When is Bayesian Optimization Beneficial? A Critical Assessment of Optimization Strategies in High-Throughput Organic Photovoltaic Manufacturing

Matthew Osvaldo, Leonard Ng Wei Tat

Track: New scientific result

Keywords: Organic photovoltaics, Bayesian optimization, High-throughput manufacturing, Self-Driving Labs

TL;DR: Environmental control trumps AI optimization in solar cell manufacturing, challenging conventional wisdom about how to improve production efficiency.

Abstract: We present a systematic evaluation of optimization strategies for high-throughput organic photovoltaic (OPV) manufacturing. Analyzing 11,587 PBF-QxF:Y6 devices across 11 manufacturing parameters through 25 optimization iterations, we compared Bayesian Optimization (BO) and Random Search (RS). While BO achieved 7.69% PCE versus RS's 7.66%, this 0.03% advantage required 20x more computational overhead. Statistical analysis revealed no significant performance difference between methods (t-stat = 0.53, p > 0.05). Environmental factors, particularly humidity (r = 0.380), showed stronger correlation with performance than optimization strategy choice. Manufacturing process control, rather than algorithmic sophistication, emerges as the critical factor for high-throughput OPV optimization. These findings suggest prioritizing robust process control systems over complex optimization algorithms in manufacturing environments.

Supplementary: https://github.com/DuelKings/BO-vs-Random-on-High-Throughput

Presenter: ~Matthew_Osvaldo1

Submission Number: 10