The rapid changes in the business environment are transforming the global manufacturing industry and driving the growth of smart manufacturing. According to the leading market intelligence provider TrendForce, the smart manufacturing market is estimated to reach US$620 billion in 2026, with an annual growth rate of 12.7%. What semiconductor players would like to do with smart manufacturing, as opposed to their peers in traditional manufacturing, is to improve production quality and efficiency through Internet of Thing (IoT), big data analytics, artificial intelligence (AI), automation, cloud computing, and other technologies, so as to move one step closer to advanced manufacturing.
The Smart Manufacturing Forum held during SEMICON Taiwan 2023 was aimed to address some of the challenges brought by advancements in manufacturing processes and increasingly complex equipment. Dr. Scott Yu, Deputy Director of TSMC and Co-chair of SEMI Taiwan Smart Manufacturing Committee, and Dr. Mark da Silva, Senior Director of SEMI opened the forum by welcoming experts from various fields and believed that AI, machine learning (ML), big data and other latest technologies would help the semiconductor industry to go up yet another level from being automated to being smart and create new possibilities for all members of the industry.
Generative AI: A Booster for Smart Manufacturing
AI which has been developing very fast in recent years has become a key driver for the industry, but most people have no idea about the differences of AI-related technologies. Traditional AI allows computers to imitate human behaviors to solve tasks through learning, analysis, and prediction, and machine learning is brought in to make AI better by processing data with statistical and mathematical methods to train the system to address specific problems. What deep learning can do is allow AI to crack complex problems with the use of data and neural networks. As for generative AI which has garnered everybody’s attention recently, it’s all about how neural networks can learn from very large data sets, understand the context, and have the ability to generate new and unique data just like human beings. It is expected to empower smart manufacturing to do much more than it does right now.
Keynote speaker Dr. Andrew Liu, Senior Data Scientist of NVIDIA pointed out that large language models (LLMs) have also come to the attention of the world along with the wave of generative AI. Compared with traditional natural language processing methods, LLM stands out for not requiring labeled data and the huge number of parameters that often reach billions or hundreds of billions could be managed by a single model without retraining, to expand the influence of AI in industries and enterprises. LLM training method could be very simple, which is to allow the AI system to have the ability to predict what the next word would be after going through a large amount of training. Followed by further customized training, it will be ready to take care of different kinds of problems in the industry. Generative AI is expected to be worth as much as US$150 billion, while global software output is about US$685 billion. Goldman Sachs even estimates that AI technology will contribute up to 7% to global GDP.
“The common development process can be roughly divided into concept development, case collection, concept validation, prototype design, optimization, and expansion, and mass production will be the last stage,” said Mr. Jixin Jia (Gin), Microsoft AI Global Black Belt, Asia Data & AI of Microsoft.“ In a world in which cybersecurity threats are literally everywhere, the semiconductor industry must have a comprehensive cybersecurity protection mechanism that prevents important data from being stolen and likewise severe incidents. Azure OpenAI is a powerful cognitive service that comes with a lot of planning done based on security, such as IM-1 for identity management, PA-3 for privacy access, NS-1 for network segmentation, and DP-4 for encryption sensitivity, allowing users to strengthen overall security and protection through the mix of functions.”
Mr. James Lin, Deputy Division Director of UMC and Co-chair of SEMI Taiwan Smart Manufacturing Committee said that both the pyramid model for smart manufacturing vision formulated by SEMI Smart Manufacturing Global Executive Committee and the levels of driving automation defined by SAE (Society of Automotive Engineers) would be incorporated into making AI-driven autonomous smart factories come true, while improving UMC's long-term competitiveness in semiconductors. As for the challenges in technical development or management thinking that would definitely come in the process of implementing smart manufacturing, he suggested companies interested in making smart manufacturing also a part of their production process or operation first have a smart manufacturing vision and complete process auditing/review, and then continue to enhance corporate value through three steps: proof of concept, replicating successful models, and large-scale expansion.
For companies that would also like to develop smart manufacturing, his suggestion was to “Generative AI could be seen as a major revolution in AI technology, and it also extends application scenarios to areas such as digital assistants and production optimization,” pointed out Mr. Charles Sevior, CTO, Unstructured Data Solutions of Dell Technologies. For example, a large amount of data collected during the manufacturing process in the semiconductor industry could be used to train designated AI models to check whether or not there are defects in the wafer design, production process, and other stages, thereby enhancing wafer yield and reducing overall costs.
