Computer Processor Power Scheme Cuts Waste, Energy Use

(Intel Corp.)

Engineers and computer scientists at Case Western Reserve University in Cleveland have developed a method for allocating power in computer processors that cuts device energy use by as much as 40 percent. The team led by computer science professor Swarup Bhunia presented their findings in January at the 25th International Conference on VLSI (Very-Large-Scale Integration) Design in Hyderabad, India where it won the best paper award.

Bhunia’s research, funded by Intel Corp., traced the use of electric power in computer processors — central components in devices from cell phones to full-size servers. They focused particularly on the datapath that performs computations and makes control decisions, and memory that stores data. They found current methods that require powering-up the entire processor are inherently wasteful, since only parts of the processor are used at any one time.

The team that includes PhD student Lei Wang and PhD alumni Somnath Paul, also found that current methods of controlling power use in processors, called coarse-gating, require shutting down whole sections of the processor. The problem with that approach, they discovered, was that some parts of those processor sections are in use at most times, which require those parts to be fully powered.

Bhunia and colleagues came up with a more fine-grained gating method that provides power to smaller pieces of the processor rather than entire parts. For example, the addition module of a processor is configured to add very large numbers. While that module is used often in a processor, it is not always crunching large numbers.

Their fine-grained gating method extends power only to the specific sections of the addition module needed at any one time, not the entire module. That same approach is applied to memory. While memory cells are capable of holding large numbers, the amount of power extended for storage is calibrated to the size of the data being stored, rather than the one-large-size-fits-all approach currently in use.

The team estimates that a high-performance system like a desktop computer could reduce its total power consumption by about 40 percent. Their method could also extend smart phone batteries from their current eight hours to 11 hours.

Read more: University Spin-Off to Develop Semiconductor Power Devices

*     *     *