Comparative analysis and evaluation of ageing forecasting methods for semiconductor devices in online health monitoring

TL;DR

The paper tackles prognostic forecasting for MOSFET ageing under thermal fatigue, focusing on bond-wire lift-off via the precursor $R_{DS_{ON}}$. It compares classical state-space trackers (EKF/UKF), statistical methods (ARIMA/Holt), NN ensembles, and Temporal Fusion Transformers (TFTs) across short- and long-term horizons, with TFTs enhanced by covariates to improve long-horizon accuracy. Short-term results favor traditional methods or NN ensembles for speed and reliability, while TFTs excel in long-term forecasting and reveal key ageing turning points through attention. Practically, the work informs selecting prognosis models for online health monitoring and suggests covariate design as a critical driver for TFT performance; future work points to edge deployment and foundation-model adaptations. Overall, TFTs emerge as a powerful tool for long-term MOSFET prognostics when covariates are available, complementing faster, simpler methods for short-term monitoring.

Abstract

Semiconductor devices, especially MOSFETs (Metal-oxide-semiconductor field-effect transistor), are crucial in power electronics, but their reliability is affected by aging processes influenced by cycling and temperature. The primary aging mechanism in discrete semiconductors and power modules is the bond wire lift-off, caused by crack growth due to thermal fatigue. The process is empirically characterized by exponential growth and an abrupt end of life, making long-term aging forecasts challenging. This research presents a comprehensive comparative assessment of different forecasting methods for MOSFET failure forecasting applications. Classical tracking, statistical forecasting and Neural Network (NN) based forecasting models are implemented along with novel Temporal Fusion Transformers (TFTs). A comprehensive comparison is performed assessing their MOSFET ageing forecasting ability for different forecasting horizons. For short-term predictions, all algorithms result in acceptable results, with the best results produced by classical NN forecasting models at the expense of higher computations. For long-term forecasting, only the TFT is able to produce valid outcomes owing to the ability to integrate covariates from the expected future conditions. Additionally, TFT attention points identify key ageing turning points, which indicate new failure modes or accelerated ageing phases.

Figures (12)

• Figure 1: MOSFET degradation cycle example calculated from the case study dataset (cf. Section \ref{['sec:CaseStudy']}).
• Figure 2: TFT architecture including Gated Residual Networks, Variable Selection Networks and encoder-decoder structure.
• Figure 3: $V_{DS}$, $V_{GS}$ and $I_{D}$ used to compute $R_{DS_{ON}}$.
• Figure 4: Obtained run-to-failure experiments after pre-processing.
• Figure 5: TFT architecture selection for short-term MOSFET ageing forecasting for different tests (T09, T11, T12, T36) with 33% of data used for training.