 (a, f()..)) d)" ~ (a, l f()..) d)..)_ ~ (a, 11 f()..) d)")+ ~ 11 (a, f()..))+ d)" (2.3)
where the Riemann integrals are taken componentwise. The proof can be found in
Marinov [1990 b].
If D C RN is a convex set and F: D >+ RN has its Frt!chet derivative F'
continuous in D, then for any x, y ED,
where the integral is also in componentwise sense (see, for example, Ortega and
Rheinboldt [1970]).
Finally some definitions are necessary. We define hyperplanes Pi, i = 1, ... ,p in
RN :
36 II Lumped parameter networks
Let us consider the decomposition RN Uf=lPi = Uf=lRi, where Ri, i = 1, ... , q :::; 2P
are disjoint, open and convex sets (called "regions") with a boundary contained in
Uf=lPj.
The function F: RN >+ RN is called "continuous piecewise continuously dif
ferentiable"  CPWCD  if it is continuous everywhere and in any region Rj the
(Frechet) derivative F' exists and is continuous.
The function F: RN >+ RN is called "continuous piecewise linear"  CPWL  if
F(x) = Ajx + B j for all x E Rj. The Figure 2.1 presents the types of non lineari ties
(in the scalar case) which appear in this chapter.
Let us consider the nonlinear network of Figure 2.2. The box represents a
resistive nonlinear multiport with independent or/and controlled sources. To its N
ports are connected P bipolar (npn and/or pnp) transistors, Q junction diodes, R
nonlinear capacitors and S nonlinear inductors, so we have N = 2P + Q + R + S
components.
The semiconductor devices are described by the dynamic response large signal
model proposed by Gummel [1968) and presented in Figure 2.3 (a) for the jth tran
sistor and in Figure 2.3 (b) for the kth diode. These models take into account the
nonlinear DC properties as well as the presence of nonlinear junction capacitances.
They include standard circuit elements with six constant parameters a}, at, T2jl,
T2j, C2jl and C2j having strict positive values and a} < 1, at < 1, and two
nonlinear functions hjl, hj. The resistances of conducting layers are included in
the resistive multiport.
The following notation is used for the current vector
2.1. Mathematical model 37
+ + +

/2j1 (U2jl)
jll i '2,
C2jl e2j
Uk
i2 +
U2jl U2j
+ +
(a) (b)
Figure 2.3 The semiconductor device models
where
and analogously for the voltage vector u. Also, the state variable will be written as
z= [~]
where
q = [ql, ... , q2P, q2P+l, ... , Q2P+Q, Q2P+Q+l, ... , Q2P+Q+R]tr
is the vector of capacitor charges (both in semiconductor device models and exterior
capacitors) and
v=
d
w = T F( v) + dt Z (2.5)
= ffi
P [ 1
T a j
j=l f
Here IQ is the Q x Q identity matrix, OR+S is a zero matrix with R +S rows and
columns, A EEl B mect.lls [~ ~] and Xj is the jth component of x.
for j = 1,2, ... , 2P + Q. We suppose that the extra device capacitors and inductors
are described by the nonlinear functions I;' i.e. Qj = Ij(Uj) for j = 2P + Q +
1, ... ,2P+ Q + Rand"pj = Ij(ij) for j = 2P + Q + R+ 1, ... ,N.
Our first hypothesis regards the scalar nonlinearities.
I. (a) For every j = 1, 2, ... , 2P + Q the function Ii : R f+ R is CPWL and in every
region (here interval) fj(x) > O.
(b) For every j = 2P + Q + 1, ... , N the function I j : R f+ R is CPWL and there
exist strictly positive constants T/j and ej
such that 0 < T/j ::; Ij( x) ::; for ej
any x in any region (interval).
2.1. Mathematical model 39
The second assumption is related to the "hybrid" description of the nonlinear re
sistive multi port.
II. (a) There exists H: RN 1+ RN and B: [0,00[1+ RN such that w = H(v) +
B(t).
(b) His CPWCD.
(c) B is continuous.
We shall use hypotheses III and/or IV and/or V regarding the Jacobian matrix of
H.
III. There exist strictly positive numbers dJ, ... , dN such that for every j = 1, ... , N
there is an Wj E R for which in any region the following holds:
aH
l (x) + ,,.!.  ' (x) I ::; Wj
N d laH (2.6)
ax' ax'
__ .
1
~d·
i=l 1 1
iotj
IV. There exist strictly positive numbers db ... , dN such that for all k = 1, ... , P
(2.7)
and for every j = 1, ... N there is an Wj < 0 satisfying inequality (2.6) in any
region.
V. There exist strictly positive numbers d 1 , ... ,dN such that for every j = 1, ... ,N
there is a (3j < 0 which in any region makes true the inequality
_aH
_J(x)_"
N .!. '(x) I '?(3j.
d laH (2.8)
ax'1 ;=1 d·1 ax'1
~
iotj
Now we shall formulate a new series of hypotheses, parallel with the previous
ones but with smoother functions.
1*. (a) For every j = 1,2, ... ,2P +Q the function fJ: R 1+ R is of C 1 type and
fj(x) > 0 for x E R.
(b) For every j = 2P + Q + 1, ... , N the function "Yj: R 1+ R is of C 1 type and
o < TJ j :S "Yi ( x) :S ej for x E R.
II". The same as II but replacing the CPWCD property by the C 1 property of H.
III", IV", V". The same as III, IV, V respectively but using "for all x ERN"
instead of "in any region" .
All our results will be stated supposing either (some of) assumptions IV or (some
of) assumptions 1* V" . The proofs will be given only for the first case, their
transposition to the second being obvious.
40 II Lumped parameter networks
If hypothesis I is valid, then we observe that for all j = 1, ... , N the functions Ii
are invertible and 1;1 = 9 is CPWL. Then we can define the function G: RN I> RN
by
G(x) = [gl(Xl), ... ,9N(XN)]tT
which has the CPWL property in RN. Similarly, if 1* is valid, 1;1 = gj exists for
all j and is of C 1 class. Also we can define as above G: RN I> RN, G E C 1 .
With the previous notation we have G(z) = v and we may define the function
A: RN I> RN by
A(z) = TF(G(z))  H(G(z)) . (2.9)
If, in addition, hypothesis II (a) holds, then from (2.5) and (2.9) we derive the
following differential equation in RN (with associated initial condition) describing
the dynamic behavior of our network:
(E(zo,A,B))
Corresponding to this problem we can formulate the steady state (or DC) equation
of the network under study:
(S(A, B))
2.2. Dissipativity
If we accept the hypotheses I and II, let us denote by R(r;), ri = 1, ... ,Pi, the
open regions (intervals) defining the CPWL structure of g;, i = 1, ... , N. That
is R = U~;=IR(r;) and g;(x;) = mT;x; + n T; for all x; E R(r;). Let also, for
k = 1, ... ,p, Pk(X) = {x E RN I 2:j:l cjXj = bk} be hyperplanes in RN on which H
is not differentiable. Let us denote
that FoG is CPWL. Therefore, we conclude that the hypotheses I and II imply
the CPWCD property of A. In addition, 1* and 11* imply the C1 property of A.
Starting from these remarks we can prove our first result, which will be essential
below.
Lemma 2.1. Let the hypotheses I, II, III (or 1*, 11*, 111*) be valid. Then,
(a) there exists W E R such that, for any region of A and for any Z there
(respectively, for any Z E RN) we have
/LV(A'(Z)) ~ W ,
Proof. (a) Let us consider j E {1, ... , 2P + Q} and dt, .. , dN the numbers from the
assumption III. Denoting by Ai the components of A and by tij the elements of T
we have:
I I ['(
N
8A j. ()
8 di 8
z + "L.J d 8Hj. (G(Z)) +
8Ai. = tjjfj gj(Zj))  8
z} i=l} Z} x}
i"#i
max [(t jj + t ~i
i=l }
Itijl)/Tj j WjfCj] = max (sjfTj j WjfCj)
i"#j
For the same numbers dt, ... , d N as above but for j E {2P + Q + 1, ... , N} we
find
(2.12)
42 II Lumped parameter networks
where
T/ for W ~ 0
Pi = { } (2.13)
~i for W <0.
If we take into account relation (2.1) together with inequalities (2.10) and (2.12)
it follows JlD(A'(z)) :s; W where
W= max { j=1,max
... ,2P+Q
[max (Sj . Wi)]
Tj 'Cj
j. max
}=2P+Q+I, ... ,N
(Wi)}
pj
. (2.14)
(b) Let us consider Pi, i = 1,2, ... ,p, the hyperplanes which define the CPWCD
property of A. Let also Xo and xm be two different points in RN not being on the
same Pi (see Figure 2.4).
(2.15)
2.2. Dissipativity 43
and
ml ml
where x>. = x~~l + ),(xie  x~~l). Hence, using (2.15) and (2.16) we derive
+ :L [A(x~e)  A(x~e)] .
k=l
With this equality, by successively applying properties ii) and vi) from Lemma 1.1,
and also (2.3), we can write
ml
1°1
(xO xm,A'(x>.)(xO _xm))+
ml
d)'+:L IIA(x~.)A(x~e)lld.
k=l
Due to the property (2.2) and the first part of the present lemma, we also have
ml ml
::; wllxO  Xmlld L (tt+l  tk  20:) +L IIA(x~.)  A(x~e)lld
k=O k=l
ml
= w(1  20:m)llxO  xmlld + L IIA(x~e)  A(x~e)lld .
k=l
The continuity of the function A implies, via the property vii) from Lemma 1.1,
the continuity of the function 0: f4 (xO  x m , A(x~.)  A(x~e))+' This is why, for
0: tending to zero, the last inequality becomes
44 II Lumped parameter networks
Lemma 2.2. If the assumptions I, II and V (or 1*, I1* and V* ) are fulfilled, then
(a) there exists f3 < 0 such that for any region Ri of A and for any z E Ri
(respectively, for any z E RN) we have
!ID( A'(z)) ~ f3
(b) for every xo, xm ERN it holds
(xo  xm,A(xO)  A(x ffi ))_ ~ f3l1xo  xffilld .
Proof. (a) The desired inequality is found in a similar way as in the proof of Lemma
2.1 (a) and the constant f3 has the value
f3 = min
. { min.
j=I ..... 2P+Q
. (rj
[min  ' f3j)]
Tj 'Cj
. .
min
'j=2P+Q+I ..... N
(f3j)}

1]j ,
(2.18)

where we have additionally denoted
1  dj+l a(,j+I)/2 for j E {I, 3, ... , 2P  I}
=1
N d]
rj = t]]  L Itijl
i=1 
1 d]l j/2 £ . {2 4
 Tar or } E
2P}
" ... ,
(2.19)
i#j )
 1 for j E {2P + 1, ... , 2P + Q} .
(b) The proof begins with relation (2.17). If we successively apply properties v),
vi), viii) from Lemma 1.1 and (2.3), we find
(xO  Xffi ,A(x~£)  A(x~£))_
~ L
ffiI
k=O
(tHI  ik  2e) 1°
I
(XO  x m , A'(x,\)(xO  Xffi))_ d)"
ml
L IIA(x~£)  A(xt.)lld .
k=1
Property (2.2) and part (a) of the present lemma then yield
ffiI
(xO  x m, A(x~£)  A(x~.))_ ~ f3(I 2em)lIxO  xffilld  L IIA(x~.)  A(xt.)lId
k=1
which gives the result when e tends to zero. o
2.3. DC equations 45
2.3. DC equations
The following result is related to the existence and uniqueness of the steady
state solution of the network under consideration.
Theorem 2.1. Under hypotheses I, II and IV (or 1*, II* and IV*) and for each
E E RN,there exists a unique solution of the problem (S(A, E».
Proof. Because IV implies III, Lemma 2.1 shows that A  wI is totally dissipative
on RN. Moreover, this operator is continuous and then the result of Webb (see
Webb [1972]) gives
On the other hand, the inequalities (2.7) from the assumption IV imply Si < 0
for any j = 1, ... , 2P + Q (see (2.11)). Adding the fact that in assumption IV, wi
is strictly negative for any j and therefore Pi = ei for j = 2P + Q + 1, ... , N (see
(2.13)), we have
which is a strictly negative number. Thus, in (2.20) we can take ,\ = l/w and we
obtain the surjectivity of A, i.e. the existence part of the theorem. The uniqueness
is an immediate consequence of the dissipativity of A and negativity of w. Indeed, if
zl and z2 are two solutions of the steady state problem, then 0 = (zl_z2, E E)+ =
(Zl  z2,A(Zl)  A(Z2))+ ::; wllz 1  z 2 11d that implies zl = z2. 0
A first remark on the above theorem regards the fact that the existence and
uniqueness of capacitor charges and inductor fluxes under steady state conditions
proven above is equivalent (via the property of G being onto in RN) to the existence
and uniqueness of the hybrid vector v which satisfies
T·F(v)+H(v)=E. (2.22)
The existence and uniqueness result formulated above for equations (2.22) and
(2.23) is a partial extension of (or is strongly related to) many published results for
the case when H is linear, for instance, Willson [1968, 1970], Sandberg and Willson
[1969 a,b], Willson and Wu [1984]. In the case when H is nonlinear, our result is
an extension of the smooth classes of functions, considered by Fujisawa and Kuh
[1971], to CPWL and CPWCD classes for F and H respectively. Our result is of
the same nature as that given in Chien [1977] (Corollaries 7 and 8), where F and
Hare CPWL.
Two properties that one might expect our network (described by (2.22) or (2.23))
to possess are: "small" changes in input IJ cause "small" changes in output v and,
a bounded sequence of input vectors yields a bounded sequence of outputs. For
the case when H is linear this problem is solved in Sandberg and Willson [1969 a]
and in Willson [1970]. By using dissipativity we can, almost directly, derive the
following result for the CPWCD case:
Theorem 2.2. (a) If the hypotheses I, II and IV (or 1*, II* and IV*) are valid,
then the solution v of (2.22) is a continuous fUIlction of the vector n.
(b) If we add the hypotheses h(O) = 0 for j = 1, ... , 2P + Q and H(O) = 0 (where
o is the zero vector in RN) then every bounded sequence IJI , IJ2, n3,... is mapped
by equation (2.22) iIlto a bounded output sequence vI, v 2 , v 3 , •..
Proof. For two inputs IJ and 73* with outputs v = G(z) and v* = G(z*), we have
wllz  z*lld :S (z  z*, A(z)  A(z*))+ = (z  z*, 73 + 73*)+ :S 1173  73*lld .
because of Lemma 2.1. Hence, the negativity of w implies that If f+ z is a continuous
function. By using the continuity of G, statement (a) follows. The additional
hypotheses from (b) imply A(O) = 0 and taking z* = 0 in above sequence of
inequalities we derive
(2.24)
Theorem 2.3. Under hypotheses I, II and III (or P, II" and lIP) the problem
(E(zO, A, B)) has a unique solution z: [0,00[1+ RN , for each Zo E RN.
Proof. The hypotheses I and II assure the continuity of the function (t, z) 1+ A( z) +
B(t) on the whole domain [O,oo[xR N . In addition, Lemma 2.1 gives, for each t,
the dissipativity of A(·) + B(t)  wI on RN. Hence, in view of Theorem 1.8, the
result follows. 0
A comparison between the hypotheses used in the preceding theorem and those
used in Sandberg [1969] allow us to formulate the following remarks:
(1) Here, the resistive multiport was supposed to be nonlinear and the condi
tion H(O) = 0 (automatically satisfied in the linear case) was not imposed.
(2) We remark also that hypothesis III which restricts the class of nonlinear
ities is implicitely verified in the linear case considered by Sandberg.
(3) The nonlinear characteristicfunctions Ii, j = 1, ... , 2P + Q and (j,
j = 2P + Q + 1, ... , N, were supposed to be CPWL, while in Sandberg's
paper they are of C I class.
(4) In the above we did not use the additional conditions fiCO) = 0, (j(O) =
which were used by Sandberg.
°
(5) The additional boundedness hypothesis imposed by Sandberg to the time
function B (which describes the independent sources) is no longer needed
here.
Let us now consider the solutions Zl and Z2 of the problems (E(zJ,A,BI)) and
(E( z~, A, B2)) respectively. Of course,
d
dt [Zl(t)  z2(t)] = A(ZI(t))  A(z2(t)) + BI(t)  B2(t)
and because the function t 1+ IlzI(t)  z2(t)lId is differentiable on [O,oo[ except a
countable set, Lemmas 1.15 and 1.1 give:
d
dtllzl(t)  z2(t)lId ~(zl(t)  z2(t),A(zl(t))  A(z2(t)))+
(2.25)
+IIBI (t)  B2(t)lId a.e. in [O,oo[ .
Analogously,
Theorem 2.4. Let hypotheses I, II and IV (or 1*, 11* and IV·) be fulfilled.
(a) If Zl and z2 are the solutions of the problem (E( z~, A, BI » and (E( z~, A, B2»
respectively and BI(t)B2(t) tends to 0 fort + 00, then limt_ex>[ZI(t)z2(t)] = o.
(b) If z is the solution of (E(zo, A, B» and Bex> = limt_ex> B(t), then there exists
Zoo E RN, independent of zo, such that z(t) + Zex> when t + 00.
(c) The solution of the problem (E( zo, A, B» is globally exponentially asympotically
stable in the Lyapunov' sense.
Proof·
(a) Taking into account Lemma 2.1, we get from inequality (2.25):
which gives
(2.28)
for all t E [0,00[. Because of assumption IV we have w < 0 (see (2.21» and the
result easily follows.
