Fast time-domain simulation through combined symbolic analysis and piecewise linear modeling

This paper presents a method for fast time-domain simulation of analog systems with nonlinear parameters. Specifically, the paper focuses on ∑-Δ analog-to-digital converters (ADC). The method generates compiled-code simulators based on symbolic analysis. Code is optimized using loop invariant elimination, and constant folding. Circuits are described as structural macromodels. Non-linear parameters are expressed using piecewise linear (PWL) models. The paper presents a technique for automatically creating PWL models through model extraction from trained neural networks (NN). As compared to existing behavioral simulation methods for ∑-Δ ADC, this technique is fully automated and more accurate. In our experiments, compiled-code simulation was about 100x faster than Spectre (numerical) simulation.