This article was written based on the presentation by Dr. David Fried, Corporate Vice President, Lam Research, titled “Chips Making Chips. How Virtualization, Digital Twins and Machine Learning are Accelerating the Spiral of Innovation” at the MES and Industry 4.0 Summit 2023 in Porto, Portugal.

MESI-Summit_2025_OnDemand_DavidFried_1200x630_5.jpg

Chips Making Chips: How Virtualization, DT and ML are Accelerating Innovation

The semiconductor industry has long been on the leading edge of manufacturing automation, and companies like Lam Research continue to drive that innovation. In a recent presentation at Critical Manufacturing’s MESI Summit, titled “Chips Making Chips: How Virtualization, Digital Twins, and Machine Learning Accelerate the Spiral of Innovation,” Lam Research shared its vision for the chips industry, based on data, MES and Industry 4.0. The presentation included some fascinating data that shows how humans and smart technologies can work together to field rapidly increasing business and technology channels.

Challenges for chip makers and users

Lam Research produces semiconductor wafer fabrication equipment, specializing in etching, strip, deposition, and cleaning markets. In addition to the ever present need to improve product quality and throughput, the industry faces unprecedented demands for precision, complexity, cost control and sustainable operations. Lam views industry 4.0 innovation as key to meeting these changes and envisions a time within this decade when smart production machinery will dispatch, operate and maintain itself to attain record high productivity and yields with almost no waste.

Intelligent equipment

Driving such goals is Lam’s vision of equipment intelligence, a concept that integrates equipment self-awareness, self-maintenance and self-adaptability.

Self-awareness entails machinery that knows its parts, history, state and supply chain, digitally, and can adapt based on analysis of operating contexts. With such awareness, the equipment can self-maintain, replacing consumable parts itself. Lam edge chambers, for example, can already replace edge rings automatically, leaving them to run for a year without being opened. This involves precise part placement and automated positioning systems, and ensures optimal performance, which can reduce etch depth variation by up to 50 percent.

Self-adaptation extends the automation from maintenance to production, enabling equipment to modify operations based on data about current conditions. Lam’s spectral reflectometer (LSR), for example, uses machine learning to cease etching at the precise depth, reducing variation and improving yields. The machines blend an understanding of physics with digital data to optimize processes.

Bringing the physical and virtual worlds together

In the Lam model, virtual and physical worlds converge in the semiconductor ecosystem across a digital thread that runs through equipment and associated data and processes. This linkage enables precise digital modeling of all equipment and assets that provide a digital twin of the primary process, enabling operations planners to experiment with a true simulation of the primary process, without impacting production. This includes virtual mockups of chambers, identifying pinch points, part overlaps, and assembly sequence issues long before any metal gets cut. It enables virtual process development and refinement, facilitating virtual builds, augmented reality (AR) for maintenance, and virtual reality (VR) for training.

Digital simulation proved its value to Lam during the COVID-19 pandemic, when Lam tech support teams used AR to assist customer engineers with complex maintenance tasks, keeping operations running despite travel restrictions. This would have been impossible without a digital representation of equipment and associated data.

Lam believes that the potential for intelligent equipment and digital twins will revolutionize process development and lead to a new paradigm of collaboration between humans and artificial intelligence. To explore this further they benchmarked the task of recipe development between human experts, human non-experts and AI.

Finding the sweet spot in human and AI collaboration

To compare the effectiveness of human and AI collaboration, Lam Research and their research team created a virtual model of a plasma reactor on a digital twin and tasked a team of process engineering experts and an untrained AI algorithm to create complex etching recipes for that reactor.

The scope of the problem was daunting, from a set of possibilities in the order of 1023 , they had to find just one that would work on the reactor, and do so at the lowest cost, based on a virtual cost structure created for the experiment.

In the first phase of the experiment, the human team outperformed the AI algorithm, coming in with a simulated cost of $105,000, most of which was labor, compared to the AI cost of $720,000, most of which was in computing time. The team concluded that without any training, an algorithm was much worse than experienced process engineers.

But when they analyzed progress along a timeline, they found that most of the human time was spent in the final stages of the solution, where computing got more demanding. And when they repeated the experiment handling of the heavy lifting early in the process, AI beat the humans to a workable specification 42% of the time. And when they handed it off even further along in the process, the algorithm could beat humans up to 99% of the time. By experimenting with different hand-off points, Lam has been using such simulations to find the ideal handoff point for various to optimize their process and the experiment, which has changed the way they now do process engineering.

Embracing the spiral of innovation

Lam Research’s journey exemplifies the virtuous cycle of innovation, where advanced technologies like virtualization, digital twins, and machine learning drive progress. And part of that progress is itself driving the innovation as the chips industry continues to evolve this spiral of innovation promises to deliver the next generation of chips and technologies, shaping the future of semiconductor manufacturing.