(b) Let Zoo be the (unique) solution of stationary regime A( z) + Bex> = 0 (see Theo
rem 2.1). Of course, Zex> is identical with the solution ofthe problem (E(zoo, A, Boo».
Then, the statement (a) gives the result.
(c) It is sufficient to put in (2.28) BI = B2, to find that
Theorem 2.5. Under hypotheses I, II and IV (or 1*, II' and IV' ) and addition
ally supposing h(O) = 0 for j = 1, ... , 2P + Q, ';'j(0) = 0 for j = 2P + Q + 1, ... , N
and H(O) = 0, the boundedness of the function t ...... B(t) on [O,oo[ implies the
same property for the solution function t ...... z( t) of the problem (E(zo, A, B)).
Proof. Because the above conditions imply A(O) = 0, in the same way as that used
in proving Theorem 2.4 (a) we can obtain
d
dt IIzlid ::::; wllzlld + IIB(t)lld a.e. in [0, oo[ .
If w < 0 (a sufficient condition for this is the hypothesis IV), this shows that the
function D: t ...... Ilz(t)  zoo(t)lld/llzo  zoolld is strictly decreasing on [0,00[,
starting from 1 and tending to o. This function describes the global behaviour of
the network between initial and steady state. If we fix A EjO, 1[ we can define "the
Adelay time" Tt, as the (unique) moment when this function equals A. The above
results allow us to frame this parameter between two bounds which can be apriori
computed.
Corollary 2.1. Let hypotheses I, II, IV, V (or 1*, II', IV', V') be valid, where
the latter two ones are satisfied for the same d = {d 1 , ... , d N }. Then,
e Pt ::; D(t) ::; ewt for all t > 0 ,
Proof. The right hand inequalities immediatelly follow from (2.30), while for the
left hand ones we use (2.26) and Lemma 2.2. Thus we derive
d
dtllz(t)  zoolld:O:: i1llzo  Zoo lid a.e. in [0,00[,
50 II Lumped parameter networks
2.5. An example
u 1K
~ r
100 K
F 10 K
;
S1 S2 S"
Figure 2.5
h(Ul) = 10 5 X [exp( 40Ul)1], !2(U2) = 1.638 X 10 5 X [exp( 40u2)1]. The
bulk emitter, collector and base resistances are 10, 20 and 100 ohms respectively.
The operator H is CPWL in R5 with the hyperplanes X5 = 6 and Xs = 10
defining three regions, k = 1,2,3, for X5 < 6, X5 E [6,10] and Xs > 10 respectively.
In each of these regions the resistive multiport is described by w = Hkv + Bk,
where Hk is a 5 X 5 symmetric matrix with the common elements for k = 1,2,3:
Htl = H:3 = 5.68451 X 10 3 , Hf2 = Ht4 = 0.99056 x 10 3 , Ht2 = Hfl = H:4 =
Ht3 = 0.94315 x 10 3 , Ht3 = H:l = 4.22539 x 10 3 , Ht4 = Htl = Hf3 =
H:2 = 0.04225 x 10 3 , Ht5 = H;1 = H:5 = H;3 = 0.47 x 10 3 , H;5 = H;2 =
Ht5 = H;4 = 0.0047 x 10 3 • These matrices differ in HJ5 = H~5 = 1.44794 X 103 ,
Hg5 = 1.19794 X 10 3 • The elements ofthe vectors Bk are Bt = B: = 98.1 x 104,
B~ = B! = 99 x 104 for k = 1,2,3 and B~ = 35.88 X 104 , B~ = 20.88 x 104,
B~ = 45.86 X 104 •
The circuit is initially in the steady state with e = 0.6 corresponding to the
transistors in the off state: VI (0) = V3(0) = 0.35, V2(0) = V4(0) = 10 and V5(0) =
0.986. The initial capacitor charges are ZI(O) = Z3(0) = 1.7599 X 10 12 , Z2(0) =
Z4(0) = 10.1638 x 10 12 and Z5(0) = 1.972 X 10 12 •
Hypotheses I and II are obviously satisfied and the same is the case with as
sumption IV if we choose d = {0.961, 1, 0.961,1, 0.823} and WI = W3 = 31 X 10 6 ,
W2 = W4 = 39 X 10 6 , Ws = 88 X 10 6 • With the same d, our circuit satisfies
hypothesis V, namely /31 = /33 = 11.34 X 10 3 , /32 = /34 = 1.94 X 10 3 and
/35 = 2.31 X 10 3 • As we see, all hypotheses of Theorems 2.12.5 and Corol
lary 2.1 are fulfilled such that all properties given by these theorems are valid for
our circuit. Also, 81 = 83 = 0.0114, 82 = 84 = 0.442, rl = r3 = 1.998,
r2 = r4 = 1.557 and P5 = e5 = 115 = 2 X 10 12 • With these, relations (2.18) and
(2.21) yield: W = 0.0114 X 10 9 and /3 = 2.26 X 109 such that we can compute the
bounds of the O.ldelay time: ,£do.1 = 1.018 X 109, T~.1 = 201.981 X 109. On the
other hand, an adhoc program numerically integrating the system (E(zo, A, B))
gives Tll = 146 X 10 9 . It seems that the reasonable tightness (especially of the
upper bounds) and the calculational simplicity of the bounds from Corollary 2.1
make them useful for initial stages of circuit design. Such bounds can be included in
so called "timing simulators", the usual handling programs for designers of digital
circuits (see Ruehli and Ditlow [1983]).
Chapter III
3.0. Introduction
The circuit consists of n voltage sources, P bipolar transistors and an infinite num
ber of capacitors and resistors, these last ones grouped together in an "infiniteport"
"R· oo . Each transistor j = 1,2, ... , P is a nonlinearly lumped parameter modelled by
a GUl11l11ei circuit as shown in Figure 2.3 with positive parameters a} < 1, at < 1,
T2jl, T2j, C2jl, C2j and fuuctions hil, hi with continuous and strict positive
52
3.0. Introduction 53
L IGkjl <
00
inf{Gjj; j E N} > a
while for the capacitors we naturally impose
According to (3.1), G is a linear function in £1 space, such that for any voltage
sequence {Uj}~_n+1 E £1 and for each k = n + 1, ... ,0, 1, ... we have
00
ik =  L GkjUj. (3.3)
j=n+1
On the other hand, denoting by tkj the elements of the matrix T = .9P [ 1j
J=l a f
for k,j = 1,2, ... ,2P and tkj = a for k,j > 2P, the circuit structure implies:
2P
ik = LtkjUj O'Yj1)(qj) + ddt for kEN. (3.4)
j=l
Here, qj is the electric charge of capacitor Cj and qj = 'YiC Uj) = CjUj+r;!j (Uj) for j =
1,2, ... ,2P while qj = CjUj for j > 2P. Then, by using (3.2)(3.4) we can derive
the following infinite system of differential equations with variable {qj} j E £1 (finite
total electric charge)
dqk(t) ~  ::
{ ~ = ~ak;q;(t)+fk(q1,q2, ... )+fdt),
k=1,2, ...
(3.5)
qj(O) = ej
54 III Infinite systems
where
ajj = Gjj/Cj for any j E N
for k::; 2P
for any kEN and where fik = {~
for k > 2P
o
]k(t)= L Gkjej(t) for any kEN and t;:::O.
j=n+1
With this example in mind, we shall study below a system of the form (3.5) with
less restrictive hypotheses than those imposed by (3.1). The mathematical interest
of our problem consists of the dissipativity properties of the linear and nonlinear
part of the system, as we show in Section 3.1. Section 3.2 includes the main re
sult regarding uniqueness and stability of £Pcontinuous solutions of (Sd. For the
quasi autonomous case an existence theorem is proved in Section 3.3 and it contains
also a result about the stability of the equilibrium solution.
It is obvious that if the transistors do not exist in our circuit (i.e. P = 0),
then system (3.5) is a linear one. In this case only, we are able to prove good
convergence properties of the truncated system solution (Section 3.4). Also, easily
computable bounds of truncation errors are infered. In Section 3.5 we verify that the
system (3.5) (governing the circuit in Figure 3.1) fulfils the conditions imposed by
the general theory developed in Sections 3.13.3. Consequently it has interesting
qualitative properties. Also, in the linear case, a very simple concrete example
illustrates the theory.
Infinite circuits with more general structures than the above one and whose vari
ables belong to a Hilbert space (finite energy) were studied by Dolezal [1977,1979].
As in our case, the dissipativity is the central concept of the work. Also Zema
nian [1976,1981,1982) treated twodimensional infinite circuits appearing in the
numerical analysis of certain boundary value problems in semiconductor devices.
Moreover, many authors dealt with countable infinite systems of equations and
their wide range of applications. The reference Deimling (1977) contains bibliog
raphy on this subject up to 1975. We also mention papers: McClure and Wong
[1975,1976,1979)' Chew, Shivakumar and Williams (1980), Miller and Michel (1980),
Marinov [1984,1986].
3.1. Preliminary results 55
Let aij, (i,j = 1,2, ... ) be complex valued functions defined on the interval
e
[0,00[, = {e;}i E fP and S E [0,00[. Let also f;, (i = 1,2, ... ) be complex
functions defined on [0, oo[ xf P • Our aim is to study the infinite system of equations,
associated with an initial condition:
dUi(t) ~
~ = J~_l aij(t)Uj(t) + Ii(t,Ul(t),U2(t), ... )
(Se) {
{
d~~t) = A(t)u(t) + F(t, u(t))
u(s)=e andtE[s,oo[
where A(t) is a "matrix" linear function defined by using functions aij and F is a
perturbation nonlinear function defined with the functions Ii. Unlike in the case of
finite dimensional systems, equation (Ee) and system (Se) are no longer equivalent.
As a consequence, the proofs here extensively use properties of dissipative functions
along with "classical" topics such as uniform convergence, derivability of series of
functions, etc.
Let E be a normed vector space and we denote
Ck = {f : E >+ fP ; I is continuous} .
The conditions of compactness of a set in CP (Lusternik, Sobolev [1974,p.167])
imply that I E Ck if its components 1; : E >+ C, i E N are continuous and
L::'l \Ii(tW < 00 uniformly on any compact subset of E (in short, u.c. E).
Let P E [1,00[ and q be its conjugate, that is q = p/(p  1) for p > 1 and
q = 00 for p = 1. Some assumptions, that are to be used in the following, are made
regarding the functions aij and Ii from (Sel:
AI. aij: [0,00[>+ C is continuous for any i, j E N.
L
00
L laij(tW < 00
00
#i
L \fi(t,xW < 00
00
u.c. [O,oo[xt'P.
i=l
e
We will denote by F : [0, oo[ x p ....... fJP the function having Ji'S as com
ponents; then, hypothesis FI is equivalent to F E C{o,oo(xt P '
F2• There is a function Cl! : [0,00[> R that is integrable on any compact
interval from [0, oo[ such that F(t,')  o:(t)I is dissipative for any t E
[0,00[.
The next result can be easily shown (for p = 1 see Taylor [1958,p.220] and
McClure, Wong [1976]).
where x = {Xj}j E fiP. Morover, IIB(t)11 ::; ~(t) with equality for p = l.
(ii) for any s E [0, oo[ ancl U E Cr.,oo(' the functiOll t ....... B(t)u(t) belongs to Cr.,oo(·
D
+ f EN} =1
1
sup{ Reajj(t) lai)(t)1 j for p
/iA(t) = i~!
'T)
Proof. Taking into account hypotheses Al and A 2 , it follows that /LA,W and f3
are lower semicontinuous and consequently almost everywhere continuous in [O,oo[
and bounded below on any compact set from [0,00[. As I'A(t) ~ wet) + f3(t) for
any t E [0, oo[ it follows from hypothesis As that /LA is also bounded above on
compacts of [O,oo[ . The proof is complete. 0
We note that 1) is a linear subspace of f.P and 15 = f.P in the topology of the norm
in f.p. From hypothesis A2 it follows immediately that
1) = {x = {xa} i E f.P {f
j=1
aij(t)Xj }. E
•
f.P for each t E [O,oon
and it results (under hypothesis A 2 ) that for each t E [O,oo[ we may consider the
linear functions:
where x = {Xi}, E 1). We remark that A(t) = D(t) + B(t) where B(t) satisfies
Lemma 3.1.
If t E [O,oo[ and n E N then we may start from the operator A(t) and, with
assumption A 2 , construct the function An(t) : f.P tt f.P with the ith component
defined by:
ifi ~ n
ifi >n
where x = {x;}i E f.p. This definition is consistent because, as it follows from
hypothesis A2 that
00
[Dn(t)X)i = { ~ii(t).1'i if i s: It
ifi > It
for i s: n
for i >n
for every x= {x;}i E €p.
We shall associat e with the sequenc e {l'd, E (P, the sequenc
e xn = {Xni in
which x? = Xi for i s:
n and xi' = 0 for i > n.
Lemma 3.3. Under hypothe sis A2 we have for .1' = {x;} i E £P, t E [O,oo[ and
n E N:
Proof.
Let us first consider p = 1. Let Y = {I, 2, ... } with count.ing
measure and
Yo = {i E Y : Xi = O}. From Lemma 1.7 we obtain:
iEYo iEYYo
t
iEY iEY
o
A similar procedure may be used to prove the next result.
Theorem 3.1. Let ~,1] E f!P and u, v E Cr.,oo[ be f!Psolutions for (Se) and (S'I)
respectively. Then, under llypotheses AI, A 2 , A 3 , F l , F 2,
(iii) if lim
t(X)
fl [J.lA(r) + a(r)] dr
s
= 00, then 11 is asymptotically stable.
(v) if 5 < 0 exists such that ILA (r) + 0:( r) ::; 5 for any r 2: 8, then u is uniformly
asymptotically stable.
d
+ F"(t, u(t))
II
dt u(t)  v(t) = An(t)[u(t)  v(t)]  FII(t, v(t)) . (3.7)
for t E [8,00[\N, where we have denoted w(t) = u(t)  v(t). From (3.8) and
Lemma 1.1 it follows:
:t IIw(t)1I11 ::; \ w(t(, AII(t)w(t)) + + \ W(t)lI, F(t, 0if)  F(t, lI(t)II)) _ + 511 (t)
(3.9)
where we have denoted
Taking into consideration Lemma 3.3, hypot.hesis F 2 , Lemma 1.4 and denoting
= 5n(l) + /3(t)llw((t)  w(t)"11. we obtain from (3.9)
I'n(t)
d n II
&111O(t) II::; [}1A(t) + o:(t)] 1111'(/) II + I'n(t) on [8, oo[\N. (3.10)
The function t ...... fttl [!lA(l') + 0(1')] d1' is absolutely continuous on every interval
[t 1 , t 2 ] C [8,00[\N. Consequently, it. is derivable almost everywhere in [t 1 , t 2 ] and
from (3.10) one obtains:
(3.11)
3.2. Properties of solutions 61
(3.13)
and the same with v instead of tl. It follows from (3.14)(3.16) that {{n(t)}n ; 0
U.c. [s,oo[. Hence, from (3.13) and (3.11) we have:
o
62 III Infinite systems
dUi(t) ~ 
+ li( 111 (t), 112( t), . .. ) + Ii (t)
z
{ ;It = ~ aij1lj( t)
)=1
L
00
'#J
A 2. w = sUp{Rcaii i E N} < 0
F~. The function ii : CP f> C is continuous for any i E Nand
L
00
VIe shall denote by P : £1' f> CP the function with i;,s as components;
hypothesis F~ statcs that F E Cfp.
F;. There is a a E R such that P  aT is totally dissipative.
3.3. QuasiautoBOI110US case 63
p
Fa. L
00 _
F~. There is a continuous function d: [s, oo[ x [8,00[>> [s,oo[ such that
 
IIF (td  F (t2 )11 ::; It 1  t2ld( t 1 , t2) for all (tl , t2) E [8, oo[ X[8, 00[.
D = {x = {X;}i E £1'
Lemma 3.5. Under assumptions Ai, A;, Fi, F; and with c = w + (3 + a the
operator A = A + F(t,.)  cI is "m"lota11y dissipative, for all t E [8,00[.
Proof. Let us first suppose that the following properties are already proved
(i) Al = D  wI is linear, closed, densely defined and "m"totallv dissipative.
(ii) A2 = B + F(t,·)  ((3 + Q)I is continuous, everywhere defined and totally
dissipative for each t E [8,00[.
Then, following a result of Webb [1972]' we have "m" total dissipativity of
A = Al + A2 •
Most of the properties assumed, as satisfied in (i) anel (ii), follow directly from the
definition of the set D, functions D, B, F, as well as from Lemma 3.1. Vve have to
verify only the underlined properties in (i) and (ii). Let {xn}n be a sequence of
elements of D such that {xn}n ; x and {AIJ:n}n ; y. There is then, > 0 with
IIDxnll P < , for any n E N. On the other hand, for any 171 E N we have
m
for n ; 00. It results in ~:1 laiix; 11' < 00 and therefore xED. But
64 III Infinite systems
Lemma 3.6. Let hypotheses A~, Ai, F~, F2 be valid and c = w + (3 + 0 < O.
Then for any y = {Yi}; E fP there is x E V SUell that its components verify the
system
00
Proof. It is sufficient to show that the equation .4.1' + F(x) = y has a solution.
