A Controlled Diagnostic Study of Hardware-Induced Distortions in Hardware-Aware Training
For researchers and engineers developing AI accelerators, this work clarifies which hardware non-idealities can be mitigated via training versus requiring hardware-level fixes, but the study is limited to a specific training paradigm and is not a universal theory.
This paper presents a diagnostic framework to evaluate which hardware-induced distortions in analog in-memory computing can be compensated by hardware-aware training (HAT). It identifies three key diagnostics that separate compensable perturbations (e.g., read noise) from those that break optimization (e.g., IR-drop), providing guidance for hardware-software co-design.
Hardware-aware training (HAT) is widely used to improve the robustness of neural networks on non-ideal AI accelerators, such as analog in-memory computing (IMC) systems. However, not all hardware-induced distortions are equally compensable by training. This paper presents a diagnostic framework that models hardware non-idealities as structured perturbations of the forward operator and evaluates their compatibility with gradient-based optimization. We analyze six representative perturbation classes--read noise, variability, drift, stuck-at faults, IR-drop, and ADC discretization--and identify three key diagnostics: gradient expectation consistency, bounded gradient variance, and non-degenerate sensitivity. Our results show a clear separation between perturbations that can be compensated by HAT and those that consistently break optimization. This provides practical guidance for hardware-software co-design, clarifying which non-idealities can be addressed at the training level and which require circuit-, architecture-, or calibration-level mitigation. This study should be interpreted as a controlled empirical analysis under vanilla forward-perturbation HAT, rather than as a universal theory of hardware-aware training.