Switch-Electrical Segmented Waveform Relaxation for Digital MOS VLSI and Its Acceleration on Parallel Computers

Switch-Electrical Segmented Waveform Relaxation for Digital MOS VLSI and Its Acceleration on Parallel Computers The Segmented Waveform Relaxation Method (SWRM) is a new and efficient waveform relaxation method for circuit-level simulation of large-scale digital MOS networks. SWRM is original in two aspects. 1) Dynamic feedback loop cutting and waveform event driven scheduling result in a varying simulation order of the subcircuits and, for digital synchronous circuits, lead to logically correct and accurate ground waveforms in one waveform iteration where the only approximation is due to Miller feedback over MOS capacitances (timing error being usually <5 percent). 2) Dynamic decomposition of large subcircuits at MOS transistors which are temporarily in a nonconducting state. SWRM has been implemented in the program SWAN and tested on large industrial circuits (up to 4000 transistors) with speed gains of 10 > 100 over SPICE2G6. A mixed-mode switch electrical implementation of SWRM together with novel acceleration methods on parallel computers will be described together with test results on real-life circuits.