TL;DR
This paper introduces a novel "multi-code state assignment" technique for sequential circuits that reduces power consumption by intelligently assigning multiple codes to certain states. This allows for fewer state variables, enabling clock-gating of unused flip-flops and leading to significant power savings, as demonstrated through simulations.
What Problem Does This Paper Address?
Designing low-power digital circuits is a critical challenge, and sequential circuits, which form the backbone of many digital systems, offer substantial opportunities for power optimization. Traditional sequential circuit design involves several steps, including state reduction and state assignment. State assignment, in particular, plays a crucial role in determining the overall power dissipation by influencing the switching activity of internal signals and the complexity of the combinational logic.
A key source of power inefficiency in sequential circuits arises from "redundant states." When the number of available state variables (k) is greater than the number of actual states (l) required by the circuit (l < 2^k), there are 2^k - l "redundant" or "unused" states. While these can sometimes simplify combinational logic, they also introduce potential reliability issues and, more importantly for this paper, represent unused flip-flop states that consume power. Existing methods for power reduction in sequential circuits often focus on clock gating or using double-edge triggered flip-flops, but this paper targets the state assignment itself as a primary optimization point.
Methods
The core contribution of this paper is the multi-code state assignment technique. Unlike traditional "uni-code" assignment where each state maps to a unique binary code, multi-code assignment allows certain states to be represented by multiple binary codes. This is achieved by strategically utilizing the redundant state codes.
The authors propose an algorithm to implement this multi-code state assignment:
- Identify Redundant States: Determine the unused state codes based on the number of state variables and the actual number of states.
- Multi-Code Assignment: Assign multiple codes to states that can cover redundant state codes. This is done by selecting "cubes" (groups of minterms) that encompass a desired state and one or more redundant states. The goal is to cover as many redundant states as possible with the fewest state variables.
- Clock Gating: When a state is assigned a multi-code, and the system is in a state where only a subset of the assigned codes is active, the flip-flops corresponding to the unused state variables can be "clock-gated." This means their clock input is blocked, preventing them from switching and thus saving power.
- Algorithm: A systematic algorithm is provided to guide the selection of multi-codes and the application of clock gating, prioritizing states with higher occurrence probabilities for more effective power saving. The algorithm iteratively refines the state assignment to minimize power dissipation.
Datasets
Not discussed in the paper. The paper uses examples of sequential circuits (a one-hot ring counter, a decimal up-counter, and an 8421 BCD detector) for simulation and demonstration, rather than standard datasets.
Key Insights and Ideas
- Redundant states are an opportunity, not just a problem: Instead of trying to eliminate redundant states entirely (which can be complex and potentially impact reliability), this paper shows how to leverage them for power optimization.
- Multi-code assignment enables efficient clock gating: By assigning multiple codes to a single state, we create situations where some state variables are not needed for certain state transitions. This directly enables clock gating for those specific flip-flops.
- Prioritizing high-probability states is key: The proposed algorithm emphasizes assigning multi-codes to states that occur more frequently, maximizing the potential for power savings through clock gating.
- Reduced state variables lead to simpler circuits: By using fewer state variables to represent the necessary states, the overall circuit complexity can be reduced.
Key Results
The paper demonstrates significant power savings through simulations using PSPICE:
- One-hot ring counter example: A one-hot ring counter, originally using 4 state variables and 16 triggering actions per cycle, was transformed into a "one-zero-hot" counter using 3 state variables and multi-code assignment. This resulted in a reduction from 16 to 9 triggering actions per cycle, theoretically a 43% saving. Actual simulation showed a 33% power saving, attributed to the overhead of the clock gating NOR gates.
- Decimal up-counter example: Applying the multi-code assignment to a decade up-counter resulted in a simplified combinational circuit and a 10% energy saving for the flip-flops themselves, despite the overhead of an additional clock gating NOR gate.
- 8421 BCD detector example: This example showed that when a system is in certain states (A or B), clock gating can be applied to the flip-flop
x, saving 50% of its energy dissipation. For all three flip-flops, an overall saving of approximately 16.7% was achieved.
Limitations and Future Work
- Overhead of clock gating: The paper acknowledges that the clock gating logic (e.g., NOR gates) introduces some additional power dissipation, which can slightly reduce the theoretical maximum savings.
- Complexity of the assignment procedure: While an algorithm is provided, the process of selecting optimal cubes for multi-code assignment can be complex, and the paper mentions that procedures like ESPRESSO can be used to handle this efficiently.
- Reliability considerations: While the paper suggests multi-code assignment can improve reliability by eliminating non-redundant states, a deeper analysis of the reliability implications of entering multi-coded states might be beneficial.
Future work could involve exploring more sophisticated algorithms for cube selection, investigating the impact of multi-code assignment on other performance metrics (like timing), and extending the technique to more complex sequential circuit architectures.
Summary updated