This article describes a method to detect and classify single-event transients (SET) to determine the originating circuit node impacted by ionizing radiation. SETs were measured via two-photon absorption (TPA) laser excitation on a custom CMOS phase-locked loop (PLL), and Convolutional Neural Networks (CNN) were used to classify the spatial dependencies of the transient responses. A clustering technique is described to identify groups of related circuit nodes and achieves over 90% identification accuracy.
Supervised deep learning is a subset of machine learning (ML) that allows for the development of phenomenological models based on measured observations through layered artificial neurons. A convolutional neural network (CNN) is one model typically used in image recognition that allows for classifying behavior in discrete sets or enabling regression analysis. This work demonstrates a one-dimensional CNN to analyze irradiated commercial off-the-shelf (COTS) electronics. COTS microcontroller units (MCU) were irradiated with 10 keV X-rays, and CNN models were trained on the resulting internally generated noise signatures from the clock module. As a result, the MCUs are accurately classified as either fresh or dosed, and the total ionizing dose (TID) is predictable through a regression model. This approach allows for part dosimetry and in situ device health monitoring without additional components.