An Early Completion technique is developed to significantly increase the throughput of NULL Convention self-timed digital systems without impacting latency or compromising their self-timed nature. Early Completion performs the completion detection for registration stage i at the input of the register, instead of at the output of the register as in standard NULL Convention Logic. This method requires that the single-rail completion signal from registration stage i+1 , Ko i+1 , be used as an additional input to the completion detection circuitry for registration stage i , to maintain self-timed operation. However, Early Completion does necessitate an assumption of equipotential regions, introducing a few easily satisfiable timing assumptions, thus making the design potentially more delay-sensitive. To illustrate the technique, Early Completion is applied to a case study of an optimally pipelined 4-bit by 4-bit unsigned multiplier utilizing full-word completion, where a speedup of 1.21 is achieved while self-timed operation is maintained and latency remains unchanged.
S. C. Smith, "Speedup of Self-Timed Digital Systems using Early Completion," Proceedings of the IEEE Computer Society Annual Symposium on VLSI, 2002, Institute of Electrical and Electronics Engineers (IEEE), Jan 2002.
The definitive version is available at http://dx.doi.org/10.1109/ISVLSI.2002.1016884
IEEE Computer Society Annual Symposium on VLSI, 2002
Electrical and Computer Engineering
Keywords and Phrases
Early Completion Technique; NULL Convention Self-Timed Digital Systems; Completion Detection; Delay-Sensitive Design; Equipotential Regions; Full-Word Completion; Latency; Logic Design; Multiplying Circuits; Optimally Pipelined Unsigned Multiplier; Pipeline Processing; Register Input; Registration Stage; Self-Timed Operation; Single-Rail Completion Signal; Speedup; Throughput; Timing; Timing Assumptions
Article - Conference proceedings
© 2002 Institute of Electrical and Electronics Engineers (IEEE), All rights reserved.