Fig. 4: Volume and time optimization campaign results of AlphaFlow.

a First absorption peak wavelength and peak-to-valley ratio (b) as a function of the full cALD cycle number for the volume and time exploration and exploitation and the cALD chemistry exploitation. c Absorption and photoluminescence spectra for the complete cALD cycles of the volume and time exploitation runs on each of the three starting CdSe nanoparticle sizes studied. d Output parameter space for the exploration and exploitation of the three CdSe nanoparticle sizes. e Average predicted reward for a single step in the volume and time optimization campaign as a function of the injection volume and reaction time of the first injection (OAm). Note that surface plot colors are a topographic guide to the eye. Injecting OAm has little immediate influence on the measured reward, but forward predicting ahead shows that the decision significantly affects downstream reward. The RL agent was trained on the full data set for the 480 nm CdSe nanoparticle volume and time optimization campaign.