By the same arguments as those used in the proof of Lemma 3.5 we can show that
A + F  cI is "m"totally dissipative and this implies the existence of the solution
of the equation x  p[Ax + F(x)  c:r] = z for any z E CP and p > O. Taking
p = l/c and z = y/c we have the proof. 0
e
Proof. As the function B + F( t, .) is continuous on p it is sufficient to verify the
above property for D instead of A + F(t,·). But :l" E V as we have obtained in
the proof of Lemma 3.5. As {aiiX;'}n > aiixi for any i E Nand IIDxll < M it
follows (cf. Lusternik, Sobolev [1974,pp.149j) that {Dxn }n~Dx. 0
Theorem 3.2. Under assumptions Ai, Ai, Fi  F4 and for any ~ E CP, there
exists U E Cr.,col' CPsolution of S;'
Proof. Let {~n}n be a sequence of V such that {en}" > ~ . Because of Lemma 3.5,
Lemma 3.7 and assumptions F;, F 4, the hypotheses HI, H2 and H3 of Theorem
3.3. Quasiautonomous case 65
1.9 are satisfied and this ensures the existence of a function un : [8,00[1+ V with
the following properties:
a) un E Cr.,oo['
b) un is weakly derivable (let (un):" be its weak derivative), Un(8) = en and
(un):"(t) = Aun(t) + F(t,un(t)) for any t E [8,00[.
c) if urn also satisfies a) and b) with urnes) = em then,
Ilun(t)um(t)II:S Ile n e mllexpc(t8) forany tE [8,00[,
where c = w + (3 + a.
From b) we have
(3.17)
for any 1* e
E if (the dual space of p ), with the unique associated sequence
{mdk E eg • Taking mk = 8i,
where i E N <mel 81
is the Kronecker's symbol,
we have from (3.17)
d
dt ui'(t) = aiiui'(t) + [Bun(t)]i + fi(t,U"(t» (3.18)
Corollary 3.2. Let hypotheses Ai, A;, Fi, F; be valid and c = a + f3 + w < 0.
Then, there exists U E D such that for any E €P, e
lim u( t) = U , where u is the
Proof. According to Lemma 3.6, let U = {Ui}i ED be the solution of the system
00
e
Let = (ei)i Eel be a sequence of complex numbers and aij (i,j = 1,2, ... ) be
complex valued functions defined on [0, oo[ and satisfying more restrictive conditions
than the above ones, namely
AI: aij is continuous for each i, j = 1, 2, ... ;
A2: L laij(t)1 < 00 uniformly on every compact subset of [0, oo[ for
i=l
each j = 1,2, ... ;
A3: aCt) = sup { ~ laij(t)l; j = 1,2, ... } is bounded on any compact subset
of [0,00[.
We shall also consider functions Ii : [O,oo[ ft C (i = 1,2, ... ) with the following
conditions:
Fl: Ii is continuous for each i = 1,2, ... ;
00
dUi(t) ~
{ ~ = ~ aij(t)Uj(t) + fiCt), t ?': 0,
with the solution denoted by u = (U1, U2, ... ) where u(t) = (U1(t), U2(t), ... ) E £1
for every t 2:: O. We associate the following truncated system:
{
du!(t) =
t j=1
t
aiAt)uj(t) + J;(t), t 2:: 0,
with the solution denoted by (ur, u~, ... , u~). We let un = (ur, u~, .. . , u~, 0, ... )
and clearly un(t) E £1 for every t 2:: O.
When the approximation of the solution of system (5{) requires the calculation
of the solution of the truncated system (5;) the evaluation of the truncation error
cn(t) = lIu(t)  un(t)lh is needed. An upper bound 6. net) and a lower bound on(t)
of cn(t) are derived below.
With the above hypotheses we can define the bounded linear operator A(t)
£1 ...... £1 with
{
d~~t) = A(t)u(t) + F(t), t 2:: 0,
x(O) = e.
Let us define a new linear bounded operator on £1, namely An(t) : £1 ...... £1 with
where pn(t) = (Mt), h(t), ... ,f,,(t), 0, ... ) and ~n = (~l' 6,· .. ,~n, 0, ... ).
According to the continuity of the solutions of (Ed and (E;), we have
L.:::llui(t)  ui(t)1 < 00 uniformly on compact subsets of [0,00[. On the other
hand, let Mi be the countable subset of [0, oo[ on which the function t rt IUi(t) 
ui(t)1 is not differentiable. On [0, oo[ \ Mi we have
1~lu;(t)Ui(t)11 ~ 1~[u,(t)ui(t)]1
~ I[A(t)u(t)];I + I[A"(t)u"(t)];I + Ifi(t)1 + l[Fn(t)]il·
By the continuity of A(t) and..1"(t) on £1 and by the assumption F2, the above
1t
inequality shows that L.:::l lu,( t) ui'( t)1 < 00 uniformly on every compact subset
of [O,oo[ \ M, where M = U::I .VI;. Hence, the function t rt Ilu(t)  un(t)lh =
fn(t) is differentiable on [0, oo[ \.VI and (see Lemma 1.15) on this set we have
dfn(t)
~ = (u(t)  un(t), A(t)u(t) + p(t)  A"(t)un(t)  pn(t))+ (3.19)
= (u(t) unCi), A(t)ll(t) + F(t)  A" (t)1/"(t)  pn(t))_
dE n(t)
~ ~ u(t)
( 1/
" (t),..1(t ) [u(t )  1l
n)])
(t + + (3.20)
+ IIA(t)  ..1"(t)lIlllu"(t)lh + IIF(t)  r(t)II!,
and from the second part
where
and where
J.Ln(S) = sup{Reajj(s) + t
•=1
la;j(s)l; j = 1,2, ... , n} .
i",j
Taking into account (3.22) and (3.23), from (3.20) we obtain on [0, oo[ \ M
where
 [[(expl x
J.L_A(S)ds)bn(x)g,b) + An(x)]dx] exp 1° J.LA(s)ds.
(3.26)
Note that J.LA(S) has the form of J.LA(S) with ajj(s) instead of ajj(s).
70 III Infinite systellls
Lemma 3.8.
(i) Under hypotheses AI, A2, A3, FI, F2, for any t ::::: 0, the sequences
{~n(t)}n and {on(t)}n are bounded.
(ii) If AI, A2, A3, FI, F2 are assumcd together with the hypothesis
Proof. The lemma follows directly from the inspection of expressions (3.25) and
(3.26) and taking into account the fact that for any nand t, Il.,(t)~ /lA(t) ~
a(t)"n(t) ~ aCt) and An(t) ~ IIF(t)II]. 0
Lemma 3.9. Assume the initialhypolheses AI, A2, A3, FI, F2 together with the
hypotheses:
AS: For any positive integer n, there is 111" ::::: a suell that ,net) ~ Mn for all
t ::::: o.
A6: There is a strictly posili1'e number I' sllch that
+L
00
Re ajj(t) laij(t)1 ~ 1/ < afor any j = 1,2, ... and any t E [0,00[.
i=l
i"/,j
Denoting
it follows that
3 ..5. Applicat.iolls 71
(3.28)
where
Ln(t)o=snp{f I.U:r) I XE[O,t]}.
l=n+l
The assertions of the Lemma 3.9 follows fwm (3.29), as Ln(t), L net) ::; N. 0
Expression (3.29) constitutes an upper bonnd for the truncation error (valid
under the assumptions given in the statement of the> lemma), which are easier to
apply than (3.25).
Let us now verify that the theory developed above really works for the net
work in Figure 3.1 described by (3.5). In this case we have constant coefficients
and p 0= 1. By (3.1), (3.2) and the continuity of functions fj we see that the
and
.
f IGkjl)
Gjj + k=1
a = max
jj + max (
_G_ 1 + a} ; ___ ~k#=J~·_ _
~ O.
l$j5,2P Cj TJ c}
2P
::; a 2::: Iz)  z;l,
j=1
because
+ 2::: Itkjl
00
tjj = 1 +aj .
k=1
k#j
o
[0,00[2 ...... [O,oo[such that for any t,s E [0,00[, L lej(t)ej(s)1 ~ Itsld(t,s).
F:
j=n+l
With this very reasonable hypothesis about the circuit sources, it is clear that
holds. Thus, from Theorems 3.2 and 3.1 we obtain the existence and uniqueness of
q = {qj}j E C[~,oo[' C1solution of (3.5). By considering
Corollary 3.1 gives stability conditions related to the diagonal dominance of the
matrix G, as in the finite dimensional case treated in the previous chapter.
Let us consider now a concrete example of an infinite linear circuit. It origi
nates from an infinite length distributed structure described by Telegraph Equations
(Ghausi and Kelly [1968]):
au(t,x)
a
x
= 1·(J:)i(t,.r)
{
ai(t, x) au(t, J')
a;: = C(.T)~  g(:r)u(t,x)
t 2: 0, x 2: °.
Here, u(t,x) and i(t,x) are the voltage and the current respectively, at the moment
t and at the point x along the line. The distributed parameters (per unit length)
will be the conductance g(x) = 9 and the capacitance c(x) = c, both constant and
°
positive, and the resistance 1'( x) > a strictly increasing function.
By discretization of the above equations with respect to the xvariable only, and
°
with a constant step h > we obtain:
(3.30)
:ru
+
!.L h,.o :rl h"l

1'1 7j_l
eo 110 he
he hg
This network is of the form of that ill Figure 3.1, with n = 1 and P = 0. The
system governing this circuit is derived from (3.30), being of the form S~ with
constant coefficients, namely
dUJ
=11
UJl • (1 1 g) +
,++ UJ+I
{ ~~o = :~12rJl } cil 2 )} ch 2 r}_1 c ch 2 rJ
where
f/A = max (   .
1'1ch2
2 1
+ 1'uch + !JC :
'j 0
2
)'1ch
+ 2
 + g)
1'2ch2 c
.
Because liII1 n _ oo t,.n(t) = limo"(t) = 0, the sollltioll of the truncated circuit
converges to the one of the infinite network. The above bounds can be useful to
choose adequate values of the numlwr of cells and of the step size with the view
of obtaining a desired error. Let us finnlly remark the boundedness of functions
t f4 ,e."(t) and t t> .sn(t).
Chapter IV
The technology of integrated circuits imposes upon their designers the need to
deal with structures with distributed parameters. Figure 4.1 shows a schematic
diagram of part of a digital integrated chip, consisting of an n MOS transistor with
gate (G), drain (D) and source (S) as terminals, and its thinfilm connection with
the rest of the chip. This onchip connection can be made by metals (AI, W),
polycristaline silicon (polysilicon) or metal silicides (W Si 2 ). Alternative materials
to oxidepassivated silicon substrates are saphire and gallium arsenide (Saraswat
and Mohammadi [1982], Yuan et al. [1982]' Passlack et al. [1990]).
Interconnection
bate G
O)(ide
/
'  _ _ ' / Conduction channel
P sutlstrat!
75
76 IV Mixedtype circuits
The performance criteria for such a digital circuit are: a high operating speed,
a high level of integration, a small chip area, the nonexistence of false switching
and low power consumption. The operating speed is given by the clock frequency,
in close relation with the rise time. Referring to Figure 4.1, "the rise time", for
this circuit, is the time required for the output voltage (the drain potential, for
example) to rise (or fall) from its initial value to 90 percent of its final value under
a stepvoltage located at point A  the output of another stage (although different
by definition, our "delay time" used in Chapter V will express also "the inertia"
of the circuit). Primary interest at the design stage is to be able to predict this
performance which is clearly dependent on the delays caused both by devices and
by interconnections. But, with the advances in technology, the cross section of
connecting wires decreases while its length increases as a result of the increase in
the number of devices on the same chip (integration scale). That is why the delay
time associated with interconnections becomes an appreciable part of the total delay
time. In certain cases when the wiring lengths are as short as 1 mm with 4 /lm
minimum feature size, or for recent advanced GaAs MESFET and GaAs HEMT
technologies, the interconnection delay dominates the global delay (Saraswat and
Mohammadi [1982), Bakoglu and Meindl [1985)).
Our goal is to study the delay caused by connecting wires. Then, it is reasonable
to consider the simplest models for MOS transistors: a resistor between drain and
source having a very small resistance in the ON state and a very high one in
the OFF state. Capacitances associated with the pullup source diffusion, contact
cuts and the gates being driven can be included by connecting respective nodes
to the ground. Our model is not restricted to MOS circuits at all. The bipolar
circuits where grounded resistors can appear (O'Brien and Wyatt [1986)) are also
approachable.
Naturally, being interested in wiring delay, we try to model the interconnections
as exact as possible. The required frequencies make valid the quasitransverse
electromagnetic wave approximation (Wohlers (1969)). At the same time, with
subnanosecond rise times, the electrical length of interconnection can become a
significant fraction of the wave length. That is why the transmission line property
of these interconnections can no longer be neglected if we desire an accurate model
and therefore we shall use the wellknown Telegraph Equations (Ghaussi and Kelly
(1968)).
av .
{ ax =n
(TE)
:! e: gv.
=
Here vet, x) and i(t, x) are respectively the voltage and the current at the moment t
at a point x on the distributed structure ("regline"). The distributed parameters
(all per unit length) are: resistance of conductive path r > 0, capacitance e > 0
4.1. Examples 77
4.1. Examples
Two examples below show that some mixed models can be wrongly formulated.
Let us consider the circuit presented in Figure 4.2 which contains an inverter
formed by T1 , T 2. It controls the gates T 4 , T5 through the lines L1 and L2 and the
pass transistor T3 •
Vin2
Vinl
Ts
=
Figure 4.2 A digital circuit example
78 IV Mixedtype circuits
VD
R2
o Ll d 1 R3 L2 d'2 R4
T
0
Rl
R7 T R5
=
(4.1 )
If VI(t,X) denotes the voltage across LI during the studied dynamic regime we
have
(4.2)
Equations (4.1) and (4.2) show that the mixed model of the circuit does not possess
a solution in the "classical sense" (i.e. a continuously differentiable function in the
x and t variables) under the initial condition VlO(X). Indeed, if such a solution is
supposed to exit, then we have to obtain
4.1. Examples 79
and
.
hm
t_O
ax (t,O) = a
aVI aVIO
x
(0),
i l (t, 0) jl
+ j2n+l in+l(t)
+ +
WI +
T
rl ~ VI(t, 0) Resistive W2n+1
  Vn
~l(t)
Cl multiport 81
<
91
W2 j = Gw+B
VI (t, dr)
+ 
+
lil(t,d l jz
1 2n 2n+m
G=
2n  
2n + mL_'_.... I
I
I
in(t,O) j2n1
hn+m in+m(~)
+'""""i=i
W~"+~" ~(t+m
Figure 4.5 The general mixedtype network
°
The first 2n terminals of the multi port are connected to n elements with param
eters distributed over their entire length: resistance ri > 0, capacitance Ci > and
conductance gi 2: 0, i =!,n. Linear capacitors (with capacitance Si > 0, i = I,m)
are connected to the last m ports.
Of course there exist resistive networks with 2n + m pairs of terminals which
are not of type (G, B, 2n + m). For instance, the following conditions imposed on
our mixedtype circuit assure the existence of the (G, B, 2n + m )type description
of its resistive part:
 one of the multiport terminals is common for all external elements (this is the
common "ground" of the rcglines and of the capacitors),
 all sources are independent,
°
 none of the or di terminals is connected directly (i.e. through a zeroresistance
branch) to the ground,
°
 there is no direct connection between and/or d;terminals. Also, two or more
terminals with capacitors (which are not "the ground") are not connected di
rectly together.
Indeed, let us consider the resistive network with all sources removed. vVe may
simplify this network by using the starmesh transformation to remove all internal
nodes. Clearly we obtain a circuit described by j = Gw. Moreover, the matrix
G is symmetric and "weakly diagonally row sum dominant" (WDRD). This means
that, for all i = 1, 2n + m we have
2n+m
G ii 2: Si = L IGijl, (4.4)
j=1
#i
where Gij are elements of the matrix G. Clearly a WDRD matrix is semipositive
definite G 2: 0. Also, for many networks in the subclass considered above, G is
"diagonally row sum dominant" (DRD):
2n+m
for k = I,m.
From (TE) we formally derive the system:
(E)
Also, from (4.3) and (T E) we obtain the following system of boundary conditions:
1 aVI Vl(t,O)
(f,O)
rl ax
1 aVI vI(t,dl )
+(t,dd
rl ax
1 a~n ( Vn(t,O)
   t,O)
(BC) rn ax = G + B(t), t EjO,T[,
1 aV n
+a(t,dn) Vn(t, dn)
rn x
dVn+1 ( ) Vn+l(t)
Sl~t
dV~+m ( )
sm~t vn+m(t)
where the capacitor equations in+k(t) = Sk dVntt(t) were also used. Finally, adding
the vector of initial conditions Vo = (VI,O,' .. ,Vn,O, . .. ,vn+m,O) we obtain
{
Vk(O, x) = Vk,O(X) ; x E jO,dk[, k = r;rt
(IC)
VnH(O) = vnH,o ; k = I,m.
Thus, our dynamic problem will be
where
(SE)
and
(SEC)
°
Let us now present some notations that are used throughout this chapter. Let K
be the set of the real (K = R) or complex (K = C) numbers. K~ is the space
Km with the euclidean norm weighted by positive constants SI,'" ,Sm' If a square
matrix is semipositive (positive) definite we shall put G ~ °
(G > 0); Gtr is the
transpose of G.
