Prof. Matt Watkins Researching Paths to Better Performance and Energy
Some people will drop everything to make sure their smartphone is fully charged.
Low-battery anxiety, a disorder of the modern age, often strikes when the charge dips below 70 percent, sending frantic users to the nearest computer port, charger, or outlet.
Matt Watkins, assistant professor of electrical and computer engineering, is researching techniques that will make a battery charge last longer, hopefully easing some of that overcharged anxiety.
For decades, performance of computer processors grew at an exponential rate. This massive growth has supported innovations that have revolutionized myriad aspects of human life. As processor performance has continued to grow and computer systems are becoming more and more mobile, the power needed to run these devices can no longer be ignored.
“We want our phone battery to last as long as possible,” Watkins says. “But we need to innovate beyond designs from the past if we are going to have systems that are both responsive and provide long battery life.”
One approach, popular in mobile devices like phones and tablets, is to have a computer chip using multiple types of processors with different levels of efficiency and performance. Some of them provide high performance, allowing your games and other applications to run quickly, but consume a lot of power, notably limiting battery life if used all the time. Others are very power efficient, but are not as fast for demanding applications.
“We want to use the different processors judiciously,” Watkins says. “Utilize the high performance processor in cases that will notably improve the user experience and employ the low power processor when it is ‘good enough’ to extend battery life.”
In reality there is more than just a single decision to be made. Watkins uses a knob analogy to describe how to fine tune battery efficiency.
“We’re looking at three to four ‘knobs’ that we adjust to provide greater performance and save power,” he says. “For some programs you can turn the knobs down and save a bunch of power and have a minimal impact on performance. For other programs, if you turn the knob down to save power it will dramatically degrade performance.”
A major challenge with these systems is that the best way to set the knobs is different for different programs and can even vary within different periods of the same program. To provide good performance while also providing the longest possible battery life, the processors and their different knobs need to be actively and intelligently managed.
A possibility in the future is to allow the user to pick how important performance is for a certain application.
“The question may come down to how much slowdown are you willing to accept if it will extend battery life? On a normal day, when you know you’ll be home to charge your phone in a few hours, you might only accept a small performance loss. If you are on a long trip and won’t be able to charge your battery for days, you might be willing to accept a 30 percent slowdown if it will double or triple battery life.”
This examination of energy efficiency (which is what leads to longer battery life) is not just for phones, Watkins says.
“It’s also for big data centers like Google and Amazon. While data centers don’t run on batteries, they do consume a large amount of energy. If an application can run more efficiently it will save energy, which saves money, and that makes them happy. And from an environmental standpoint, we are all happier if we consume less energy and burn fewer fossil fuels.”