Radiation Hardened Power Efficient Artificial Intelligence and Machine Learning (AIML) Processor

Radiation Hardened Artificial Intelligence and Machine Learning (AIML) processors are essential to NASA applications due to unavoidable evolution in technology that brings computational complexity. Traditional radiation hardening methods incur high size, weight, and power (SWaP) cost. We propose a radiation hardening scheme that employs intelligent analysis of vulnerabilities in the AIML processor that enables hybrid selective redundancy implementation to allow self-healing from radiation effects. The proposed radiation analysis method will use statistical and probabilistic models that can capture the stochasticity of AI, ML and Neural net inference. In Phase I, we will design the radiation hardened AIML processor using a defined testbed and simulation process. We will calculate the performance characteristics, including error mitigation and SWaP cost, and derive guidelines for improved design. In Phase II, we will optimize the design to meet NASA requirements, build a prototype, and experimentally verify the performance of radiation hardened AIML processor.