A Sparsity-Promoting Time Domain Evaluation Method for Thermal Transient Measurement of Power Semiconductors

  • Yi Zhang
  • , Anton Evgrafov
  • , Shuai Zhao
  • , Sven Kalker
  • , Rik W. De Doncker

Research output: Journal article publicationJournal articleAcademic researchpeer-review

Abstract

This article investigates evaluation methods of thermal transient measurements to obtain the internal thermal structure of semiconductor devices. First, the study uncovers the limitations of a widely accepted standard method that uses frequency-domain deconvolution. An important finding is that the sideband of the time constant spectrum by the standard method has no physical meaning despite it beeing interpreted as a continuous spectrum for a long time. Second, by understanding the limitations of the existing method, the article proposes an alternative method that remodels the frequency-domain deconvolution as a regularized least squares problem in the time domain. With the benchmark of the true values of several thermal networks based on simulation, the proposed sparsity-promoting method demonstrates several advantages, including a better ability to identify adjacent parameters in the time-constant spectrum and the obtained structure function reducing relative error by an order of magnitude. The influence of varying noise levels has also been evaluated. Finally, a proof-of-concept experiment using a commercial power semiconductor device validates its effectiveness.

Original languageEnglish
Pages (from-to)7525-7535
Number of pages11
JournalIEEE Transactions on Power Electronics
Volume39
Issue number6
DOIs
Publication statusPublished - 1 Jun 2024

Keywords

  • Reliability
  • semiconductor device packaging
  • system identification
  • thermal analysis
  • transient analysis

ASJC Scopus subject areas

  • Electrical and Electronic Engineering

Fingerprint

Dive into the research topics of 'A Sparsity-Promoting Time Domain Evaluation Method for Thermal Transient Measurement of Power Semiconductors'. Together they form a unique fingerprint.

Cite this