Prediction of solar particle events with SRAM-based soft error rate monitor and supervised machine learning. (November 2020)

• Record Type: Journal Article
• Authors: Chen, J.; Lange, T.; Andjelkovic, M.; Simevski, A.; Krstic, M.
• Is Part Of: Microelectronics and reliability. Volume 114(2020)
• Publication Date: 2020-11
• Subjects: Electronic apparatus and appliances -- Reliability -- Periodicals; Miniature electronic equipment -- Periodicals; Appareils électroniques -- Fiabilité -- Périodiques; Équipement électronique miniaturisé -- Périodiques; Electronic apparatus and appliances -- Reliability; Miniature electronic equipment; Periodicals; 621.3815
• DOI: 10.1016/j.microrel.2020.113799
• Languages: English
• ISSNs: 0026-2714

Abstract

Abstract: This work introduces an embedded approach for the prediction of Solar Particle Events (SPEs) in space applications by combining the real-time Soft Error Rate (SER) measurement with SRAM-based detector and the offline trained machine learning model. The proposed approach is intended for the self-adaptive fault-tolerant multiprocessing systems employed in space applications. With respect to the state-of-the-art, our solution allows for predicting the SER 1 h in advance and fine-grained hourly tracking of SER variations during SPEs as well as under normal conditions. Therefore, the target system can activate the appropriate mechanisms for radiation hardening before the onset of high radiation levels. Based on the comparison of five different machine learning algorithms trained with the public space flux database, the preliminary results indicate that the best prediction accuracy is achieved with the recurrent neural network (RNN) with long short-term memory (LSTM). Highlights: Embedded approach for the prediction of solar particle events in space application Real-time soft error rate measurement from on-board single event upset monitor Offline trained supervised machine learning model with the historical soft error rate Prediction the soft error rate one hour in advance and fine-grained hourly tracking variations during solar particle events Recurrent neural network with long short-term memory model has the best accuracy