Unified Characterization Platform for Emerging NVM Technology: Neural Network Application Benchmarking Using off-the-shelf NVM Chips

TL;DR

This paper introduces a unified FPGA-based platform for electrical characterization and benchmarking of emerging NVM chips, demonstrating their application in neural network inference and comparing their performance to traditional memory technologies.

Contribution

It presents a versatile FPGA test-bench for detailed electrical analysis of various NVM chips and showcases their use in neural network applications, enabling direct comparison with SRAM and Flash.

Findings

NVM chips exhibit distinct current, endurance, and error profiles.

Emerging NVMs can serve as effective active weight memory in neural networks.

Benchmarking reveals potential advantages of NVMs over traditional memories.

Abstract

In this paper, we present a unified FPGA based electrical test-bench for characterizing different emerging NonVolatile Memory (NVM) chips. In particular, we present detailed electrical characterization and benchmarking of multiple commercially available, off-the-shelf, NVM chips viz.: MRAM, FeRAM, CBRAM, and ReRAM. We investigate important NVM parameters such as: (i) current consumption patterns, (ii) endurance, and (iii) error characterization. The proposed FPGA based testbench is then utilized for a Proof-of-Concept (PoC) Neural Network (NN) image classification application. Four emerging NVM chips are benchmarked against standard SRAM and Flash technology for the AI application as active weight memory during inference mode.