Machine Learning Assisted Revelation of the Best Performing Single Hetero-junction Thermophotovoltaic Cell

TL;DR

This paper employs machine learning to optimize and analyze single-heterojunction thermophotovoltaic cells, leading to the design of a high-efficiency cell with 16.50% efficiency using Ge and InGaAsSb materials.

Contribution

It introduces a machine learning-based approach to identify optimal material combinations and device structures for high-performance TPV cells, including a novel p-Ge/n-InGaAsSb design.

Findings

Ge identified as the optimal emitter layer.

Achieved 16.50% efficiency in the TPV cell.

Optimized parameters significantly improved cell performance.

Abstract

In this work, Machine Learning (ML) techniques have been employed to explore the highest performing single-heteronunction thermophotovoltaic cell. Initially, traditional homo junction TPV cells have been explored using ML methodologies for the optimal material combinations. ML methods have notably been devoted to analyze the importance of each parameter in the model, thereby improving the comprehension of the system's behavior and facilitating design optimization. Following this investigation, it has been found that Ge emerged as the most effective emitter layer when paired with the optimal base layer, InGaAsSb compound that possesses a direct bandgap of 0.53 eV. Subsequently, a p-Ge/n-InGaAsSb single-heterojunction TPV cell is introduced executing a device transport model featuring a p-n structure. This cell operated at black body (TBB) and cell temperatures of 1578 K and 300 K, respectively. Through meticulous optimization efforts, the performance of the TPV cell is significantly enhanced resulting in an impressive efficiency of 16.50%. This efficiency is accompanied by a short circuit current, JSC=15.53 A/cm2, an open-circuit voltage, VOC=0.47 V, and a fill factor FF=79.5%. These findings suggest that this structural configuration holds considerable promise for the development of high-performance TPV cells.