A novel machine learning workflow to optimize cooling devices grounded in solid-state physics.

Area of Science:

• Solid-state physics
• Nanotechnology
• Computational materials science

Background:

• Solid-state cooling devices are crucial for integrated-chip thermal management.
• Accurate modeling requires coupling quantum non-equilibrium Green's function for electrons with the heat equation (NEGF+H).
• NEGF+H simulations are computationally intensive, hindering rapid design optimization.

Purpose of the Study:

• To develop a machine learning (ML) workflow to accelerate the design optimization of solid-state cooling devices.
• To reduce the computational burden associated with NEGF+H simulations.
• To identify optimal heterostructure designs balancing cooling power and electron temperature.

Main Methods:

• A novel machine learning (ML) workflow was proposed and trained using data from NEGF+H simulations.
• The ML workflow explored a vast design space of [Formula: see text] device configurations.
• The methodology optimized heterostructures for the best trade-off between cooling power (CP) and electron temperature (T_e).

Main Results:

• The ML workflow achieved prediction relative errors below [Formula: see text] for CP and [Formula: see text] for T_e.
• Optimal device designs were identified within a large search space.
• Computational time was drastically reduced from two days per simulation to 10 seconds for design optimization.

Conclusions:

• The proposed ML workflow significantly accelerates the design process for solid-state nanocooling devices.
• This approach alleviates the high computational cost of traditional NEGF+H methods.
• The ML workflow enables efficient discovery of high-performance cooling device designs.