Synthesis of energy-efficient configurable processor arrays

We present a methodology to synthesize processor arrays that can be configured within an application domain. These configurable processor arrays are much more energy-efficient than general purpose processor arrays, and are comparable in performance to their hard-wired counterparts. The synthesis methodology is applicable for CMOS technology and is based on elementary architectural transformations like retiming, slowdown, holdup and folding. As an illustration of the methodology, the technique is applied to synthesize an energy-efficient processor array for finite-impulse-response filtering, configurable for sample-rate and filter-order. In order to achieve energy-efficiency, a variable clock and power supply, generated by a phase-locked-loop, is used.