Achieving Dependability of AI Execution with Radiation Hardened Processors

The reliance on radiation-hardened hardware, essential for domains requiring high-dependability such as space, nuclear energy and medical applications, severely restricts the choice of components available for modern AI-intensive tasks, particularly for real-time AI-based classifications. To address this challenge, we propose leveraging the High Performance Data Processor (HPDP) as a radiation-hardened and low-power co-processor in conjunction with an optimized AI framework for efficient data processing. The HPDP's dynamic reconfiguration capabilities and dataflow-oriented architecture provide an ideal platform for executing AI-driven applications that demand low-latency, high-throughput streaming data processing. To fully utilize the co-processor's capabilities, we utilized Klepsydra's AI-runtime inference framework, which, due to its lock-free execution and efficient resource management, significantly enhances data processing throughput without increasing power consumption. Our approach entails programming the HPDP as a dedicated mathematical backend, enabling the AI framework to execute workloads directly on this co-processor without requiring additional hardware-specific coding. This paper presents the preliminary results of our implementation, describing the application domain, AI pipeline, key features of the HPDP architecture, and performance evaluation. Our solution demonstrates a significant advancement in deploying AI on radiation-hardened platforms by using the HPDP as a dependable, efficient, and reprogrammable co-processor, making it highly suitable for any application requiring dependable execution in any environment.