In-Place Zero-Space Memory Protection for CNN

TL;DR

This paper presents a novel in-place zero-space ECC method with a new training scheme to protect CNNs from memory faults, achieving reliable inference without additional memory overhead.

Contribution

It introduces the first zero-space cost memory protection for CNNs using an in-place ECC and a weight distribution-oriented training scheme.

Findings

Achieves memory protection without additional memory overhead.

Maintains CNN inference reliability under memory faults.

Introduces a new training scheme for weight distribution.

Abstract

Convolutional Neural Networks (CNN) are being actively explored for safety-critical applications such as autonomous vehicles and aerospace, where it is essential to ensure the reliability of inference results in the presence of possible memory faults. Traditional methods such as error correction codes (ECC) and Triple Modular Redundancy (TMR) are CNN-oblivious and incur substantial memory overhead and energy cost. This paper introduces in-place zero-space ECC assisted with a new training scheme weight distribution-oriented training. The new method provides the first known zero space cost memory protection for CNNs without compromising the reliability offered by traditional ECC.