For T E ]0,00[, m a positive integer and X a normed space, cm(o, T; X) will
be the space of functions, defined on ]0, T[, with values in X and with continuous
derivatives (up to and including order m). Let Cg"(O, di ; K) be the functions from
°
COO(O, di ; K) with compact support. For lJ E]O, 1], CV(O, T; X) will denote the space
of functions for which there exists M> such that Ilf(t 1 )  f(t 2)11 S Mltl  t21V
for all tI, t2 E ]0, T[ (Holder continuity);
with Ilflli~ = 2::7=1 111;111,,; where I; are the components of f. Hm,i is the Sobolev
space of functions from L 2 ,i with generalized (or distributional) derivatives (up to
and including order m) also in L 2 ,i; H;:'(K) = 07=1 Hm,i == H;:'.
84 IV Mixedtype circuits
In the sequel, it is convenient that the first 2n equations from (BC) are homo
geneous. This is why we shall take
Vk = Uk + Uk, k = 1, m +n , ( 4.6)
where
u~(t,x) = O'k(t)X 3 + f3k(t)X 2 + 5k(t)X, k = l,Ti'
{ (4.7)
ukCt,x) = 0, k = n+,.....,..l,n,.+m.
(4.8)
for all k = l,Ti'. To simplify the writing below, the following notations will be used
for k = l,Ti'
and
1 {)un ( ) 1 {)u n ( )) tr
,{) t,O, {) t,d n
rn X rn X
N · 1
= dlag, P = d'lag,
gk
rkck Ck
 =(
G21 1)~
diag Sk G2 1,  =(
G 22 1)~
diag Sk G 22
where G21 ,cJ 22 together with Gl l and G12 are block matrices composing G (see
Figure 4.5).
With these notations, and after a change of variables for (4.6), the problem
(P(B, vo)) = (E)+ (Be) + (IC) becomes
= t lad;
1=1 0
Cdi(X)gi(X)dx + nf
i=n+l
Sin/;"g;.
Here 1'0U a and 1'1U a have the same definitions as before except the time variable is
absent. Then, the problem (El) + (BG 1) + (IG 1) suggests that we formulate the
following Cauchy problem on the space XR:
du 
(GP(B,uo)) { T=Au+B(t)
u(O) = Uo == v(O)  u*(O) E XR ,
where
 _ [Ba(t)]
B(t)  Bb(t) .
In view of studying the problem (GP(B,uo)), the following two lemmas give
properties of operator A.
(Au, u) x .. = t
i=110
r (~r, dd2~i
d
;
x
 9iUi) Ui dx  ( G21l'OU a + G22Ub , u b ) ~.
K,
t ~I
;=1 ri 0
d
; ddui Ui =
X
(Yl u a , 1'0U )K2n a =  (Glll'OU a+ G12 u b , l'ou a )
K2n
,
we obtain
(Au, u}x .. = t
i=1 ri
~ 10r'l ddui 12 dx  (Glll'OU a+ G12 Ub , l'ou
X
a)
K2n
(4.12)
Therefore
Lemma 4.2.
a) If G == Gtr :::: 0 and K == C, then A is "m"dissipative on XI(.
b) If G :::: 0 and K == R, then A is "m"dissipative on XI(.
Proof. We have to prove that for all ,\ > 0, RC,\:!  A) = XI( i.e. for
and
( 4.15)
Above, 'I is the identity operator in XI( and 'I a is the nth order unity matrix.
Let us take the space YI( = Hf(K) x K: with the norm II hili = Ilhalik. +
Ilhbllk;n. For u,v E YI( let us denote 1
On the other hand, it is well known (cf. Agmon [1965]) that on Hf the following
norms are equivalent
where 1 ::; p ::; nand (Xl,"" Xp )tr is arbitrarily taken in TIf=l [0, d;J.
Then, there clearly exist positive constants K; such that
88 IV Mixedtype circuits
and this shows that the form a(,·) is bounded. On the other hand,
al(u,u)=Real(u,u)~ m,in
l~':5n riCi
_1_II~uaI12
x L;
a2(u,u)=Rea2(u,u)~ l:O;':O;n
m.in (~+'\)lIualliR'
C; 2
So a(·,·) is a coercive form. Due to the LaxMilgram lemma (see, for instance,
Fattorini [1983 p.214]), there exists a unique u E YK such that
Ln
;=1
C; l d''''d~;+
0
[ ~ 1
r,c, x
( !!i+,\ )U;
c,
] w;dx=Lci
n
;=1
id''j;wi dx
0
(4.18)
h
were ~u;.
dx 2 IS t h e d'lstn'b utlOn
. al denvatIve ax
' . 0 f du; E L 2,i. It £;0 11ows th a,
t (see
Yosida [1974 p.48])
 1 ~Ui
d2 (x) +
(9i +,\ ) Ui(X) = J;(x) a.e. in ]O,di[ and for all i = r,n.
riCi x Ci
n
and because II CoCO, di; K) is dense in L2 we obtain (4.14).
i=1
4.3. Existence and uniqueness 89
Now we return to (4.17) and integrate by parts (see proof of Lemma 4.1), to
obtain
a
( N d;;2 ,w a ) Ln + (/'1 u\ /'OW a )K2n + a2( u, v) + a3( u, v) = (f, W)XK (4.19)
2
i.e. U E D(A). Finally, using (4.19) with (4.14) and (4.20) we find
\
~
G 21 /,OU
a
+ (..\dmgs;
.
+ Gn)U
~ b
,W
b)
Km
= (f b ,w b )Km' for all w b E K:" .
Therefore (4.15) holds and this ends the proof of Lemma 4.2. o
The above lemmas allow us to prove existence and uniqueness results.
Below, according to the similar definitions for the solutions of (CP(E, uo» used as
in Section 1.4, we mean by strong solution of our dynamic problem (P(B,vo» a
function in C(O, T; X R ) which is absolutely continuous on each compact interval of
]0, T[ and with components VI, V2, ... 'v n + m ' satisfying (P(B, vo» for a.e. t and x.
It is clear from this definition that time derivatives of the solution are considered in
the X R norm, while the space derivatives are of a distribution (generalized) type.
Also, by generalized solution of (P(B, vo» we mean a function v E CI(O, T; XR)
satisfying (P(B, vo» for all t and a.e. x. Clearly a generalized solution is a strong
solution.
Theorem 4.1.
a) Suppose that there exists v E]O, 1] such that b; E CHV(O, T; R) for i = r,n and
b; E CV(O,T;R) for i = n + 1,n + m, G = Gtr 2: and that Vo E X R . Then
there exists a unique generalized solution of (P(B, Vo
°».
b) Suppose that b; E C 2 (0,T;R) for i = r,n and b; E CI(O,T;R) for i =
n + 1, n + m, b;(O) = 0, for i = r,n, G 2: °
and that Vo E D(A). TIlen
there exists a unique generalized solution of (P( B, vo»).
Proof. a) According to Lemma 4.1, Lemma 4.2 part a) and according to the fact
that on a Hilbert space a linear maximal dissipative operator is densely defined (see
Brezis [1973,Prop.2.3]), the hypotheses of Lemma 1.13 are fulfilled (with 5 = 0).
Then A generates an analytic Cocontraction semigroup. On the other hand, the
90 IV Mixedtype circuits
hypotheses imply (via (4.9) and (4.10» iJ E C"(O, T; XR). Then Theorem 1.7 gives
a unique classical solution for (CP(iJ, uo» with which (4.6), (4.7) and (4.9) assure
the generalized solution v of (P(B, vo» belongs to the space Cl(O, T; XR).
°
b) Lemma 4.1 and part b) of Lemma 4.2 show that A is a linear, densely defined,
"m"dissipative operator on XR. Also iJ E Cl(O, T; XR)' From bi(O) = for
i = I;n" we deduce that u*(O) E V(A) and then uo E V(A). Therefore, (CP(iJ,uo»
has a unique classical solution (Theorem 1.4), and the result b) of Theorem 4.1
follows. 0
Let us try to discuss the significance of the above result. The hypotheses in
the literature usually require very smooth sources (such as a constant or sinus
functions). In our results, the initial conditions may be of "squareintegrable type"
(therefore, including discontinuous functions) and not even satisfying the boundary
conditions (part a». This allows us to include primitive models for transistors
following specific goals, as we have discussed in Section 4.1. This is in accordance
with the request G = Gtr i.e. the network must not contain controlled sources. The
main constraint is the semi positivity of G. If we use improved models for transistors
assuring that the initial conditions satisfy the boundary ones, then naturally we have
to accept nonreciprocal networks (G i Gtr). With a little price paid (bi(D) = D),
the constraints for the model validation are the same, G:::: D.
On the other hand, let us note that, if we consider smoother initial condi
tions, from the wellknown apriori estimates for parabolic problems combined with
Sobolev's imbedding theorem (see for example Lions and Magenes [1972] or, for
a more classical approach Hellwig [1967]), we can obtain a solution in the clas
sical sense, i.e. with differentiability properties commonly used. In fact we have
Marinov, Neittaanmaki [1988]:
Theorem 4.2. Let us consider all independent sources having simultaneous step
variation at t = D. Then, if G :::: D and Vk,O E C2(D, d k ; R), k = I;n" together
with Vn+k,O, k = 1, m satisfy the boundary conditions (SBC), then the problem
(P(B, vo» has a unique solution in classical sense.
Referring to the first example above (Figures 4.2 and 4.3) we find
G l +G 2
G3 +G6 G 3 G6
G 3 G3
G= G4 G 4
G4 G4 + Gs
G6 G6 +G7
and B = (VdG 2 ,D, ... ,D)tr, where we have denoted Gi = I/Ri. Because G = Gtr >
D, the conclusion of Theorem 4.2 holds for VD =constant and any initial condition
in X R .
4.3. Existence and uniqueness 91
o
l/R 0
l/R 1 andB=(J,O,O)fT.
1/R 1/R
Clearly G = GfT ~ 0 and that is why the generalized solution of the dynamic process
exists for a source J E CI+V(O, T; R) and for square integrable initial conditions.
Let us consider now the usual case in digital circuits which have nonsmooth
sources, such as square pulses (i.e. sequences of Heaviside functions) or trapezoid
pulses (i.e. differentiable functions with discontinuous derivatives). In this case we
shall not expect a solution in the above sense. However, we can find a sequence
of associate problems with (P(B,vo)) which "converges" to (P(B,vo)) and has a
convergent sequence of strong solutions. The limit of this sequence of solutions
will be called a "weak solution" of our problem. More precisely, a function v E
Ll(O,T;XR) is a weak solution of (P(B,vo)) if there exist sequences {Bi}i and
{Vni with the properties
 {Bi}i~B in Ll(0,T;R2n+m)and{vni~VO in X R .
 For each i the problem (P( B i , v~)) has a unique strong solution vi.
 {Vi}i ~ v in L 1 (0, T; XR).
Due to this theorem, both our circuit examples have unique weak solutions if
they are excited with discontinuous inputs such as the usual sequences of pulses.
Again, the semipositivity of G is essential.
Proof. Let uo = Vo  u*(O) where u*(O) is given by (4.7) and (4.9). It results
in Uo E X R and because D(A) = XR, for all i there exist u~ E D(A) such that
{Uh}i ~ Uo in X R . We shall choose v~ = u~ + u*(O). On the other hand, let
Bi be step functions such that {Bi}i ~ B in Ll(0,T;R2n+m) and Bi(O) = B(O)
for all i. Let us consider the problem (CP(.ai,uh)) where iJi is a step function
obtained through (4.9) and (4.10) and by extension in the nondifferentiable points
of Bi. Let 0 = ao < al < ... < an = T be the partition of [0, T] such that
.ai(t) == Zk on [akl,ak[. If we denote by Sk(t) the semi group generated by the
maximal dissipative operator A + ZkJ, and we define ui(t) by ui(O) = ub and
ui(t) = Sk(t  akl)u(akJ) for t E [akl,ak], it is clear that u i is a unique strong
solution of (CP(iJi,ub)). From Lemma 1.15 and Lemma 4.1 we find:
d··   .
dt lIu'(t)  uJ(t)llx. ~ IIB'(t)  BJ(t)lIx. a.e. in [0, T] .
92 IV Mixedtype circuits
Consequently,
If we suppose the existence of the inverse G221 , the time independent problem
(SP(B)) = (SE) + (SBC) can be rewritten in the form
(4.21 )
(4.22)
V
b=G~ n1 (b
B G~ 211o v a) , ( 4.23)
where ,OV a and II va have the same meaning as in the preceding section and
Ba E R2 n, Bb E Rm are two vectors such that B = [~:].
For the convience of future references let us list some hypotheses:
HI: There exists at least an index i, 1 :=; i :=; n, such that gi = 0,
HI: For all i = r;n, gi > 0,
H2: G22 is invertible
H3: Gl l  GI2 G;;}G 21 > 0,
1
H3: G l l  G12 G22 G21 2': 0.
~ ~ ~ ~
Theorem 4.4.
a) Let the assumptions HI + H2 + H3 or HI + H2 + H3 are valid. Then
the problem (SP(B)) has a solution v E II COO(O, T; R) x Rm.
° i=l
b) If G > and if (SP(B)) has a solution in a classical sense this solution
is unique.
4.4. The steady state problem 93
Proof·
a) Let us take gk = 0 for k = 1, k. and gk > 0 for k = k. + 1, n. The following
functions
Vk(X) = MkX + Nk, k = I;"""ko
{ (4.24)
Vk(X) = Mke a • r + Nkea.r, k = k. + 1, n
verify equations (4.21), where we have denoted ak = .,jrkgk' Let us take also the
vector w E R2 n with components W2kl = Nk and W2k = Mkdk + Nk for k = 1, k.
and also W2kl = Mk + Nk and W2k = Mkea.d. + Nke a • d• for k = k. + 1, n.
By checking the boundary conditions (4.22) for the functions (4.24) we obtain
a linear system of equations in R2n:
for k = k. + 1, n.
Furthermore, in (4.25) we have denoted
~ n
(Pw, W)R2n ~ Ml ~)W2kl  W2k? + M2 L (W~k_l + W~k)
k=1 k=ko+l
for all w E R2n and where Ml and M2 are strictly positive constants. Hence,
the matrix P + G is positive definite under each of the two sets of hypotheses.
Consequently, the equation (4.25) has a solution which fixes the constants in (4.24).
The existence is proved.
94 IV Mixedtype circuits
b) In the same way as in Section 4.3, the problem (SP(B)) can be written as a
problem on the space XR, namely
0= Au + il , ( 4.26)
where A and il have the same definition as quoted in Section 4.3, but Ilk, (Jk, Ok are
independent of time. Of course, for the present context, the constants Ck and Sk
are artificially introduced to keep the previous notations.
If u E 'D(A) is another solution of (SP(B)) (and, hence of (4.26)) then,
On the other hand, if G > 0 then, by Lemma 4.1, A is strongly dissipative. This
means that, supposing u oF u, we have
Corollary 4.1. If gi 2: 0, i = r,n and G > 0 then (SP(B)) has a unique classical
solution.
If we try to apply these results to our above examples of Section 4.1 we find
 for the first example, where G > 0, we have a unique steady state solution even
if 9i = 0 for i = lor/and i = 2,
 for the second example, G 2: 0, a:;} exists and Gl l  G12G:;21G21 = O. The
above results assure the existence of direct current solution only for 9 oF 0 and
does not affirm anything in the case 9 = O. Because we know that in this last
case the solution does not exist (see Section 4.1), we conclude that our sufficient
conditions for existence are very close to the necessary ones.
A good model of a real process is one whose accuracy can be improved at will by
more and more precise measurements of the inputs and parameters. In this respect,
the following theorem will give sufficient conditions for the continuous dependence
of the solution upon the data, the sources (vector B) and the initial conditions
(vector vo).
4.5. Other qualitative results 95
Theorem 4.5. Let G 2 0 and let {Bi}i be the sequences with elements in
L 1 (0, Tj R2n+m) and {V~}i a sequence with elements from XR such that {Bi}; + B
in L 1 (0,TjR 2 n+m) and {Vni + Vo in XR when i + 00. Ifv and vi are the weak
solutions for the problems (P(B, vo)) and (P(B i , v~)) respectively, then {vi} + v
in Ll (0, Tj XR).
Proof. We may consider Bi and B as step functions. Otherwise, the assertion can
be obtained by passing to the limit the result with step varying functions.
Let u and u i be weak solutions (see Brezis [1973 p.64]) for the problems
(CP(Bi,um and (CP(B,uo)) respectively (obtained by using (4.6), (4.7), (4.9)
and (4.10)). The dissipativity of A combined with Lemma 1.15 gives
Using the fact that {Ui*(O)}i + u*(O) in R 2n+m and {Ui*}i + u* in L 1 (0,TjXR)
we obtain the result of Theorem 4.5. 0
At last, a property which the digital circuits might have is that if the sources
are constant functions suddenly connected then the outputs must tend to constant
values. As the following theorem shows, this property is obtained by strengthening
the hypotheses.
Theorem 4.6. Let us consider gi > 0 for all i = l,Tl, B(t) = B is constant for
t2 0, Vo E XR and suppose there exists a > 0 such that
If v is a strong solution of (P(B, va)) and Voo is a solution (in a classical sense) for
(SP(B)), then for all t 2 0 we have
duoo 
{ d.t = Au oo +B
uoo(O) = U oo .
96 IV Mixedtype circuits
On the other hand, (P(B, vo)) is associated with the problem (ep(E, uo)) with the
strong solution u = v  u·. Thus, we obtain
{
~(uuoo)=A(uuoo) a.e. in [O,T]
(4.28)
(u  uoo)(o) = u  U oo E XR .
