In this paper an efficient technique for estimating the average number
of signal transitions in CMOS logic circuits is presented. A logic si
gnal is modeled as a two state, discrete-time Markov chain, that is ch
aracterized by two parameters. The steady-state average number of tran
sitions on a signal is expressed as a function of the Markov chain par
ameters. Formulas for the two parameters associated with the gate outp
ut as a function of the corresponding parameters associated with the g
ate input are derived. This allows estimation of signal transitions at
all gates by local propagation of the parameters. For circuits withou
t reconvergent, the local propagation provides very accurate estimates
and the time to obtain the estimates is a negligible fraction of the
time required to perform logic simulation. For circuits with reconverg
ent fanout, the estimation procedure is based on the (local) use of OB
BDs. This scheme has the flexibility of trading off between speed and
accuracy by allowing the user to specify the number of levels of recon
vergent fanout that are to considered. Results of experiments carried
out on a large number of benchmark circuits are given. The theoretical
estimates of switching activity are compared with those obtained via
simulation. The experiments also examine the number of levels of logic
that need to be examined in the presence of reconvergent fanout.