With the rapid advancements in technology scaling, the occurrence of both transient and permanent faults has increased, even as MP- SoCs (multiprocessor system-on-chips) achieve higher performance levels. Despite the development of numerous fault detection tech- niques, some faults still go unnoticed, leading to silent data corrup- tions and system failures. In this work, we propose a novel approach to monitor bus transactions and detect anomalies that could lead to critical data corruption or system failure, leveraging different deep learning models.

Our research focuses on data extracted from the AMBA-AHB bus in an MPSoC environment based on NOEL-V processors. By evalu- ating different data representations we demonstrate the effectiveness of using image representations for anomaly detection.

The results highlight that utilizing deep learning with optimized data representations improves the detection of anomalies, offering a robust framework for identifying faults in MPSoCs. This study pro- vides a foundation for future research in fault detection and con- tributes to the development of more reliable MPSoC systems.