UCLA faculty members merge data processing and memory to increase computing performance

Jul 20, 2018

By UCLA Samueli Newsroom

Two UCLA computer scientists are part of a new multi-university microelectronics research center that aims to dramatically speed up computer performance by integrating data processing into memory and storage for future computer systems.

The Center on Research in Intelligent Storage and Processing in Memory (CRISP) is headquartered at the University of Virginia and includes research groups from eight universities. The lead researchers from UCLA are Jason Cong, Distinguished Chancellor’s Professor of Computer Science, and Song-Chun Zhu, professor of statistics and computer science.

The physical separation between a computer’s memory components that store data and the processors that operate on that data have been established in mainstream computer architectures since 1945, when John Von Neumann first outlined the design of a computer.

While that design has worked well for a long time, that separation, known as the “memory wall,” has become a bottleneck, holding up computing speed and performance. In particular, this is becoming a sticking point with the advent of big data applications. There’s so much data, in fact, that processors often have to wait for the transfer of data, which can take more time and energy than computation itself. Even at the tiny scales of the integrated circuits used in modern computers, these travel times start to add up. This slows down processing and also heats up the computer’s components.

CRISP is looking to address this major bottleneck, by designing new computing systems where memory and processors are integrated much closer together. This could be in a single “3-D stack” or computing elements would be distributed in a memory and storage hierarchy so that they can work in harmony. Cong and Zhu will receive $1.85 million in research funding for projects aimed at opening up that bottleneck.

Cong will design new computer chip architectures and affiliated software and hardware systems to improve computing and speed in memory and storage. He will also demonstrate the advantage of those systems on big-data applications, such as machine learning and computational genomics.

Zhu will design new algorithms for emerging intelligent systems, such as computer vision in video analysis, and for cognitive robots, which use artificial intelligence to learn from experiences. He will also analyze the components in computer sensing, perception, reasoning and planning, with an emphasis on speeding up the movement of data through computer systems.

CRISP, funded with $27.5 million, is one of six new Joint University Microelectronics Program (JUMP) centers managed by the Semiconductor Research Corporation with the support of the Defense Advanced Research Projects Agency (DARPA).