ANN-based Standard Cell Switching Waveforms Characterization Model for PVT Variations

VLSI circuit designs at advanced processes often necessitate addressing diverse requirements for PVT variations. In this work, we proposed an artificial neural network (ANN)-based method to characterize switching waveforms of standard cells for process-voltage-temperature (PVT) variations. Unlike traditional methods such as CCS and ECSM, our approach eliminates the prerequisite of library files for multiple PVT scenarios. Instead, every single ANN model is trained to characterize a unique standard cell with capturing its characteristics for various PVT corners. The experimental results indicate that, compared to SPICE simulations, the evaluation metrics of the switching waveforms obtained by our method have only a maximum error of 3%, an average error within 0.5%, but a minimum of 30+X of runtime improvement. Hence it will be feasible and applicable to implement such model into a commercial signoff EDA tool to reduce iterations and accelerate convergence by precisely simulating the timing, power, and EM/IR of a chip at the standard cell level.