EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

2941. Variability reduction in semiconductor manufacturing using machine learning and explainable AI

Invited abstract in session MD-23: Scheduling for sustainability, stream Circular Economy, Remanufacturing and Recycling .

Monday, 14:30-16:00
Room: 82 (building: 116)

Authors (first author is the speaker)

1. Venugopal Ramadasu
Department of Information Systems and Operations Management, Vienna University of Economics and Business
2. Fabian Lindner
Faculty of Business Administration and Engineering, Zittau/Görlitz University of Applied Sciences
3. Gerald Reiner
Department of Information Systems and Operations Management, Vienna University of Economics and Business
4. Germar Schneider
Infineon Technologies GmbH & Co. KG
5. Gerhard Luhn
SYSTEMA Systementwicklung Dipl.-Inf. Manfred Austen GmbH

Abstract

The impact of variability on a production system in terms of flow time and work in progress is a well-known problem. Recent research works investigating the application of machine learning (ML) models on production data to improve process performance have shown promising results. Drawing inspiration from such works, in the first part of the study we do partial replicative research and apply existing ML models to identify the sources of variability such as machine parameters, defects, product routing and so on that lead to increased flow time in a semiconductor manufacturing plant. Using the production data for a work center in the wafer production process, we identify the critical parameters in the work center that influence the flow time using ML algorithms such as random forest, neural networks and gradient boosting. To enhance the interpretability of the results, we apply explainable AI (XAI) approaches such as Shapley Additive Explanation (SHAP). As a second step, the results will be presented to process experts to derive actions that can reduce the variability in the process and thus enhance the efficiency in resource utilization. Through our work, we expect, on the one hand, to reinforce the potential benefits of using ML models to understand and improve production processes, and on the other hand, to increase the trust of experts on the use of ML models and to leverage their knowledge to tackle the issues they face in manufacturing.

Keywords

Status: accepted

Back to the list of papers