Long-Term and Short-Term Transistor Aging in Deep Neural Networks: Impact and Mitigation
A new research paper details the impact of transistor aging on the accuracy of deep neural networks (DNNs) used in applications like image classification. Transistor aging slows down switching speeds, leading to timing violations and reduced inference accuracy. The paper proposes an aging-aware retraining methodology to create more resilient DNNs that maintain accuracy even with aggressive timing guardbands, and also touches on using short-term aging for hardware Trojan detection. AI
IMPACT This research could lead to more reliable AI hardware implementations, ensuring sustained performance over time.