Fig. 3: cALD sequence selection campaign results. | Nature Communications

Fig. 3: cALD sequence selection campaign results.

From: AlphaFlow: autonomous discovery and optimization of multi-step chemistry using a self-driven fluidic lab guided by reinforcement learning

Fig. 3

a Illustration of all injection sequences in the exploration runs with the exploited sequence highlighted in red. b First absorption peak wavelength of CdSe/CdS hetero-nanostructures as a function of injection number for the RL-selected cALD chemistry and the manually input conventional injection sequence. c UV-Vis photoluminescence and absorption (d) spectra after each full cALD cycle for the RL-discovered and conventional cALD sequences. e Schematic of reagent addition sequences for conventional cALD cycles and the AlphaFlow-selected sequence. f Frequency histograms of the forward predicted reward for the four reagent injection options in the cALD chemistry exploration campaigns. The red arrow indicates the path taken when exploiting the agent. Upstream injections assume prior injections followed the exploited path. The learning agent was trained on the full data set of the cALD chemistry exploration campaign.

Back to article page