Vaibhav Verma aims to develop a new processor for artificial intelligence applications at the edge of the “internet of things.” Verma is a Ph.D. student of electrical and computer engineering advised by Mircea R. Stan, Virginia Microelectronics Consortium Professor at the University of Virginia.

“There are many novel AI accelerators proposed in the literature, but very few are used in real applications,” Verma said. Accelerators are very difficult to program and integrate in a computer stack. With colleagues in Stan’s High-Performance Low-Power research lab, Verma seeks to change the paradigm from cloud-based to on-chip AI solutions. This shift will enable open-source, low-power processing of AI algorithms that move, retrieve and store data throughout the internet of things. 

“We realized we needed to develop the software infrastructure along with the hardware, if we want our novel processor to be used in real-life applications,” Verma said. The team’s end-to-end, system-level solution offers the flexibility of a central processing unit with higher speed and better energy-efficiency. A grant from the Semiconductor Research Corporation supports this research. The team is building a complete software development kit to make it easier for end-users to use their novel processor.

Verma presented the methodology and benefits of team’s novel processor at the 2020 Design Automation Conference. Verma was among a select group of students invited to participate as a young student fellow. “I virtually met a lot of students from different universities across the globe, including some who are working in the same research area. It was a good experience learning from peers,” Verma said. Verma earned a young fellows poster presentation award for AI-RISC:  Scalable open-source processor for AI applications at the edge of IOT. Conference organizers selected 21 out of 287 submissions for this honor.  

Verma brings real-world experience to his research. After earning his bachelor’s degree at Birla Institute of Technology and Science, Hyderabad, India, Verma worked in the memory R&D group at Synopsys in India, securing two patents on improving read and write circuits in Static Random-Access Memories (SRAM) at advanced technology nodes. More recently, Verma completed a research internship at Qualcomm.

Verma joined the Ph.D. program at UVA to broaden his research horizons from circuits research to architecture research for emerging applications. In the first two years of his Ph.D. at UVA, he successfully led a team of UVA students to fabricate a low-power system-on-chip consisting of a RISC-V processor and a deep neural network accelerator in advanced 14 nm technology node. This gave him the confidence to tackle bigger research problems in the domain of AI hardware. Following the successful completion of his degree program, Verma plans to continue research in the field of energy efficient architecture design for machine learning and artificial intelligence applications.