Use of machine learning and artificial intelligence (ML/AI) is on an exponential rise across fields1 including all aspects of the semiconductor industry. In the last decade, the use of ML/AI exploded in the areas of speech recognition, facial recognition, smart phone features, search engines and now large language models like ChatGPT, Bard AI, and CoPilot. The ML/AI growth has been enabled by massive data storage capacity and increased compute performance, leading to projections for the semiconductor industry to reach over $1 trillion in annual revenue by 2030, with about 50% of the industry’s growth related to GenAI2. Figure 1: McKinsey Company on GenAI driving semiconductor industry growthAs semiconductor manufacturing drives toward Industry 4.0, SEMI member companies have a vision of Industry 5.0, truly adaptive manufacturing, integrating human creativity with robotic precision enabled by AI. Along that path, automation and data exchange in every step of manufacturing is essential, with data acquisition, data integrity and relevance, and operational Digital Twins3 as defined steppingstones to the factory of the future.Based on growing member interest in ML/AI, in 2019, SEMI assembled technology communities that quickly engaged in AI discussions and proofs of concept, discovering gaps in the path to Industry 4.0. Successful demonstrations of the value of AI in chip manufacturing process development and factory efficiency, not to mention GenAI uses in society, hastened the pace to produce faster, more powerful chips to accommodate the computation and communication requirements. Recognizing the industry opportunity and the mounting role AI plays in the semiconductor supply chain, SEMI initiated several thought leadership efforts, namely the Smart Manufacturing Initiative, Smart Data-AI Initiative, and the Future of Computing think tank.Smart Manufacturing According to the SEMI World Fab Forecast, over 100 new and expanded wafer fabs will begin volume production by 2027. This massive capacity expansion will need to achieve the highest possible operational efficiency and performance. To this end, the Smart Manufacturing Initiative is a technology community with over 120 member companies collaborating pre-competitively to transform manufacturing. The SEMI Smart Manufacturing Global Executive Committee (GEC), outlined a roadmap vision for the cognitive factory of the future based-on technology, sustainability and future talent. The GEC has been working with members to realize that vision. Figure 2 describes this vision in terms of the technology progression needed and the approximate timeline for implementation by most manufacturers. The proliferation of this vision through Smart Manufacturing Forums at SEMICON events around the globe, newsletters and blogs has garnered enormous interest and participation in the initiative and is central to the mission of connecting and raising awareness within the ecosystem. Figure 2: AI-Driven Smart Factory (Point Systems to Autonomous Solutions) To move the needle on this vision, industry experts in the initiative successfully created and launched the Industry 4.0 Readiness Assessment Model (IRAM) to help assess technology deployment progress. IRAM adoption is steadily growing. Modern front-end and back-end lines produce an extraordinary amount of multi-modal data from a variety of sources, and this is key to success in unlocking the potential of AI in manufacturing environments. The initiative’s global working groups on Data Architectures and Smart Control Room among others are working towards a holistic Cognitive Factory framework uniting the vertical and horizontal flow of information. Integral to the Cognitive Factory are smart manufacturing standards, that will accelerate the vision outlined above, and without which local solutions are unlikely to scale.In 2023, the Smart Manufacturing Initiative brought together industry leaders in a unique Digital Twin workshop to align on the state of semiconductor development and usage. The key takeaways from this workshop are captured in a white paper that highlighted the need to accelerate efforts in multiple areas including standards. Along with SEMI International Standards, Smart Manufacturing supports other standards development organizations (SDOs) and NIST standards development, for example, to identify and drive critical standards for Cognitive Factory implementation. The initiative is planning future workshops on Cognitive Factory Framework requirements, Digital Twins, and Smart Data AI in the coming months. that highlighted the need to accelerate efforts in multiple areas including standards. Along with SEMI International Standards, Smart Manufacturing supports other standards development organizations (SDOs) and NIST standards development, for example, to identify and drive critical standards for Cognitive Factory implementation. The initiative is planning future workshops on Cognitive Factory Framework requirements, Digital Twins, and Smart Data AI in the coming months.The GEC has identified critical interrelationships in addition to the technology focus. At the intersection with sustainability, the initiative has formed a collaborative task force with the SEMI Semiconductor Climate Consortium (SCC) to develop a bottom-up technology roadmap that can be used as a blueprint for device makers to meet their proclaimed sustainability goals faster. The task force organized a technical session at SEMICON West 2024 and will be releasing a white paper in the near future. Similarly, the initiative is working with the SEMI Foundation to identify necessary future skills and to make training available through SEMI University. Smart Data AI – Applying AI to Semiconductor OperationsSEMI’s Smart Data-AI Initiative started by assembling a group of interested companies to explore the pivotal role AI could play in the industry and to address the criticality of data. All stakeholders agreed that a formidable challenge was (and still is) the integrity of that data and the security of sharing that data, which is considered IP to most. The optimal implementation of ML/AI techniques can only be gained by access to the comprehensive data set which is owned by numerous supply chain partners. Consequently, semiconductor R D, process and design have not yet realized the full benefit of Data-AI advances. In response, the initiative developed a framework to create value for members and support industry progress. Four pillars underpinning the strategy are:Educating stakeholdersBuilding