| NULL Convention Logic (NCL) provides an asynchronous design methodology employing dual-rail signals, quad-rail signals, or other Mutually Exclusive Assertion Groups (MEAGs) to incorporate data and control information into one mixed path. In NCL, the control is inherently present with each datum, so there is no need for worse-case delay analysis and control path delay matching. This dissertation focuses on optimization methods for NCL circuits, specifically addressing three related architectural areas of NCL design.; First, a design method for optimizing NCL circuits is developed. The method utilizes conventional Boolean minimization followed by table-driven gate substitutions. It is applied to design time and space optimal fundamental logic functions, a time and space optimal full adder, and time, transistor count, and power optimal up-counter circuits. The method is applicable when composing logic functions where each gate is a state-holding element; and can produce delay-insensitive circuits requiring less area and fewer gate delays than alternative gate-level approaches requiring full minterm generation.; Second, a pipelining method for producing throughput optimal NCL systems is developed. A relationship between the number of gate delays per stage and the worse-case throughput for a pipeline as a whole is derived. The method then uses this relationship to minimize a pipeline's worse-case throughput by partitioning the NCL combinational circuitry through the addition of asynchronous registers. The method is applied to design a maximum throughput unsigned multiplier, which yields a speedup of 2.25 over the non-pipelined version, while maintaining delay-insensitivity.; Third, a technique to mitigate the impact of the NULL cycle is developed. The technique further increases the maximum attainable throughput of a NCL system by reducing inherent overheads associated with an integrated data and control path. This technique is applied to a non-pipelined 4-bit by 4-bit unsigned multiplier to yield a speedup of 1.61 over the standalone version.; Finally, these techniques are applied to design a 72 + 32 × 32 multiply and accumulate (MAC) unit, which outperforms other delay-insensitive/self-timed MACs in the literature. It also performs conditional rounding, scaling, and saturation of the output, whereas the others do not; thus further distinguishing it from the previous work. The methods developed facilitate speed, transistor count, and power trade-offs using approaches that are readily automatable. |
