Department Personnel
Associate Professor Antonios Armaou | Research
Multiscale optimization
Many existing and emerging chemical processes are distributed in space, including plug-flow, packed-bed reactors and Metal-Organic Vapor Phase Epitaxy. In the last decade, due to the advent of ever increasing computational power, research intensified on the design of model-based controllers that account for the fact that basic process variables vary with both time and space. The underlying assumption in the specific controller designs is that a model, albeit a non perfect one, does exist.
There are a host of problems however, where the yield of the process entails the definition of metrics of the resulting products' "quality". In such cases, the formulation of the control objective is often a very involving operation and requires knowledge of the product microstructure, which requires a finer accuracy than most process models provide. Statistical simulations (kinetic Monte Carlo, Lattice Boltzmann) provide an estimate of the expected process behavior at the cost of increased computational requirements. A further complication arises from the lack (in principle) of real-time measurements of the control objective; in most applications where finer spatial-scale properties need be measured, probes operate at discrete time and moreover there is a time delay associated with the processing of the raw data.
To put the research problem presented in context let us consider a showerhead Plasma enhanced chemical vapor deposition reactor, depicted in Figure 1.
Figure 1: Plasma enhanced chemical vapor deposition reactor with showerhead arrangement.
In the insert surface topology features are named.
(Select the image to enlarge)
As the reactant gas enters the chamber through the showerhead, plasma reactions take place leading to the formation of radicals in the bulk of the plasma region; the produced radicals transport towards the wafer and adsorb to its surface, forming a new layer of material on the substrate wafer. One important process objective is minimizing the roughness of the deposited film. Once adsorbed, surface reaction and transport mechanisms determine the surface structure; the structure in turn defines the local roughness of the deposited material.
Dr. Armaou's research focuses on optimal operation of chemical processes when the process objective requires quantitative descriptions at a finer accuracy than the available continuum-level process models can capture. The lack of practically implementable models in the finer scale is addressed by combining coarse-graining methods with model reduction techniques to derive multiscale process models that are able to estimate the dynamic behavior of all the process variables and at the same time are computationally tractable. The derived models are subsequently used to formulate optimization problems that can be solved using standard search algorithms. The proposed theoretical research is applied to specific industrially important microelectronics fabrication processes formulating the associated process operation problems and subsequently applying the developed optimization methodologies.
For example, consider metal-organic vapor phase epitaxy, used to deposit composite thin-film heterostructures on wafer surfaces for the production of optoelectronic devices. We are focusing on identifying operating conditions that enforce radial uniformity of the deposition rate across the wafer surface, as well as enforcing composition uniformity, minimizing impurity incorporation in the film. The need for efficient control of the film structure at the molecular level also becomes necessary during the change of deposition material for the growth of the heterostructures, where the abrupt change of material at the atomic interface is critical for the resulting device performance.