X‐Ray TID Suppresses Inhibitory Plasticity of SnO‐Based Memristors and Its Impact on Neuromorphic Computation

ABSTRACT X‐ray total ionization dose (TID) significantly modifies synaptic plasticity in SnO‐based volatile memristors, a device class promising for neuromorphic computing in radiation‐exposed environments. We further observe that synaptic operation becomes more stable when training is initiated after a reset, indicating a protocol‐dependent mitigation of variability. Device‐parameterized convolutional neural network simulations indicate that nonlinearity and epoch‐to‐epoch variation both decrease in irradiated devices, leading to higher training accuracy (≈80% after 200 epochs) than pristine counterparts (<50%); confusion‐matrix analysis further highlights nonlinearity as a more critical determinant of performance than raw dynamic range. To interpret this phenomenon, we propose a conduction mechanism that accounts for potentiation, relaxation, and depression dynamics under irradiation. These findings show that radiation‐induced modifications—often considered detrimental—can be harnessed to strengthen key synaptic metrics in volatile memristors. We highlight a design trade‐off in which linearity and stability may be prioritized over raw dynamic range for learning accuracy, and this positions TID‐tuned memristors as candidates for robust neuromorphic hardware operating in extreme environments.