A Predictive Framework for Fault Vulnerabilities in Edge Hardware Accelerators

Modern Hardware accelerators, such as Graphics Processing Units (GPUs), are increasingly being deployed in edge computing safety-critical applications, including autonomous and semi-autonomous automotive systems. However, evaluating fault vulnerabilities and reliability remains highly resource-intensive and time-consuming when using simulation-based fault injection (FI) campaigns. This challenge is amplified by the architectural complexity of modern devices and the huge number of workload scenarios, making traditional FI approaches impractical. This work introduces and evaluates a high-level modeling framework for the early fault propagation analysis and reliability estimation of hardware accelerators for edge computing. Our framework uses Colored Generalized Stochastic Petri Nets (CGSPNs) to map the most crucial features of a design into a model that allows the prediction of fault vulnerabilities in a design. The framework aims to support designers in estimating key reliability metrics, such as fault propagation rates, before low-level implementation (e.g., gate-level), thereby facilitating informed architectural decisions and the development of more robust systems.