| Applying asynchronous methods to the design of high performance systems opens up a whole new region of design space. The goal of this dissertation is to explore some of this new space, showing the promise of asynchronous logic for creating high-speed circuits and examining some of the trade-offs that occur. The use of an asynchronous rather than a synchronous design style changes significantly the choices that can be made at several design levels. It affects the logic families used, the implementation of larger blocks such as state machines, and the high level architectural design of a system.; Gallium arsenide (GaAs) is used as the implementation technology, since it currently offers speeds comparable to the best CMOS or BiCMOS processes and is available to the academic community through the MOSIS program. The combination of a self-timed design method and GaAs technology is a synergistic one; GaAs provides the raw speed while the self-timed techniques described here make that performance usable at the system level. As an example, in a synchronous system the higher performance offered by GaAs will be greatly reduced if clock skew increases as a percentage of the cycle time, and faster gates do not help much if the overall cycle time is limited by interconnect dominated paths. The local interconnects possible in most self-timed circuits allow faster gates to result in higher performance, and long wires slow down only the affected signals, rather than the entire chip.; Narrowing the design space to self-timed GaAs circuits makes the subject tractable as the focus of a Ph.D. dissertation. Nevertheless it is an extremely large area to cover, and mostly virgin territory. Therefore, this dissertation tends to be more broad than deep, but specific examples and actual test circuits provide the detail that might otherwise be missing in such a wide sweep. |
