Machine learning delta-T noise for temperature bias estimation.

Area of Science:

• Condensed matter physics
• Quantum electronics
• Nanotechnology

Background:

• Delta-T shot noise arises in temperature-biased electronic junctions at the atomic scale.
• It exhibits a quadratic dependence on temperature difference and nonlinear transmission coefficients.

Purpose of the Study:

• To demonstrate the use of delta-T noise for estimating temperature bias in atomic-scale junctions.
• To develop a machine learning approach for predicting temperature differences.

Main Methods:

• Supervised machine learning with a neural network trained on synthetic and experimental data.
• Input features included scaled electrical conductance, delta-T noise, and mean temperature.
• Ensemble averaging of delta-T noise measurements across atomic junctions.

Main Results:

• A neural network trained on synthetic data accurately predicted temperature biases from experimental data.
• The mean bias was less than 1 Kelvin for junctions up to 4G0 conductance.
• Ensemble averaging improved prediction accuracy, overcoming limitations of single measurements.

Conclusions:

• Delta-T noise, when averaged over an ensemble of junctions, enables accurate estimation of temperature bias.
• Machine learning offers a viable method for estimating temperature biases and other stimuli in electronic junctions.