A 2.72GOPS/11mW low power reconfigurable accelerator with a highly parallel datapath consisting of combinatorial circuits in 65nm CMOS

Abstract

CMA (Cool Mega-Array) is a high energy-efficiency reconfigurable accelerator for battery-driven mobile devices. It consists of a large processing element (PE) array without memory elements for mapping the data-flow graph of the application being executed, a small simple programmable micro-controller for data management, and a data memory. Unlike traditional coarse grained reconfigurable processors in which each PE provides registers and context memory, a CMA rduces power consumption by doing away with that for switching of hardware context and storing intermediate data in registers and their clock distribution. Although the data-flow graph mapped on the PE array is static during execution, various application programs can be implemented by making the best use of flexible data management instructions in the micro-controller. When the delay time of the PE array is shorter than the data handling time taken by the micro-controller, the supply voltage for the PE array is scaled to reduce the power consumption without degrading the performance. In contrast, when the delay time of the PE array is longer, wave pipelining is applied to enhance performance of the PE array. A prototype CMA chip (CMA-1) with 8 × 8 PE array with 24-bit data width was fabricated on the basis of 2.1× 4.2-mm 65-nm CMOS technology, and achieves sustained performance of 2.5-GOPS/11.2-mW. This energy efficiency is comparable to that of the most-energy-efficient accelerators that have been reported.