Fast and Accurate Error Simulation for CNNs Against Soft Errors

The great quest for adopting AI-based computation safety-/mission-critical applications motivates the interest towards methods assessing robustness of application w.r.t. not only its training/tuning but also errors due to faults, in particular soft errors, affecting underlying hardware. Two strategies exist: architecture-level fault injection and application-level functional error simulation. We present a framework reliability analysis Convolutional Neural Networks (CNNs) via an simulation engine that exploits set validated models extracted from detailed campaign. These are defined based on corruption patterns output CNN operators induced by faults bridge gap between simulation, exploiting advantages both approaches. compared our methodology against SASSIFI accuracy injection, TensorFI terms speedup strategy. Experimental results show achieves about 99% effects SASSIFI, ranging 44x up 63x TensorFI, implements limited models.