Green Computing -- Rethinking Computer Design and Deployment

When most people think of sustainability, the efficiency of a transistor isn't the first thing that comes to mind. If we think of computers at all, we might consider the vast amounts of energy used in data centers of Google and Amazon, but not the energy consumed by our smartphones, notebooks, and personal computers.
 
But Houman Homayoun isn't like most people. Assistant Professor Homayoun, a faculty member in the Department of Electrical and Computer Engineering at the Volgenau School of Engineering, believes that adjustments to these small components can have an exponential impact. The overarching theme of his research is green and sustainable computing. His research targets how to use computing resources in a more energy efficient and environmentally responsible manner. 
 
 "The current state of computing is not sustainable technologically, economically, or environmentally," says Homayoun. "The energy consumed by computers grows as we ask them to do more and more. We need to rethink the way we design, deploy, and utilize them."
 
The basic design of computers has remained the same for several decades. The brain and brawn of the computer is housed on a chip made of silicone. This microprocessor contains various components and one of the key components is its core. Simply stated, the core is the part of the computer that reads and executes instructions. For years the computer industry has relied on adding more cores to increase computer power¬–from single core, to dual, to quad and even eight and sixteen-core processors. Now the microprocessor industry faces a challenge called "dark silicone."

"Dark silicone is a way to describe the part of the computer that must be powered off to keep the energy demands reasonable," said Homayoun. "Sometimes as much as fifty percent of the silicone chip is not being used.  The more cores that we add to the chip, the more darkness we will have." A good way to think about dark silicone is to imagine an office building with half of the lights turned off at night—only some of the rooms are functional.
 
Homayoun and his team of researchers aim to address this problem by developing heterogeneous architectures, or systems that use more than one kind of core, to increase chip performance, efficiency, and lifetime reliability. Heterogeneous designs gain performance not just by adding cores, but also by incorporating specialized processing abilities to handle specific tasks. For example, one core could be designated for one task and when that function is dormant that section of the silicone could remain dark to conserve energy. Because the other cores would continue to work, performance would not be affected. 
 
By enabling more diverse and customized designs, the researchers hope to push the efficiency envelope even further. "Our work seeks to build better computing systems with the goal of saving energy and improving performance. It focuses on the design challenges in high-performance computing systems."