Duke Receives $2 Million to Innovate Semiconductor Tech

The National Science Foundation today announced a new round of funding, including support from the CHIPS and Science Act, to speed creation of new semiconductor technologies and manufacturing and strengthen workforce development efforts. The projects are supported by the NSF Future of Semiconductors (FuSe) program through a public-private partnership spanning NSF and four companies: Ericsson, IBM, Intel and Samsung. 

Two proposed projects including Duke researchers captured $2 million of the funding; these projects will bolster the development of hardware and software for resource-savvy machine learning and sustainable manufacturing techniques. 

Duke electrical and computer engineering (ECE) faculty member Tania Roy will lead one of the two projects alongside colleagues Aaron Franklin and Hai “Helen” Li. Her three-year, $1.4 million project is titled “Co-designed Systems for In-sensor Processing with Sustainable Nanomaterials (COSMIC).” The project zeroes in on AI’s inefficiencies and proposes innovative ways to trim its massive power requirements and natural resource consumption through small, specialized, local hardware that is independent from the cloud.   

“Let’s imagine running a large language model on a new type of hardware that’s smaller and lighter than an Easter egg and powered by a coin battery.”

Tania Roy, DUKE ASSISTANT PROFESSOR of electrical and Computer engineering

According to Roy, many engineering students believe that semiconductor design and manufacture has already peaked—but that perception is far from truth. “Training large language models (LLMs) like ChatGPT requires the same amount of energy that it takes to power 120 US homes for a whole year,” said Roy. “Leaving that kind ofcarbon footprint is not sustainable. But let’s imagine running an LLM on a new type of hardware that’s smaller and lighter than an Easter egg and powered by a coin battery. This is the focus of our proposal, in the context of vision sensing and image processing. It’s very exciting.” 

Developing the innovative materials and devices and co-designing circuits and systems to enable this new technology is predicted to help reinvigorate the industry and create a wealth of job opportunities. “Enthusiasm for new devices and circuits is building, and we expect to draw more high school and college students toward the semiconductor industry, helping to fill the jobs proliferating through the CHIPS Act,” said Roy. 

A second project, led by Arizona State University,is targeting AI’s data processing energy use, replacing graph processing units with processing-in-memory (PIM) systems that offer 100 times the efficiency. Duke ECE professor and Athena director Yiran Chen will contribute his expertise in software-hardware co-design to the project, which is expected to supercharge applications like autonomous driving, robotics, personalized cognitive speech and smart connected health. Duke will receive $600,000 over three years for the project, titled “Efficient Situation-Aware AI Processing in Advanced 2-Terminal SOT-MRAM.” 

“Duke ECE is pressing forward with renewed commitment to become a linchpin of semiconductor-centric scientific research and education, through the avenues opened by the NSF and the CHIPS and Science Act,” said Li. “This promising initiative not only leverages our current strengths but also sets a solid foundation for future endeavors. We are eager and optimistic about reaching these ambitious and unprecedented milestones.”