Roughening Particle Filter Based Prognosis on Power MOSFETs Using ON-Resistance Variation

Prognosis on microelectronics devices has gained wide attention due to its usability in wide range of applications. In this work, we have explored the degradation of Power MOSFETs due to thermal cycling and used the ON-resistance variation as the failure precursor for prognosis. The particle filter algorithm has been adopted for this purpose and the drawbacks of these algorithms such as particle degeneracy and particle impoverishment have been dealt with. An independent Gaussian jitter noise is purposely added to the resampled particle with the intention of roughening. The results indicate an improvement in accuracy over the standard particle filter algorithm. The impact of different degrees of roughening on the accuracy of RUL prediction is explored and the performance is evaluated using prognostic metrics such as relative accuracy and computational time.