Plasma-thermal atomic layer etching (ALE) is an emerging direct metal etch (DME) method that can potentially enable integration of many metals previously considered impossible to pattern. By combining a low-temperature plasma activation with thermal removal, metal layers can be patterned directionally and selectively. However, little is understood about the surface chemistry that leads to directionality and selectivity. In this dissertation, a suite of modeling tools is developed, ranging from thermodynamic screening of potential process chemistries, to development of accurate interatomic potential models, large-scale atomistic simulation of plasma processes, and finally the analysis of trajectories. The thermodynamic screening model is applied to combinations of oxygen/nitrogen plasma activation on nickel/copper surfaces, using formic acid/formamidinate as the etchant. In total 8 processes were screened computationally, predicting a nitrogen-plasma based process on nickel metal etch will yield similar etching characteristics as previously demonstrated oxygen-plasma based process. On all combinations of modifier/substrates it is predicted that higher surface coverage generally leads to more favorable etching, and there exists a critical coverage below which etching is unfavorable thermodynamically. It is found that inserting ions into the subsurface sites (possible through the impact ion energy) makes etching highly favorable. This protocol can be readily extended to other combinations of metals/modifiers/etchants to allow for a rapid screening of etching chemistries. The demonstrated complex site-dependence of the etching energetics is accounted for explicitly in molecular dynamics simulations enabled by a machine learning interatomic potential for copper and oxygen trained on \textit{ab-initio} calculation data. A large scale simulation protocol for atomisitc plasma-surface interaction simulation is developed, and used to obtain atomically-resolved trajectory of copper oxidation. The simulation results show that for a low-energy plasma (kinetic energy < 20 eV), the ions do not penetrate into the substrate lattice. The oxidation process in the bulk of the film is still diffusion limited. The effect of ion energy lies in delivering additional, depth-dependent thermal energy that promotes diffusion within the oxide film. It is confirmed that at 80C the oxidation is not self-limiting. The chemical identity of the oxide film is determined to be mainly CuO. The crystalinity is further studied in a separate set of simulations that increased ion flux 4-fold, effective accelerating simulation. Repeated, prolonged ion-impact on the already-oxidized film (2 nm thick oxide) leads a layer of crystalline CuO beneath an atomically-thin, but rough top layer of amorphous CuO. This is not observed in other non-accelerated simulations, which always gave amorphous CuO structures. Presumably this is due to crystallization being a slow process dominated by rare events. By investigating the effect of process-relevant parameters (temperature, ion-to-neutral ratio, and ion kinetic energy), it is found that self-limited growth may be possible by lowering the subtrate temperature. In such conditions, the limiting thickness is controllable by tuning the ion energy distribution function (IEDF) in the plasma. This is due to the limited range of energy delivery through collision cascade in the oxide film.

A framework to study the thermodynamics of chemistries of the etching step in ALE processes is presented. The effects of two modifiers(O & N), two substrates(Cu & Ni) and two etchants(formic acid and formamidine) on etching energy are studied with two models. The bulk model uses the bulk formation energy to provide a quick estimate of the etching energy, while the layer model incorporates effects of surface stability and adsorbate configurations. The effect of etchants is independent of models used. Etching by formic acid is favored by 0.45 eV on Ni but slightly disfavored by 0.07 eV on Cu. The two models produce the same qualitative results: on the bare metal, all modifier/etchant pair is unfavorable; on activated substrates, etching is unfavorable for Ni but favorable for Cu. The bulk model shows that the effect of modifiers is insignificant on Cu, but on Ni the O activation is favored by 0.20 eV. With the layer model, a difference of ∼ 0.3 eV favoring the N activation is observed for Cu substrate. On Ni substrate this effect depends on coverage. Below ∼ 0.07 A^-2 , N is the favored modifier. The reverse is true above ∼ 0.07 A−2 . The layer model is applied to metastable structures on Ni/O/FAmd system formed by placing O atoms in sublayer interstitial sites, resulting in a favorable etching energy consistent with the experiments. The validity of the layer model is established, but the structures fed into the model must be carefully selected to match with the actual substrate.