The awarded paper addresses one of the key challenges in analog in-memory computing: the temporal drift of phase-change memory (PCM) devices, which can significantly degrade neural network accuracy over time. The work introduces a drift-aware training framework combined with systematic hyperparameter optimization that enables neural networks to maintain robust performance despite long-term hardware-induced variations. By integrating physics-based drift-regularization strategies into the training process, the proposed approach improves both accuracy and stability with minimal training overhead.

IEEE ICM is a well-established international conference covering microelectronics, hardware systems, and emerging computing technologies, with a strong emphasis on hardware-aware machine learning and neuromorphic computing.

This research contributes to SECAI's mission of advancing reliable and energy-efficient AI hardware, with direct relevance to edge intelligence and embedded AI systems requiring sustained multi-year operation without periodic recalibration.