CrossNAS: A Cross-Layer Neural Architecture Search Framework for PIM Systems

TL;DR

CrossNAS is an automated framework that explores multiple design layers to optimize machine learning deployment on PIM systems, achieving superior accuracy and energy efficiency.

Contribution

It introduces a novel cross-layer NAS framework using single-path one-shot and evolutionary search for PIM system optimization.

Findings

Outperforms previous PIM NAS methods in accuracy.

Achieves better energy efficiency.

Maintains comparable or shorter search times.

Abstract

In this paper, we propose the CrossNAS framework, an automated approach for exploring a vast, multidimensional search space that spans various design abstraction layers-circuits, architecture, and systems-to optimize the deployment of machine learning workloads on analog processing-in-memory (PIM) systems. CrossNAS leverages the single-path one-shot weight-sharing strategy combined with the evolutionary search for the first time in the context of PIM system mapping and optimization. CrossNAS sets a new benchmark for PIM neural architecture search (NAS), outperforming previous methods in both accuracy and energy efficiency while maintaining comparable or shorter search times.