Stochastic Variations in Nanoscale HZO based Ferroelectric finFETs: A Synergistic Approach of READ Optimization and Hybrid Precision Mixed Signal WRITE Operation to Mitigate the Implications on DNN Applications

TL;DR

This paper presents a combined digital and analog approach to optimize read and write operations in HZO-based ferroelectric finFETs, addressing stochastic variations to improve DNN application performance.

Contribution

It introduces a hybrid precision mixed-signal method integrating high-precision digital and low-precision analog units for variation mitigation.

Findings

Enhanced robustness of Fe-finFETs in DNNs

Effective reduction of stochastic variation impacts

Improved accuracy in neural network computations

Abstract

This paper reports a synergistic approach of READ and WRITE optimization by deploying a high-precision digital computation unit along with a low-precision ferroelectric finFET (Fe-finFETs) based analog vector-matrix multiplication block for mitigating the impact of stochastic device variations in hafnium zirconium oxide (HZO) based Fe-finFETs.