w-bit Phase 1 Validation

This directory contains the reproducible simulation suite for the w-bit architecture (older docs may say "w-bit"—same project).

Project Identity

w-bit is an alternative non-binary, weighted, noisy logic substrate intended for ML-style inference/learning and approximate control, not a drop-in binary CPU replacement. Phase 1 is a controlled validation suite (Exp A/B/C) with reproducibility safeguards (map invariance, path hygiene, fixed outputs under results/).

Overview

Three experiments are implemented to validate the core claims:

• Experiment A (Router Sweep): Statistical verification of routing efficiency on a 10x10 grid.
• Experiment B (Noise Breakdown): Robustness testing of logic gates under increasing analog noise.
• Experiment C (Learning Search): Structural search to confirm non-linear learning capability (XOR).

Quick Start

To run the full Phase 1 validation suite with default parameters:

python3 run_phase1.py

Results will be generated in results/expA/, results/expB/, and results/expC/.

For a quick validation run, use:

python3 run_phase1.py --smoke

Phase 2 scaffold (binary baseline comparison) is available via:

python3 run_phase2.py --smoke --run_expB_grid

Running Individual Experiments

You can run each experiment manually with custom parameters.

Experiment A: Router Sweep

python3 experiments/exp_a_router_sweep.py --trials 100 --sigma 0.1 --grid 10

• --obstacle_density: List of densities to sweep (e.g., 0.1 0.2 0.3). Default is [0.1, 0.2, 0.3].

Experiment B: Noise Breakdown

python3 experiments/exp_b_noise_breakdown.py --trials 100 --R 3

• --sigma: Specific noise level. If omitted, runs a sweep from 0.0 to 1.0 (step 0.1).

Experiment C: Learning Search

python3 experiments/exp_c_learning_search.py --trials 10 --R 2

• --H: List of hidden neuron counts to sweep (e.g., 0 1 2 4). Default is [0, 1, 2, 4].

Metrics

All experiments log the Relaxation Cost Proxy (RCP), defined as: RCP = (N_cells * N_steps * R^2) / I_out

Logs are saved as results.csv in the respective result directories.

Paths & Outputs

• Repo root: wdit_project/
• Outputs are under results/
  • Exp A → results/expA
  • Exp B (noise breakdown) → results/expB
  • Exp B (grid, optional) → results/expB_grid
  • Exp C → results/expC
  • Aggregated report → results/phase1_report.csv

run_phase1.py uses these paths by default; pass --smoke for fast checks and --run_expB_grid to include the grid sweep.

Phase 2 Plan (Baseline Comparisons)

Phase 2 now includes a binary baseline scaffold to compare against w-bit:

• Runner: run_phase2.py (supports --smoke, --run_expB_grid)
• Outputs: results/phase2/wbit/... and results/phase2/binary/...
• Aggregation: analysis/aggregate_phase2_report.py
• Goal: side-by-side summaries/plots showing regimes where w-bit is more noise-robust or achieves lower RCP at comparable success. Binary mode now runs with enforced binary states; plotting via analysis/plot_phase2_comparison.py.

Local Venv (optional)

Create and activate a local virtualenv:

python3 -m venv .venv
source .venv/bin/activate
python -m pip install -U pip