(140d) Multiscale Simulation and Optimization of An Atomic Layer Deposition Process in a Nanoporous Material
AIChE Annual Meeting
2009
2009 Annual Meeting
Computing and Systems Technology Division
Control and Optimization of Smart and Multi-Scale Plants
Monday, November 9, 2009 - 4:15pm to 4:35pm
Atomic Layer Deposition (ALD) is a thin film deposition process in which the growth surface is exposed to reactive precursor gases in an alternating fashion. A feature of the surface adsorption and reaction mechanisms is that they are normally self limiting, allowing for atomically accurate control of film thickness and uniform deposition over complex surface topographies.
ALD is an inherently dynamic process characterized by multiple time scales: the faster corresponding to molecular events taking place during each exposure cycle and the slower to nucleation and steady growth. Likewise, multiple length scales are found in these systems where the macroscopic length corresponds to gas phase transport effects, and the microscopic to the atomistic nature of the film growth. An approach to coupling modeling elements across these scales to simulate ALD growth of alumina films inside nanopores of high aspect ratio is the topic of this paper. We consider Al2O3 ALD, one of the most widely studied ALD systems [1,2], grown from alternate exposures of the growth surface to TMA (Al(CH3)3) and water.
Microscopic Lattice MC ALD simulation
In our ALD reaction simulation approach, we focus on a microscopic region of the growth surface and discretize this region into a two dimensional grid of lattice sites, one dimension representing spatial position along the growth surface (with periodic boundary conditions), and the other the depth of the deposited film. This coarse-graining approach allows us to investigate amorphous film growth at the molecular scale and keeps the problem computationally tractable by limiting the number of configurations the film and reacting species can take. The potential reactions growth surface groups can undergo with the two precursors are enumerated, and transition probabilities dictating surface reaction rates are chosen so that our simulation results are consistent with the experimentally observed reaction rates of [1]. This kinetic Monte Carlo modeling strategy results in a microscopic simulator that predicts Al2O3 ALD growth per cycle (GPC) values consistent with measured rates [2].
Pore-level modeling
Nanostructured membranes have numerous chemical reaction, storage, and separation applications. They can be created using an anodic aluminum oxide scaffold, with pore sizes and surface properties subsequently modified to high precision using ALD processes. We now focus on developing a multiscale (continuum transport at the pore scale coupled to our lattice Monte Carlo simulation of the film growth process at the atomistic scale) simulator of the ALD process taking place within these nanoporous materials.
We consider a description of gas phase transport and surface reaction of the TMA and water precursors within a 250μm by 400nm nanopore, where gas phase transport of each precursor is described by Knudsen diffusion. Because ALD is a cyclic process, one simplification we use is to average each precursor material balance over its exposure (half) cycle, giving a local precursor dosage as a function of spatial position along the pore. The resulting boundary value problems, subject to a specified exposure level at each end of the pore, are discretized using orthogonal collocation. The dose-averaged rate terms defining spatially localized precursor consumption must be evaluated at each collocation point using the lattice Monte Carlo simulators, resulting in a numerical problem combining a pore-scale continuum description of transport in the nanopore and a sequence of microscopic simulator elements for film growth. The pore transport and reaction models are solved over each exposure half cycle, and the pore film thickness profile is updated after each full cycle.
Results
When each end of the pore is exposed to the same precursor dosages, the TMA and water precursors diffuse into the open ends of the nanopore during each exposure step in a symmetric manner, and a portion is adsorbed onto the pore walls. Early in the deposition process, we observe only a small amount of precursor depletion in the innermost regions of the pore. However, the depletion effects grow with increasing number of deposition cycles, leading to preferential deposition near the pore mouths, further accelerating the development of the cycle-averaged precursor partial pressure gradients. The simulations were performed for a total of 200 ALD cycles, after which the pore mouths essentially close, preventing further deposition within the pore. This example illustrates the potential for our numerical methods to capture the spatially and time varying reaction and transport phenomena of the ALD process.
It is interesting to compare the previous case to simulations correspond to the asymmetric boundary condition case. Exposing the nanomembrane to the precursors in this fashion results in a deposition profile that grows from the center outwards, illustrating the potential for controlling the uniformity by manipulating the pore mouth boundary conditions. It is these novel processing modes that will be the focus of this paper, together with interpretation of the simulation results against experimental observations.
[1] Dillon, A. C., A. W. Ott, J. D. Way, and S. M. George, Surf. Sci. 322 (1995) 230-242.
[2] Puurunen, R. L., Appl. Phys. Rev. 97 121301 (2005).