“My journey has essentially been a progression from understanding fundamental light and matter interactions, to asking a much more practical question,” explains Dr. Wei Zhou. “How do we actually build and scale these systems to solve real-world problems?”

Zhou wears multiple hats in his capacity at Virginia Tech, serving as both associate professor of electrical and computer engineering and director of micro- and nanofabrication cleanroom and laboratory. Alongside this, he has roles as faculty associate for the microsystems and nanotechnology division of the National Institute of Standards and Technology (NIST) and as associate editor for IEEE Photonics Technology Letters.

Spanning both the semiconductor and photonic realms in his work, as well as between academia and industry, Zhou therefore sits at an interesting vantage point when it comes to scaling nanomaterials, be they nanophotonics, nanoelectronics or otherwise. “Over time, my realisation has been that generating a brilliant concept in a lab is only probably 10% of the battle,” he explains. “The other 90% is the very painstaking process of manufacturing it and making it practically useful.”

At Virginia Tech, Zhou and his team focus on advancing AI-native hybrid optical-electrical nano-bio interfaces. Breaking this concept down helps explains Zhou’s mission in going beyond the likes of CMOS (complementary metal-oxide semiconductor) and designing scalable platforms for electronics better and faster – and which can truly make the most of breakthroughs across key industries.

“A major challenge in fields like healthcare, agriculture, and environmental monitoring is moving from a reactive posture to a proactive one, which requires translating very noisy, complicated biological-related signal molecules into clean, digitised data in a continuous, real-time fashion,” says Zhou. “Our group and our labs build micro nano-devices that bridge these different information and data by co-sensing information electrically and optically.”

Where does AI come in? “What makes this nano-scale information conversion device AI-native [is] that we design the nano scale, the sensors or transducer devices, in terms of the solid-state physics, as well as the data conversion layer in the same time, synergistically, from day one,” explains Zhou.

“We’re not just measuring something,” adds Zhou. “We are generating the spatially and temporally correlated high-fidelity structured data, converting from the biological domain into the computational cyber-physical domain, that AI models can use, and now desperately need to learn and adapt, to distinguish some molecular information in the cellular network in real-time for real applications.

“This is a co-design principle with both science, manufacturing, and the business in mind in day one with [an] AI enabler layer.”

Another important element in transforming the scientific process, from experiment to design, is through autonomous, or self-driving, laboratories. In terms of bridging the gap between the lab and the real-world context, Zhou notes the Genesis Mission from the US Department of Energy, featuring a who’s who of tech and which aims to 'develop an integrated platform that connects the world's best supercomputers, experimental facilities, AI systems, and unique datasets across every major scientific domain.'

Zhou sees the autonomous lab ecosystem currently across three levels, in a classification not dissimilar to the highest three levels as published by Royal Society Open Science in a recent paper. Level 1, Zhou notes, is individual automation steps. Level 2 connects multiple tools in the workflow, while level 3 is end-to-end, from design, to fabrication, to data collection and simulation, to business-driven product.

“There is a long way to go,” admits Zhou of the latter. Yet he adds it is an ‘exciting direction’ for society. Much as those at Genesis are keen to stress that the mission is to enable scientists, not replace them, Zhou also notes the human aspect. “Ultimately, human[s] will play a key innovative layer in terms of defining the mission, the purpose, the trust to build the business, the social economic layer, the policy,” he says.

So, what does this new blueprint look like for retaining the innovative human layer with autonomy? Ultimately, it all goes back to the need to move beyond traditional electronics design.

Zhou describes the need for what he calls ‘physics-informed AI innovation’. In thermal dynamics and quantum mechanics, physics is used as a backbone to develop the AI model to co-design tools or chips with the fabrication information, he notes. This technological approach can therefore help build the future workforce. “This is the future, and then we need to use this kind of technology to train the future talent, instead of an old thing,” he says. “Old things just do old things. We need to fuse the physics-informed AI innovation with the talent development and next-generation leadership in the scientists and engineers.”

With an eye on the CHIPS Act and a need for a more joined-up approach to workforce strategy, Zhou argues there is a growing need for a workforce who can move across physical systems and AI, with more hands-on, integrated training environments. He calls this role a ‘scientific AI engineer.’

“The CHIPS Act actually is a phenomenal catalyst for mobilising the capital and expanding the baseline training programme across the US,” says Zhou. “The next challenge is not producing more engineers, but producing the AI plus X plus business. That means this engineer needs to use AI-native tools, and also have a domain knowledge across all quantum, optics, photonics, biosensors, as well as have the business mindset.

“We can no longer just sustain traditional cleanroom technicians for semiconductor or isolated data centers,” Zhou adds. “We need to develop the key pillar, AI plus X (domain expertise), Y (business and socioeconomics perspectives), type of talents.”

Zhou will be speaking at Microelectronics US in Austin later this month on these themes, on a panel exploring cross-industry perspectives on the evolving microelectronics landscape. “The real bottleneck in AI-driven science innovation and the business, for semiconductor or other areas, is no longer the algorithms or computation power, or a semiconductor factory. It actually is a lack of the integrated, interoperable innovation,” he concludes. “Workforce translation; infrastructure; connecting different people and resources beyond their silo is the key.”