Artificial Intelligence Performance and Radiation Effects in Neuromorphic NorthPole Hardware

One computing concept that is pushing forward efficient applications of AI to real world solutions is the neuromorphic model. Borrowing ideas from biological neural computations, neuromorphic hardware promises to deliver great performance at low power consumption. These characteristics are ideal for space computing applications. The hope is that low power consumption will allow for artificial intelligence techniques, such as object detection, to be deployed in space. It follows that performance under radiation hard environments must be known to be able to assess deployment success in space. This work explores the performance of the You-Only-Look-Once version 4 implemented in IBM NorthPole hardware under heavy ion radiation. The main metric considered is the mean average precision since we are focusing on overall system performance. Results show resilience to radiation up to a threshold after which the probability of failure is high.