Speaker
Description
After the seminal discovery of two-dimensional (2D) topological insulators (TIs) in CdTe/HgTe single-layer quantum wells (QWs), these fascinating systems have emerged as promising candidates for various devices in topological electronics and spintronics. One of the key limiting factors for practical applications of TIs is a small band gap, which results in low operating temperatures, usually not exceeding 15 K for CdTe/HgTe QWs. Hence, the search for application-suitable alternatives is in high demand.
One possible direction is provided by InAs/Ga_{1−x}In_{x}Sb bilayer QWs. Perhaps the most promising results have recently been obtained in InAs/Ga_{0.65}In_{0.35}Sb/InAs triple QWs, where stable helical edge transport has been observed at temperatures up to 60 K. Importantly, the geometry of 2D TIs operating at elevated temperatures is becoming increasingly complex. As a result, the design space – material composition, layer thicknesses, etc. - becomes high-dimensional and strongly nonlinear. This makes conventional trial-and-error approaches impractical and motivates the need for automated discovery strategies.
In this talk, we will discuss how two modern automated approaches - Evolutionary Algorithms and Machine Learning (specifically Reinforcement Learning) - can be applied to the search for advanced topological insulators based on quantum well heterostructures. We will outline the advantages and disadvantages of these strategies. Focusing on the example of a triple quantum well, we will demonstrate how the combination of advanced numerical tools with Evolutionary Algorithms can accelerate the discovery of optimized structures for advanced TIs.