Abstract
Power balancing techniques are provided for improving power efficiency of pipelined processors. A design-level implementation can be incorporated during synthesis of pipeline clocks in which a register transfer level (RTL) code, operating frequency, and available voltage domains are used to perform cycle time stealing with, and optimize for, power efficiency. A test-level implementation can be incorporated during testing of a chip in which delay and power measurements are used to perform calculations based on cycle time stealing and optimization of power efficiency. The calculations are then used to perform voltage scaling and/or adjust tunable delay buffers. Process variations may also be corrected during test time. A run-time approach can be incorporated for dynamic power balancing in which the operating system keeps track of one or more performance indicators such as a count of floating point instructions and uses a look-up table to provide the appropriate delays.
Original language | English (US) |
---|---|
U.S. patent number | 8806410 |
State | Published - Aug 12 2014 |