Source-linked AI summary
Beyond Ternary OPV: High-Throughput Experimentation and Self-Driving Laboratories Optimize Multi-Component Systems
Stefan Langner, Florian Häse, José Darío Perea, Tobias Stubhan, Jens Hauch, Loïc M. Roch, Thomas Heumueller, Alán Aspuru-Guzik, Christoph J. Brabec
TL;DR
OPV optimization increasingly requires exploring multi-component blends, but conventional ternary studies often vary only the additive while holding the host donor–acceptor ratio constant. The paper develops robotic high-throughput experimentation and Bayesian self-driving optimization for quaternary blends, screening broad composition spaces and identifying photostable compositions with substantially fewer samples and less material.
Problem
Ternary OPV studies commonly hold the host donor–acceptor ratio constant and vary only additive content because experimental resources limit broader composition searches, while device stability remains a central challenge.
Method
The study combines robotic ink formulation and drop-cast film formation with automated characterization and ChemOS-based Bayesian optimization for closed-loop quaternary-blend experiments.
Results
Over 2000 quaternary active layers were tested in 7 days with less than 15 mg per component, while self-driving optimization achieved about 93% sample reduction for equivalent stability information.
Takeaways & Limitations
P3HT- and PBQ-QF-rich blends showed improved stability over PTB7-Th-rich blends, whereas PCBM and oIDTBR could destabilize each other dramatically.
Abstract
from arXiv · showhide
Fundamental advances to increase the efficiency as well as stability of organic photovoltaics (OPVs) are achieved by designing ternary blends which represents a clear trend towards multi-component active layer blends. We report the development of high-throughput and autonomous experimentation methods for the effective optimization of multi-component polymer blends for OPVs. A method for automated film formation enabling the fabrication of up to 6048 films per day is introduced. Equipping this automated experimentation platform with a Bayesian optimization, a self-driving laboratory is constructed that autonomously evaluates measurements to design and execute the next experiments. To demonstrate the potential of these methods, a four-dimensional parameter space of quaternary OPV blends is mapped and optimized for photo-stability. While with conventional approaches roughly 100 mg of material would be necessary, the robot based platform can screen 2,000 combinations with less than 10 mg and machine learning enabled autonomous experimentation identifies the stable compositions with less than 1 mg.
Introduction
The paper combines automated high-throughput film formation with Bayesian optimization to explore and optimize photo-stability across quaternary OPV blends. This approach maps complex composition spaces while reducing material use and experimental burden.
- High-throughput experimentation: 6,048 films per day can be fabricated by robot-based drop-casting, enabling scalable screening of multi-component OPV compositions.The method uses 0.6 mg/mL inks and 96-well glass substrates.
- High-throughput experimentation: The automated platform combines robotic ink formulation, drop-casting, annealing, absorbance characterization, and 18-hour light-treatment degradation tests.It investigates two quaternary systems containing P3HT, PCBM, oIDTBr, and either PTB7-Th or PBQ-QF.
- Photo-stability mapping: 68% relative absorbance loss for PTB7-Th contrasts with around 19% for P3HT, while the two acceptors show minimal absorbance change.Degradation is quantified from the integral change in absorption spectra before and after illumination.
- Photo-stability mapping: ~10 wt.-% P3HT stabilizes PTB7-Th completely, whereas blends containing both oIDTBR and PCBM form a drastically destabilized region.The PCBM:oIDTBR mixture reaches absorbance loss up to 74%, which the authors associate with a possible morphology-mediated oxygen interaction.
- Self-driving laboratory: Bayesian optimization reconstructed the grid-HTE stability distribution with only 7% as many samples and identified comparable photostability with an average of 27 samples in virtual experiments.Within budgets of 30 and 60 samples, SDA exceeded HTE with chances of about 96% and 97.5%, respectively.
- Conclusion and outlook: Over 2000 quaternary active layers were tested in 7 days using less than 15 mg per component, while machine learning enabled about 93% sample reduction.The conclusion reports improved stability for P3HT- and PBQ-QF-rich blends relative to PTB7-Th-rich blends.
