INDEX20 RESEARCH · MATHEMATICAL MODEL 01

PhaseFlow 18

A reproducible experiment in hierarchical state decomposition for event-driven controllers.

18 logical states3 equivalent RTL variants1,080 gate-level casesUpdated November 18, 2026
GENERIC CELLS68 / 82 / 147hierarchical / binary / flat
LOGIC ACTIVITY7,596 / 11,484hierarchical / binary · sparse workload
VALIDATION1,080 / 1,080synthesized transition cases passed
Scope: PhaseFlow 18 is an independent Index20 Research model inspired by hierarchical state decomposition. It is not the production Index20 scoring algorithm and is not presented as an exact reconstruction of L. F. Marakhovsky's circuitry.

MODEL

18 states as a product, not a flat list

The logical state is factored into a strategy and a processing phase:

S = G × P
|S| = 3 × 6 = 18

During phase t, command x(t) changes the work phase. During phase Δ, command e(Δ) selects the strategy. Both operations form one external cycle.

SleepCapturePreprocessInferTransmitFaultEcoNormalUrgent

INTERACTIVE MODEL

Run one external cycle

STARTEco / Sleep
→
AFTER tEco / Sleep
→
AFTER ΔEco / Sleep

The reference model is ready.

COMPARISON

Same behavior, three representations

ImplementationGeneric cellsFlip-flopsCombinational cells
Hierarchical 3 × 6681553
Independent binary fields82874
Flat 18-state one-hot14736111

Generic Yosys synthesis, without a target standard-cell library. The binary version uses fewer flip-flops; the hierarchical version uses fewer generic cells in this specific controller.

WORKLOADS

Stable combinational-output transitions over 10,000 cycles

Sparse events7,596 hierarchy11,484 binary14,286 flat
Active operation102,639 hierarchy154,605 binary192,150 flat
Mixed commands88,288 hierarchy138,887 binary170,516 flat

This is a switching proxy. It excludes glitches, capacitance, routing, clock-tree power and physical energy.

REPRODUCIBILITY

Evidence and method

  1. All 360 combinations of 18 starting states, five phase commands and four strategy commands were compared in the software reference.
  2. Three RTL variants were synthesized to generic gates with YoWASP Yosys 0.70.62-dev.1236.
  3. The exported gate networks were executed through both phases for 1,080 total cases.
  4. Three deterministic 10,000-cycle input profiles were used for activity comparison.

What this result does and does not establish

Established for this model

  • Equivalent observable behavior over the tested state/input domain.
  • 68 generic cells for the hierarchical RTL versus 82 binary and 147 flat.
  • Lower stable combinational-output activity for the hierarchy in all three test profiles.

Requires further measurement

  • Standard-cell area, critical timing and maximum clock rate.
  • Dynamic/leakage power, physical energy and post-layout wiring.
  • Advantage on a customer's real controller and workload.

QUESTIONS

Frequently asked questions

Does PhaseFlow improve Index20 website rankings?

No. This is a published research model and is not used by the production scoring engine.

Does 68 versus 82 cells prove lower silicon area?

No. Generic cells are an early synthesis result. Technology mapping, constraints, placement and routing are required for a physical area claim.

Can the method be tested on another controller?

Yes. A valid pilot needs a fixed behavior contract, the best applicable baseline, one target library and representative event traces.