Impact of Radiation on Spatio-Temporal Pattern Recognition in Memristor-Based Neuromorphic Circuits

Computers today generally rely on the von Neumann architecture and are extremely susceptible to bit errors as well as component damage caused by environments with heavy radiation. Moore's Law is also nearing an end due to fundamental physical constraints. Thus, advancements in computation are being made less by improved device design or scaling and more by architectural optimization. Artificial neural networks (ANNs) operate very differently from digital computers and rely on a high degree of interconnectivity between components as well as integrated memory. Specifically, neurons connect to thousands of other neurons via synapses, which alter the connection strengths depending on learning rules such as spike timingdependent plasticity (STDP). ANNs are able to solve many complex problems and perform cognitive tasks much more efficiently than digital electronics, and are likely to be implemented in future computational systems. They are also expected to be more robust and reliable. This work focuses on understanding how radiation affects STDP learning behavior and pattern recognition capability in circuits with memristive synapses. Measured data will be used as a guide for development of memristor and transistors SPICE (simulation program with integrated circuit emphasis) models that capture the physical effects of strikes by gamma, neutron, proton, and beta radiation produced by a variety of sources. The study will include both high fluence events (caused by nuclear fission and fusion detonations) and long-term exposure from fallout or other types of radiological weapons. Network schematics and associated layouts will be created, with Monte Carlo simulations guiding the probable location, timing, and dose of strike events during operation of the network. These events will then be included in transient SPICE simulations of learning and recognition, where factors such as pattern detection success, failure, and false positive rates will be analyzed to determine the overall impact of radiation. Based on the findings, final recommendations will be made regarding best practice device and circuit design to mitigate the threats posed by radiation exposure