University of Michigan
Electrical & Computer Engineering
1301 Beal Ave., 2417C EECS
Ann Arbor, MI 48109
Tel: 734 763 4526
Fax: 734 763-4617
- Energy-efficient near-threshold computing
- On-chip power supply boosting
- Near-threshold computing to maximize performance of heterogeneous workloads given a power constraint. Power consumption of a core can be reduced super linearly by lowering supply voltage while throughput can be maintained through parallelism.
- Circuit and CAD techniques to enable near-threshold voltage operation.
- Fast core boosting to efficiently adjust operating voltage to maximize energy efficiency. An intermediate power rail is used to raise the core supply voltage without disturbing other cores on a many-core processor.
- Random number generation by using time-dependent oxide breakdown (OxiGen). Most prior true random number generator architectures rely on post-processing to remove bias in the generated stream. OxiGen repeatedly forces soft gate oxide dialectric breakdown under voltage stress and uses the time to breakdown to generate a random bit sequence, without requiring a post-processor.