Variability-Aware Training and Self-Tuning of Highly Quantized DNNs for Analog PIM

TL;DR

This paper introduces a variability-aware training algorithm for highly quantized DNNs on analog PIM architectures, significantly improving accuracy under fabrication variability, and proposes a self-tuning inference method to further reduce accuracy loss.

Contribution

It presents a novel joint variability- and quantization-aware training method and a self-tuning inference architecture for robust DNN deployment on analog PIM hardware.

Findings

Outperforms prior models on multiple datasets and models.

Achieves up to 35.7% accuracy gain in low-bitwidth models with high variability.

Self-tuning reduces accuracy loss to below 10%.

Abstract

DNNs deployed on analog processing in memory (PIM) architectures are subject to fabrication-time variability. We developed a new joint variability- and quantization-aware DNN training algorithm for highly quantized analog PIM-based models that is significantly more effective than prior work. It outperforms variability-oblivious and post-training quantized models on multiple computer vision datasets/models. For low-bitwidth models and high variation, the gain in accuracy is up to 35.7% for ResNet-18 over the best alternative. We demonstrate that, under a realistic pattern of within- and between-chip components of variability, training alone is unable to prevent large DNN accuracy loss (of up to 54% on CIFAR-100/ResNet-18). We introduce a self-tuning DNN architecture that dynamically adjusts layer-wise activations during inference and is effective in reducing accuracy loss to below 10%.