Machine Learning-Based COTS Device Process Identification and Total Ionizing Dose Degradation Prediction via Enhanced Electrical Stress Features

Abstract: Commercial Off-The-Shelf (COTS) devices are widely used in aerospace electronic systems, but their process design does not meet the application requirements of the space radiation environment, leading to performance degradation risks caused by Total Ionizing Dose (TID) effects. However, traditional sampling-average radiation hardness (RH) assessment methods are costly and time-consuming, and fail to effectively address lot-to-lot and within-lot fluctuations in the RH consistency of COTS devices. This paper proposes a machine learning model based on physical feature enhancement. A high-quality dataset is constructed via irradiation experiments on devices from multiple manufacturers and lots. By introducing electrical parameter responses under multiple electrical stress conditions as enhanced feature parameters, the model realizes non-destructive identification of device manufacturing processes and prediction of total-dose radiation degradation. Results show that the model achieves identification accuracy of approximately 0.965 for device manufacturers and 0.842 for lots; at four specific dose points, the coefficient of determination (R²) for radiation degradation prediction is above 0.838, outperforming the sampling-average prediction method. Incorporating electrical parameter responses under multiple electrical stress conditions improves the model’s performance in manufacturing process identification and TID radiation degradation prediction. This study reveals that differences in the pre-irradiation initial electrical parameters of devices have an implicit correlation with their radiation hardness characteristics. Compared with a single test condition, the responses of device parameters to multiple electrical stresses contain richer RH feature information. In addition, the model is verified to have certain generalization ability on new lot samples not included in the training set. This method provides a new approach for the efficient screening and assessment of the radiation hardness of COTS devices.