The central task in heterogeneous catalysis research and development is to develop suitable solid catalysts for various chemical reactions to achieve the best performance in activity, selectivity, and stability. However, because of the multiple parameters from catalyst fabrication to testing, conventional search methods are time-consuming and labor-intensive. High-throughput workflows offer a transformative solution by enabling highly integrated and programmed in-parallel experiments to generate large, reproducible experimental datasets.
In a review published in ENG. Chem. Eng., researchers at Xi‘an Jiaotong University trace the evolution of high-throughput technology from its germination in 1909–1913—when Alwin Mittasch and coworkers performed approximately 20,000 systematic experiments to develop the iron-based catalyst for ammonia synthesis—to today‘s autonomous systems. The development progressed through six phases: germination and pioneer, early foundation, initial commercialization, technology breakthrough, maturity and quantification, and intelligence and convergence.
The state-of-the-art high-throughput workflow comprises five key successive steps in a closed and autonomous cycle: experiment design using design of experiments, genetic algorithms, or machine learning-guided active learning; catalyst synthesis on robotic platforms executing methods such as impregnation, co-precipitation, or hydrothermal synthesis; catalyst testing and screening in parallel continuous-flow or batch reactors with up to 96 channels; catalyst characterizations including physisorption, chemisorption, temperature-programmed techniques, X-ray diffraction, transmission electron microscopy, infrared and Raman spectroscopy, and X-ray photoelectron spectroscopy; and data handling and analysis using machine learning methods such as random forests, neural networks, Bayesian optimization, and active learning.
High-throughput workflows have demonstrated significant advantages in time and resource efficiency. Catalyst research and development time can be reduced from decades to months or even weeks. However, several critical pitfalls remain. First, results generated by high-throughput experiments often cannot reflect realistic reaction phenomena under industrial conditions, as feeds are typically high-purity and catalyst samples are powders or small particles far smaller than industrial catalyst shapes. This can lead to deviations in catalyst performance between kinetic and non-kinetic control conditions. Second, most high-throughput reactors operate at near-ambient pressure, while many industrially relevant reactions require high pressures of 15 to 100 bar. Third, the funnel-shaped screening approach—primary screening of hundreds to thousands of samples, secondary screening of medium samples, and tertiary in-depth analysis—can lead to the discard of so-called negative catalyst candidates based on a single or a few activity measurements, potentially eliminating catalysts that would be excellent under elevated pressures or require induced activation over time.
Looking ahead, the trajectory of high-throughput technology points toward fully autonomous, integrated, and predictable systems through deeper cooperation with self-driving laboratories, multimodal data integration, high-throughput operando characterizations, and hybrid models combining density functional theory calculations and microkinetic modeling with experimental constraints.

DOI
10.1007/s11705-026-2691-1