Rapid Quantification of Threading Dislocation Networks in Topologically Insulating Bi 1− x Sb x Thin Films via Deep‐Learning Analysis of Etch Pits

BiSb alloys are of prime interest because of their intriguing thermoelectric properties, as well as being attractive topological insulator materials for quantum and spintronic devices. Their properties have mostly been studied by transfer of flakes or by studying bulk materials, while epitaxial integration with industrial substrates remains challenging. Mismatches in crystalline structures and lattice parameters lead to the formation of dislocations at the substrate interface, which then propagate through the heterostructure. The effects of these dislocations on the functional properties of BiSb thin films have not been explored experimentally. A wet‐etching method with dilute HCl followed by scanning electron microscopy observation is presented here, which greatly simplifies and accelerates the quantification of the surface threading dislocations density, thus allowing for correlating crystal defects to functional properties. An off‐the‐shelf instance segmentation neural network (Mask‐R‐CNN) is fine‐tuned to rapidly detect and quantify the etch pits with a recall of 97% and an accuracy of 79%. (Scanning) transmission electron microscopy observations confirm that etch pits correspond to threading dislocations, and allow visualization of the (a/3) <2‐1‐10> lattice distortion. A platform is thus presented to study the effects of the dislocations’ density on the functional properties of BiSb thin films with a simple technique.