Automated design of nonreciprocal thermal emitters via Bayesian optimization

TL;DR

The paper tackles designing broadband nonreciprocal thermal emitters that break Kirchhoff's law by optimizing multilayer structures built from InAs and magnetic Weyl semimetals. It introduces a general workflow that couples Bayesian optimization with a reparameterization scheme to reduce design dimensionality while enforcing practical constraints. The approach discovers high-performance emitters with far fewer layers than state-of-the-art designs, achieving broadband nonreciprocity from 5 to 40 μm and demonstrating significant gains when combining InAs with Weyl semimetals. The findings highlight the power of combining numerical optimization with material heterostructures for practical nonreciprocal radiative devices and point toward experimental validation and multi-fidelity extensions.

Abstract

Nonreciprocal thermal emitters that break Kirchhoff's law of thermal radiation promise exciting applications for thermal and energy applications. The design of the bandwidth and angular range of the nonreciprocal effect, which directly affects the performance of nonreciprocal emitters, typically relies on physical intuition. In this study, we present a general numerical approach to maximize the nonreciprocal effect. We choose doped magneto-optic materials and magnetic Weyl semimetal materials as model materials and focus on pattern-free multilayer structures. The optimization randomly starts from a less effective structure and incrementally improves the broadband nonreciprocity through the combination of Bayesian optimization and reparameterization. Optimization results show that the proposed approach can discover structures that can achieve broadband nonreciprocal emission at wavelengths from 5 to 40 micrometers using only a fewer layers, significantly outperforming current state-of-the-art designs based on intuition in terms of both performance and simplicity.

Figures (12)

• Figure 1: (a) Illustration of radiative heat exchange between two blackbodies (A and B) and an emitter (E) under thermal equilibrium. (b) Illustration of a multilayer structure consisting of InAs layers on top of Weyl semimetal layers on a reflective substrate.
• Figure 2: Design parameters ${\bf p}$ for a nonreciprocal emitter of interest and their reparameterization into ${\bf x}$. The structure consists of $n_1$ InAs layers with carrier concentrations ${\bf N}$ and thickness $t_1$, and $n_2$ Weyl semimetal layers with Fermi levels ${\bf E}$ and thickness $t_2$. After reparameterizing ${\bf N}$ and ${\bf E}$, the number of parameters reduces from $n_1+n_2+2$ to $p_1+p_2+2$, where $p_1 \le n_1$ and $p_2 \le n_2$ are the numbers of columns of ${\bf D}$ and ${\bf F}$, respectively.
• Figure 3: Schematic illustration of three consecutive iterations of BO for optimizing a univariate function $f(x)$. In each iteration, BO constructs a probabilistic model for the objective function $f(x)$ using the current data, formulates an acquisition function $q(x)$ from the model, and maximizes $q(x)$ to identify a new location to query the objective function.
• Figure 4: Optimization results for 3-layer InAs structure in comparison with the state-of-the-art 10-layer InAs structure (reference) LiuM2023. (a) Optimization histories from LCB and PI. (b) Comparison of contrast values of the initial and final structures from the first and last iterations of the first LCB trial, and the state-of-the-art 10-layer InAs structure.
• Figure 5: Absorptivity ($\alpha$) and emissivity ($\varepsilon$) of the initial and final 3-layer InAs structures from the first and last iterations of the first LCB trial, and the state-of-the-art 10-layer InAs structure (reference) LiuM2023.