Multi-Algorithm Ensemble Learning Framework for Predicting the Solder Joint Reliability of Wafer-Level Packaging.

Area of Science:

• Materials Science
• Computer Science
• Mechanical Engineering

Background:

• Traditional design-on-experiment methods for packaging reliability are costly and time-consuming.
• Machine learning (ML) offers potential for efficient prediction but faces challenges with limited training data and high computational costs.
• Validated finite element analysis (FEA) can generate simulation data, but optimizing ML training with reduced datasets is crucial.

Purpose of the Study:

• To develop an efficient prediction approach for packaging reliability using validated FEA and ML.
• To address the challenges of reduced training dataset size and maintain prediction accuracy in ML models.
• To propose an ensemble learning framework for robust and accurate packaging reliability predictions, using Wafer-Level Packaging (WLP) as a case study.

Main Methods:

• Utilized validated finite element analysis (FEA) to generate simulation data for packaging reliability.
• Applied machine learning (ML) techniques to predict packaging reliability based on FEA-generated data.
• Developed and implemented an ensemble learning framework integrating multiple ML algorithms to enhance predictive robustness.

Main Results:

• The ensemble learning framework demonstrated improved generalization capabilities for packaging reliability prediction.
• Accurate predictions were achieved even with constrained training data, overcoming limitations of small data ML.
• The proposed approach effectively enhances predictive robustness by leveraging complementary strengths of diverse ML algorithms.

Conclusions:

• Ensemble learning provides a robust solution for predicting packaging reliability with limited data.
• This approach enhances design efficiency by reducing reliance on costly experimental methods.
• The framework offers a computationally efficient and accurate method for packaging reliability assessment.