Double nanowire quantum dots and machine learning.

Area of Science:

• Condensed Matter Physics
• Materials Science
• Computational Physics

Background:

• Accurate prediction of electronic properties in nanostructures is crucial for quantum technologies.
• Quantum dots offer tunable electronic properties based on their size and shape.
• Computational methods for simulating these properties can be computationally intensive.

Purpose of the Study:

• To develop an efficient computational method for estimating single-particle energies in double InAs/InP nanowire quantum dots.
• To leverage machine learning and transfer learning to reduce the computational cost of these estimations.
• To provide a framework for solving the inverse problem of linking nanostructure properties to their spectra.

Main Methods:

• Combining an atomistic tight-binding approach with machine learning algorithms, specifically neural networks.
• Utilizing transfer learning to capitalize on results from smaller-scale computations.
• Employing a small training set to predict energies across a multidimensional parameter space.

Main Results:

• Accurate recovery of ground state energies with a root-mean-square deviation of approximately 1 meV.
• Demonstrated the effectiveness of the machine learning approach even with a training set that is a small fraction of the total search space.
• Successfully applied the method to double InAs/InP nanowire quantum dots.

Conclusions:

• The proposed machine learning approach significantly enhances the efficiency of predicting single-particle energies in nanowire quantum dots.
• Transfer learning proves effective in reducing data requirements for accurate energy estimations.
• This technique offers a valuable tool for researchers investigating the relationship between nanostructure morphology and spectral properties.