Carnegie Mellon University has announced plans to spend $2.6 million in grant money on the development of smart, ultra-efficient silicon chips for future devices.
Cash from the National Science Foundation is to be used to run a three-year initiative at the University's Centre for Silicon System Implementation, which will look to dramatically boost the efficiency of silicon semiconductors by allowing them to improve themselves through machine learning.
'Ironically, for a technology that has imbued nearly every object with some level of intelligence, the management of a silicon chip's internal resources and infrastructure is largely ad hoc and piecemeal, leading to sub-optimal utilisation of chip resources,' claimed Shawn Blanton, head of the CSSI and professor of electrical and computer engineering at the University. 'This lack of sophistication means that today's chips actually waste a significant fraction of the power they burn, leading to shorter battery lives, higher data centre power demands, and ultimately less efficient use of the world's shrinking energy supply.'
The solution, Blanton and his team claim, is SLIC: Statistical Learning in Chip, a machine-learning system which allows a circuit to monitor its performance and conditions and make adjustments as required to ensure peak efficiency. Using technology developed in partnership with the Robotics Institute, the team hopes to create 'smart' processors which can dramatically boost their efficiency automatically and without user interaction.
The SLIC technology would go far beyond existing power management systems, adjusting the silicon circuit in far finer-grained ways than is currently possible and based on far more data than the crude temperature- and load-sensitive triggers used today.
SLIC isn't just limited to improving performance and lowering power draw, however. 'The high-speed, high efficiency SLIC engines also will enhance applications outside the chip,' claimed Blanton, 'like improving new smart systems, such as sensors that predict blood sugar levels for controlling diabetes or streamlining brain-computer interfaces for controlling prosthetic limbs.' The technology is also claimed to have implications for infrastructure supervisory systems including power grids, air traffic control, and the telecommunications infrastructure.
'The research has tremendous impact as the team works essentially to develop a new paradigm in IC technology where the chips themselves have a kind of intelligence and thus provide far superior performance to the systems they are part of,' added Ed Schlesinger, head of the electrical and computer engineering department at the University.