SPINE: A Fault Injection Profiler for Quantized Neural Networks under Accumulated Faults
This work addresses the need for realistic fault characterization in quantized neural networks deployed on edge devices, providing a tool for reliability analysis without retraining.
The paper presents SPINE, a fault injection profiler for quantized neural networks under accumulated bit-flips, and evaluates it across multiple topologies and memory layouts to guide selective hardening strategies.
Deploying deep neural networks at the edge demands efficient inference under strict cost and power constraints. Quantized neural networks address these demands by replacing floating-point parameters with low-precision integers, yet their weights remain continuously exposed to radiation-induced bit-flips during inference. Fault Injection can be used to simulate those environments, but existing studies fail to characterize how accumulated upsets translate into mispredictions under realistic memory layouts. This paper presents a GDB-driven profiling framework that injects cumulative weight bit-flips directly onto the target binary of edge CPUs, generating per-layer fault profiles without requiring model retraining or code modification. Evaluated across multiple topologies, quantization efforts, and memory layouts, the results indicate how selective hardening strategies should be applied to effectively protect neural networks.