This research is on asynchronous, or self-timed, design. Such designs eliminate the global synchronizing clock, thus permitting higher performance, scalability, and lower power. The project is investigating problems in datapath design: (1) the design of high-performance and low-power datapath components, and (2) a system-level application to compressed- code embedded processors. In (1), a technique, "speculative completion", for designing asynchronous datapath components which give high performance, yet have little area and power overhead is being developed. This technique allows use of existing synchronous components, with small amounts of added circuitry and a multi-slotted completion signal, to allow early completion. As part of this research, (a) the technique is being applied to design fast adders, multipliers, and comparators; and (b) new optimizations are being developed, which target statistically-skewed (i.e. non- random) input data that appear in real-world applications. In (2), research activity being carried out is focused on a system-level application: the design of embedded systems with compressed memories. The investigation includes (a) exploration and evaluation of a variety of memory compression techniques; (b) design of a small and fast asynchronous decompression circuit, which can decompress such memories, on the fly, at acceptable rates.
