(545b) Optimization of a Carbon Dioxide-Assisted Nanoparticle Deposition Process Using Sequential Experimental Design with Adaptive Design Space

Optimization
of a carbon dioxide-assisted nanoparticle deposition process using sequential experimental
design with adaptive design space

Michael
J. Casciato*^,?, Sungil Kim^?, J.C. Lu^?, Dennis
W. Hess^?, and Martha A. Grover^?

^?School of Chemical & Biomolecular
Engineering, Georgia Institute of Technology, Atlanta, GA 30332

^?H. Stewart Milton School of Industrial
& Systems Engineering, Georgia Institute of

Technology,
Atlanta, GA 30332

*E-mail:
michael.casciato@chbe.gatech.edu

Abstract

A sequential design of experiments
methodology with an adaptive design space is proposed and implemented to
optimize a nanofabrication technique that possesses significant uncertainty in
terms of design region and model structure along with an engineering tolerance
requirement. This method is termed Layers of Experiment (LoE) with Adaptive
Combined Design (ACD) and is developed specifically for advanced technology
processes where there is little or no foreknowledge of the design region and/or
model structure from either mechanistic understanding or empirical studies.
This technique was applied to optimize an elevated pressure, elevated temperature
carbon dioxide (epet-CO₂) nanoparticle deposition process where
silver nanoparticles were deposited directly from an organometallic precursor
in the fluid phase onto a silicon wafer surface.

A target mean nanoparticle size of 40nm
was chosen, with surface enhanced Raman spectroscopy (SERS) as the motivating
application. Using the LoE/ACD method, it was possible to find the process
optimum for the epet-CO₂ process and build a reliable model using
only 12 experiments conducted in two sequential layers. With temperature as the
design variable, the first layer of experiments was conducted in the region [60°C, 150°C], while the
second layer was conducted in the region [98°C, 128°C], indicating that the LoE/ACD algorithm
reduced the size of the design region by 60°C while building a reliable model for statistical
inference. In the first layer, a purely space-filling design was used since the
model structure was unknown; in the second layer, the ACD approach yielded a
design weighted between space-filling and D-optimal, favoring the D-optimal
design. The optimized temperature for fabricating 40nm mean silver
nanoparticles in this system was 117°C.