gld@cunixd.cc.columbia.edu (Gary L Dare) (10/13/88)
At ICCD 1987, Saleh, Webber, Xia and Sangiovanni-Vincentelli presented a paper on "Parallel Waveform-Newton Algorithms for Circuit Simulation". This paper discussed the propogation of waveforms from one subcircuit to another, as they are generated on a processor element. This method was proven successful by J. White with PRELAX, a waveform relaxation based circuit simulator; these people presented it for use with the Waveform-Newton method where only one Newton iteration takes place rather than iterating to convergence as on (P)Relax. In their paper, they state that the single time step (user specified) used by all subcircuits in the first WRN iteration makes time point pipelining impossible. However, they are basically using a Gauss-Seidel approach to take into account the order of circuits to be processed (but, taking advantage of independant subcircuits to achieve parallelism). After a discussion with a couple of people, this statement doesn't make sense. Is anyone else familiar with this topic, and could they tell me what the authors were trying to say? gld ~~~~~~~~~~~~~~~~~~~ je me souviens ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Gary L. Dare > dare@eevlsi.ee.columbia.EDU "Too old to rock'n'roll, > gld@cunixd.cc.columbia.EDU too young to die!" -- Jethro Tull > gld@cunixc.BITNET