Cryogenic plasma etching is a leading strategy to mitigate aspect-ratio-dependent etching (ARDE) in high-aspect-ratio (HAR) semiconductor fabrication, yet the underlying neutral-transport mechanisms remain debated. While conventional models incorporate surface diffusion as a critical transport channel, its efficacy at cryogenic temperatures is quantitatively uncertain. In this work, we employ machine-learning molecular dynamics using the SevenNet potential to calculate a comprehensive transport-parameter dataset—including desorption barriers and diffusion coefficients—for 16 neutral species on cryogenic ammonium fluorosilicate passivation layers. Our analysis reveals a mechanistic regime shift: below 250 K, lateral surface diffusion becomes kinetically frozen for most species, causing total transport to collapse into a purely Knudsen-limited regime. Consequently, the mitigation of ARDE is not driven by diffusion-assisted flux recovery but by the exponential scaling of surface residence time. We demonstrate that interhalogen species possess optimal desorption energies (0.70–0.85 eV) that balance deep-feature penetration with sufficient reaction probability, whereas hydrides desorb too rapidly and metal halides cause entrance clogging. These findings refine the physical understanding of the “total-transport” model and establish desorption-controlled residence time as the governing descriptor for optimizing cryogenic HAR processes.
Um et al. (2026) studied this question.