UNIVERSITY PARK, Pa. — Vijay Narayanan and Anand Sivasubramaniam, distinguished professors of computer science and engineering in the School of Electrical Engineering and Computer Science at Penn State, are part of the Center for Research in Intelligent Storage and Processing in Memory (CRISP), one of six centers chosen across the nation by the Semiconductor Research Corporation’s Joint University Microelectronics Program to undertake high-risk, high-payoff research that addresses existing and emerging challenges in microelectronic technologies.
Of the $200 million allocated among the six centers during a five-year term, CRISP has received $27.5 million to remove the separation between memories that store data and processors that operate on the data.
“The ‘Big Data’ era is necessitating real-time processing of massive volumes of data across a diverse set of computing devices, from resource-constrained mobile devices at the edge to the back-end high performance servers in the cloud. Rather than the conventional Von Neumann style of bringing the data to the compute engines, this new research will transform the style into one where computation is performed very close to the data,” Sivasubramaniam explained. “This will be very important to sustain the real-time processing needs to avoid expensive data transfers, and also alleviate data movement energy costs.”
Narayanan elaborates, “Imagine the day when you end up spending more time commuting to your work than the productive hours at work. A similar problem exists with our computing systems — they spend a lot of energy and time in shuttling the data from the memory to the processor to manipulate and extract useful information from the data. This project explores how one can shift most of that data manipulation and extraction to the memory, reducing the traffic to the processor. This would be akin to being able to work from home most days and commute to work only when you do not have the resources to get things accomplished at home.”
CRISP’s researchers — Narayanan and Sivasubramaniam, in collaboration with researchers from seven other universities — will seek to build computer processing capabilities into memory storage at the chip level and pair processors with memory chips in 3D stacks. Once addressed, users would be able to perform previously unattainable computations on massive amounts of information, ultimately enabling rapid advances in national security, medical discovery, and beyond.
“Though ‘near-data computing’ ideas have been around for a while, our research will address the issues needed to make this practical by considering the entire computing stack holistically, from devices and circuits in the hardware, to the micro-architecture and the high level programming, compiler, applications and system software layers,” said Sivasubramaniam. “They will all work synergistically to leverage each other’s capabilities.”
Narayanan and Sivasubramaniam have been allocated approximately $2.325 million of the center’s $27.5 million to work on aspects related to hardware, systems software and applications to make intelligent memory systems scale for a wide range of end-application platforms.
“Imagine sifting through all your photo albums to find a specific friend’s face,” said Narayanan. “Instead of transferring all the photos from the computer’s memory to the processor, one-by-one, to search for your friend’s face, many photos at different memory locations can be searched concurrently. We have been developing circuits that integrate such computational capability within memory systems to help video analytics. They have many applications including surveillance, retail shopper behavior, and self-driving vehicles.”
The researchers will explore other applications such as genome matching and database applications.
Led by the University of Virginia, CRISP also consists of the following universities: Cornell University; the University of California, Los Angeles; the University of California, San Diego; the University of California, Santa Barbara; the University of Washington; and the University of Wisconsin. For more information on CRISP, as well as the other university research centers, visit http://bit.ly/2snbcvI.