MICSim: A modular pre-circuit simulator for mixed-signal compute-in-memory accelerators in CNNs and transformers

Abstract

This work introduces MICSim, an open-source, pre-circuit simulator designed to assist circuit designers to evaluate early-stage chip-level software performance and hardware overhead of mixed-signal compute-in-memory(CIM) accelerators. MICSim features a modular design, allowing easy multi-level co-design and design space exploration. Modularized from the state-of-the-art CIM simulator NeuroSim, MICSim provides a highly configurable simulation framework supporting multiple quantization algorithms, diverse circuit architecture designs, and different memory devices. This modular approach allows MICSim to be effectively extended to accommodate new designs.

MICSim natively enables evaluating accelerators’ software and hardware performance for convolutional neural networks (CNNs) and Transformers in Python, leveraging the popular PyTorch and Hugging Face Transformers frameworks. MICSim can be easily combined with optimization strategies to perform design space exploration and can be used for evaluating chip-level Transformers CIM accelerators, making it highly adaptable to different networks. Also, MICSim can achieve $9\times \sim 32\times$ speedup of NeuroSim through a statistic-based average mode proposed by this work, without significant error across various networks. MICSim is available as an open-source tool on GitHub (https:// github.com/ MICSim-official/MICSim_V1.0.git).