communitiesExecuting proof-of-concept projectsDeveloping industry standardsTo explore the data challenges the subject matter experts highlighted, a collaborative proof-of-concept (POC) project was proposed in 2019 and accepted by the initiative's partners at Army Research Laboratories4 along with academic and industry partners. The project has completed two phases and is starting on its third phase under the expert guidance of an Industry Advisory Council (IAC) comprised of leaders in the Smart Data-AI community.The POC project, being conducted by principal investigators at Cornell University, demonstrated significant accomplishments from the first two phases, including:An AI model to predict device geometry by optimizing photolithography and plasma etching processesInitial demonstration of secure data-sharing techniques with software-hardware co-optimizationInnovative metrology ideas to train AI algorithms rapidlyStudents trained in cross-disciplinary skills to address the industry’s critical talent shortageFurthermore, the visionary objectives laid out at the initial stages of the POC proved to be synergistic with the strategic goals of the CHIPS Act5, which articulates the need for “collecting, aggregating, and sharing data sets that enable benchmarking and operational improvements, tools development, the creation of digital twins, and training AI models,” and that “the NSTC could develop a methodology for the voluntary sharing of data that protects the proprietary component and national security while enabling access to appropriate performance data.” Phase 3, to be completed by August 2025, will advance the state-of-the-art toward the following specific objectives:A framework to create and integrate Digital Twins of semiconductor R D and manufacturing process toolsAbility to explore processes and generate virtual devices swiftlyDefined interfaces to combine models for each process module or toolAccurate AI-based models for executing virtual process flows to build virtual devicesAdvanced solutions for secure data-sharing across the ecosystem – for example, federated learning where raw data is protected for each entity by building models locally, and only the outputs of the local models are used to build flow-level AI modelsFoundation for future industry standards for secure data-sharing and for interfaces in the virtual innovation environmentSEMI continues to build the collaborative community for Data-AI and strives to synergize with broader efforts such as the Digital Twin Manufacturing Institute, NSTC, and NAPMP in the U.S., and international standards development. Smart Data AI – System-level Innovation for AI – Future of ComputingThe cross-collaborative and synergistic objectives of Smart Manufacturing, the Smart Data-AI proof-of-concept work, and SEMI Standards merge to advance the state-of-the-art. The objective is to help members realize the full value of technology and innovation. In addition to improving semiconductor operations using AI, the efforts also strive to enable SEMI members to participate in, and ultimately profit from, market growth opportunities. Continued progress in AI is crucial both for the industry’s march towards $1 trillion in annual revenue, and for continuing to realize AI’s benefits to society.There are some hurdles to overcome in such a dynamic market. AI models, and the data they process, are outpacing hardware advances, posing a major roadblock for continued progress. As GenAI becomes more pervasive, the performance and power challenges continue to multiply, and require significant innovation in both hardware and software. While individual companies will develop competitive products in this domain, the entire ecosystem needs to evolve in a synergistic manner. As a global industry association, SEMI can play an important role in ensuring this. SEMI started a series of workshops and technology sessions to develop the community and identify opportunities and challenges. The first in this series was a joint workshop with McKinsey Co., held in October 2023, with a focus on innovations in “Domain-Specific Architectures.” Strategically, it brought together thought leaders from three diverse communities - start-ups, investors, and SEMI member companies across the supply chain. This was followed by an overcapacity audience at the Future of Computing session at SEMICON West 2024, where we explored AI-specific hardware with leaders in academia and industry. The Initiative’s next planned event in October 2024 is a focused workshop that is designed to be highly interactive and bring together visionaries and thought leaders from across the value chain – materials, devices, architectures, algorithms, and critical enabling technologies such as photonics, chiplets, advanced packaging, and 3D and heterogeneous integration. The overarching goal is to identify pre-competitive collaborative actions that would help the entire industry. The “Future of Computing” is the broad path to the industry’s future success. While AI systems are the current major wave on this path, future waves may be about heterogeneous integration of photonics and other components, and ultimately, quantum technologies joining the mainstream. SEMI continues to monitor these future trends, strengthen the ecosystem and enable innovation through pre-competitive collaboration, and accelerate implementation through standards.SEMI is fostering today’s collaborations while helping the industry navigate the future of electronics.Melissa Grupen-Shemansky is CTO at SEMI, Pushkar Apte is a Strategic Technology Advisor and Leader of the SEMI Smart Data-AI Initiative, and Mark da Silva is Senior Director of the SEMI Smart Manufacturing Initiative.Definitions and References:1https://arxiv.org/abs/2405.15828 Eamon Duede, William Dolan, Andre Bauer, Ian Foster, Karim Lakhani2McKinsey Company3Digital Twins for semiconductor manufacturing operations are dynamic, predictive, data-driven virtual models of a physical asset, process, or an entire factory, constantly synchronized with its real-world counterpart through real-time data streams and analytics4Research was sponsored by the Army Research Laboratory and was accomplished under Cooperative Agreement Number W911NF-19-2-0345. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the Army Research Laboratory or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for Government purposes notwithstanding any copyright notation herein.5“A Vision and Strategy for The National Semiconductor Technology Center (NSTC)” published by the CHIPS R D Office.
