Analysis of Single-Event Transients (SETs) Using Machine Learning (ML) and Ionizing Radiation Effects Spectroscopy (IRES)

A methodology for automating the identification of single-event transients (SETs) through ionizing radiation effects spectroscopy (IRES) and machine learning (ML) is provided. IRES enhances the identification of SETs through statistical analysis of waveform behavior, allowing for the capture of subtle circuit dynamics changes. Automated identification of SETs is facilitated by a k -nearest neighbors ( k NNs) ML algorithm with IRES data. One-hundred thousand waveforms were measured from CMOS phase-locked loop (PLL) circuits irradiated at the Naval Research Laboratory's two-photon absorption (TPA) laser facility. Known SET signatures were used to train various k NN models based on statistical features derived from several standard circuit metrics and eight moment-generating functions. Results show that SETs can be automatically identified by the k NN models, with several features resulting in greater than 98% correct identification of SETs. The tradeoffs in ML-based anomaly detection, based on the size of available training sets, choice in signal metric, and the number of included statistical moment-generating functions are discussed, along with opportunities for the future development of specific event-type classification, in situ measurement, and real-time classification of data.