On the other hand, the hypothesis about the matrix G implies (see relation (4.13)
in the proof of Lemma 4.1)
a.e. In [0, T] and for u(t) i= U oo . From here we obtain the desired inequality. 0
The above result is an asymptotic stability property (of the global exponential
type) of the d.c. solution: regardless of the initial conditions, the solution tends to
the same steady state value.
Our first circuit example satisfies the conditions of the theorem with a =
min(Gs , G 7 ), while for the second circuit from Fig 4.4, (even with 9 > 0, when
the steady state solution exists) the asymptotic convergence is not warranted by
the above result.
The results presented in this chapter were obtained in Marinov and Lehtonen
[1989] and Marinov and Neittaanmiiki [1988]. A different approach leading to a
variational solution can be found in Marinov and Moro§anu [1991]. On the other
hand, because in practice the nonlinearity of pullup transistor in a MOS driver
significantly influences the delay time (Wyatt [1985]), a similar problem to the above
one but containing a nonlinear resistive part (i.e. nonlinear boundary conditions) is
also interesting: Moro§anu, Marinov and Neittaanmiiki [1989,1991]. Some results
on nonlinear parabolic systems with very general nonlinear boundary conditions
can be found in Moro§anu and Petrovanu [1986] and Moro§anu [1988].
The reader has probably remarked that the mathematical interest in our problem
(E) + (BC) + (IC) lies on very special boundary conditions. On the one hand they
are of "crossed type", i.e. the value of a derivative at a boundary point depends
on the value of the function at all boundary points. On the other hand, boundary
4.6. Bibliographical comments 97
conditions contain time derivatives. The fact that this type of boundary condition
appears in transmission line problems was observed a long time ago: Brayton and
Miranker [1964], Cooke and Krumme [1968}. They refer to the complete Telegraph
Equations, which are of the hyperbolic type. Other qualitative studies on nonlinear
hyperbolic equations with nonlinear boundary conditions (of crossed type and even
with time derivatives) are: Barbu [1977}, Barbu and Moro§anu [1981]' Moro§anu
[1981a,b,1982,1988]. Although above we consider degenerate Telegraph Equations
(neglecting the inductance), our results can not be derived by those referring to the
hyperbolic case.
Other recent studies regarding the correctness of distributed parameter models
for integrated circuits are Showalter and Snyder [1986], Bose and Showalter [1990],
Showalter and Xu [1990].
Chapter V
5.0. Introduction
In the preceding chapter we have shown that the delay time problem in in
tegrated circuits leads us to consider mixedtype circuits with distributed ele
ments described by Telegraph Equations and lumped resistive and capacitive el
ements (Figure 4.5). Moreover, the wellposedness of the mathematical model
(P(B, va» = (E) + (BC) + (IC) has been studied, various conditions for the exis
tence, uniqueness and L 2 stability of different kind of solutions being formulated.
Let us consider now an integrated circuit whose delay time (especially caused by
interconnections) we want to evaluate. We assume that the corresponding mixed
type circuit has unique dynamic and steady state solutions in a classical sense.
(Sufficient conditions for this are given in Theorems 4.2 and 4.4.) The first problem
solved below (in Section 5.2) is to choose precise supplementary conditions in which,
irrespective of initial conditions, the dynamic solution tends to the steady state (this
is the so called "global asymptotic stability" of the steady state). The convergence
here is in the space of continuous functions of t and x, and the transient regime is
provoked by a step variation of a part of the sources, the other sources remaining
constant. It is exactly the interesting case in the study of the switching speed
in integrated structures (see Marinov and Neittaanmiiki [1989J and the examples
in Sections 4.1 and 6.6). In Section 5.1 we define "the global delay time" as a
performance parameter which expresses the rate of evolution of the whole network
from initial conditions towards steady state. Because the stability theorem found
here assures even the exponential type of stability, it is possible to infer an upper
bound of the global delay time. As many examples show (see Section 6.6) this
upper bound is sufficiently tight and can be itself considered as a global delay time.
It is a very attractive parameter for circuit designers due to its closed form and the
simplicity of the calculus. The inclusion of this delay time in a CAD (Computer
98
5.1. Remarks on delay time 99
Aided Design) timing analyzer is perfectly possible. Let us remark that, from the
mathematical point of view, the dissipativity of the abstract operator governing
our problem (in the space of continuous functions) again plays the main part in our
reasoning. The fact that the diagonal dominance of the matrix G is a condition for
our results to work, does not significantly reduce the area of their applicability. In
this respect the examples given in Section 6.7 are relevant.
A nonlinear case in which lumped nonlinearly modelled bipolar transistors are
interconnected by reglines is treated in Section 5.3.
The framing of our study in abundant engineering and mathematical literature
devoted to the stability and delay time problem is analysed in Section 5.4.
We deal with the dynamic process of the general network from Figure 4.5, after
switching the constant sources at t = O. So we have the problem (P(B,vo» =
(E) + (Be) + (Ie) from the previous chapter, which we rewrite here:
(E)
_~ O~n(t,O) vnCt,O)
(Be) rn ox =G +B, t ~ 0,
1 OV n
+T(t,dn) vn(t, dn )
rn x
dV n +l ( ) Vn+l (t)
Sl~t
Above, B is a constant vector obtained from the constant value of sources. The
steady state to which the transient regime tends, is described by
(SBC)
o
o
To facilitate the writing below, we shall consider even the capacitor voltages as
space dependent functions:
We shall suppose throughout this chapter that the problem (P(B, vo» as well
as (SP(B» has a unique solution in the classical sense. We have denoted these
solutions by v = (VI, .•. , v n + m ) and by v = (VI,"" vn + m ) respectively. Hence,
our assumptions in what follows must be consistent with existence and uniqueness
conditions (see Theorem 4.2 and Theorem 4.4).
The global dynamic behaviour of our network can be described by a function
D : [0,00[+ R, named "delay" and defined by
When the asymptotic stability conditions are fulfilled, the delay variation begins
from 1 (corresponding to the initial conditions) and tends to 0 (corresponding to
the steady state) when time indefinitelly grows. If we fix A E (0,1), the speed of
this evolution (and therefore the speed of signal propagation in our network) can
be expressed by the last moment when the delay equals value A. In this way, we
are conducted to define "the (global) A delaytime" as
1 aUI uI(t,dd
+(t,dt)
rl ay
1 aUI UI(t,O)
(t,O}
rl ay
1 aU2 U2(t,0)
(t,O}
r2 ay
1 aU2
+(t,d2} u2(t,d2)
r2 ay
(Bel) =G +B, t ~ 0,
_~ OUn(t,o} un(t,O}
rn ay
1 au n ( Un(t, d n )
+  8 t,dn)
rn Y
dUn+1
Sl~ Un+l(t)
dU n+m
Sm~ un+m(t}
102 V Asymptotic behaviour
{
Uk(O, Yk) = Vk,O(Xk), Yk EjO, dk[, k = r,n
Un+k(O) = Vn+k,O, k = 1,m.
Similarly, if we denote the new voltages in the direct current problem by
ih(yd = vl(d  Yl) = Vl(Xl), U2(Y2) = V2(X2), ... un+m(Yn+m) = vn+m(Xn+m),
we obtain new boundary conditions (SBC l ) which, confronted by (SBC), and have
the same differences as (BC I ) confronted by (BC).
The change of variables we made, gives
max IUk(t,Yk)  Uk(Yk)1 = max IVk(t,Xk)  Vk(Xk)1
y, E[a,d.] x. E[a,d.]
for k = 1, n + m and all t > 0. This shows that the delay (and consequently, the
delay time) is invariant at lineterminals inversion.
Now, we shall try to put the problem (PI(B, va»
= (El) + (Bel) + (leI) in
a similar form to problem (P( B, vo».
To this goal let us denote by M Ell N the
matrix [~ ~], where M and N are matrices even with different dimensions.
Let h be the unity matrix with the dimension k, and for every k = r,n we denote
P;' = { I 2k  2 Ell [~ ~] Ell I2n+m2k when ak = 1
hn+m when ak = 0.
_~ B~n(t,O) un(t,O)
rn By = Pl]GPl +plB.
~ Du n
B (t, d)
n un(t, dn)
rn Y
dUn+l
s]~ Un+l(t)
d~n+m
Sm~ Un+m(t)
5.2. Asymptotic stability 103
Therefore, the problem (EI) + (Bel) + (leI) has the same form as the problem
(E) + (Be) + (Ie), in which the matrix G is replaced by PIGPI and the vector
B by PI B. Because the delay time is the same for the two problems, when we
compute the upper bounds of this parameter (depending on G and B with respect
to PIGPI and PI B) clearly we must take the minimum value to get the best upper
bound.
Extending the above arguments, the delay time remains unchanged if in the
problem (E) + (Be) + (Ie) we replace the matrix G with the matrix
G" = p:;n P;::'l' ... P:'GP:' ... P;::'l' p:;n
and the vector B with the vector
B" = p;:n P:~l' ... P;' B ,
where we have denoted 0" = (0"1, 0"2, ... , O"n). Depending on the values of 0", this
new problem comprises of all possibilities for lineterminals inversion (for the above
example 0" = (1,0,0, ... ,0)). This is why throughout the following we shall consider
our problem with the matrix G" (whose elements are Gfj) and with the vector B"
(with elements br). For each 0" we shall find an upper bound 1\(0") of the delay
time, after which we shall minimize this value with respect to 0" i.e. with respect to
the 2n possibilities appearing when we interchange the lineterminals. Therefore
1\. = min1\(0") . (5.1)
"
5.2. Asymptotic stability. Upper bound of delay time
For reasons that will become clear below, in our problem (E) + (Be) + (IC)
with G" and B", we shall make a change of functions
a"x
{ Vk(t,X) = Uk(t,X)cos :k ' (5.2)
vkCt,X) = Uk(t,X), k =n+1,n+m,
where a~ will be conveniently chosen. Also, we shall extract the time derivatives
from the boundary conditions and attach them to the system; so we shall obtain
for t :2: 0
aUk 1 a2Uk [2a k akx] aUk
7ft = rkCk 8x 2  rkCkd k tan d; Tx

(ak)2 + 9k] Uk, x EjO,d [, k = 1,n
[ :i'T k
rkCkak Ck
8
;t =  Sk
1
n
L G~+k.n+ jU j(t, 0) + L
[n+m
j=n+l j=l
n
G~+k.2j1 Uj( t, 0)+
(5.3)
_~ aUl(t,O) UI(t,O) 0
rl dx UI(t,ddcosaf
1 aUl(t,dI) iT al Ul (t, dd. iT
rl ax cos a 1 rl 1
d smal
un(t,O)
=cr
un(t,dn)cosa~ + +F,
Un+1(t,O) 0
anun(t, dn). iT
un+m(t,O) d sma n
rn n
(5.4)
where cr and lr
are formed with the first 2n rows from GiT and BiT respectively.
The initial conditions become:
aiTx
{ Uk(O,X)=VkO(X)/cos ;k'
(5.5)
Uk(O, x) = VkO(X), x E]O, d k [, d k = 0, k = n + 1, n + m .
n+m
We denote Y = II [0, d;] and define on the Banach space C(Y; Rn+m) the subset:
;=1
fICO)
fI (dd cos af
D(A) = =cr
du = Au
{ dt (5.6)
u(O,') = Uo = a function with components given by (5.5).
The following lemma is essential for deriving our result. According to (4.5) we
shall denote Sr
= E~::tj;.!i IGfil·
Lemma 5.1. Let us suppose that Gis DRD (see (4.5)), and for every j = r;n let
us consider 'Y'! E]O, 11" /2[ such that
If we choose oj = 'Yi  e where e > 0 is such that oj E]O, 11" /2[ and if
where w = J 1 and
M(w) = {(p,y) I 1 ~l~n+m
max max IWi(X)1 = Iwp(y)1 = Ilwll}.
O~x~dj
(5.10)
Let us consider, already proven, that if (p, y) E M( w) and p = r;n:, then y E]O, dp[.
If for p = r;n: we suppose wp(y) ~ 0, (5.10) shows that y is a maximum point
of wp in ]0, dp[, i.e. d: (y) = 0 and ~;p (y) ::; 0 and then
If for p = r;n: we suppose wp(y) < 0 where (p, y) E M( w), then y is a minimum
point of wp in jO,dp[ and (5.11) is valid again, as we can easily observe.
Finally, if (p, y) E M(w) and p = n + 1, n + m (that means y = 0) then
IWp(O)1 = Iwp(y)1 ~ IWi(X)1 for any i = 1, n + m and any x E [OA]. It follows
 sgn wp(y) [ "
[AJ  AJ]p(y)sgnwp(Y) =  Gn+p,n+pwp(y)+
Spn
n+m n n ]
n m
 LG~P_1,2jWj(dj)wp(0)cosaj  LG~p_1,2n+jWn+j(0)Wp(0).
j=1 j=1
5.2. Asymptotic stability 107
But, (p,y) E M(w) implies IWp(O)1 ~ IWi(Xi)1 for all i = l,n+m and Xi E [O,di]'
and therefore
n m
+ L IG~p_l,2jllwp(0)12 + L IG~p_l,2n+jllwp(0)12
;=1 ;=1
i.e.
1 dwp(O)
;:~Wp(O) :::; (G2p
tT
1 ,2pl + S2p_l )1 Wp ()12
tT
0 .
p
But, as we can easily observe, the DRD property given for G, implies (in fact is
equivalent to) the same property for GtT. That is why the last inequality gives
(5.13)
If wp(O) < 0, then wp(O) ~ Wi (X ) which again implies (5.14). But (5.14) contradicts
(5.13) and this means that the initial assumption is false, i.e. y I O.
Now, let us suppose that p = r;n, y = dp and (p,y) E M(w). By utilizing even
rows in the definition of V(A) we find as above:
1 dWp(d p) (d)
t
dwp p cosaptT
rp x
which implies
 (G 2p,2p
~ d cos 2 Qp~ + S~2pCOSQp~ + 1d < O.
+ 1)
rp p rp p
From here, with inequality 01; < tan 01; valid for 01; EJO, H we derive
+ S~2p +  d
Q~
 G 2p,
~ ~
2 pCOSQp
p •
SIllQ p <
~ 0.
rp p
1 dw (d )
ppw (d )COSQ~ < 0 . (5.16)
rp dx p p p
On the other hand, if wp(dp) ~ 0, then wp(dp) ~ Wi(X) for any i = 1, n +m and
x E [0, d;J. Thus,
1 w(d)w(d x)
lim p p p P wp(dp) cos 01" ~ 0. (5.17)
rp xo+ X p
converts the original steady state problem (S E) + (S BC) into the abstract equation
Ail = 0 on the space C(Y; Rn+m). This equation combined with (5.6) gives
~(u  u) = Au  Au (5.19)
{ dt
(u  u)(O) = uo  U .
on [0,00[, where the C(Yj Rn+m) norm was considered. Solving this differential
inequality we obtain
,"  f
m.in cos,i ~ cos 'dXi ~ 1, for all Xi E [0, d i ],
l~l~n i
we obtain
"
Wo = max { max [Clj)2
     gj] j max (Gn+j,n+j+Sn+ j )} (5.21)
l~}~n clJrjCj Cj n+l~}~n+m Sjn
we get
_ ew~.t
D(t) ~ Du(t) = . u (5.22)
mml~i~n cos ' i
for all a and t 2:: O.
Since from (5.20) we see that D is strictly decreasing (w~ < 0), we find T>. =
Dl(A). Also if we define
Theorem 5.1. Let us consider the mixed type network from Figure 4.5 with a
resistive multiport of (G, B, 2n + m )type where G has the D RD property. Then,
i) the delay D: [0, =[> R is a strictly decreasing function,
ii) the direct current solution of our problem is globally exponential asymp
totically stable, and
iii) the upper bound of the Adelay time is
In A min cos ,i
T>. = min 1~I~n (5.23)
u Wo
where Wo and cos ,i are given by (5.21) and (5.7) respectively.
110 V Asymptotic behaviour
The second statement above assures that, regardless of initial conditions, all
dynamic solutions in the classical sense tend in C(Y; Rn+m) to the same constant
value, after the simultaneous connection of constant sources. In fact, in the same
way we can obtain the stability of any solution (Marinov, Neittaanmiiki [1988]) and
its boundedness as well.
The third result above is an upper bound for ,xdelay time. The formula (5.23)
implies 2n times application of relations (5.21) and (5.7), which involve all param
eters of the circuit: rj,cj,gj,dj,sj and Gjj. The simplicity of calculus makes this
formula proper for fast simulators, used in digital network design. Of course, it is
necessary that the upper bound is tight enough. This fact will be verified by the
numerical computation of T).. in Chapter 6 where several examples will be given.
As we have seen above, the essential tool for delay time evaluation in linear
mixedtype networks was the dissipativity of the operator governing the dynamic
evolution. But the linearity of this operator plays no part in mathematical rea
sonings. From here derives the idea to extend the above approach to a nonlinear
case.
Let us now consider the general network from Figure 5.1, where a (G, B, 2p+2n)
type resistive multiport connects p bipolar transistors T1  Tp and n distributed
parameter elements ("rcglines"), L1  Ln.