S.1 Experiment planning with robotics constraints
The experiment-planning workflow adapts ChemOS suggestions to the finite set of compositions that the dispensing robot can distinguish. This accounts for dispensing accuracy and prevents nominally different suggestions from being treated as fully distinct experiments.
- Experiment planning with robotics constraints: The dispensing robot has approximately 1 µL accuracy, so realized blend compositions can differ from ChemOS suggestions.This can cause two requested blends to have almost identical compositions and repeat a previous experiment.
- Experiment planning with robotics constraints: An added ChemOS module transforms continuous planner parameters into the grid of distinguishable experiments realizable by the automated platform.The module integrates with ChemOS without interfering with its existing modules.
S.2 Results of high-throughput experimentation and self-driving laboratory
The study compares high-throughput experimentation and self-driving approaches for mapping and optimizing multicomponent OPV blend stability. Supplementary figures document the automated platform, film reproducibility, material and blend stability, and optimization comparisons.
- Automated platform: The semi-automatic robot system integrates pipetting, spectroscopic characterization, heating, stock-solution handling, microplate vessels, waste collection, and heat sealing.
- Film and material stability: Fresh films are compared with aged films after 18 hours of continuous illumination to assess photo-stability.
- Film and material stability: The supplementary stability figures cover the constituent materials and binary, ternary, and quaternary mixtures in the studied blend systems.
- Optimization comparison: Table S1 lists the 10 best compositions found by the self-driving approach and grid-HTE.
- Optimization comparison: Figure S7 compares grid-based high-throughput experimentation with ChemOS optimization using distance to the best found HTE composition.
- Optimization comparison: The tetrahedral plot identifies most stable compositions in blue and most unstable compositions in green or red across HTE and SDA datasets.
- Optimization comparison: Virtual-robot traces compare HTE with ChemOS/Phoenics and quantify SDA acceleration and confidence of improvement within 30- or 60-sample budgets.
S.3 Construction of virtual robots
The authors construct Bayesian-neural-network virtual robots to emulate blend photo-degradation and assess optimization behavior in silico. The models reproduce measured test-set behavior for both studied blend systems with substantial transferability.
- Virtual-robot construction: Virtual robots use probabilistic machine-learning models to emulate experimental procedures and estimate the merit of new experiments in silico.
- Virtual-robot construction: Bayesian neural networks predict expected photo-degradation while also inferring experimental noise and reducing overfitting through a Bayesian framework.
- Training and evaluation: The networks were trained using 850 of 1,041 HTE experiments, with 190 experiments reserved as a test set.
- Training and evaluation: The blend ratios were transformed from the 4-simplex to a three-dimensional unit cube, while photo-degradations were normalized by the cross-validation-set average.
- Training and evaluation: Coefficients of determination were 0.88 for PBQ-QF and 0.87 for PTB7-Th, with similar test and training performance indicating good transferability.
S.4 Statistical analysis of virtual robot optimizations
The statistical analysis uses virtual-robot simulations to compare self-driving optimization with conventional high-throughput experimentation. It evaluates experiment counts, acceleration, and the probability of outperforming HTE under fixed budgets.
- Simulation framework: The expected benefits of self-driving experimentation over conventional HTE were estimated through simulations using a constructed virtual robot.
- Optimization metrics: ChemOS was evaluated by estimating experiments needed to find a blend at least as photostable as the average most photostable HTE blend.
- Optimization metrics: The analysis computed experimentation acceleration by relating HTE experiment counts to ChemOS experiment counts.
- Optimization metrics: The simulations also estimated the probability that ChemOS identified a more photostable blend than HTE after a specified experiment budget.
S.5 Experimental
The experimental section specifies the materials, solvents, suppliers, and substrate fabrication used for the OPV experiments. Multi-well glass substrates were patterned with an ultraviolet-curable adhesive.
- Materials: The experiments used chlorobenzene, oIDTBR, PTB7-Th, P3HT, PCBM, and PBQ-QF obtained from the listed commercial or collaborating suppliers.
- Multi-well substrates: The substrate fabrication patterned 96 wells on a 125x85 mm² glass substrate using a dispensing robot and UV-curable adhesive.