Machine Learning Assisted Soft Error Variation Analysis of Radiation Hardened MRAM Circuit

Radiation-induced soft errors are becoming more challenging to the reliability of conventional memories with the downscaling process technology and increasing miniaturization of transistors. Spintronic-based memory is a viable contender for addressing this challenge due to its non-volatility, high speed, and radiation tolerance. However, radiation induced charges can still affect hybrid spintronic/CMOS peripheral circuitry. Machine learning (ML) can be a powerful approach for accelerating the analysis of radiation-hardened circuits, especially in the context of magnetic random-access memory (MRAM) technology. This paper proposes a novel machine learning (ML)-assisted approach for investigating the variability of radiation induced charges on radiation hardened MRAM circuit to shorten the loop of technology pathfinding. Our approach is based on training ML based algorithms to predict the probability of soft errors occurring in the radiation hardened (RH)-MRAM circuit. ML algorithms are trained using a combination of simulated data, allowing it to accurately model the effects of radiation-induced disturbances on the circuit. Our ML assisted approach provided a more accurate and efficient analysis of soft errors in the MRAM circuit, with R2 scores of 0.99. Furthermore, significant improvement in prediction time is observed as compared to the SPICE simulation time.