For the transistors we shall consider the nonlinear Gummelmodel (presented in
Figure 6.3) and reproduced in Figure 5.2 to specify some different notations:
According to physical reality, the transistor model contains the functions hk1,
hk : R  t R with strictly positive derivatives and six strictly positive parameters
112k1, 112k, S2kl, S2k, T3k1, T2k· From Figure 5.2 we easily derive
(5.24)
(5.25)
T T
~y ~:r
il i2
+ ZI  Z2 +
jl + VI  V2 + j2
1
Resistive
multiport
j == Gv + B
. ( )+
I)p+n t,O 
(+i 2P+ n (t, d)
n
 iu2p+n(~&+n t, d"
L"
Figure 5.1 The network under study
rI_.~......_{_.~..,
ie = hkl(Z2kt)
.
T2kJ . T2k .
le '1,(,
s2kJ 82,..
+ +
Figure 5.2 Large signal model of the kth transistor
112 V Asymptotic behaviour
_~ OU2~+n(t,0) U2p+n( t, 0)
rn Ox
+~ OU2p+n (t, dn) 0
rn ox U2p+n(t, dn)
for i = 1, 2p, where tik are elements of T. Our result will be derived supposing that
the parameters satisfy:
For each i = 1,2p, Li < 0
Al {
For each i = 2p + 1, 2p + n, Gji + Si < 0 .
5.3. A nonlinear circuit 113
Sip+2i + r~i
(G2P+2i,2P+2i + ri~J
COS'Yi= ~ __~__ (5.27)
2 ( G2p+ 2 i,2p+2i + ri~i)
Also, for every E ~ ° with 'Yi  f EjO, 71' /2[, we denote
As in the preceding section, we shall consider all voltages as time and space variables
Ui :
2p+n
°
[0, oo[ x [0, dj  2p j ...... R for i = 1, 2p + n where di  2p = for i = 1, 2p. Let Y =
II [0, di  2p j. We suppose that for Uo E C(Y; R2p+n) with Uk,O E C 2 (]0, dk[, R)
1=1
there exists a solution in the classical sense of (E) + (BC) + (IC), U E C(Y; R2p+n)
with Uk(t) E C2(jO,dk [,R) for t 2:: 0. Also, we suppose the existence of a steady
state solution il = (ill, ... , il 2p+ n )tr E C(Y; R2p+n) where ilk E C 2 (]0, dk[, R).
If we define, as in Section 5.1, "the delay" as D: [0,00[> R, D(t) = lIu(t)illi/
lIuo  illl where the norm of space C(Y; R2p+n) was used, then we can prove the
following result:
Theorem 5.2. Suppose that for the problem (E) + (BC) + (IC), Ui,O oF iii for
every i = 1,2p + n, and that Al is valid. Then lim Ui(t, x) = ii;(x) uniformly in
too
x for i = 1, 2p + n. Moreover, the delay is a strictly decreasing function and
Proof. The method of proof will be the same as above: we formulate the problem
as a Cauchy one in C(Y; R2 p+n) and the assymptotic stability will be implied by a
dissipative property.
First of all, a change of functions appearing in (E) + (BC) + (IC):
Uk(t,X) = Wk(t, x), k=1,2p
{ cos f3 k  2p X n:..., (5.30)
Uk (
t, )
x = Wk (t, )
x d , k = 2p + 1, 2p + n ,
k2p
114 V Asymptotic behaviour
W2p+n(0)
W2p+n( dn ) cos /3n
+
o
(AWh(x) h~I(Wl(Xt})
{ [
= diag(8~1, ... , 8 2p1 ) _G ll :
h 2:(W2P(X2p»
W2p+l(0) ft(h~I(Wl(xd»
W2p+l (d 1 ) cos /31
_G 12 T
W2p+n(0)
W2p+n( d n ) cos /3n hp(h2: (W2p(X2p»)
5.3. Comments 115
It is straightforward to show that (E) + (BC) + (IC) together with (5.30) give
the following differential equation in C(Yj R2p+n):
dw(t, .) = Aw(t .)
dt '
{ for k = 1,2p (5.31)
,,()/'~ ~ 2~) ~
Uk,O
Wk(O, .) = fh,2 . .
{ •• d. foc k 2.+1,2.+ n ,
Let us take w, wE 'D(A) with 1/J = w  wand
m!lX
2p+1EO;.EO;2p+n xE[O,di_2,]
max l1/Ji(X)1 = l1/Jq(Y)1 = 111/J1I} .
For (qj y) E M(1/J) and q = 1, 2p we obtain
[Aw  AW)q(y)sgn1/Jq(Y) ~
Tk
L IG qkll1/Jq(Y)1 ,
2 +2n] P
where we have applied the mean value theorem and denoted by IJ. the (positive)
derivative of function Ik in the intermediate point from jhk"I(Wk)' hk"l(Wk)[' So,
for q = 1, 2p we have
(5.32)
For (qj y) E M(1/J) with q = 2p + 1, 2p + n, we choose f3j = 'Yj  f where 'Yj is given
by (5.27) and f > 0 is such that 'Yj  f EjO, 7r /2[. Reasoning as in the proof of
Lemma 5.1 (we omit the details) we obtain
Now, (5.32) and (5.33) give the total dissipativity of A (see Lemma 1.6) for all
above chosen e:
(AwAw,ww)+ ~w.llwwli.
If we remark that Al implies w. < 0, from here we derive the result. o
If we adopt for the Adelay time the same definition as in preceding section, from
(5.29) we derive
T>. ~ '1.\ = (In A m.in cos 'Yi)/WO , (5.34)
lEO;.EO;n
where cos 'Yi and Wo are easily computable (see (5.27) and (5.28».
116 V Asymptotic behaviour
5.4. Comments
The stability of the circuits with distributed structures has been studied by
many authors using different methods. Prada and Bickart [1971] use a Lyapunov
theory for a functionaldifferential equation of retarded type that describes a large
class of such circuits. A small signal stability criterion (in terms of the roots of
characteristic equations) is derived in Brayton [1968] while inputoutput stability
results are given in Desoer [1977] for circuits with parasitic elements, by applying
the theory of singular perturbations.
The extensive bibliographies of Ghausi and Kelly [1968] and Kumar [1980] sum
marize the work on distributed recircuits up to 1980. We have to remark that
for a nonuniform open circuited reline, Protonotarios and Wing [1967] show the
step response to be monotonously increasing and Singhal and Vlach [1972] obtain
bounds of this response. As regards the delay time, a pioneering work in this field
is Elmore [1948] who called the first moment of the impulse response the delay.
The explosive increase in the work on transmission lines in the past decade has
been mainly motivated by the preoccuppation with the delay time in MOS inter
connections (see Section 4.0) and for the performances of microwave transmissions.
Some authors have worked on the transient analysis of a single transmission line
giving exact analytical time domain expressions for voltage and current at any
point on the line: Cases and Quinn [1980], Preis and Shlager [1988], Zurada and
Liu [1987].
Many other authors (Gao et al. [1990], Passlack et al. [1990]' Araki and Naito
[1985]) tried to implement a transmission line model in a general purpose CAD
circuit simulator such as SPICE. But, for delay time prediction at the design stage,
a much faster simulator is needed (see Section 6.0). To this goal, one of the most
accepted methods is to use a very simple RC lumped model of the whole network as
a basis to infer easily computable bounds of the delay time. The first result in this
direction is due to Rubinstein, Penfield and Horowitz [1983] on RC "tree" networks.
Extensions of these bounds to nonlinear RC networks and to RC mesh networks
were given by Wyatt [1985 a,b]. An interesting extension of Elmore's delay to RC
networks was found by Chan [1986 a,b]. Bounds which can be improved iteratively
were developed by Zukowski [1986 a,b] while the delay time sensitivity is treated in
Jain et all. [1987]. Also, RC mesh type circuits have been studied by Lin and Mead
[1984], Chan and Schlag [1989], Harbour and Drake [1989]. Relating to all these
papers, we observe that the accuracy of approximation of interconnections by RC
ladder networks is not clear, being studied only for one line: Sakurai [1983]. For
instance, for a open circuited reline of length d, the rise time is red 2 , while for a
RC cell with R = rd and C = ed the rise time is 2.3 red2 • That is why our bound
of the delay time in a network with exactly modelled reglines is probably welcome.
Of course, a lower bound of the same type is desirable. We have done it recently
5.4. Comments 117
but working in a completely different way, Marinov and Neittaanmiiki [1991 a,b].
On the other hand, we have to observe that the delay time notion in all the
above papers is related to a given inputoutput path, while our delay time is a
global one, describing the speed of signal propagation after switching of a part or
all sources. Of course, one of our bound shortcomings is the apriori necessity to
describe the resistive part of the network as a (G, B, 2n + m) multiport.
Finally, let us mention that the results of this chapter are obtained in Marinov
[1987], Marinov and Neittaanmiiki [1986, 1988, 1989, 1990 b]. Another nonlinear
case can be found in Marinov and Neittaanmiiki [1990 a].
Chapter VI
6.0. Introduction
118
6.1. The mathematical model 119
Let us consider again the general mixedtype circuit of Figure 4.5 which has n
lines and m capacitors connected to a resistive multi port. The mathematical model
comprises of
 a system of parabolic equations for the voltage v = (VI, ... , V n ) on lines
1 aVl (t, 0)
 rl ax
1 aVl(t,dl)
rl ax
1 a~n(t, 0) Vn(t,O)
 rn ax =G
vn(t, dn )
+ B(t), t ~ 0, (6.2)
~ avn(t,dn) Vn+l(t)
rn ax
i2n+l(t)
k = 1, ... ,n
(6.3)
= Vn+k,O, k = 1, ... ,m.
In general, the current, the voltage and the electric charge of the kth capacitor
are related by
= ffAP(qk(t))
_ dqk(t) (6.4)
di
Taking into account that our theory in preceeding chapters was given for the linear
case, we shall describe below only this case, namely
(6.5)
120 VI Num. approx. of mixed models for digital into circuits
where Sk is capacitance. (Let us mention that our simulator is more general, and
also works for the nonlinear case.) It follows
. ( ) dVn+k ( )
Z2n+k t = Sk~ t (6.6)
(6.7)
(6.8)
In (6.7) and (6.8) we have used the following notations: G j denotes the jth row
of the matrix G of boundary conditions, Bj(t) denotes the jth component of the
vector B(t) and the voltages at the end points of lines and in capacitors are denoted
by
u(t) = (Ul(t), ... , U2n+m(t»tr
G = (Gij);,j!;"
G(2n),(2n) G(2n),(2n+m) ]
11 1,(2n+l)
(6.10)
[
= G(2n+m),(2n) G(2n+m),(2n+m)
(2n+l),1 (2n+l),(2n+I)
with blocks
G. ij
G~ = [ (6.11)
Gkj
6.2. Construction of the system of FEMequations 121
The system of the differential equations in time is obtained from the partial dif
ferential equations of lines and the differential equations of the boundary conditions
by using the following principles:
 FEMmodel is obtained by discretizing the lines in space by using the
finite element method. If line k (k = 1, ... , n) has N /.: discretization
n
intervals (N/.: +1 discretization points) we get 2: (N/.: + 1) differential
"=1
equations
 The differential equations (6.8) describing capacitors are added to the
FEMmodel for lines. Consequently, the problem (6.1)(6.3) leads to an
n
initial value problem with 2: (N/.: + 1) + m differential equations.
/.:=1
6.2.1. Space discretization of rcglines
Let us take the ith (i = 0, ... , N,,) basis function of line k (k 1, ... ,n),
tPi: [0, dkJ
1+ R defined by (see Figure 6.1)
0, elsewhere
which is used in a space discretization of the partial differential equations of lines.
tPo
°
Xil Xi Xi+l
~
t!.h.
Figure 6.1
122 VI Num. approx. of mixed models for digital into circuits
the length of the discretization interval. So the values of basis functions at the
discretization points are as follows
if Z = J
if i # j
If we multiply the partial differential equation (6.1) with the basis function <Pi( x)
and integrate over line k we get
(6.14)
(6.15)
if i = 0
if i = Nk .
(6.16)
6.2. Construction of the system of FEMequations 123
As we see, tPi joins the boundary conditions to the differential equations only at
the first (start of line) and the last (end of line) discretization points of each line.
We approximate the voltage Vk(t,X) of the line k by
N.
Vh,k(t, x) =L V(k(t)tPj(x) (6.17)
j=O
where vi k(t) is the voltage at the jth discretization point Xj and at time t. This
approxi~ation is of order O(h~) (see KflzekNeittaanmiiki [1990], for example).
If we replace in (6.15) v by Vh we get
If we calculate the integrals afj and bfj in equation (6.18) for all combinations of i
and j we get for line k
2 1
1 4 1
J
~ k N. flhk
Ak = (a 1].. ).1,]=
. 0=  6 (6.19)
1 4
1
and
1
1 2 1
~ k N 1 [ 1
Bk = (b ij )i.1=O = flhk (6.20)
1 2 1
1 1
(6.22)
(6.23)
and
(6.24)
The vector VL(t) contains the values of voltages vk(t,j~hk) (j = 0, ... Nk and k =
1, ... , n) at all discretization points of lines at time t and its size is also N(n) and
VOL = (VOl, ",VON(n» is the pointwise initial condition corresponding to condition
(6.3). The square matrix Ch (size N(n) x N(n» has the following form
. GL. L
In
1 (6.26)
.. : G~';.
where GrjL (size (Ni + 1) x (Nj + 1» is a submatrix of the form
G(2iI),(2jl) 0 0 G(2iI),(2j)
0 0 0 0
G,;!=..!..
IJ Ci
(6.27)
0 0 0 0
G(2i),(2jl) 0 0 G(2i),(2j)
The corner terms of GrjL correspond to the start and the end points of the lines and
they are elements of matrix G.
6.3. FEMequations of the model 125

Bdt) = (BI B2 ... ,
,0, ... ,0,,
CI CI
B2k 1 B2k
,0, ... ,0,, ... ,
Ck
.
line k
Ck (6.28)
Cn
° °
B 2n  1 , , ... , , B2n)tr
Cn
Let us join the capacitors equations (6.8) to the system of FEM equations of
lines (6.21 1 ). We get
(6.29I)
YeO) = Vo
OCAP (size m) is a zero square matric. The square matrix G can be written as
(6.30)
GLL
nl GLL
nn GLC
nl GLe
nm
GeL
11
GeL
In Gec
11
Gee
1m
GeL
ml
GeL
mn
Gee
ml
Gee
mm
where GLe , GeL and Gce come from the boundary condition terms related to ca
pacitors.
Let us first consider submatrices which come from the boundary conditions of
the lines. Line connections are the same as for matrix GrJ in (6.27). We can
126 VI Num. approx. of mixed models for digital into circuits
write lines connection with the capacitors (a vertical vector whose size is Ni + 1;
i = 1, ... , n and j = 1, ... , m)
(6.31)
Also, the terms corresponding to the connections between capacitors and lines
(a horizontal vector whose size is Nj + 1; i = 1, ... , m and j = 1, ... , n) are grouped
together in
GfjL = (G(2n+i),(2jI), 0, ... ,0, G(2n+i),(2j») (6.32)
Finally, a block corresponds to capacitorcapacitor (one term; 1, ... ,mand
j = 1, ... , m) connection is defined by
(6.33)
We can write the source vector (size N(n) + m), see (6.28))

B(t) =
BI
( ,0, ... ,0,, B 2n 1
B2 ... , ,0, B 2n
... ,0, ,
CI CI Cn Cn
, .,. .J
= (VI (t, 0), ... , VI (t,jl6.h l ), ... , VI (t, dJ), ... , (6.35)
vn(t, 0), ... , v n(t,jn6.h n ), ... , vn(t, d n),
tr
Vn+l(t ) , ... ,vn+m(t) ) ,
When the FEM is applied to the model (6.1), (6.2), (6.3), we obtained the
initial value problem as described above. There arc several integrators available
6.4. Residual evaluations 127
for solving the numerically obtained problem (in the simplest case we could apply
implicit Euler or CrankNicholson methods with a fixed time step). We have applied
the subroutine D02NGF of NAG Fortran Library. D02NGF is a general purpose
routine for integrating the initial value problems of stiff systems of implicit ordinary
differential equations
A(t,y)y' = g(t,y) (6.36)
and it has been developed from SPRINT package (Software for problems in time,
Brezins, Dew, Furzeland [1989]).
Because in the case of this chapter, A( t, y) does not depend on t and y, one could
use the integrators designed for solving stiff systems of explicitely defined ordinary
differential equations
(6.37)
Consequently, other subroutine like D02NBF and D02NDF of NAG could as well
be applied, instead of D02NGF. In Appendix I we have outlined in detail the basic
steps on how to solve the obtained initial value problem by integrator D02NGF.
The usersupplied RESID routine of D02NGF defines the system of differential
algebraic equations to be solved. The integrator supplies approximate vectors for
the solution y and its time derivative y'. The main purpose of the RESID routine
is to compute the residual vector r
r=g(t,y)A(t,y)y' . (6.38)
with the initial condition V(O) = Vo. The vector R(t) is the value of the residual
at the discretization points and at time t.
