14.1 A 22nm 104.5TOPS/W µ-NMC-Δ-IMC Heterogeneous STT-MRAM CIM Macro for Noise-Tolerant Bayesian Neural Networks

Compute-in-memory (CIM) macros [1]–[5] for convolutional neural networks (CNNs) [6]–[7] and vision transformers (ViTs) [8] enable high-performance computing for energy-efficient edge-AI devices. Image recognition applications are constrained by inference accuracy degradation or $\text{misjudgments}$ due to environmental noise. Bayesian neural networks (BNNs) [9], which represent weights using its mean ($\mu$-weight) and difference from mean tF $\Delta$-weight) outperform CNN and $\text{ViT}$ models in terms of noise tolerance, making them promising candidates for edge-AI devices dealing with noisy inputs. Using digital circuits to implement BNN [10]–[12] imposes a tradeoff between inference accuracy and performance, power and area (PPA).