Spatiotemporal-Aware Bit-Flip Injection on DNN-based Advanced Driver Assistance Systems

2026-04-04

AI Summary

This work addresses the safety risks in advanced driver-assistance systems (ADAS) arising from DRAM bit flips that distort deep neural network (DNN)-based driving decisions. To this end, the authors propose STAFI, a spatiotemporal-aware fault injection framework that jointly models the spatial criticality of weight bits and the temporal context in which faults occur. STAFI employs a progressive metric-guided bit search (PMBS) strategy coupled with a critical fault timing identification (CFTI) mechanism to efficiently pinpoint the most hazardous combinations of faulty bits and their activation moments. Experimental evaluation on production-grade ADAS models demonstrates that STAFI identifies 29.56 times more safety-critical faults than the strongest baseline method, substantially enhancing the efficiency of high-risk fault detection.

Abstract

Modern advanced driver assistance systems (ADAS) rely on deep neural networks (DNNs) for perception and planning. Since DNNs' parameters reside in DRAM during inference, bit flips caused by cosmic radiation or low-voltage operation may corrupt DNN computations, distort driving decisions, and lead to real-world incidents. This paper presents a SpatioTemporal-Aware Fault Injection (STAFI) framework to locate critical fault sites in DNNs for ADAS efficiently. Spatially, we propose a Progressive Metric-guided Bit Search (PMBS) that efficiently identifies critical network weight bits whose corruption causes the largest deviations in driving behavior (e.g., unintended acceleration or steering). Furthermore, we develop a Critical Fault Time Identification (CFTI) mechanism that determines when to trigger these faults, taking into account the context of real-time systems and environmental states, to maximize the safety impact. Experiments on DNNs for a production ADAS demonstrate that STAFI uncovers 29.56x more hazard-inducing critical faults than the strongest baseline.

Problem

• bit-flip
• deep neural networks
• advanced driver assistance systems
• fault injection
• safety-critical

Innovation

• Spatiotemporal-Aware Fault Injection
• Progressive Metric-guided Bit Search
• Critical Fault Time Identification
• DNN reliability
• ADAS safety