CRASHLESS

(Contributing research activities without direct consortium partnership involvement) CRASHLESS is a research initiative focused on the reliability and dependability of AI hardware accelerators operating under hardware faults and uncertain runtime conditions. The project investigates scalable methods for reliability assessment, fault-aware AI design, and resilient accelerator architectures targeting deep neural network inference. Research contributions associated with RSS include: Fault injection methodologies Reliability-aware AI accelerator optimization Sensitivity analysis of neural networks Hardware-aware robustness evaluation Approximate and fault-tolerant computing techniques