This paper develops an ultra-low power asynchronous circuit design methodology, called Multi-Threshold NULL Convention Logic (MTNCL), also known as Sleep Convention Logic (SCL), which combines Multi-Threshold CMOS (MTCMOS) with NULL Convention Logic (NCL), to yield significant power reduction without any of the drawbacks of applying MTCMOS to synchronous circuits. In contrast to other power reduction techniques that usually result in large area overhead, MTNCL circuits are actually smaller than their original NCL versions. MTNCL utilizes high-V_t transistors to gate power and ground of a low-V_t logic block to provide for both fast switching and very low leakage power when idle. To demonstrate the advantages of MTNCL, a number of 32-bit IEEE single-precision floating-point co-processors were designed for comparison using the 1.2 V IBM 8RF-LM 130 nm CMOS process: original NCL, MTNCL with just combinational logic (C/L) slept, Bit-Wise MTNCL (BWMTNCL), MTNCL with C/L and completion logic slept, MTNCL with C/L, completion logic, and registers slept, MTNCL with Safe Sleep architecture, and synchronous MTCMOS. These designs are compared in terms of throughput, area, dynamic energy, and idle power, showing the tradeoffs between the various MTNCL architectures, and that the best MTNCL design is much better than the original NCL design in all aspects, and much better than the synchronous MTCMOS design in terms of area, energy per operation, and idle power, although the synchronous design can operate faster.

This paper proposes a radiation hardened NULL Convention Logic (NCL) architecture that can recover from a single event latchup (SEL) or single event upset (SEU) fault without deadlock or any data loss. The proposed architecture is analytically proved to be SEL resistant, and by extension, proved to be SEU resistant. The SEL/SEU resistant version of a 3-stage full-word pipelined NCL 4 × 4 unsigned multiplier was implemented using the IBM cmrf8sf 130 nm 1.2 V process at the transistor level and simulated exhaustively with SEL fault injection to validate the proposed architectures. Compared with the original version, the SEL/SEU resilient version has 1.31× speed overhead, 2.74× area overhead, and 2.79× energy per operation overhead.