Electrical Engineering and Computer Science Department Chair Alex K. Jones and Professor Bryan Kim Receive NSF Grant to Develop Energy-Efficient Chiplets for Data Centers

August 25, 2025 -- Alex K. Jones, Department Chair of Electrical Engineering and Computer Science (EECS), and Bryan Kim, assistant professor in EECS, have received a grant from the National Science Foundation (NSF) to use chiplet technologies to improve the efficiency and capability of data centers. These data centers are needed for tasks like training large AI models. The $2M project is a collaboration between the University of Arkansas, the University of Central Florida, and Syracuse University. Jones and Kim serve as co-principal investigators on this project.  

Chiplets are small, modular circuits designed for specific tasks like processing, memory, input and output, or communication, and are an evolution of large single chips known as systems on a chip. Chiplets are assembled into a package using a method called 2.5D integration, which allows for the integration of different semiconductor technologies in a compact way. 

“It’s more than just the next generation of the printed circuit board,” says Jones, who leads the Syracuse University team.  

Using 2.5D integration, smaller chiplets help reduce defects that often occur in the chip-making process. This method improves the efficiency of semiconductor production while still maintaining the benefits of their close integration.  

“Of course, chiplets allow us to save on waste during fabrication, which benefits cost and even reduces greenhouse gas emissions,” says Jones.  “But the real game changer is that using chiplets, we can integrate semiconductors that would have been fundamentally incompatible in the same fabrication process, giving a system designer access to the best material for power conversion, sensing, computing, memory, and communication all in the same package.”  

“Even for Silicon-based fabrication, the rules to make compute logic, DRAM memory, and Flash storage are all different and highly tuned to those functions,” says Kim.  “By building these components individually but allowing their integration through chiplets, we can break down data movement limitations that currently limit our biggest applications like training large language models that form the basis for ChatGPT.” 

Jones and Kim will create computer systems using next-generation chiplets. Their collaboration with the University of Arkansas, which has expertise in new power delivery methods, and the University of Central Florida, which specializes in disaggregated computing, will combine memory, processing, and storage from different computers to meet the specific needs of various applications.