Thermal-Substrate Computing

Simulation code and reproducibility data for the paper:

Decentralized Thermal-State Load Routing and an ENAQT-Inspired Circuit Design Principle for Energy-Efficient Manycore Architectures Cleber Barcelos Costa (2026).

ORCID: 0009-0000-5172-9019

---

Overview

This repository contains the code and result files needed to reproduce the two simulation phases of the paper:

• Phase 1 — ENAQT-analog efficiency model. Computes signal-transport efficiency η(T) for the proposed ENAQT-inspired topology versus conventional CMOS across the operating temperature range T ∈ [25 °C, 80 °C].
• Phase 2 — Decentralized thermal routing agent. Simulates a 16-node 4×4 grid under uniform and hotspot loads, comparing the H6 decentralized routing algorithm against thermally-oblivious centralized scheduling.

The paper presents two independent contributions:

1. A decentralized thermal-state load-routing algorithm validated by the Phase 2 simulation (92.1% thermal variance reduction, 14.2 °C peak reduction under 4× localized overload, O(1) communication per node).
2. An ENAQT-inspired circuit topology presented as a theoretical design proposal, with the Phase 1 simulation demonstrating the mathematical behavior of an ENAQT-analog efficiency model. Physical validation in fabricated silicon is required and is the subject of the proposed H7 experimental program.

---

Repository structure

thermal-substrate-computing/
├── README.md               This file
├── LICENSE                 MIT License (code)
├── CITATION.cff            Machine-readable citation metadata
├── requirements.txt        Python dependencies
├── .gitignore              Python ignore patterns
├── simulation_phase1.py    Phase 1 — ENAQT-analog efficiency
├── simulation_phase2.py    Phase 2 — Decentralized thermal routing
└── data/
    ├── phase1_results.json Reference output (Phase 1)
    └── phase2_results.json Reference output (Phase 2)

---

Quick start

Requires Python 3.10 or later.

# Clone
git clone https://github.com/gallori-ai/thermal-substrate-computing.git
cd thermal-substrate-computing

# Install
pip install -r requirements.txt

# Reproduce Phase 1 (Table 1 of the paper)
python simulation_phase1.py

# Reproduce Phase 2 (Table 2 of the paper)
python simulation_phase2.py

Both scripts are deterministic. Phase 2 uses a fixed random seed (42) for symmetry-breaking initial conditions.

Note: The default output path inside the scripts (knowledge/papers/H6-thermal-substrate/) reflects the author's internal project layout. Either create that directory before running, or edit the path argument in each save_results(...) call.

---

Reference results

Reference outputs for both phases are checked into data/. After running the simulations, your generated JSON should match these files bit-for-bit (modulo float formatting).

Phase 1 highlights

T (°C)	η_H6	η_conv	Δη rel.	Energy advantage
25	0.750	0.720	+4.2 %	4.0 %
50	0.809	0.660	+22.6 %	18.4 %
75	0.859	0.604	+42.2 %	29.7 %
80	0.868	0.594	+46.1 %	31.6 %

Phase 2 highlights

Scenario	Routing	Mean variance (°C²)	Mean T_peak (°C)
Uniform load	H6	0.008	35.04
Uniform load	Central	0.008	35.04
Hotspot (4× load)	H6	12.46	49.71
Hotspot (4× load)	Central	158.88	63.88

---

Authoritative parameters

All simulation parameters are documented in Appendix A of the paper.

Phase 1

• Coupling parameter κ_m = 3.0
• Bath correlation time τ_bath = 1 ps
• CMOS baseline η₀ = 0.72, α = 3.5×10⁻³ K⁻¹
• Gate capacitance C = 5 fF, supply voltage V = 0.8 V

Phase 2

• N = 16 nodes, 4×4 grid, von Neumann 4-neighborhood
• T_ambient = 25 °C, T_initial = 35 °C, T_threshold = 38 °C
• Q_per_task = 0.8 °C/step, α_diss = 0.08
• Hotspot multiplier = 4.0 (nodes {0, 1, 4, 5})
• 300 simulation steps, random seed = 42

---

Citation

If you use this code or build on this work, please cite the paper:

@misc{costa2026thermal,
  author       = {Costa, Cleber Barcelos},
  title        = {Decentralized Thermal-State Load Routing and an
                  ENAQT-Inspired Circuit Design Principle for
                  Energy-Efficient Manycore Architectures},
  year         = {2026},
  publisher    = {Zenodo},
  version      = {1.0},
  doi          = {10.5281/zenodo.19857070},
  url          = {https://doi.org/10.5281/zenodo.19857070}
}

A CITATION.cff file is included for GitHub's automatic citation feature.

---

License

• Code (*.py): MIT License
• Paper and result data (data/*.json, PDFs): CC BY 4.0

---

Contributing

Issues and pull requests are welcome — replications, parameter explorations, ports to other languages, or extensions to larger N.

The natural next step after this work is H7: an FPGA prototype of the routing algorithm (Contribution 1) and a fabricated test structure for the ENAQT-analog topology (Contribution 2). If you're working on either, please get in touch.