Improve Quality and Optimize Manufacturing Processes with Smart Manufacturing
It is true that chip manufacturing technology has continued to evolve and go beyond its limits due to cross-sector collaboration in the industry in recent years, but unprecedented challenges are also coming. Firstly, it takes the entire industry to establish multiple correlations between material parameters, yields, or electrical properties to move steadily toward advanced manufacturing. Secondly, a more precise control and monitoring method is also needed to reduce costs and improve efficiency. That explains why the semiconductor industry has long been trying to use innovative technologies such as AI and ML to create smart platforms that can collect and analyze large amounts of data to make automation and digitization possible, to move the industry along more rapidly to advanced manufacturing processes.
Dr. Jae Yong Park, VP of Technology, Mechatronics Research of Samsung stressed the importance of an AI-driven data analysis function that enables the manufacturing industry to mine useful information from a large amount of manufacturing data and to enhance decision-making and optimizing processes. It is widely used in semiconductor fabs for real-time monitoring, fault detection, predictive maintenance, and output forecasting, which is of great help in increasing output, reducing defects, and enhancing overall operational performance.
“Merck is a major global supplier of semiconductor materials,” said Dr. Safa Kutup Kurt, Head of Operations for Digital Solutions at the Electronics business of Merck KGaA, “what you can get from us are very comprehensive digital solutions that help different industry players address problems that arise with wafer size reduction and process changes and many other pain points. The smart manufacturing strategy launched exclusively by Merck KGaA uses advanced analysis and modeling tools, coupled with complete subsequent analytical capabilities, to help clients increase output and facilitate robust manufacturing.
“Smart manufacturing is a method of analyzing data generated by equipment to improve equipment’s health monitoring and process control. It is not only suitable for the most advanced fabs or factories, but can be widely used in other manufacturing industries.” Mr. Chan Pin Chong, Executive Vice President & General Manager, Products & Solutions of Kulicke & Soffa explained: “The manufacturing industry today continues to improve manufacturing processes, and finds itself faced with challenges in chip miniaturization, high density and more. It would be nice to bring smart manufacturing technology into making the back-end packaging stages more smooth with heterogeneous integration(HI) and 2.5 D/3D packaging technology, extending the application time of Moore’s Law.”
Dr. Joseph Ervin, Senior Director, Semiconductor Software Products at Coventor, A Lam Research Companywalked the audience through the challenges facing semiconductors in the process towards advanced manufacturing, saying that it takes virtual process modeling, virtual metrology, virtual process optimization, monitoring of image measurement and other technologies to get it done. Take ML technology, which is widely used in different fields as an example. It has shown considerable benefits in real-time feed-forward and feedback optimization, and is one of the indispensable technologies for optimizing semiconductor manufacturing processes.
“As the world tries to double the pace of energy efficiency progress, digital transformation not only is key to further economic growth, but also matters a lot to whether or not humans can reduce energy consumption,” highlighted by Ms. Katharina Westrich, Global Head of Vertical Management Semiconductor Digital Industries, Siemens. She believed that for advanced manufacturing to take root in semiconductors, smart manufacturing plays a vital role that increases energy usage as well as cuts material waste. It is quintessentially one of the most important steps in making ESG goals become reality.
Mr. Gabriel Villareal, Director of Applications Engineering, Mask and Smart Manufacturing (MSM) of Synopsys shared his take on the emergence of all kinds of new chip architectures and graphics technologies, saying that they are brought to the market to meet various applications/service needs and therefore drive the semiconductor industry to move towards advanced manufacturing. Compared with 10 years ago, today's wafer manufacturing process is more complex and has to rely on AI, ML, and other tools to analyze large amounts of data to solve different problems in the production process, and at the same time takes the speed of mass production of advanced chips to the next level.
The semiconductor industry is not only among the most heavily invested businesses in automation, but also the earliest one to carry out smart manufacturing and has the widest range of applications that allow a lot more consumers to enjoy the convenient life made possible by advanced manufacturing. This approach of leveraging smart manufacturing to develop advanced manufacturing is definitely worth learning for other manufacturing industries.
The ‘SEMI Smart Manufacturing Initiative’ was launched to help chipmakers and suppliers solve problems that might be happening when adopting smart manufacturing applications, to shorten the time it takes to have a return on investments, improve the manufacturing process, and optimize the ‘sensing-connection-prediction chain’ operating efficiency. SEMI has long been committed to promoting industry development, and will continue to work with the SEMI Smart Manufacturing Committee on cross-border resource integration and exchanges in the future and reach the goals of smart manufacturing and digital transformation together with Taiwan's semiconductor industry chain players upstream to downstream.