Late Breaking Result: FPGA-Based Emulation and Fault Injection for CNN Inference Accelerators

TL;DR

This paper introduces an FPGA-based platform for rapid fault tolerance analysis of CNN inference accelerators, significantly speeding up fault injection testing compared to traditional software methods.

Contribution

It presents a novel FPGA-based emulation framework that automates fault injection for CNN accelerators, enabling faster and more efficient fault tolerance analysis.

Findings

Fault injection into multipliers affects classification accuracy.

FPGA emulation is an order of magnitude faster than software.

Framework supports various CNN models and hardware architectures.

Abstract

A new field programmable gate array (FPGA)-based emulation platform is proposed to accelerate fault tolerance analysis of inference accelerators of convolutional neural networks (CNN). For a given CNN model, hardware accelerator architecture, and FT analysis target, an FPGA-based CNN implementation is generated (with the help of the Tengine framework), and fault injection logic is added. In our first case study, we report how the classification accuracy drop depends on the faults injected into multipliers used in Multiply-and-Accumulate Units of NVDLA inference accelerator executing ResNet-18 CNN. The FT analysis emulated on Zynq UltraScale+ SoC is an order of magnitude faster than software emulation.