Mary Jane Irwin, Evan Pugh Professor and A. Robert Noll Chair in Engineering in the Department of Computer Science and Engineering at Pennsylvania State University, is as committed to her research as she is to serving the computer science community, in particular women in the field. Her interests and accomplishments span computer architecture, multicore systems design, and energy-aware design. She has also been active in the Computing Research Association’s Committee on the Status of Women in Computing Research (CRA-W), the Grace Hopper Celebration of Women in Computing, the Board on Army Science and Technology, and the National Academy of Engineering’s Membership Policy Committee. She has been deeply involved with ACM as well, co-founding the Journal on Emerging Technologies in Computing Systems and serving as an elected member of ACM’s Council, as vice president from 1997 to 1998, and as editor-in-chief of ACM’s Transactions on the Design Automation of Electronic Systems from 1998 to 2004.
You grew up in Memphis [Tennessee], where your dad was a professor at what was then Memphis State University. What drew you to computer science?
I was good at math, but it was early in the days of computing and there was no degree in computer science at Memphis State at the time. So I majored in math and took all the computing-related courses that they had. Then I went to graduate school at the University of Illinois at Urbana-Champaign (UIUC) to become a professor, although at the time I really had no idea what that entailed, other than college-level teaching. UIUC was the luck of the draw because my husband got a job there, so that’s the only place I applied. In that case, my draw was very lucky!
At UIUC, you worked with James E. Robertson, and you wrote your dissertation on computer arithmetic.
It’s a challenge to build arithmetic components so that they run as fast as possible without consuming a lot of power, and of course they’re a central part of what’s going on in the CPU. Jim Robertson was very famous in computer arithmetic. I was attracted to that research area, but I also think there was some empathy between us, because I was one of the very few women in UIUC’s computer science graduate program, and Jim was of American Indian descent. He’d spent time on the reservation growing up, and he knew what it was like to be from an underrepresented group in a majority population.
"It’s a challenge to build arithmetic components so that they run as fast as possible without consuming a lot of power, and of course they’re a central part of what’s going on in the CPU."
And you’ve been at Penn State ever since?
Next year I celebrate my 40th anniversary at Penn State. I used to tell my students, "I’ve been on the faculty longer than you’ve been alive." Now I can say, "much longer than you’ve been alive."
How did your research interests evolve?
I was also really interested in hardware design, in building advanced circuits and special processors. At Penn State, I had a number of strong grad students, one of who completed his Ph.D., worked for IBM, and then came back as a faculty member. We had a very productive collaboration until he passed away.
You’re referring to Robert M. Owens, who died in 1997.
We started building projects and they were a wonderful experience. We built two different special-purpose signal processors, including custom chip design, board design, algorithm design, software design—all sorts of things that we didn’t know we were getting into when we started the projects.
The first, called the Arithmetic Cube, was a high-speed programmable VLSI processor for solving linear digital signal processing problems.
We designed and implemented a custom architecture that could run at the speed of the commercial parts that were available at the time, even though it was an academic project and not implemented in the most current technology node. It had custom CMOS chips that we had fabricated through MOSIS (Metal Oxide Semiconductor Implementation Service), a board-level design, and novel signal processing algorithms.
What we discovered along the way is that the design tools we needed to support the design process either didn’t exist, or existed only in companies that didn’t distribute them. So we had to build the design tools as well. We built the logic synthesis tools, a high-level synthesis tool, automatic cell-generation tools, and the software we needed to test the design. It turned out to be a much bigger project than we anticipated.
After that, you built a processor known as the MicroGrain Array Processor, or MGAP, designed to address complex signal and image processing problems.
We wanted to build a more programmable processor than the Arithmetic Cube. That led to the MGAP implementation, which was very FPGA-like, looking back at it, except that it only allowed nearest-neighbor communication. There were two generations of that. We used a lot of the design tools that we had built for the Arithmetic Cube, and of course we designed new tools.
"Some design with FPGAs, and that’s great, but the number of people building custom hardware in universities has dwindled to almost nothing."
You also came up with a new hardware description language: Gate LangUagE, or GLUE.
This was in the early days before there was Verilog or VHDL, so we came up with our own HDL which, looking back, was very much like a rudimentary version of Verilog. We got some things pretty close to right in that regard.
Several of the tools you developed over the course of your work on those projects were pioneering in their own right—for instance, the architectural-level power simulator that eventually became SimplePower.
At the end of the MGAP Project, we started working on an accurate power simulation tool. Until then, we just focused on performance and area, but we were beginning to realize, as had many, that power was going to be the next big problem. Bob passed away in the very early stages of its design, so I continued on that work with a new Penn State colleague, Vijay Narayanan, and once again a bunch of graduate students. Most of our work since then has been this simulation-based work. There’s been very little building.
Is that because of cost?
It’s really, really expensive these days to build custom hardware at the current or near-current technology node. You just can’t afford to do it at a university, so most architects do simulation-based research. Some design with FPGAs, and that’s great, but the number of people building custom hardware in universities has dwindled to almost nothing.
What are you working on now?
Right now, some colleagues and I are working on a grant that examines reducing the power consumption of the non-core components of a chip. There are a lot of people who are looking at how to make the cores more energy efficient. We’re looking at how to make the components that support the cores more energy efficient. We’re focusing now on the cache space—how much of the cache you really need to have turned on, especially when some of the cores are turned off, and how you manage it—but there are a lot of non-core components to work on.
Throughout your career, you’ve been involved in advocating on behalf of women in the field.
I’ve been involved with outreach to women both through ACM (ACM-W) and in particular through the Computing Research Association’s Committee on Women (CRA-W). And I still am involved in local campus activities and try to attend the Grace Hopper Celebration about every other year.
What’s changed?
Unfortunately, the numbers at the undergraduate level have not come up like we had hoped. There are some locations where they’ve improved—Harvey Mudd, for example, and Carnegie Mellon, though I haven’t seen their latest figures. But at many places, the number of undergraduate women in computer science and computer engineering is still in the low teens. My sense is that we do have more women in leadership positions.
At the graduate level, the numbers are better because we have women from offshore, which is great. Most of these women come to the U.S. because they want to get a Ph.D. If they’re going to be adding to the infrastructure either here or in their home country, it’s a great thing.
But the number of native-born U.S. women going into computer science and engineering is still way too low. It’s a pipeline issue. So that’s always a little frustrating for those of us who have spent a lot of time working to change it.
Join the Discussion (0)
Become a Member or Sign In to Post a Comment