"I'm looking at how to make the processor more power efficient, or how to get more performance from the same power so, for instance, you can record or edit your video on your phone while you're talking or multitasking in some other way."
Assistant professor of electrical engineering
Technology consumers want faster, lighter, smaller, and more-power efficient electronics. Matthew Watkins is working hard to meet their demands.
"I'm not directly designing the next craze, but what I'm working on enables the next craze," says the assistant professor of electrical engineering. "You want your phone to last you all day or all week and provide good performance, while also being small and thin. This means less space for the battery, which requires better power efficiency. I'm looking at how to make the processor more power efficient, or how to get more performance from the same power so, for instance, you can record or edit your video on your phone while you're talking or multitasking in some other way."
Watkins says there's been a shift in thinking since about 2005. Prior to then, designers pushed the envelope by increasing processor frequency. "People thought if they got a processer with a higher frequency, it was better and faster. But as frequency continued to increase, the chip started running so fast and consuming so much power that — this is a bit of an exaggeration — it would eventually melt," he explains.
Now, instead of ramping up frequency, designers are putting multiple cores on a chip, in a sense going from a single brain on a processor to multiple brains. The only problem is that not all software is designed to make use of more than one core. "Do we put down more cores and hope software programmers make use of them? Parallel programming is a difficult task and so far most programs haven't used more than a handful of cores at a time," he says.
To explore new possibilities, Watkins builds his own multiprocessors in a computer simulator. "It's not practical for a single person to build a state-of-the-art multiprocessor, costing hundreds of thousands or millions of dollars to fabricate, so I build simulators to create a model that mimics what might actually happen on a real processor," explains Watkins. Through simulation, he's experimented with optics, using light to communicate across the chip, and incorporating reconfigurable hardware, which is hardware that can perform different tasks at different times, into a traditional microprocessor. Both provide better performance and power efficiency than standard microprocessors.
Technology behind that next craze is achievable, but the solution has to be cost-effective for manufacturers and useful for a variety of applications for end users. Says Watkins, "It comes down to a balance between performance and power and the cost and to market time implications of changing the design of today's existing microprocessor."
Posted October 2012