Digital twins aren’t a new tool for the chip industry, but they are getting democratized to a point where they are more accessible for a broader set of commercial applications.
The ability to ingest more, higher-quality data from a wider array of sources and the application of artificial intelligence is helping to extend digital twins beyond product design to virtually envision manufacturing environments, which will reduce waste as well as contribute to meeting sustainability goals. And as the chip industry ramps up U.S. onshore manufacturing in the wake of the CHIPS Act, digital twins are poised to be a critical tool for workforce development while also accelerating productivity.
In early February 2024, the National Institute of Standards and Technology (NIST) announced its intent to create a new semiconductor manufacturing institute that will use digital twin technology for production, packaging and assembly. NIST is looking to corral curriculum and best practices through its CHIPS Research and Development Office to launch a competition for a new public-private Manufacturing USA Institute.
Michael Grieves, executive director of the Digital Twin Institute, pioneered the concept of digital twins in 2002. He proposed the digital twin as the conceptual model underlying product lifecycle management with a different name—it was NASA’s John Vickers who wrote “digital twins” into the agency’s roadmap in 2010.
Grieves liked that term better, he told EE Times in an interview, and about five years later, computing capabilities allowed the adoption of digital twins to hit exponential growth as use cases in manufacturing were identified, he said. “We started to see the whole idea of being able to use information to replace wasted resources as a value driver for digital twins.”
Digital twins are more effective when silos of information are eliminated, Grieves added. “Depending on what functional area they were in, you wound up with huge inefficiencies or the inability to optimize the entire process.”
Expanding the use of digital twins out to manufacturing facilities is the natural evolution of the virtual approach the chip industry has used for decades, Michael Munsey, VP of semiconductors at Siemens, told EE Times in an exclusive interview. “The concept of a digital twin, even though we may not have called it a digital twin, has been core to the semiconductor industry for a very, very long time.”
Digital twins are being extended to visualize equipment and manufacturing facilities before they are built, then optimized once they are in production. (Source: Siemens)
Optimizing manufacturing facilities for sustainability
Munsey said digital twins can be used to build anything, so why not the manufacturing facilities that produce semiconductors—not just the devices themselves? “You can actually build a virtual representation and simulate it and optimize it long before you ever build the physical thing,” he said.
Building a digital twin of a fab allows for the modeling of the manufacturing of the chip, making it possible to optimize a facility before concrete is ever poured, Munsey said.
A digital twin does more than simulate the manufacturing process, he added. It also optimizes all the electricity, water and chemicals being used, which helps to achieve sustainability goals. Mix in internet-of-things sensors from an actual manufacturing environment, he said, and you can bring in real-time data back to the digital twin to test further optimizations.
Munsey explained it’s now possible to have a complete, closed loop where manufacturing data can be taken back all the way to the early design stage to improve processes, methodologies and decisions.
Lam Research’s vision for digital twins is that they are a full representation of all semiconductor manufacturing systems and processes, David Fried, corporate VP of the company’s Semiverse Solutions, told EE Times in an exclusive interview. “A complex system may even have many layers of digital twins, each containing the relevant and required data to achieve a specific purpose.”
Building a complete, holistic picture
Lam is involved in the creation of digital twins in four key areas, Fried said, including the device scale for detailed integrated modeling of a device to reduce cycles of silicon learning, as well as the process scale by using simulation to streamline process development.
At the reactor scale, he explained, digital twins can simulate operation conditions in the chamber to predict and optimize process behaviors, while equipment and expertise can be built faster and more effectively and reduce tool downtime with AI-enabled troubleshooting.
As the semiconductor industry drives toward net-zero transition, Fried said Lam and its customers are looking for smart ways to achieve more while using less materials and energy. Extending the physical infrastructure into the virtual world with digital twins can help to reduce the consumption of physical materials, gas and water, he said.
Equipment-scale twins are also important tools in training the service engineering workforce. In the past, this training had to be done at a physical site, but could not begin until a new tool was built, shipped and installed. Lam’s technical training centers—located around the world—are equipped with virtual-reality systems to allow for easier access to training on new tools and closer proximity to customers.
Jerry Chen, head of global business development in the industrials and manufacturing sector at Nvidia, told EE Times in an exclusive interview that the use of digital twins has been significantly democratized far beyond just being used for large, complex and mission-critical systems like spacecraft and other aerospace systems.
Better data, AI augment digital twin capability
More recently, Chen said, digital twins have been able to take advantage of more sophisticated AI models, which allows them to be more accurate and allow for more experimentation. “Not only can you do more of them without expending the infrastructure and the material consumption to do these experiments, but you can also do them at super-real time,” he said.
Eric Brecken, director of technology policy for Nvidia’s government affairs team, said increased access to data has been accompanied by improved interoperability between systems, which has been driven by AI applications with features that translate text into image, speech and video. “Dissimilar data can now move more seamlessly between layers,” he said.
This is especially beneficial for the semiconductor industry because there are many layers that require comprehensive simulations, Brecken said. “It requires a lot of data handoffs at very different layers.” The democratization of digital twins means there are opportunities to share best practices in an open-source model, he added.
Chen said building off a common technology platform to share data could reduce marginal costs for the ecosystem by sharing data in a secure way across all the disparate subsystems that they’re all creating these models for. “We have been on this journey now for building connective technologies for digital twins,” he said.
Generic, shared models improve industry productivity
Nvidia’s Omniverse platform has been largely focused on the factory level, Chen said, “but increasingly, we’re seeing that extending to other areas in the layer cake as well.” The intention isn’t to create a singular solution for the market, he explained, but to build out a platform of enabling technologies to allow the ecosystems to build up and connect and communicate between the disparate synthetic models.
Chen said AI has been a game-changer for digital twins in that it aids the development of foundation models that are relatively generic that can be added to with domain-specific and proprietary information.
Platforms like Nvidia’s Omniverse platform are being extended across various layers to allow the ecosystems to build up and connect and communicate between the disparate synthetic models that take advantage of the many sources available. (Source: Nvidia)
Increased accessibility of digital twins can improve productivity, which is why their use is being discussed in relation to workforce development efforts related to the CHIPS Act.
Lam’s Fried said talent shortage faced by the chip industry is putting demands on education that appear daunting. “It is cost-prohibitive for academic institutions to provide physical access to the most advanced nanotechnologies.”
Lam added that experimenting with volatile chemistries critical in the development and creation of semiconductors can be dangerous for students who are learning to work with semiconductor manufacturing equipment. “Virtually simulating real-world labs provides greater democratization of engineering skills training, heightened safety, improved sustainability and greater access to new talent pools around the world,” he said.
Grieves said the biggest benefit of digital twins augmented by AI is the gift of time. AI can be set loose to try different combinations and come up with solutions that humans don’t have the bandwidth for, allowing for more innovation and shortened timelines—which aligns with the goals of the CHIPS Act. “The most valuable resource we have are the ticks of the clock,” he said.