To address the challenge of accurately predicting the etched structures and surface morphology during the wet etching process of indium phosphide (InP), this paper employs the surface activation energy theory to elucidate the mechanism underlying the anisotropy observed in InP etching from a microscopic atomic energy perspective: different crystal facets exhibit distinct activation energies due to variations in atomic arrangement and chemical bonding structures, thereby giving rise to anisotropic etching rates. Building upon this foundation, by systematically analyzing the atomic structural characteristics of different crystal facets, this study proposes three substrate modeling strategies: the silicon-like model, which ignores the difference between In and P atoms and determines the removal probability solely based on the coordination type; the equivalent effect model, which distinguishes between the differential effects of heterogeneous and homogeneous atom pairs on etching activity but assumes that In and P atoms with identical bonding structures exhibit equal activity; and the original model, which fully retains the differences between In and P atoms, assigns separate removal probability parameters corresponding to each coordination type, and incorporates the influence of atomic size differences on stability, thereby fully capturing the structural characteristics of the InP binary atomic system. Based on these models, this paper further develops an atomic removal probability determination function (InP-RPF), which links the macroscopic etching rate of a crystal facet to the microscopic atomic removal probabilities.
Subsequently, this paper integrates evolutionary algorithm (EA) with the Monte Carlo method (MC) to construct an etching simulation model for InP-EMC. The model simulates four crystal planes: (100), (110), (111), and (211), using the minimization of the total etching rate error as its fitness function. Through population-based evolution, the energy parameters within the InP-RPF are automatically optimized to bring the simulation results closer to experimental values. The evolutionary process employs binary encoding, roulette wheel selection, single-point crossover (with a probability of 0.5), and basic bit mutation (with a probability of 0.001), along with an optimal individual retention strategy to prevent the degradation of superior genetic traits. The termination criterion is set as the ratio of the total simulation error to the total experimental rate being ≤ 6%; upon meeting this condition, the optimized energy parameters and the simulated etching rates for the target crystal planes are automatically output. A comparison of the simulation accuracy among the three models under etching conditions at 25 °C with HBr:CH₃COOH = 1:1 reveals that the original model not only accurately reflects the structural characteristics of the InP binary atomic system but also yields etching rates and morphological simulation results that closely match experimental data. Regarding etching rates, the relative errors for typical crystal planes—(100), (110), (111), (211), (221), (411), and (310)—are all less than 8%. For the three-dimensional morphology simulation, the original model successfully reproduces the experimental sidewall crystal plane configurations (e.g., 111, 110) and corner chamfering features observed on square concave/convex films and rhombic concave films under different etching solution systems. The half-sphere etching simulation further elucidated the distribution pattern where the 100 plane exhibits the fastest etching rate (approximately 8.5 μm/min), while the 111 plane shows the slowest rate (approximately 2.3 μm/min); moreover, the difference in etching rates between the front and back surfaces of the (111) plane arises from the distinct nature of the In and P atoms. Additionally, the simulation results obtained under both HCl:HNO₃ = 2:1 and pure HBr etching conditions align well with experimental data, validating the model's excellent universality.
In summary, the InP-EMC simulation model developed in this study can serve as a virtual process verification platform, providing predictions for sidewall morphology and etching progress for typical device fabrication processes—such as via etching, mesa-type APD isolation, and ridge waveguide fabrication—prior to actual tape-out, thereby effectively shortening the process development cycle and reducing material costs.