CS professor Saugata Ghose wins 2023 Intel Rising Star Faculty Award.
Intel bestows this award annually to fifteen early-career academics worldwide for their research in the fields of computer science, electrical engineering, computer engineering, and chemical engineering.
The award summary commended Ghose’s efforts to redesign computer hardware to efficiently process data, notably the strides he has made in advancing the realization of implementing widespread processing-in-memory (PIM) technology.
Ghose will receive a one-time award of $50,000 to support research or curriculum development, computer lab infrastructure, and other professional development opportunities within the university.
Ghose’s approach focuses on designing programmer-friendly software tools and hardware platforms to allow for the easy and widespread integration of PIM into computers. “Our goal is to make computers smarter about the data they’re handling and to be able to manipulate that data more efficiently,” Ghose said.
For years, scientists pushed for speed in computer processors, but that speed came at a price, particularly as the amount of data generated by society has grown rapidly. “That cost can be small things, like how long batteries last on our phones, to big questions, like what’s the environmental impact of what we’re doing with computers today. What we need to do is change the way that we design computers, and that’s a lot of what inspires me because it’s a little bit of uncharted territory,” Ghose said.
One big challenge is integrating these new technologies to adapt existing programming models for PIM. Many researchers have proposed how to design hardware improvements for PIM, but have not yet been able to implement those changes in a way that can be familiar enough for computer developers to easily integrate into their current software and systems.
Ghose’s work aims to bridge this existing gap between change and implementation. “Disruptive changes in hardware by default can require disruptive changes in software. What we’re trying to do is avoid that software disruption as much as possible,” Ghose said.
“Computers have been in society for decades. We have a lot of existing infrastructure, tons of code that’s written, and tons of programs that people have already bought and used. We hope that people can simply take their existing programs, run it on our hardware, and see the benefits,” Ghose said.
Improved PIM efficiency would create more agile solutions for data analysis. Currently, it is inefficient for small computers such as smartphones to process data internally. Instead, they collect data and send it to a large cluster of servers in the cloud for processing, which can waste large amounts of energy sending the data across the Internet.
Ghose hopes that his work will eliminate this inefficient transfer. “Some of the things that we’re doing can make that data analysis easier and cheaper to do in terms of energy since data will now be processed right where it is being collected. That way, data can be saved and processed without having access to the network. As an example, our work will put smart sensing into domains to detect failures in larger infrastructures in ways that we just can’t do today,” Ghose said.
Ultimately, the changes Ghose hopes to make will lead to more convenience and greater accessibility for everyday users. For example, implementing PIM in cell phones will enable users to still access the same functions without relying so heavily on network connectivity to complete those same tasks. In addition, PIM can potentially extend phone battery life beyond twice its current duration.
Improvements in PIM will also bring greater accessibility for all users. “We’d be able to bring data processing services to places that don’t currently have good internet connectivity, both in rural parts of the U.S. and in third world countries, where that type of service is too expensive to provide daily,” Ghose said.
The detailed, experienced research support from Intel will provide Ghose and his students with the vital context needed to tackle these big problems. “We’ll be getting help from teams that have been looking at how they’ve built CPUs for a long time and getting help from people who understand the types of applications that companies big and small are trying to run today. That’s a big help in understanding limitations,” Ghose said.
“On the academic side, there’s only so much that a research group of a dozen students can do. There are only so many corner cases and exceptions they can explore. Intel brings thousands and thousands of expert researchers across many years that have been looking at all of these specific corner cases and their nuances, which helps guide us and enables us to dig at what we see as the biggest obstacles to making our adaptations,” Ghose said.
Ghose believes it is precisely that type of collaborative expertise that will be vital for students to embrace in the rapidly evolving field of computer science. “We can’t just work in isolation in architecture these days. We have to think about how architecture relates to field X and that can be somewhere else in the computing stack, that can be actual applications and how people use computers,” Ghose said.
In the classroom, Professor Ghose provides training and research beyond the conventional CPU so that students are best prepared to be innovative, productive scientists. “We’re teaching students how architecture intersects with the software further up the computing stack and with the hardware further down the stack so that they understand how to design across multiple disciplines,” Ghose said.
To gain broader, deeper practical experience, Ghose encourages his students to explore multiple domains. “If you look at my group, everyone’s got a different direction that they’re stretching into, which allows us to cover a much wider swath of problems and topics rather than another group that has been focused in the middle with everyone examining only their area of expertise,” Ghose said.
It is this spirit of collaboration that Ghose believes has led to his success. “ The combination of the group that we have and other people who believe in this type of philosophy but who are in different disciplines has combined to help us stretch the kind of things that we want to explore, and I think that’s a large reason we’ve been able to make as much progress as we have today,” Ghose said.