Coupling InP QD synthesis with Self-Driving Fluidic Labs (SDFLs)–automated experimental platforms which utilize machine learning (ML) algorithms and closed-loop workflows to autonomously explore parameter spaces–shows great promise for advancing QD manufacture. In this work, we leverage the experimental method known as Dynamic Flow Experiments (DFEs), in which input process parameters (e.g., flow rates and concentrations) are continuously ramped throughout the duration of an experiment while taking careful consideration of the process output parameters (e.g., optical spectral data). Compared with traditional Steady State Flow Experiments (SSFEs), DFEs enable data collection at rates up to an order of magnitude higher, by mapping transient data (otherwise discarded in SSFEs) to steady state equivalent conditions. This high data throughput establishes a pathway toward the long-term, sustainable manufacture of heavy-metal free QDs.
The InP SDFL platform integrates computer-control of process parameters, as well as fully-automated, in-situ characterization of QDs via an in-line UV-Vis spectrometer. Custom LabVIEW programs are used to control inputs such as temperature and flow rate, while custom-built, Python workflows streamline data acquisition, processing, and ML model updating. Complete automation of this material synthesis process enables consistent identification of optimal experimental conditions, yielding the best in-flow InP QDs for targeted applications (first excitonic peak-to-valley ratio = 1.65).
This closed-loop process begins with experimental testing of user-provided input parameters. Datasets from these initialization experiments are processed through the autonomous Python workflows. Latin Hypercube Sampling (LHS) is then employed to optimize sampling distribution, followed by a series of Bayesian Optimization (BO) experiments, aimed at optimizing desired optical parameters such as peak-to-valley (PV) ratio. PV ratio is a critical indicator of the success of a synthesis experiment in producing uniform, low-defect InP QDs, however, the desired model output can be customized to match user-defined performance criteria. The resulting input parameters are fed back into the closed-loop workflow, and this process is iterated until the experimental budget has been exhausted, yielding optimal synthesis conditions.
This robust experimental platform offers significant capability for industrial applications, potentially providing manufacturers the ability to precisely tune advanced functional materials to their desired specifications. SDFL’s can facilitate the efficient transition from unknown parameter spaces to precise manufacturing conditions, while substantially reducing material budgets and experimental timelines – on the order of days and weeks, rather than months and years.