Developing chemical plasma processes for semiconductor chip fabrication is costly and time-consuming. This paper investigates the use of Bayesian optimization algorithms to reduce this cost. A virtual process game was created to benchmark human engineers and computer algorithms. Results show that human engineers excel in early stages, while algorithms are more efficient near the target tolerances. A hybrid "human first-computer last" (HF-CL) strategy significantly reduced cost-to-target. The study also highlights cultural challenges in human-computer collaboration.

Keren J. Kanarik, Wojciech T. Osowiecki, Yu (Joe) Lu, Dipongkar Talukder, Niklas Roschewsky, Sae Na Park, Mattan Kamon, David M. Fried, Richard A. Gottscho