Let us note by N ( k) the number of the discretization points from the first line to
line k (N(O) = 0) and use the abbreviations (see (6.18), (6.19) and (6.22)  (6.24))
~ 2 4t:.h
kg k
ak=
t:.hkTkCk
+
6Ck
bk = 1 + t:.hkgk (6.40)
t:.hkTkCk 6 Ck
~ t:.hk
Ck = 6
128 VI Num. approx. of mixed models for digital into circuits
We can write the matrix equation of residuals (6.39) component wise and by doing
this we have for line k (k = 1, ... ,n):
 a residual equation 1= N(k  1) + 1 for the start point
RI(t) =  2Ci, Vi'(t)  Ck Vi'tl (t)
ak ~
 '2Vi(t)  bkViH(t)
1
 ~L
n
(G(2kl,2il) VN(il)+l(t) + G(2kl,2i) VN(i)(t»)
k i=l (6.41 )
(6.43)
t i=l
(G(2nH,2i!)VN(il)+1(t) + G(2nH,2i)VN(i)(t»)
m
(6.44)
i=l
6.7. Examples 129
The discrete steady state model for (6.1), (6.2) and (6.3) reads
(6.45)
where V eo denotes the steady state solution. This has been obtained by leaving out
the time derivative terms from the model of the circuit (6.29). We calculate this
steady state solution by using subroutine C05PCF from NAG Fortran Library.
We can calculate the apriori upper bound for delay time according to relation
(5.23). The delay in the discretized model is obtained by the formula
(6.46)
gives voltages of the FEM model at starting time t = 0 and we can see that
o ::; D(t) ::; 1 for every time t.
Now let us give the formula of the Adelay time for the FEM model
where D(t) is given by (6.46). The Adelay time gives the last time when the
difference between the discretized dynamical model and the discretized steady state
model has dropped to a Ath of the original difference.
6.7 Examples
The purpose of this section is to illustrate the usage of the proposed numerical
method for solving some typical mixed circuits. At the same time, we shall numeri
cally compute the delay time given by (6.47) in order to compare it with its apriori
upper bound derived in Chapter V.
130 VI Num. approx. of mixed models for digital into circuits
,A ,A
IA.
,
I
I I
E! I
,...,,
I
I I
!~
0+,... I
I I
,
I I
E: ,
I
__________ +____ I
..E
e~.~L ~
Output
o_J
I
L_J~~d___ _
Example 6.1. The circuit shown in Figure 6.2 is a multiplexer realised with NMOS
pass transistors.
We want to study the effcct of interconnection lines LI and Lz on the signal
delay after step sources (with e = 1 value) are connected. We suppose that during
the transient process the pulses A and A, applied to the address lines, maintain the
transistors TI , T z , Ts and T7 in the ON state (having the drainsource resistance
R1 = 1) and the transistors T3, T4 , T6 in the OFF state (R z = 10). The parameters
of the lines are: 7"1 = I, CI = 10, 91 = 0.1, d 1 = 1 for the first line L1 and 7"z = I,
C2 = I, 92 = 0.1, d 2 = 1 for the second line L 2 • If the output load is modelled by
s = 1 and R3 = 10, we obtain the mixed type circuit from Figure 6.3:
In Figure 6.4 we present the same circuit with the resistive part grouped in a
multiport.
With notations in figure (compatible with the general case in Figure 4.5), Kirch
hoff's first and second law give, respectively
i~ + i~  h = 0; i; + i~ ]4 = 0;
i~ + i;  i~ = 0; i~ + i~ + ]s = 0,
and
Ll RI
T =
R2
HI
I
H3
L2 Hz
T
=
Hz
=
Figure 6.3 NMOS multiplexer, Equivalent network
L,
r  1
I + WI  Wz + + W3  W4 + I
I I
1 iI iz }3 j, }s I
1 .. I
1 I
R3 wsl
'.
1
1 R, R, R, R, " 1
I
i'3
, I
"
"
I
"
1 " "
I e R, + R, R,
I "
1 's
I
I
I
I I
L _________________________________ ~
Figure 6.4 Equivalent network with resistive part viewed as a multi port.
By the elimination of i~ i~ we obtain a linear relation between j = (it ,h, h, j4,jS)
and w = (Wl,W2,W3,W4,WS) of the form j = Gw + B where
G1 0 0 0 0
0
Gi + 3G 1 G 2 + G~ 0
G 1 G 2 G 21 
2G 1 + G 2 2G 1 + G 2 2G 1 + G 2
0 0 G1 0 0
G 1 G 2 4GIG2+G~ G 1 G 2
0 0
2G 1 + G 2 2G 1 + G 2 2G 1 + G 2
Gi G 1 G 2 Gi + G 1 G 2 + 2G 1 G 3 + G 2 G3
0 0
2G 1 + G2 2G 1 + G 2 2G 1 + G 2
and B (eG 1 , 0, eG 1 , 0, oyr, Here we have put Gi = 1/ R i. The initial values
132 VI Num. approx. of mixed models for digital into circuits
are all zero: Vl,O(X) = V2,O(X) = V3,O = O. We choose for the two lines the same
space discretization step: t:,h 1 = t:,h 2 = dr/20 = d2/20 = 0.05, while for the
273 time steps the minimum and maximum time steps are 2.155 x 10 6 and 3.547
respectively, in NAG subroutine D02NGF.
The Figs 6.5, 6.6 and 6.7 show the voltage variation along lines and on the
capacitor, respectively.
Line 2
t = 97.1
1 = 97.1
t=4
t = lG t=3
t= 2
t = 12
t= S 1=1
1=4
1= n
Lt'Ilgth Length
Capacitor 1
Steady state
O.5G +::::;::;oo~
o "rrr""T"""?
Time 97
Figure 6.7 V3(t)
6.7. Examples 133
As we see, all components of the solution tend to the steady state. Because G
has the DRD property, this is the confirmation of the global stability result from
Theorem 5.1 ii). The speed of the evolution to the steady state differs from one
element to the other. So, from Figures 6.5 and 6.6 we observe that the voltage along
the first line varies (approximately) ten times slower than the second line voltage.
That is in accordance with the "engineering feeling", because RI C I = 10R2 C2 ,
where RI = rId}, R2 = r2d2, C I = cldl , C2 = C2d2 are the "lumped equivalent"
resistances and capacitances of the lines.
A good image of the global speed of evolution of the circuit is given by the delay
D(t) from Fig 6.8. This gives the delay time values: T o.s = 6.11 sand T O•I = 21.3 s.
Time 97
Figure 6.8 The delay
In the same picture we see the upper bound D(1,o)(t) of the delay. For its calcu
lation, the program uses the formulae (5.7), (5.21) and (5.22) with the intermediate
results from Table 6.1.
Table 6.1
134 VI Num. approx. of mixed models for digital into circuits
According to (5.23) we find 1'0.5 = 14.5 and 1'0.1 = 37.0, for (f = (1,0), which
are very close to the "exact" values To.5 and TO.I from above.
T
R, R, R,
r
1 8J
1
R, H, R, R, R,
E. R, R, R,
1 8
•
1 s.
The voltage values of inputs are EI = E2 = 1 and the lumped elements have
the values RI = 1/2, R2 = 1, 81 = 3, S2 = 3/2, 83 = 3 and 84 = 3. The four
rcglines have the same length d = 1, the same resistance (per unit length) r = 1,
the same conductance g = 0 and CI = 3/2, C2 = 1, C3 = 5/4, C4 = 4/3. All
dimensions are coherent. This results in a 12 x 12 symmetric G matrix whose
diagonal elements are G ll = 2, G22 = G33 = G99 = 3/2, G44 = G55 = G71 =
G10,1O = 14/9, G 66 = G 88 = Gll,ll = G12 ,12 = 6/5 and the nondiagonal, nonzero
ones are G23 = G32 = G29 = G92 = G39 = G93 = 1/2, G45 = G54 = G47 =
G74 = G4,10 = G IO ,4 = G57 = G75 = G5,10 = GIO ,5 = G7,10 = G IO ,7 = 4/9,
G6,12 = G12 ,6 = G 8 ,1l = Gll ,8 = 4/5. The nonzero elements of Bare Bl = 2 and
B2 = B3 = B9 = 1/2. The initial state is zero for all lines and all capacitors.
The choosen space step is D.h = d/IO and the 229 time steps vary between
5.75 x 10 6 and 9.37 x 10 1 • We shall reproduce here, in pictures, only some of the
results. Figures 6.10 and 6.11 show the voltage along the first and the fourth lines
respectively for different moments.
Analogously, Figures 6.12 and 6.13 show the voltage evolution for the first and
the fourth capacitor, respectively.
6.7. Examples 135
. 1~!!~~~~~~~~~
O 97 Steady statc 0.17
2~.~
tt == 2.·)
t = 1 I)
t = :.0
:;.I:
 t = (1.2) 1 = 24.7
1=5
t = 3.7.'i
1=0 t = 2.5
1 = 1.2.j
_1=0
OL"r'~ () Lr.rr~
Length LengIh
OLr._r_ _.....~
Timf' 25 Time 25
First of all, we remark that all state values advance to a steady state, as our
theoretical result predicted. Then we find that it is a process of propagation of
signals from inputs to outputs. So, while for the line 1 (which is near to inputs)
the steady state has the values Vl,oo(O) = 0.966 and vl,oo(d) = 0.899, for the line
4 (which is close to output) these values are V4,oo(0) = 0.166 and v4,oo(d) = 0.099.
The same happens with capacitors 81 and 84 (0.866 compared to 0.066). The above
pictures also clearly show that the elements which are close to sources (inputs) tend
to the steady state more quickly than those that are far away, near to outputs. The
global switching speed of the circuit is expressed by the moment when the delay
D(t) (see Figure 6.14) pass through an imposed A value. This is the Adelay time
and in our case the numerical integration gives To.s = 1.96 and TO.l = 6.38.
With regards the upper bound of the delay time, this was found for the best
136 VI Num. approx. of mixed models for digital into circuits
Time 25
Figure 6.14 The delay
permutation a = (0,1,0,0), giving To.s = 9.3 and TO.I = 18.4. Again the tightness
of these bounds is reasonable.
Example 6.3. Now let us consider a mesh network. This kind of network arises
in models for the gates of large transistors (used in the final stages of clock or pad
drivers) or for CMOS transmission gates, Wyatt [1985]. The Manchester carry
adder with carrybypass circuitry, Chan and Schlag [1989], Chan and Karplus
[1990], provides another example of a digital network containing closed loops. Let
us consider the (very symmetrical) network in Figure 6.15.
T
E
T
Figure 6.15 A simple mesh network
G S2 = G36 = G63 = G45 = G54 = 1/4 and the others being zero. All six elements
of vector n are 1/4. We have used the discretization space step 6..h i = d;j20,
i = 1,2, while the minimum and the maximum time steps in the NAG subroutine
D02NGF are 1.049 x 10 4 and 2 respectively, the number of time steps being 1515.
The numerical results, graphically presented in Figures 6.166.19, confirm the
good work of our program. This is because, the symmetry of the circuit apriori
shows that the steady state must have the same voltage =1 both on capacitors and
on all points of the two lines.
Liut' 1 Line 2
t = 2730
"<1l "
bO
bD
:l
= 82.2 ~
t =82.2
~ = 54.8
= 27.4
t =
54.8
t = 27.4
0
L<'llgth Length
o ~~~~~~ O~r.~'~
Time 2.7£3 Time 2.7E3
Also, the speed of the transient regime differs from one element to the other
according to engineering intuition. The global delay is shown in Figure 6.20 together
with its apriori upper bound.
14
OL,==""'""'.T~
Time 2.7E3
This second one is computed with the "best" permutation a = (0,0) glvmg
wg = 0.705 X 10 3 and mini=I,2 cos,i = 0.6991. As a consequence we find
'1'0.5 = 1490 and '1'0.1 = 3770 as upper bounds for the "exact" delay times To.s = 295
and TO.I = 935, respectively.
Let us underline here the main applicative aspects of the models and methods
presented in Chapters IV, V and VI.
The problem of delay time prediction is one of the crucial aspects in digital cir
cuit design. In modern technology the improvement of this performance becomes
mainly limited by interconnection parasitics. This is why we have introduced an
accurate and general model for a network in which connecting wires are modelled
more exactly as distributed parameter elements while the devices are lumped ap
proximated. To use such a "mixed type" mathematical model, care should be taken
in obtaining the conditions for its validity. As we have seen, our conditions are in
good agreement with the engineering necessities, assuming the large applicativity
of our new model. The second problem solved above was to derive an upper bound
of the delay time for our mixedtype circuit. The formula giving this bound is
6.S. Concluding remarks 139
simple and therefore appropriate for engineering use in design. The third problem
about our mixedtype model, was its numerical realization and implementation.
The discretization was based on variational formulation of the problem and on the
use of the finite element method in space discretization. By this method it is easy
and natural to handle the "crossed" boundary conditions. The semi discrete model
(called FEMmodel above) leads to an initial value problem for systems of differen
tial equations. Typically this system is stiff. As described in Appendix I, one can
combine the FEMmodel with a general purpose subroutine library such as NAG.
Otherwise, such a sophisticated subroutine library makes it possible to build up a
fairly general electrical simulator including nonlinear capacitors and even lumped
nonlinearly modelled transistors (not described here). Having our circuit simula
tor built, we have verified our delay time bound for some typical examples. A
reasonable tightness was found.
Appendix I
END
We set up the full matrix linear algebra (D02NSF) and choose either Backward
Differentiation formulae (D02NVF) or BLEND formulae (D02NWF) as the inte
gration method with adaptive time stepping. The numerical results in Chapter 6.6
have been obtained by using Backward Differentiation Formulae because it copes
fewer iterations than BLEND formulae. The integration routine D02NGF calls
userdefined subroutines RESID to calculate 1"(t, V), JAC to calculate the Jacobian
of the residual system and MONITR to cope with error situations and to change
some integration parameters. The form of those residuals is given by formulae
140
Appendix I 141
(6.41)~(6.44).
.. ) Br(t,y) 0 f t he resl'd
., ~
The (l,) th component of the JacobIan matnx
ual system must be given in the form
Bri(t,y) _ A h [B!Ji(t,y)
BYj  'J + (a Yj ' (A3)
~
FORMULA e.?
FEM DISCRETIZATION
I
FOR SPACE VARIABLES.
SEE SECTION 8.2.1.
~
ci1~~~i~I'i~.g~~ %1~~~C:1
n UNES) TIME DEPENDENT
DIFFERENTIAL EQUATIONS.
SEE FORIdULA 8.16.
~
BOUNDARY CONDITIONS OF LINES
'7
I
ARE CONNECTED TO FIRST AND LAST
DIFFERENTIAL EQUATIONS OF LINES.
SEE FORIdULA 6.16 AND SECTION 6.2.2.
~ ~7
I I
THAT SUBROUTINE LIBRARIES (FOR EXAMPLE
NAG) DEYAND. SEE SECTION 6.4.
Figure Al
Appendix I 143
THE INITIALIZATION
OF THE PARAMETERS.
I I
INITIALJZINO AND CALCULATlNC
V .... LUES or VARIABLES. CHECKING
ACHIEVED VALUES (REAl)lN).
~
READINa DR CALCULATINC (IMICCN) INITIAL VALUES. I
J
CALCULATINC CALCULATINC VALUES
~ ~
STEADY NAO: OF STEADY STATE:
STATE C05Pcr FUNCTION JSTEFUN)
SOLUTION' OR JACOBIA ~STEJAC)
(STESTA). OP' THA.T FU CTlaN.
l
CAJ..CUlATINO
T1WE ~
.1 VALUES FoR CAPACITORS
ARE CALCULATED DIRECTLY.
DERIVATIVES
AT INITIAL
TIME POINT
.1'" WANTED
I
(INYDDT).
I CALCULATINC VALUES P'OB
I LINES (NAO:COGPCF).
~ i
CALCUlATlNQ VALUES OF FUNCTIONS
OF DIFFERENTIAL EQUATIONS OR
I
JACOBIAN OF THOSE ruNCTIONS
AT INITIAL TIME POINT (YDOTP'N).
Figure A2
144 Appendix I
CALCULATINC CALCUl.Jt..TING
VALUES OF VALUES OF
FUNCTIONS OF JACOBIAN OF
DIFFERENTIAL FUNCTIONS OF
DIFFERENTIAL
E~:t:Jlg)~S EQUATIONS (.lAC).
Figure A3
List of symbols
145
146 List of symbols
CPWL 36
accretive 12
"m" 12 CrankNicholson method 127
operator 12 CPWCD 36
approximation DCequation 45
FEM 123 quasi autonomous 62
Yosida 17
delay 49, 100, 113, 129
apriori upper bound 109, 129
delay time 49, 100, 129
asymptotically stable 48 global ,\ 101
derivative
ball 2
Fn\chet 35
basis functions 121 Gateaux 8
lefthand 24
bipolar transistor 52
righthand 24
boundary conditions
differential equation
system of 82
abstract 55
crossed type 96, 119
nonlinear 96, 112
dissipative 1
hyper maximal 13
capacitor maximal 13
nonlinear 36 mdissipative
Cauchy problem 19, 23, 105 operator 1, 12
strongly 12
circuit 37, 77 totally 12
infinite 71 w 12
classical solution 19 drain 75
convex DRDcondition 81
strictly 3
dual space 2
convergence
weak 2 duality
weak' 2 type functional 3
mapping 3
Cosemigroup 14
dynamical behaviour 47
147
148 Index
inductor operator
nonlinear 36 accretive 12
bounded 14
infinite network 52 dissipative 1, 12
hyper maximal dissipative 13
interconnection 76
wdissipative 11
maximal 13
junction diodes 36 "m"accretive 12
"m" dissipative 12
Kirchhoff's law 130 "m" totally dissipative 13
Index 149
weak
convergence 2
convergence' 2
weak' 2
WDRD 81
Yosidaapproximation 17
References
Agmon, S., "Lectures on elliptic boundary value problems," Van Nostrand, New
York, 1965.
