Rajaguru, Pushparajah ORCID: https://orcid.org/0000-0002-6041-0517 (2025) Anand Model parameter estimation for the aluminium wirebond in power electronic module and Lifetime prediction by combining the finite element analysis and Machine learning. Microelectronics Reliability. ISSN 0026-2714 (Print), 1872-941X (Online) (In Press)

PDF (Author's Accepted Manuscript) 51238 RAJAGURU_Anand_Model_Parameter_Estimation_For_The_Aluminium_Wirebond_In_Power_Electronic_Module_(AAM)_2025.pdf - Accepted Version Restricted to Repository staff only Download (1MB) | Request a copy

Abstract

This report focuses on the estimation of Anand viscoplastic model parameters for aluminum wirebonds, a critical component in Power Electronic Modules (PEMs). These complex PEM inhomogeneous structures are prone to thermo-mechanical failure due to heat generation and material Coefficient of Thermal Expansion (CTE) mismatches. The wirebond failures account for approximately 70% of total PEM failures. The study addresses a gap in existing literature by deriving Anand model parameters for aluminum wirebonds from experimental tensile data. This involved of conducting isothermal uniaxial tensile tests on pure aluminum wire at various temperatures and strain rates and measuring the stress strain profile of each sample specimens. The nine Anand model parameters were then determined through a four-step non-linear fitting process. The accuracy of these estimated parameters was validated by comparing stress-strain curves from Finite Element Analysis (FEA) simulations with experimental data, showing a good fit across various conditions. The research proceeded to predict the fatigue lifetime of wirebond structures under various thermal cyclic loading scenarios, adhering to JEDEC standards. Accumulated plastic strain at the wirebond heel was identified as a key lifetime prediction parameter, utilizing the Coffin-Manson relationship. The analysis revealed an exponential decrease in wirebond lifetime with increasing temperature difference (ΔT) and upper thermal cycle temperature. Finally, the study explored using tree-based machine learning (ML) regressors (Random Forest, Decision Tree, and XGBoost) to predict accumulated plastic strain, aiming to mitigate the need for computationally expensive FEA simulations. Trained on a small dataset from 11 FEA simulations, the Decision Tree model exhibited a reasonable prediction error of 2.4%, suggesting the potential for ML to provide efficient and reasonably accurate lifetime predictions in power electronics.

Item Type:	Article
Uncontrolled Keywords:	reliability, fatigue
Subjects:	Q Science > Q Science (General) Q Science > QA Mathematics Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Faculty / School / Research Centre / Research Group:	Faculty of Engineering & Science Faculty of Engineering & Science > School of Computing & Mathematical Sciences (CMS)
Last Modified:	13 Nov 2025 17:57
URI:	https://gala.gre.ac.uk/id/eprint/51238

Actions (login required)

View Item

Downloads