Nonvolatile Tunable Wavelength‐Selective Emitter with Phase‐Changing Material Ge₂Sb₂Te₅ Designed by Bayesian Optimization Method

Abstract

The ability to tune the emission wavelength of infrared emitter while maintaining wavelength‐selectivity remains a challenge. By incorporating the nonvolatile phase‐changing material Ge2Sb2Te5 (GST), a nonvolatile tunable wavelength‐selective emitter (gold (Au)‐GST‐Au structure) based on Fabry–Perot (FP) resonance theory is demonstrated. An algorithm based on Bayesian optimization (BO) and transfer matrix method (TMM) for optimal design is proposed. First, the influence of amorphous proportion on tunability is investigated. It is found that the peak emission wavelength shifts from 5.3 to 3.8 μm and peak emissivity increases from 0.75 to 0.98 with an increase of the amorphous proportion. Second, the influence of the thickness of the top and middle layer on the emission spectrum is studied. It is illustrated that the thickness of each layer has a significant impact on the emission spectrum. Finally, an objective function considering wavelength‐selectivity in atmospheric and nonatmospheric windows as well as tunability is proposed. The optimal structure can be realized within calculations for less than 0.027% of total candidate structures. When the weight w = 0, the optimal structural parameters are dAu = 0.0018 μm and dGST = 0.4318 μm; and the effect of the weight factor in objective function is also investigated.

Machine Learning Aided Design and Optimization of Thermal Metamaterials

“…Remarkably, the algorithm achieved a significantly elevated evaluation factor of 3.98 × 10 5 W/m 2 by examining a mere 2.6% of the samples encompassing a vast pool of 1.0 × 10 8 W/m 2 candidate structures drawn from the database. By incorporating the nonvolatile phase-changing material Ge 2 Sb 2 Te 5 (GST) for designing the wavelength-selective emitter (Au-GST-Au configuration), Song et al hybridized Bayesian optimization with the transfer matrix method, thereby orchestrating high emissivity within the 5–8 μm wavebands and concomitant low emissivity within the 3–5 μm and 8–14 μm regimes. Impressively, this algorithm discerned the optimal structure by evaluating less than 0.027% of the total candidates, with remarkable peak emissivity of 0.95 at a wavelength of 5–78 μm. …”