THERMO-NET Results

Home Dashboard Results Documentation GitLab

91.3%

Mean η_T

5-regime cross-validation

87.9%

Entropy Production Reduction

vs uncontrolled baseline

7.4×

Qubit Coherence Extension (T1)

47.3 μs → 351.4 μs @ 15 mK

7.5×

Qubit Coherence Extension (T2)

31.8 μs → 236.0 μs @ 15 mK

93.8%

Carnot Efficiency Approach

6.2% gap remaining

< 5.1%

Cross-Regime Drop

Zero retraining generalization

η_T by Thermal Regime

Entropy Production Reduction (%)

Dataset Composition

THERMO-NET vs Classical MPC

Regime	THERMO-NET η_T	Classical MPC η_T	Improvement	σ Reduction
Sub-2nm CMOS Node (R1)	92.1%	75.3%	+16.8%	88.4%
Photonic Crystal Reservoir (R2)	91.7%	74.8%	+16.9%	86.9%
Cryogenic Qubit Array (R3)	93.4%	76.2%	+17.2%	91.3%
Heat Engine (R4)	89.6%	72.1%	+17.5%	85.2%
Thermoelectric Harvester (R5)	90.8%	73.4%	+17.4%	87.7%

η_T = 0.913 ± 0.014

Mean Thermal Efficiency

5-regime validation · 15 mK – 1200 K

σ/σ₀ = 0.121 ± 0.018

Mean Entropy Production Ratio

87.9% average reduction

Δ(η/η_Carnot) = +16.8 pp

Improvement over Classical MPC

Physics-informed AI advantage

Performance by Thermal Regime

Regime	Platforms (n)	η_T	σ Reduction	Primary Noise	Dominant Construct
Sub-2nm CMOS Node (R1)	12	92.1%	88.4%	Ballistic phonon transport	NHTO + LEPM
Photonic Crystal Reservoir (R2)	10	91.7%	86.9%	Phonon–photon scattering	NHTO + TICT
Cryogenic Qubit Array (R3)	8	93.4%	91.3%	Quasiparticle thermal noise	LEPM + TICT
Heat Engine (R4)	7	89.6%	85.2%	Viscous & conductive irreversibility	LEPM + NHTO
Thermoelectric Harvester (R5)	6	90.8%	87.7%	Joule heating & Seebeck mismatch	TICT + NHTO

Four Key Scientific Findings

KEY FINDING 1

NHTO resolves ballistic phonon transport at sub-nanometer resolution

SIREN architecture captures phonon mean free path heterogeneity at 0.4 nm resolution

KEY FINDING 2

LEPM achieves 87.9% mean entropy production reduction

Second Law as hard constraint enables near-reversible operation

KEY FINDING 3

7.4× qubit coherence extension at 15 mK

Suppression of quasiparticle generation events via predictive thermal control

KEY FINDING 4

Carnot efficiency approach within 6.2% of theoretical maximum

93.8% of Carnot limit achieved in heat engine regime (R4)

Dataset Summary

Thermal Regimes

Sensor Stations

Years (2022-2026)

15mK–1200K

Temperature Range

Quick Links

Code Repositories

Resources