Radiation Resilience: Taming the Cosmic Storm with Brain-Inspired Computers

Abstract

The number of satellites in orbit is increasing at an accelerating pace. One major issue that currently hinders the range of satellite applications is data analysis. Most data is transmitted back to ground stations for analysis resulting in bandwidth-related issues, or the processing is performed on board, necessitating large computational power. Spacecraft, particularly miniaturized ones, face stringent energy constraints due to the scarcity of resources and the harshness of the space environment. Processing data at the edge using Spiking Neural Networks (SNN) applied on mixed-signal Neuromorphic Computing (NC) processors has been proposed as a potential solution. Neuromorphic devices focus on low power and energy-efficient operation, which aligns well with the requirements of space applications. Nevertheless, the response of these devices to the harsh conditions of the space environment remains only partially understood. The primary uncertainty lies in the impact of cosmic radiation, which can present substantial challenges, even in low orbits. Radiation not only has the potential to damage hardware but also to interfere with its operation, leading to potentially detrimental software failures. The objective of this work is to describe and analyze the effects of space radiation on NC processors. A behavioral model of different types of radiation effects is composed and experimental verification is performed at a proton beam facility using a mixed-signal NC prototype. The radiation effects are analyzed, and their influence on network operation is discussed. It is demonstrated that mixed-signal NC is resistant to perturbations caused by radiation. Additionally, software mitigation strategies are proposed to further increase the applicability of SNNs in radiation environments.