Students Named Qualcomm Innovation Fellows

Four CMU students have been named 2024 Qualcomm Innovation Fellows, a prestigious honor that recognizes students who embody Qualcomm’s core values of innovation, execution, and teamwork in their computing research. This year’s group of awardees also included students from MIT, Stanford, Cornell, and Princeton. 

Two sets of Ph.D. research partners received the nod, combining their knowledge from CMU’s electrical and computer engineering and computer science programs. 

Improving Diagnostics with Physically Aware Fault Models

Ruben Purdy and Wei Li, both electrical and computer engineering Ph.D. students, were recognized for developing a new way to diagnose silent data corruption (SDC) that occurs during chip manufacturing. This research was pursued with guidance from their advisor, Shawn Blanton. 

As the world is increasingly dominated by electrical systems, fortuitous detection can no longer be relied upon as the go-to method. In need of something more precise, Purdy and Li introduced a tool called Chef that is able to identify signs of defects on the industrial silicon used to make chips. Early results are promising, indicating that Chef can flag defects with roughly 98 percent accuracy, while the researchers predict that the tool could be leveraged to extract more specific information with continued developments.

Learned Virtual Memory for Heterogeneous Architectures

Research partners Kaiyang Zhao and Hilbert (Yuang) Chen are Ph.D. students studying computer science and electrical and computer engineering, respectively. Both are members of the CAOS group, led by advisor Dimitrios Skarlatos, an assistant professor of computer science. 

Zhao and Chen’s research looks at a key issue in computer architecture: As the memory capacity of computer servers increases, memory tools lag behind. This is an urgent problem because heterogeneous architecture is becoming more popular — and this class combines different processors into a single system, all of which use the same memory. 

“Our goal is to radically rethink virtual memory from the ground up and resolve fundamental overheads that plague its design,” they said. 

With support from their advisor, Zhao and Chen turned to learned virtual memory, an AI-infused solution, to reduce the latency that often occurs during information retrieval within heterogeneous architectures. This method uses machine learning to approximate the key distribution in the virtual address space, speeding up virtual memory translation process.

Students selected for the Qualcomm Innovation Fellowship earn a one-year fellowship and are mentored by Qualcomm engineers to facilitate the success of the proposed research. The fellowship comes with $100,000 to fund that research.

The Qualcomm Innovation Fellowship has awarded more than $20 million in funding since 2009. The fellowship receives nearly 400 submissions each year and funds about 45 students annually. This year there were 16 winning teams in North America.

Learn more about the Qualcomm Innovation Fellowship on the company's website.