Analog compute in memory improves accuracy despite noise
NOVA-CIM: Noise- and Correlation-Tolerant Stochastic Interfaces for Analog Compute-in-Memory
Hardware Architecture
Summary
Analog compute-in-memory (CIM) chips speed up calculations but often suffer from noisy readouts that hurt accuracy. The authors developed a new way called NOVA-CIM to read these noisy signals by converting them into simple yes-or-no comparisons many times and then averaging results, which reduces the effect of noise. Their method avoids the need for complex and precise converters, keeping accuracy high even with noisy signals. Tests on vision transformer models showed almost no loss in accuracy compared to traditional digital methods.
What this means in practice
- •For hardware engineers: Design energy-efficient analog CIM chips with improved noise tolerance and reduced ADC complexity for machine learning inference.
- •For machine learning system integrators: Implement NOVA-CIM interfaces to maintain deep learning inference accuracy on noisy analog accelerators for vision applications.
Authors
Jiachen Ren, Wenshuai Yao, Haobo Liu, Xincheng Feng, Chenxi Hu, Zhengwu Liu, Kechao Tang, Wenyong Zhou, Ngai Wong
Abstract
Analog compute-in-memory (CIM) enables energy-efficient model acceleration, but its reliance on ADC-based readout, which directly quantizes noisy column currents, makes inference accuracy highly sensitive to analog read noise, active-row scaling, and ADC precision. In this paper, we present NOVA-CIM, a noise- and correlation-tolerant stochastic interface for analog CIM by replacing multi-bit ADC readout with random-reference 1-bit sensing and reconstructing results through lightweight counting. By converting column currents into comparison probabilities, this probability-domain readout averages zero-mean dynamic read noise over stochastic samples while reducing dependence on high-resolution ADCs. We provide a unified robustness analysis showing that dynamic read noise is suppressed through temporal averaging and that spatial input-bitstream correlation increases instantaneous current variance rather than introducing first-order MAC bias. MAC-level experiments and end-to-end evaluation on ViT-Base validate the analysis: under read noise, Top-1 accuracy remains 84.48% near the 84.51% bfloat16 (BF16) baseline; under stochastic number generator (SNG) reuse, MAC bias stays near zero while root-mean-square error (RMSE) and stochastic cross-correlation (SCC) grow as predicted.