High-Throughput Experimentation
High-throughput experimentation runs many small, parallel measurements, commonly in multiwell plates read by an automated instrument, so that a large composition or condition space can be screened quickly. Its value depends on deciding which samples to put in the limited number of wells, and on analysis methods that extract physical parameters robustly from many noisy, individually cheap measurements. It is closely related to design of experiments and is a common building block of self-driving laboratories.
Composition spread alloy films with a 100-channel microreactor (2015)
Gumuslu and co-workers used a CuxPd1-x composition spread alloy film (CSAF), a thin film containing a continuous composition gradient on a single substrate of about 1 cm2, as a catalyst library. A 10 × 10 microreactor array measured H2-D2 exchange kinetics at 100 compositions (x = 0.30 to 0.97) between 333 and 593 K, while spatially resolved XPS and UPS mapped composition and valence band structure at the same locations. Microkinetic fits at every composition gave barriers that could be correlated directly with the electronic structure. The authors describe the approach as general for gas-solid reactions with turnover above about 1e-5 per second, below 600 K and 1 atm, that can be analyzed by mass spectrometry (Gumuslu et al. 2015). Summary: gumuslu2015.
Mapping alloy phases by XPS on composition spread films (2015)
Boes et al. used composition spread alloy films, Cu-Pd libraries with a continuous composition gradient over about 1 cm, and moved a 200 um XPS spot across them to measure core level shifts at nearly continuous composition from x = 0 to 1 (Boes et al. 2015). An anomaly in the Cu core level shift identified the B2 phase window, giving a high-throughput way to map phase boundaries as a function of annealing temperature, potentially for ternary alloys too. Summary: boes2015.
Composition, structure and size spread libraries (2016)
In an AIChE Journal Perspective, Kitchin and Gellman frame high-throughput experimentation around libraries with continuous gradients: composition spread alloy films (CSAFs) made with a rotating shadow mask tool that co-deposits up to four metals, surface structure spread single crystals (S4C) polished to a spherical dome so that one sample exposes a continuous range of surface orientations, and nanoparticle size gradients (Kitchin 2016). Examples include a single AlxFeyNi1-x-y CSAF oxidized at 700 K that resolved four oxidation regimes and the composition-dependent critical Al concentration for passivation, a Cu S4C on which oxygen uptake depended only on step density, and the CuxPd1-x microreactor and UPS studies that correlated H2 adsorption barriers with valence band energy. The outlook identifies three experimental requirements (the library, careful spatially resolved characterization, and parallel measurement of the functional property) and calls for better data analysis tools, design of experiments, multicomponent simulation methods and data archiving. Summary: kitchin2016.
Combinatorial substrate epitaxy (2017)
In combinatorial substrate epitaxy, a thin film is deposited on a polished, large-grained polycrystal and each grain acts as an independent single-crystal substrate, so a single deposition samples epitaxial growth across all of orientation space; EBSD then maps the phase and orientation of each film grain against its substrate grain. Wittkamper et al. used this to classify 135 SnO2/CoNb2O6 grain pairs by phase and orientation relationship at two growth rates, and showed that automated EBSD indexing can separate two closely related polymorphs when checked by whole-pattern dictionary matching (Wittkamper et al. 2017). Summary: wittkamper2017.
A 96-well photoreactor with colorimetric H2 detection (2020)
Lopato et al. introduced a 96-well parallel photoreactor in which each vial is sealed under a chemochromic H2-sensitive tape, and photographs of the darkening tape, processed by a computer vision script, give H2 kinetic traces for every well (Lopato et al. 2020). Calibration with injected H2 gave a linear response, and a plate of identical reactions showed standard errors of 9.8% in maximum H2 and 12.3% in maximum rate. Varying two metal salt concentrations across rows and columns allowed more than 700 bimetallic compositions to be tested. Summary: lopato2020.
Pitfalls of in situ catalyst formation in parallel photoreactor screens (2022)
Simon et al. revisited compositions screened in a 108-well photoreactor that tracks H2 by imaging colorimetric detection tape, where wells producing more than 2 µmol H2 are counted active (Simon et al. 2021). They showed that metals judged inactive in earlier screens (Ni, Co, Pd, and Au/Cu after repeated cycling) failed because in situ formed particles aggregated, and became active when a thiol capping ligand was added. The lesson for high-throughput screening is that a negative result may reflect a confounding variable such as particle formation and stability, which must be controlled before results are compared with computational predictions. Summary: simon2022.
Ternary composition screening with plate normalization (2023)
Bhat et al. used a robot-dispensed 108-well photoreactor with tape-based H2 detection to map the Cu-Ru-Fe binary and ternary composition space for photo-driven H2 production (Bhat et al. 2023). To compare plates they scaled each by internal-standard wells containing a molecular catalyst and by the average activity of Cu wells present on every plate, which put the effective detection limit at 1.1 scaled µmol H2, and they reported the maximum H2 rather than rates to avoid errors from differentiating noisy time series. The work also illustrates a limit of in situ synthesis in screens: the salt ratio added to a well is not the composition of the particles that form. Summary: bhat2023.
Surfactant mixture screening in a 96-well plate (2025)
Chilkunda, Kitchin and Tilton fit a full binary surfactant study into one 96-well plate: 95 wells of SDS and C8E4 mixtures placed by a regular solution theory model, plus one internal fluorescence standard (Chilkunda et al. 2025). They then treated the readout as a binary classification problem, labeling wells as below or above the critical micelle concentration with an empirical fluorescence threshold and fitting the physical boundary to all labels simultaneously. The authors present this as a general way to regress physical parameters from data with a classifier, adaptable to other models or to ternary mixtures. Summary: chilkunda2025.
Relation to self-driving laboratories (2025)
Canty et al. distinguish SDLs from established high-throughput and cloud laboratories by the automated, judicious choice of experiments, the adaptation of experimental methods, and workflows that integrate several tools (Canty et al. 2025). They also note that the option of brute-force high-throughput experimentation will remain attractive to investors, since advances in one approach often carry over to the other. Summary: canty2025.