Stream Processor Development using Multi-Threshold NULL Convention Logic Asynchronous Design Methodology

Decreasing transistor feature size has led to an increase in the number of transistors in integrated circuits (IC), allowing for the implementation of more complex logic. However, such logic also requires more complex clock tree synthesis (CTS) to avoid timing violations as the clock must reach many more gates over larger areas. Thus, timing analysis requires significantly more computing power and designer involvement than in the past. For these reasons, IC designers have been pushed to nix conventional synchronous (SYNC) architecture and explore novel methodologies such as asynchronous, self-timed architecture. This dissertation evaluates the nominal active energy, voltage-scaled active energy, and leakage power dissipation across two cores of a stream processor: Smoothing Filter (SF) and Histogram Equalization (HEQ). Both cores were implemented in Multi-Threshold NULL Convention Logic (MTNCL) and clock-gated synchronous methodologies using a gate-level netlist to avoid any architectural discrepancies while guaranteeing impartial comparisons. MTNCL designs consumed more active energy than their synchronous counterparts due to the dual-rail encoding system; however, high-threshold-voltage (High-Vt) transistors used in MTNCL threshold gates reduced leakage power dissipation by up to 227%. During voltage-scaling simulations, MTNCL circuits showed a high level of robustness as the output results were logically valid across all voltage sweeps without any additional circuitry. SYNC circuits, however, needed extra logic, such as a DVS controller, to adjust the circuit’s speed when VDD changed. Although SYNC circuits still consumed less average energy, MTNCL circuit power gains accelerated when switching to lower voltage domains