Araki, K. and Naito, Y., Computeraided analysis of coupled lossy transmission
lines, in Proc. Intern. Symp. on Circ. and Syst., 423426, 1985.
Bakoglu, H.B. and Meindl, J.D., Optimal interconnection circuits for VLSI, IEEE
Trans.Electr.Der. ED 32 No 5, 903909, 1985.
Barbu, V., "Nonlinear Semigroups and Differential Equations in Banach Spaces,"
Noordhoff, Leyden, 1976.
Barbu, V., Nonlinear boundaryvalue problems for a class of hyperbolic systems,
Rev. Roum. Math. Pures et Appl. 22, 155168, 1977.
Barbu, V. and Morosanu, G., Existence for a nonlinear hyperbolic system, Non
linear Anal. T.M.A. 5, 341353, 1981.
Berzins, M., Dew, P.M. and Furzeland, R.M., Developing software for timedepen
dent problems using the method of lines and differentialalgebraic integrators,
Appl. Numer. Math. 5 (1989), 375397.
Bose, M.P. and Showalter, R.E., Homogenization of the layered medium equation.
Appl. Anal., 1990.
Brayton, R.K., Small signal stability criterion for electrical networks containing
lossless transmission lines, IBM J. Research Dev. 12,431440, 1968.
Brayton, R.K. and Miranker, W.L., A stability theory for nonlinear mixed initial
boundary valueproblems, Arhch. Rational Mech. Anal. 17,358376, 1964.
Brezis, H., "Operateurs Maximaux Monotones et Semigroupes de Contractions
dans les Espaces de Hilbert," NorthHolland, Amsterdam, 1973.
Brezis, H., "Analyse Fonctionelle. Theorie et Applications," Masson, Paris, 1987.
Browder, F.E., Nonlinear accretive operators in Banach spaces, Bull. Amer.
Math. Soc. 73, 470475, 1967.
Butzer, P.L. and Berens, H., "SemiGroups of Operators and Approximation,"
SpringerVerlag, Berlin, 1967.
151
152 References
Chan, P.K., An extension of Elmore's delay, IEEE Trans. Circ. Syst., CAS 33
No 11, 11471149, 1986 a.
Chan, P.K., An extension of Elmore '8 delay and its application for timing analysis
of MOS pass transistor networks, IEEE Trans. Circ. Syst., Cas 33 No 11,
11491152,1986 b.
Chan, P.K. and Karplus, K., Computing signal delay in general RC networks by
tree/link partitioning, IEEE Trans.Comp.Aided Design CAD 9 No 8, 898902,
1990.
Chan, P.K. and Schlag, M.D.F., Bounds on signal delay in RC mesh networks,
IEEE Trans. Circ. Syst. CAS 8 No 6, 581589, 1989.
Chew, K.W., Shivakumar, P.N. and Williams, J.J., Error bounds for the trun
cation of infinite linear differential systems, J. Inst. Math. Applic. 25,3751,
1980.
Chien, M.J., Piecewiselinear theory and computation of solutions of homeomor
phic resistive networks, IEEE Trans. Circ. Syst. CAS 24, 118127, 1977.
Chua, L.O. and Green, D.N., A qualitative analysis of the behaviour of dynamic
nonlinear networks  stability of autonomous networks, IEEE Trans. Circ.
Syst., CAS 23 No 6, 355379, 1976 a.
Chua, L.O. and Green, D.N., A qualitative analysis of the behaviour of dynamic
nonlinear networks: steadystate solutions of nonautonomous networks, IEEE
Trans. Circ. Syst., CAS 23 No 9, 530550, 1976 b.
Ciesielsni, M.J., Layer assignment for VLSI interconnect delay minimization,
IEEE Trans.Comp. Aided Design CAD 8 No 6, 702707, 1989.
Cooke, K.L. and Krumme, D.W., DifferentialDifference equations and nonlinear
initial boundary value problems for linear hyperbolic partial differential eqa
tions, J. Math. Anal. App!. 24,372387, 1968.
Coppel, W.A., "Stability and Asymptotic Behavior of Differential Equations,"
D.C.Heath, Boston, 1965.
Deimling, K., "Ordinary Differential Equations in Banach Spaces," Lecture Notes
in Math., 596, SpringerVerlag, Berlin, 1977.
Desoer, C.A., Distributed networks with small parasitic elements: inputoutput
stability, IEEE Trans. Circ. Syst., CAS 24, 17, 1977.
Desoer, C.A. and Haneda, H., The measure of a matrix as a tool to analyze
computer algorithms for circuit analysis, IEEE Trans. Circ. Syst., CT 19 No
5, 480486, 1972.
Desoer, C.A. and Katzenelson, J., Nonlinear RLC networks, Bell Syst. Tech. J.
44 No 1, 161198, 1965.
Desoer, C.A. and Kuh, E.S., "Basic circuit theory," McGrawHill, New York,
1969.
References 153
Dolezal, V., "Nonlinear networks," Elsevier Sci. Pub!. Co., New York, 1977.
Dolezal, V., "Monotone operators and applications in control and network the
ory," Elsevier Sci. Pub!. Co., New York, 1979.
Dunford, N. and Schwartz, J.T., "Linear operators, Part I," J. Wiley, Intersience
Pub!., New York, 1958.
Elmore, W.C., The tranjient rejponje of damped linear networb with particular
regard to wideband amplijierj, J. App!. Phys. 19,5563, 1948.
Gao, D.S., Yang, A.T. and Kang, S.M., Modelling and jimulation of interconnec
tion delays and cro33talb in high3peed integrated circuitj, IEEE Trans.Circ.Syst.
CAS 37, 19, 1990.
Ghausi, M.S. and Kelly, J.J., "Introduction to Distributed Parameter Networks,"
Holt, Rinehart and Winston Inc., New York, 1968.
Glasser, L.A. and Dobberphul, D.W., "Design and Analysis of VLSI Circuits,"
AddisonWesley, New York, 1985.
Glez. Harbour, M. and Drake, J .M., Simple RC model for integrated multitermi·
nal interconnections, lEE Proceedings, Part G, 135, No 1, 1923, 1988.
Goldstein, J.A., "Semigroups of operators and abstract Cauchy problems," Lec
ture notes, Tulane university, 1970.
Grondin, R.O., Porod, W. and Ferry, D.K., Delay time and jignal prapagation in
largejcale integrated circuits, IEEE J. SolidState Circ., SC 19 No 2, 262263,
1984.
Gummel, H.K., A chargecontrol tranjistor model for network analysis programs,
Proc.IEEE 56,751752,1968.
Jain, N.K., Prasad, V.C. and Bhattacharyya, A.B., Delay time sensitivity in
linear RC tree, IEEE Trans. Circ. Syst., CAS 34 No 4, 443445, 1987.
Jouppi, N.P., Timing analysis and performance improvement of MOS VLSI de
sign, IEEE Trans. Compo Aided Design, CAD 6 No 4, 650665, 1987.
Kato, T., Nonlinear semigroups and evolution equations, J. Math. Soc. Japan
19, 509520, 1967.
Kenmochi, N. and Takahashi, T., On the global existence of solutions of differ
ential equations on closed subsets of a Banach space, Proc. Japan Acad. Ser.
A Math. Sci. 51, 520525, 1975.
Kim, Y.H., "ELOGIC: A relaxationbased switchlevel simulation technique,"
UCB/ERL M86/2, Univ of California, Berkley, 1986.
Kolmogorov, A. and Fomine, S., "Elements of the theory of functions and func
tional analysis," Graylock Press, Baltimore, 1957.
Krizek, M and Neittaanmaki, P., "Introduction to finite element method for vari
ational problems with applications," Pitman Monographs in Pure and Applied
Mathematics 50, Longman, Essex, 1990.
Kumar, U., A bibliography of distributed RC networks, IEEE Circ. Syst. Mag. 2
No 2,919, 1980.
Lin, T.M. and Mead, C.A., Signal delay in general RC networks, IEEE Trans.
Compo Aided Design. CAD 3 No 4, 331349, 1984.
Lin, T.M. and Mead, C.A., A hierarchical timing simulation model, IEEE Trans.
Compo Aided Design. CAD 5 No 1,188197,1986.
Lions, J.L. and Magenes, E., "Nonhomogenous boundary value problems and
applications," Vol II, SpringerVerlag, Berlin, 1972.
Lovelady, D.L. and Martin, R.H.Jr., A global existence theorem for a nonau
tonomous differential equation in a Banach space, Proc. Amer. Math. Soc. 35,
445449, 1972.
Lumer, G. and Phillips, R.S., Dissipative operators in Banach space, Pacific
J.Math 11,679698, 1961.
Lusternik, L.A. and Sobolev, V.J., "Elements of functional analysis," Hindustan
Publ. Corp. John Wiley, DelhiNew York, 1974.
Marinov, C.A., The delay time for a reg line, Int. J. Circ. Th. Appl. 15,7983,
1987.
Marinov, C.A., Sandberg type properties for nonlinear transistor networks, Proc.
Eur. Conf. Cire. Th. Design, 5358, Paris, 1987.
Marinov, C.A., On assymtotic stability of nonlinear networks, IEEE Trans. Circ.
Syst. CAS 37, No 4" 571573, 1990 a.
Marinov, C.A., Dissipativity as an unified approach to Sandberg Wilson type prop
erties of nonlinear transistor networks, Int. J. Circ. Th. Appl. 18, No 6,
575594, 1990 b,.
Marinov, C.A. and Lehtonen, A., Mixedtype circuits with distributed and lumped
parameters, IEEE Trans. Circ. Syst. CAS 36 No 8, 10801086, 1989.
Marinov, C.A. and Moro§anu, G., Consistent models for electrical networks with
distributed parameters, Casopis pro pestovanl matematiky, to appear 1991.
Marinov, C.A. and Neittaanmaki, P., Delay time for general disrtibuted networks
with applications to timing analysis of digital MOS integrated circuits, in "Sim
ulation of Semiconductor Devices and Processes," K. Board (ed.), Pineridge
Press, Swansea, 322326, 1986.
Marinov, C.A. and Neittaanmiiki, P., A theory of electrical circuits with resistively
coupled distributed structures. Delay time predicting, IEEE Trans. Circ. Syst.
CAS 35 No 2, 173183, 1988.
Marinov, C.A. and Neittaanmiiki, P., Global delay time for general distributed
networks with applications to timing analysis of digital MOS integrated circuits,
Int.J.Comp.Math.Electrical Electronic Eng., COMPEL 8 No 1, 1737, 1989.
Marinov, C.A. and Neittaanmaki, P., Asymptotical convergence evaluation for a
parabolic problem arising in circuit theory, Z. Angew. Math. Mech. 70 No 8,
344347, 1990 a.
Marinov, C.A. and Neittaanmiiki, P., A delay time bound for disrtibuted param
eter circuits with bipolar transistors, Int. J. Circ. Th. Appl., 18, 99106, 1990
b.
Marinov, C.A. and Neittaanmiiki, P., Bounds for the solution of a system of
parabolic equations arising in circuit theory, 13th IMACS World Congress on
Compo Appl. Math., Dublin, 1991 a.
Marinov, C.A. and Neittaanmiiki, P., Both sided estimates for distributed struc
tures arising in MOS interconnections, Eur. Conf. Cire. Th. Design, ECCTD,
Copenhagen, 1991 b.
Marinov, C.A., Neittaanmaki, P. and Hara, V., Signal delay in general distributed
networks, Proe. XXX Int. Conf. ETAN 3, 3543, HercegNovi, 1986.
Marinov, C.A., Neittaanmaki, P. and Hara, V., Numerical approach for signal
delay in general distributed networks, Proe. Vth Int. Conf. Num. Anal.
156 References
Semicond. Dev. Integr. Circ., J.J.H. Miller ed., 307312, Boole Press, Dublin,
1987.
Marinov, C.A., Neittaanmaki, P. and Hara, V., A consistent model for the wiring
delay of the MOS inverter, Proc, ECCTD, 8993, Brighton, 1989.
Martin, R.H.Jr., The logarithmic derivate and equations of evolutions in a Banach
space, J. Math. Soc. Japan 22, 411429, 1970 a.
Martin, R.H.Jr., A global existence theorem for autonomous differential equations
in a Banach space, Proc. Amer. Math. Soc. 26, 307314, 1970 b.
Martin, R.H.Jr., "Nonlinear Operators and Differential Equations in Banach
Spaces," John Wiley, New York, 1976.
McClure, J.P. and Wong, On infinite systems of linear differential equations, Can.
J. Math. 27,691703,1975.
McClure, J.P. and Wong, Infinite systems of differential equations, Can. J. Math.
28, 11321145, 1976.
McClure, J.P. and Wong, Infinite systems of differential equations II, Can. J.
Math. 31,596603, 1979.
Mikhailov, V., "Partial differential equation," Mir, Moscow, 1978.
Miller, R.K. and Michel, A.N., Stability theory for countable infinite systems of
differential equations, Tohoku Math. J. 32, 155168, 1980.
Minty, G.J., Monotone networks, Proc. Royal Soc. (London), Ser A 257,
194212, 1960.
Minty, G.J., Solving !!teady!!tate nonlinear networks of monotone elements, Trans.
Circ. Th, CT 8, 99104, 1961.
Moro§anu, G., Stability of solutions of nonlinear boundaryvalue problems for
hyperbolic system!!, Nonlinear Anal., TMA 5,6170, 1981, a.
Moro§anu, G., Mixed problems for a clas!! of nonlinear differential hyperbolic
systems, J. Math. Anal. Appl. 83, 470485,1981, b.
Moro§anu, G., On a class of nonlinear differential hyperbolic systems with nonlocal
boundary conditions, J. Diff. Eq. 43, 345368, 1982.
Moro§anu, G., "Nonlinear Evolution Equations and Applications," D. Reidel
Publ. Co., Dordrecht, 1988.
Moro§anu, G. and Petrovanu, D., Nonlinear monotone boundary conditions for
parabolic equations, Rend. 1st. Matern. Univ. Trieste 18, 136155, 1986.
Moro§anu, G., Marinov, C.A. and Neittaanmiiki, P., Wellposed nonlinear prob
lems in the theory of electrical networks with distributed and lumped parameters,
Proc. of Int. Conf. Simul. Modelling, lASTED, Lugano" 345348, 1989.
Moro§anu, G., Marinov, C.A. and Neittaanmaki, P., Wellposed nonlinear prob
lems in integrated circuits modelling, Circ. Syst. Sign. Proc., 10, 5369, 1991.
References 157
Protonotarios, E.N. and Wing, 0., Theory of nonuniform RC lines, Part II:
Analytic properties in the time domain, IEEE Trans. Circ. Th., CT14,
1320, 1967.
Putatunda, R., A UTODELAY: A second generation automatic delay calculation
program for LSI/VLSI chips, in Proc. IEEE Int. Conf. Compo Aided Design,
188190,1984.
Rubinstein, J., Penfield, P.Jr. and Horowitz, M.A., Signal delay in RC tree
networks, IEEE Trans. Compo Aided Design 2, 202211, 1983.
Ruehli, A.E. (ed.), "Circuit analysis. Simulation and design," North Holland,
Amsterdam, 1987.
Ruehli, A.E. and Ditlow, G.S., Circuit analysis, logic simulation and design ver
ification for VLSI, Proc. IEEE 71, 3448, 1983.
Sakurai, T., Approximation of wiring delay in MOSFET LSI, IEEE J. Solid State
Circ., SC 18 No 4, 418426, 1983.
Sandberg, LW., Some theorems on the dynamic response of transistor network,
Bell System Tech. J. 48,3554, 1969.
Sandberg,I.W., Theorems on the analysis of nonlinear transistor networks, Bell
Syst. Tech. J. 49, 95114, 1970.
Sandberg, LW. and Willson Jr., A.N., Some theorems on properties of DC equa
tions of nonlinear networks, Bell Syst. Tech. J. 48, No 1, 134, 1969 a.
Sandberg, I.W. and Willson Jr., A.N., Some networktheoretic properties of non
linear DC transistor networks, Bell Syst. Tech. J. 48, 12931311, 1969 b.
Saraswat, K.C. and Mohammadi, F., Effect of scaling of interconnections on the
time delay of VLSI circuits, IEEE Trans. Electr. Dev., ED 29 No 4, 645650,
1982.
Sattinger, D.H., Stability of nonlinear parabolic systems, J. Math. Anal. Appl.
24, 241245, 1968.
Sato, K., On the generators of nonnegative contraction semigroups in Banach
lattices, J. Math. Soc., Japan 20, 423436, 1968.
Schwartz, A.F., "Computer aided design of microelectronic circuits and systems,"
Acad.Press, London, 1987.
Showalter, R.E. and Snyder, C.H., A distributed RC network model with dielectric
loss, IEEE Trans. Circ. Syst., CAS 33 No 7, 707710, 1986.
Showalter, R.E. and Xu, X., Convergence of diffusion with concentrating capacity,
J. Math. Anal. Appl., 1990.
Singhal, K. and Vlach, J., Approximation of nonuniform RCdistributed networks
for frequency and timedomain computations, IEEE Trans. Circ. Th. CT19,
347354, 1972.
References 159
So, H.C., On the hydrid description of a linear nport resulting from extraction
of arbitrarily specified elements, IEEE Trans. Circuit Th. 12,381387, 1965.
Zemanian, A.H., Infinite electrical networks, Proc. IEEE 64, 617, 1976.
160 References