Attacking AI Accelerators by Leveraging Arithmetic Properties of Addition

2026-03-28

AI Summary

This work addresses the long-term reliability of AI accelerators, which suffer from degraded inference accuracy due to hardware aging. The authors propose a novel aging attack that exploits the commutative property of addition to reorder inputs to adders, thereby inducing unbalanced transistor stress without altering functional behavior. This technique selectively accelerates delay degradation in multiply-accumulate (MAC) units while incurring negligible area and performance overhead and requiring no additional hardware Trojans or software modifications. The method is compatible with diverse multiplier architectures, bit-widths, and general-purpose arithmetic units. Experimental results demonstrate that the attack can reduce inference accuracy by up to 64% across multiple AI models and datasets within four years, posing a significant threat to the sustained reliability of AI accelerators.

Abstract

The dependability of AI models relies largely on the reliability of the underlying computation hardware. Hardware aging attacks can compromise the computing substrate and disrupt AI models over the long run. In this work, we present a new hardware aging attack that exploits commutative properties of addition to disrupt the multiply-and-add operation that forms the backbone of almost all AI models. By permuting the inputs of an adder, the attack preserves functional correctness while inducing unbalanced stress among transistors, accelerating delay degradation in the circuit. Unlike prior approaches that rely on input manipulation, additional trojan circuitry, etc., the proposed method incurs virtually no area or software overhead. Experimental results with two types of multipliers, different bit widths, a mix of AI models and datasets demonstrates that the proposed attack degrades inference accuracy by up to 64% in 4 years, posing a significant threat to AI accelerators. The attack can also be extended to arithmetic units of general-purpose processors.

Problem

• hardware aging attack
• AI accelerators
• adder commutativity
• delay degradation
• multiply-and-add operation

Innovation

• hardware aging attack
• commutative property
• AI accelerator
• delay degradation
• adder permutation