AI Predicts Cosmic Ray Effects on Electronics Millions of Times Faster

In the harsh environment of space, electronics are constantly bombarded by heavy ions that can cause malfunctions known as 'single event effects' (SEE). Ensuring components like Silicon carbide (SiC) MOSFETs can withstand these events is critical for aerospace reliability. Traditionally, predicting these effects required highly accurate but computationally intensive simulations, a process that is extremely slow.

Now, researchers have created a vastly more efficient data-driven alternative. They constructed a dataset of over 52,000 simulated SEE events and used it to train two specialised deep learning models. One model, a Residual Deep Neural Network, predicts the peak electrical current and charge, while another predicts the entire current pulse over time.

The results are remarkable. The AI models achieved accuracy scores above 99.7% while performing the predictions five to six orders of magnitude—that is, from 100,000 to one million times—faster than the old methods. This breakthrough significantly reduces computational cost and offers a powerful new tool for modelling the reliability of crucial semiconductor devices.