Optimizing Semiconductor Probe Precision Forming: Integrating Design of Experiment, Partial Least Squares, and Genetic Algorithms for Advanced Quality Control


The manufacturing of integrated circuit (IC) involves complex design and precision that necessitate rigorous testing using semiconductor probe cards. Probe cards perform electrical tests by making contact with each die on a wafer, identifying chips that meet the electrical specifications from defective ones that do not. However, the precision forming of these probes, which involves embedding a micron-level mold into a machine to shape metal wire into probes, is susceptible to numerous variabilities that can degrade quality. The industry traditionally relies on expert knowledge and trial-and-error to adjust manufacturing parameters, which has become less effective with the decreasing size of ICs. There is an urgent need to develop an adaptive solution that enhances the efficiency and effectiveness of parameter optimization under fluctuating production conditions. To this end, a new study published in Computers & Industrial Engineering and conducted by Assistant Professor Wenhan Fu from the University of Shanghai for Science and Technology alongside Professor Chen-Fu Chien from the National Tsing Hua University and Chi-Hang Chen from the Artificial Intelligence for Intelligent Manufacturing Systems Research Center in Taiwan, the researchers performed a real-time optimization of process parameters for semiconductor probe precision forming, an area that until now, has seen limited research. The study employs a robust methodological framework combining design of experiment (DOE), partial least squares (PLS), and genetic algorithms (GA) for advanced quality control.

The team employed a full factorial design covering all possible combinations of selected process parameters at multiple levels to systematically investigate the effects of various process parameters on the quality characteristics of semiconductor probes. The key parameters included machine pressing speed and stop position of the mold pressing, reflecting the operational settings that directly influence the probe’s formation. They collected data from each combination of parameters in terms of multiple quality characteristics such as offset, tail length, tip length, bending length, straightness, angle, and three measurements of width. Moreover, they tested each parameter combination multiple times to ensure the reliability of the results and to facilitate robust statistical analysis. The authors used PLS regression to handle potential multicollinearity among parameters and to model complex relationships between a large set of variables. This is to develop a predictive model that relates process parameters to probe quality characteristics where they constructed a PLS regression model using the data generated from DOE. The model aimed to predict quality outcomes based on the variations in process parameters. On the other hand, the authors applied a genetic algorithm (GA) to find the optimal setting of process parameters that maximized the desirable quality characteristics. The optimization considered multiple objectives, transforming them into a single fitness function using techniques like gray relational analysis for multi-response optimization. The aim was to optimize the process parameters based on the PLS model to achieve the best possible quality characteristics. The GA used population-based search strategies with operations such as selection, crossover, and mutation to evolve the set of process parameters towards optimal settings.

The authors found that the PLS model effectively captured the relationships between process parameters and quality characteristics, with a significant portion of the variance explained by the model. This demonstrated the capability of PLS to handle high-dimensional data and complex variable interactions in semiconductor manufacturing. Moreover, the genetic algorithm successfully identified process parameter settings that led to an optimal balance of all considered quality characteristics. The optimized parameters significantly improved quality metrics such as straightness and tip length, which are critical for the functional performance of semiconductor probes. Furthermore, the optimized parameters were tested in a production environment, and confirmed the real-world applicability of the model. The implementation of these parameters led to a noticeable improvement in production yield and a reduction in defect rates. Additionally, the approach was found to be both efficient and effective, reducing the time and cost associated with the trial-and-error methods previously used. It also allowed for quicker response times to changes in production conditions, highlighting the adaptability of the integrated approach.

The results demonstrate the effectiveness of the proposed approach in improving the precision of probe manufacturing. The study reports significant improvements in yield and operational efficiency by integrating real-time data and adaptive algorithms into the process parameter optimization. The research conducted by Professor Wenhan Fu, Professor Chen-Fu Chien and Chi-Hang Chen is significant because it can improve the yield and quality of semiconductor probes by enabling more precise and efficient parameter adjustments. This is particularly critical as the semiconductor industry continues to scale down IC dimensions, where traditional methods become less effective. Moreover, the proposed method reduces the dependency on trial-and-error methods and minimizes human error in parameter adjustments, which can significantly cut down operational costs. Improved efficiency also leads to better resource utilization, further driving down costs. Furthermore, the integration of DOE, PLS and GA provides a sophisticated toolkit for real-time optimization. This multidisciplinary approach allows for handling complex, multivariate systems with high inter-correlations among variables, which are common in semiconductor manufacturing. Additionally, the new study addressed the need for adaptable quality control systems that can dynamically respond to changes in production conditions. This is vital for maintaining product quality in an environment where production parameters can vary frequently.

Overall, the new study concludes that the application of an integrated DOE-PLS-GA framework for real-time parameter optimization can significantly enhance the quality control processes in semiconductor manufacturing. The new approach reduces the reliance on expert judgment and mitigates the inefficiencies associated with traditional trial-and-error methods. It is noteworthy to mention that the study also discussed the scalability of the new method and its potential application in other areas of semiconductor manufacturing for instance the authors said it can also impact the broader field of manufacturing optimization through advanced statistical and computational techniques.

