In high-speed digital circuits, domino logic can reduce transistor count and input capacitance compared with static CMOS, making it useful for wide fan-in gates. Its benefits come with a design challenge: the keeper transistor that protects the dynamic node from leakage and noise can contend with the pull-down network during evaluation. A modified domino logic keeper offers a way to manage that contention while targeting lower power and circuit area.
Why the keeper matters in domino logic
A domino gate operates in two phases controlled by a clock. During precharge, a PMOS transistor charges the dynamic node. During evaluation, an NMOS pull-down network either discharges that node or leaves it high. An inverter at the output produces a non-inverting logic result and helps make it possible to cascade dynamic stages.
The dynamic node is vulnerable to leakage and charge sharing, so a PMOS keeper is commonly used to maintain its high level when the pull-down network is not conducting. But the keeper can remain on briefly at the start of evaluation. If the inputs activate the pull-down network at that moment, the keeper tries to charge the node while the network tries to discharge it. This contention creates a current path from supply to ground and can increase power or slow evaluation.
Reducing keeper size can limit contention, but a weaker keeper also reduces the circuit’s ability to preserve the node’s logic level. Keeper design therefore involves balancing power, speed and noise tolerance—not simply minimizing transistor width.
A modified domino logic keeper structure
The proposed approach places an NMOS transistor in series with the PMOS keeper. The NMOS gate is driven by the clock, while the PMOS gate is connected to the gate output. This changes when the keeper path can conduct, without adding sleep-mode control circuitry or relying on multiple threshold voltages.
During precharge, the clock-controlled NMOS is off, so the keeper path is disabled while the PMOS precharge device charges the dynamic node. At the start of evaluation, the keeper remains off, reducing the opportunity for contention. If the pull-down network conducts, the output changes state and the PMOS keeper is off. If the network does not conduct, the output keeps the PMOS on; with the clock high, the series NMOS also turns on and helps maintain the dynamic node.
The design uses a single threshold voltage and adds one NMOS to the keeper arrangement. In the comparison presented, the proposed module has six transistors apart from the pull-down network—the same count reported for SSDD and RDVD, and fewer than several other compared techniques. The design’s area advantage is therefore tied to its simpler control structure, rather than to a claim that every circuit implementation will use less silicon.
What the simulations show
The designs were simulated in Ngspice using a 16 nm predictive technology model, with a 0.9 V supply. The comparison covered 8- and 16-input OR gates and keeper aspect ratios from W/L = 1 to 6. Evaluated measures included dynamic and static power, propagation delay and power-delay product (PDP), which combines power consumption and delay as a comparative circuit metric.
For the 8-input OR gate at W/L = 6, the reported PDP for the modified keeper was 3.37 × 10-18 watts-seconds. That was reported as a 38.09% reduction relative to SSDD and a 99.99% reduction relative to CDD. For the 16-input OR gate at the same keeper ratio, the reported PDP was 2.76 × 10-18 watts-seconds, a 33.46% reduction relative to CDSMD. These are results for the simulated configurations, not universal performance guarantees.
The results also illustrate why PDP should not be read in isolation. For the 8-input gate, the proposed design’s delay at W/L = 6 was lower than that of FDSTDL, but higher than that of SSDD. Its dynamic power was lower than CDD, HSD and MHSD in the reported comparison, while some other approaches had lower dynamic power. The best choice depends on which circuit constraints matter most.
Design trade-offs for engineers
For engineers evaluating wide fan-in dynamic logic, the series NMOS changes the keeper’s timing behavior: it prevents the keeper path from being active at the beginning of evaluation, while allowing the keeper to support the node when the pull-down network is inactive. This can be relevant when contention and leakage are significant design concerns.
However, noise immunity remains an important consideration. The study reports a unity noise gain of 0.510 V for the proposed design in the 16-input comparison. Several alternatives had higher values, including HSD at 0.761 V and FDSTDL at 0.755 V. A lower reported unity noise gain indicates that the proposed circuit does not lead every compared technique on this measure. Validation should therefore include noise-margin requirements and expected operating conditions, not just power and area comparisons.
The reported simulations also include Monte Carlo analysis for the proposed circuit. For the 8- and 16-input OR gates, the reported variability values (standard deviation divided by mean) were 0.510 and 0.561, respectively. These results provide a view of the simulated circuit under modeled parameter variation, but do not replace process-specific validation or physical implementation checks.
Conclusion
A modified domino logic keeper can reduce the opportunity for keeper-versus-pull-down contention by using a clock-controlled NMOS in series with the PMOS keeper. The reported 8- and 16-input OR-gate simulations show promising PDP results and a compact transistor count, while also revealing trade-offs in delay and noise immunity. For practical design decisions, compare the full set of metrics at the intended fan-in and keeper sizing, then verify the chosen implementation against the circuit’s noise, timing and power requirements.