Multi-Kernel Execution: Applying Traditional Multi-Core Processing Methodology to GPU Architecture for AI Computation in Safety Critical Systems

This paper introduces Multi-Kernel Execution (MKE), a novel framework developed by the US Army DEVCOM AvMC MCP-AI Laboratory for correlating parallel kernel execution on GPUs with that of traditional multi-core systems. This correlation allows existing safety analysis methodologies, typically applied to traditional multi-core processors, to be readily adapted for contemporary AI hardware. The increasing use of AI in defense and security, particularly in aviation, necessitates highperformance parallel processing architectures like GPUs. However, the architectural differences between GPUs and traditional multi-core processors present challenges in establishing unified safety methodologies. MKE directly addresses this challenge, streamlining the adaptation process and minimizing resource investment. By leveraging MKE, developers can enhance the reliability and effectiveness of AI-driven systems in mission-critical applications, improving the safety and performance of AI in complex computing tasks. This innovation significantly reduces the barrier to entry for integrating advanced AI into safety-critical real-time systems.