About the author

Wenhan Fu is an Assistant Professor in the Department of Industrial Engineering, Business School, University of Shanghai for Science and Technology (USST), Shanghai, China. He is also a Researcher in the School of Intelligent Emergency Management, USST. He received the B.S. degree with double majors in Mathematics and Financial Engineering from Sichuan University, Chengdu, China, in 2014, and the Ph.D. degree in Industrial Engineering and Engineering Management with the Phi Tau Phi Honor from National Tsing Hua University (NTHU), Hsinchu, Taiwan, in 2020. He was a Postdoctoral Researcher in the Zhen-Ding Tech & NTHU Joint Research Center and Visiting Scholar in National University of Singapore. His research interests include decision analysis, big data analytics, quality control and smart manufacturing. His published more than 10 research papers in the international journals including Computers & Industrial Engineering, Journal of Intelligent Manufacturing and Applied Soft Computing. Dr. Fu received the TSC Excellent Ph.D. Thesis Award from the Taiwan Management Institute, Honorable Mention Ph.D. Thesis Award from the Operations Research Society of Taiwan, and Best Paper Award at the 2019 Annual International Conference for Chinese Scholars in Industrial Engineering.

About the author

Chen-Fu Chien is Tsinghua Chair Professor and Executive Vice President, National Tsing Hua University (NTHU), Hsinchu, Taiwan. He is now the President of Asia Pacific Industrial Engineering & Management Systems Society (APIEMS). Since 2018, he has been the Director of Artificial Intelligence for Intelligent Manufacturing Systems (AIMS) Research Center that is one of four national AI centers sponsored by National Science and Technology Council (NSTC), Taiwan. He is the founder and Director for Decision Analysis Laboratory (DALab), the NTHU-TSMC Center for Manufacturing Excellence, and the Zhen-Ding Tech & NTHU Joint Research Center in Taiwan. He received B.S. with double majors in Industrial Engineering and Electrical Engineering with the Phi Tau Phi Honor from NTHU in 1990. He received M.S. in Industrial Engineering and Ph.D. of Decision Sciences and Operations Research at UW-Madison, in 1994 and 1996, respectively. He was a Fulbright Scholar in the Department of Industrial Engineering and Operations Research, UC Berkeley, from 2002 to 2003. From 2005 to 2008, he had been on-leave as the Deputy Director of Industrial Engineering Division in Taiwan Semiconductor Manufacturing Company (TSMC). He received the Executive Training of PCMPCL from Harvard Business School in 2007. He was a Visiting Professor in Institute for Manufacturing, Cambridge University (sponsored by Royal Society, UK), Visiting Professor in Beijing Tsinghua University (sponsored by Chinese Development Foundation), Visiting Professor in Waseda University (sponsored by Japan Interchange Association Young Scholar Fellowship), and Visiting Professor in Tianjin University and Zhejiang University, China.

His research efforts center on decision analysis, big data analytics, modeling and analysis for semiconductor manufacturing, manufacturing strategy, and manufacturing intelligence. Dr. Chien and his DALab Associates have conducted in-depth university-industry collaborative research projects with the leaders of different industrial segments to validate developed solutions and served as senior consultant for leading companies including TSMC, MediaTek, Delta, and AUO. Dr. Chien has received 12 USA invention patents on intelligent manufacturing and published 6 books, 12 case studies in Harvard Business School, and more than 220 journal papers with Google citation number over 10466 and H-index 51. He has been listed as world Top 2% Scientists. He has been invited to give keynote speech in various conferences including APIEMS, C&IE, FAIM, IEEE, IEEM, IML, ISMI, ISSM, leading universities and international companies worldwide. He is a Fellow of APIEMS, CIIE, and CSMOT. Dr. Chien received the National Quality Award, the Executive Yuan Award for Outstanding Science & Technology, three Distinguished Research Awards and Tier 1 Principal Investigator (Top 3%) from NSTC, Distinguished University-Industry Collaborative Research Award from the Ministry of Education, University Industrial Contribution Awards from the Ministry of Economic Affairs, the TECO Award, the 2011 Best Paper Award of IEEE Transactions on Automation Science and Engineering, and the 2015 Best Paper Award of IEEE Transactions on Semiconductor Manufacturing.

About the author

Chi-Hang Chen is a Research Assistant in Artificial Intelligence for Intelligent Manufacturing Systems Research Center, Hsingchu, Taiwan. He received the B.S. degree in statistics from National Cheng Chi University, Taipei, Taiwan, in 2016, and the M.S. degree in industrial engineering and engineering management from National Tsing Hua University, Hsinchu, Taiwan, in 2018. He is the maintainer of R language package KeyboardSimulator, and still dedicate in open source communities. His research interest includes manufacturing intelligence, data mining and application of machine learning.


Wenhan Fu, Chen-Fu Chien, Chi-Hang Chen, Advanced quality control for probe precision forming to empower virtual vertical integration for semiconductor manufacturing, Computers & Industrial Engineering, Volume 183, 2023, 109461,

Go to Computers & Industrial Engineering

Check Also

Enhancing Photovoltaic Performance of Quaternary Cu-Ag-Bi-I Compounds: The Role of Annealing and Impurities in Cu2AgBiI6 Thin Films - Advances in Engineering

Enhancing Photovoltaic Performance of Quaternary Cu-Ag-Bi-I Compounds: The Role of Annealing and Impurities in Cu2AgBiI6 Thin Films