Information Technology
Mobile Devices Conserve Energy by Adjusting Accuracy
WARF: P130132US01
Inventors: Nam Sung Kim
The Wisconsin Alumni Research Foundation (WARF) is seeking commercial partners interested in developing a method for improving efficiency using a multiplier circuit that adjusts computing accuracy during run time.
Overview
Achieving energy efficiency is important in mobile computing devices like smartphones and tablets because they rely on battery power but are constrained by size and weight. Moreover, these devices increasingly are using sophisticated human machine interfaces (HMIs) that involve recognizing speech, gestures and handwriting.
Such ‘recognition’ tasks require large numbers of multiplication operations, for example, matrix multiplication. Specialized hardware multipliers that typically handle these high-speed operations are impractical for portable devices because they use too much energy.
Such ‘recognition’ tasks require large numbers of multiplication operations, for example, matrix multiplication. Specialized hardware multipliers that typically handle these high-speed operations are impractical for portable devices because they use too much energy.
The Invention
A UW–Madison researcher has developed multiplication circuitry that dynamically changes its accuracy (and energy usage) in response to operating demands. Accuracy is adjusted to meet particular computation tasks, power management strategies or error thresholds.
Specifically, a shift and accumulate multiplication circuit precomputes multiplicand shift amounts rather than computing them on the fly with a ‘leading-one detector.’ The circuit prestores the values in a coefficient memory. A controller adjusts accuracy according to processor needs.
Precomputation is possible in many recognition tasks associated with HMIs, where relatively static multiplier coefficients are used.
Specifically, a shift and accumulate multiplication circuit precomputes multiplicand shift amounts rather than computing them on the fly with a ‘leading-one detector.’ The circuit prestores the values in a coefficient memory. A controller adjusts accuracy according to processor needs.
Precomputation is possible in many recognition tasks associated with HMIs, where relatively static multiplier coefficients are used.
Applications
- Implementation in mobile devices, especially cell phones
Key Benefits
- Up to eight-fold improvement in energy efficiency
- Dynamic
- Adjusts to accuracy needs
- Suitable for portable devices
Stage of Development
The method has demonstrated accuracy between 93 and 96 percent while consuming as much as 11 times less energy per multiplication.
Additional Information
Related Technologies
For current licensing status, please contact Jeanine Burmania at [javascript protected email address] or 608-960-9846