KINGSTON, R.I. – Aug 12, 2026 – A guiding principle of the University of Rhode Island’s College of Engineering is tracing a line directly from academia to real-world application. Last month that connection was made even stronger.

FastFlowLM, founded by URI professor of computer engineering Ken Qing Yang and his former University of Rhode Island colleagues Tao Wei, a professor of engineering at Clemson University who remains on the URI faculty as an adjunct professor, and Alfred Xu, assistant research professor at Clemson and former research assistant at URI, have recently joined California-based semiconductor company AMD.

FastFlowLM, was founded by URI professor of computer engineering Ken Qing Yang and his former colleagues Tao Wei, a professor of engineering at Clemson University who remains on the URI faculty as an adjunct professor, and Alfred Xu. (Credit: URI)

FastFlowLM is a technology company delivering software designed to handle large language models and AI workloads.

The FastFlowLM team joining AMD is significant given AMD’s influence in the technology sector. AMD plays a central role across the computing landscape, providing the CPUs, GPUs and AI accelerators that power everything from consumer PCs and gaming platforms to large-scale AI and cloud infrastructure.

Many newer computers include a dedicated neural processing unit, or NPU, designed to handle AI workloads. FastFlowLM’s software makes more efficient use of those chips, allowing workloads to use 2 orders of magnitude larger context—the units of information used by large language models—more quickly while consuming less power. The result is faster performance, reduced energy consumption, and the ability to run long context AI applications locally without sending data to the cloud.

Today’s agentic workloads have grown beyond simple Chat, and are now typically long sessions with growing memory, and context. This requires a highly efficient technology that can support these large session context sizes without sacrificing performance.

“AI comes in two parts, model training and inferences. When you use ChatGPT, the response you get is inference. We developed a highly optimized inference software that runs on an NPU,” said Yang. “A state-of-art software can probably handle 2,000 tokens on an NPU. We developed FastFlowLM to speed up that AI inference—with the ability to handle up to half a million tokens.”

The idea for FastFlowLM emerged from research conducted by Yang, Wei, and Xu while the three were at URI. Several papers on their research were published at top-notch conferences including the International Symposium on Computer Architecture, the premier annual academic conference on computer architecture, ACM International Symposium on Field Programmable Arrays, and International Conference on Programmable Logic and Applications.

Published prior to the company’s founding, the study showed that Ryzen NPU chips could run large language models far more efficiently. While existing software on AI inferences has limitations in handling large context lengths and consumes a large amount of energy, the research results demonstrated that parallel dataflow algorithms tailored specifically to the NPUs maintained consistent high performance and energy efficiency.

That research eventually evolved into FastFlowLM, with URI Innovations and Clemson University playing key roles in protecting the intellectual property and supporting its commercialization.

URI Innovations works with faculty to help move discoveries beyond the lab through intellectual property protection, licensing, startup support, and industry partnerships. That was the case for Yang and his colleagues, whose research showed strong potential for real-world impact.

In June 2025, Yang, who has founded multiple companies, approached URI Innovations to explore protecting the team’s software.

What elevated the company’s profile, however, was putting the software into users’ hands. The team uploaded it to GitHub for broader testing, and the response mirrored the promise of the research itself.

Within hours of the software release, thousands of users pulled FastFlowLM from GitHub and ran it on their own Ryzen AI laptops. The champion of Global AI PC Dev Contest used FastFlowLM to win the competition.

That attention drew AMD’s interest.

“They tested our software having downloaded it from GitHub, and at the beginning of this year, the company approached us about us joining them,” said Yang.

After the FastFlowLM team joined AMD, two founders and their team became part of AMD’s AI Group, productizing AMD hardware solutions and maintaining commitment to the open-source developer community. This team further strengthens the ties between academia and industry.