Electrocatalytic CO₂ reduction reaction (CO₂RR) is widely recognized as a promising strategy for converting greenhouse gas CO₂ into value-added chemicals and fuels, thereby contributing to carbon neutrality and renewable energy storage. However, CO₂RR involves complex multi-electron and multi-proton transfer processes, leading to diverse reaction pathways and product distributions. The catalytic performance is fundamentally governed by the adsorption and transformation of key intermediates (e.g., *COOH, *CO,*OCHO), which are highly sensitive to the local structure and electronic properties of active sites.

Supported metal catalysts emerge as an important class of CO₂RR catalysts due to their structural tunability and versatile catalytic behavior. Recent studies reveal that catalytic reactions predominantly occur at localized active centers rather than the entire catalyst surface. These active centers can be described as "functional units" with specific geometric and electronic characteristics, including single-atom sites, dual-atom sites, and nanoparticle sites. The spatial arrangement and interaction of these functional units form a "structural ordering" feature, which plays a decisive role in determining adsorption behavior, reaction pathways, and energy barriers. Understanding the structure-performance relationship from the perspective of functional unit ordering is thus essential for rational catalyst design.

This review summarizes recent progress on supported metal catalysts for CO₂RR from the viewpoint of functional unit structural ordering. First, the structural characteristics and catalytic mechanisms of typical functional units are discussed. Single-atom sites exhibit highly tunable electronic structures through coordination engineering, enabling a precise control over intermediate adsorption and offering opportunities to break conventional scaling relationships. Dual-atom sites introduce metal-metal electronic coupling and spatial synergy, which can optimize intermediate binding and promote complex reactions such as C-C coupling. Nanoparticle sites, characterized by continuous surfaces and multiple active centers, provide cooperative platforms where size effects, facet engineering, and metal-support interactions collectively affect catalytic performance. In these systems, the synergy between geometric configuration and electronic structure governs the activity, selectivity, and stability of CO₂RR.

Despite significant progress, a rational catalyst design remains challenging due to the vast configurational space and complex coupling between geometric and electronic effects. To address these challenges, machine learning (ML) is increasingly integrated into catalyst research. This review further highlights recent advances on ML-assisted design of functional units. Descriptor-based ML models establish quantitative relationships between structural features and catalytic performance, enabling efficient high-throughput screening of candidate catalysts. In addition, machine learning potentials (MLPs) also provide near density functional theory (DFT) accuracy with a significantly reduced computational cost. These methods enable large-scale structure exploration and dynamic simulations, offering insights into the evolution of active sites under realistic reaction conditions.

Finally, some challenges and future perspectives are discussed. Key issues include the lack of realistic dynamic models for active sites under operating conditions, the absence of unified multiscale frameworks linking atomic-scale properties to macroscopic performance, and the limited transferability of ML models due to descriptor and data constraints. Future research should focus on developing multiscale theoretical-experimental approaches, constructing physically interpretable descriptor systems, and integrating ML with high-quality databases and advanced characterization techniques. Such efforts are expected to enable a precise regulation of functional unit ordering and accelerate the transition from empirical optimization to predictive catalyst design.

Overall, the paradigm shift from "global structure optimization" to "precise control of functional unit ordering," combined with the integration of theory, machine learning, and advanced characterization, provides a promising pathway for the rational design of high-performance CO₂RR catalysts and the advancement of sustainable energy technologies.