Single-Event Burnout in Semiconductor Devices: Efficient Classification of Ion Species.

Santos, Julia T.; Bianchi, Reinaldo A. C.; Guazzelli, Marcilei A.; de Oliveira, Juliano A.; Alberton, Saulo G.; Giacomini, Renato C.

Single-event burnout (SEB) represents a significant challenge to the reliability of semiconductor devices, especially, but not only, in radiation-rich environments such as space and nuclear facilities. This study presents a novel method for classifying SEB events in power semiconductor devices, leveraging ion species identification through Deep Neural Networks (DNN) and Transfer Learning (TL). Adapting a pretrained DNN originally developed for non-destructive radiation effects enables accurate SEB classification in a different transistor model. This approach is highly effective for designing more resilient power devices, minimizing the need for large datasets and reducing processing time by 39%. With an accuracy of 88%, the method offers a faster and cost-effective solution for gathering insights into the underlying physics, requiring fewer devices and fewer computational resources.

ARTIFICIAL neural networks; POWER semiconductors; IONIZING radiation; SPACE environment; NUCLEAR facilities

Journal of Integrated Circuits & Systems, 2024, Vol 19, Issue 3, p1

1807-1953

Academic Journal

10.29292/jics.v19i3.933