Direct Laser Interference Patterning of Functional Metal Surfaces: From Written Geometry to Functional Interfaces

This review examines the field of direct laser interference patterning (DLIP) of functional metal surfaces, arguing that the functional response of a surface cannot be predicted from its written geometry alone. The central proposition is that "functional response develops through two consecutive stages. The written relief first creates the possibility for coupling with an interacting agent. The realised interface then determines how that interaction is translated into measurable performance. Geometry therefore defines the opportunity for interaction, whereas the interfacial state determines its consequence."

The paper distinguishes between the written geometry, which is the periodic structure prescribed by the optical interference field (with the lateral period defined by Λ = λ/(2 sin θ)), and the realised functional interface, which is the actual surface that develops during processing. This realised interface is defined by four variables: period, depth and aspect ratio, hierarchy (including LIPSS and redeposited particles), and surface state (including oxide formation, hydrocarbon adsorption, contamination, coatings, and ageing).

The review uses a period–depth–state–function framework to analyze evidence across multiple application domains, including tribology, wetting and anti-icing, antibacterial and biomedical surfaces, optics, electrochemistry, and energy devices. A key finding is that period alone is a weak predictor across application domains. For example, Structural colour typically uses periods of 1.7-2.6 µm with an optimum depth near 0.3 µm, while Lubricated and dry tribology use periods of about 5-18 µm and depths of 0.1-1.13 µm, and State-dependent functions occupy the same broad range, including 2.6-2.7 µm anti-icing textures, 0.85-3 µm antibacterial textures and 3-17 µm biomedical textures. This overlap demonstrates that nominally similar period–depth combinations can support different functions and different levels of inference.

The analysis reveals that the transferability of geometric variables depends on the application domain. "Geometry remains highly transferable in optical systems and controlled mechanical contacts, whereas state-dependent, biological and device-level functions additionally depend on surface state, operating conditions and system architecture. For instance, in optics, period defines diffraction geometry, whereas depth controls efficiency and colour uniformity. In lubricated tribology, performance is governed by an aspect-ratio window, with one study finding the greatest friction reduction, about 25-65% relative to unstructured surfaces, at aspect ratios of approximately 0.07-0.11 rather than at maximum depth."

In contrast, for state-dependent functions like wetting and anti-icing, the realised surface state is critical. The review notes that Copper wettability therefore cannot be interpreted from geometry alone; storage history and surface chemistry are part of the functional interface. Similarly, for antibacterial surfaces, Retention, colony-forming units, live/dead imaging and long-term biofilm formation... measure different biological processes, and the same geometry can have opposite effects depending on material activity. For example, a 3 µm pattern on copper enhances Cu-mediated bacterial killing, while bacteria-scale features on an inert polymer can increase bacterial adhesion.

The review also addresses manufacturing translation, arguing that Manufacturing maturity is therefore measured not only by the ability to reproduce periodic structures, but by the ability to reproduce the mechanism through which those structures generate function. It emphasizes that high-throughput processing is only meaningful if the depth distribution and surface state are controlled and verified at the production rate.

The paper concludes by proposing a Prospective DLIP Design Atlas which provides a framework for designing and interpreting functional surfaces. This involves identifying the interacting agent, nominating the variable expected to create coupling, perturbing a measurable gate independently, and recording the direct functional endpoint. The review stresses that "A predictive relationship requires more than a correlation between morphology and performance. It requires experiments that distinguish the variables responsible for coupling from those governing its expression, identify the relevant interfacial gate through controlled perturbation, and evaluate the resulting function using direct, application-appropriate endpoints. Ultimately, Predictive surface engineering and industrial translation therefore depend not only on writing periodic structures reproducibly, but on controlling and verifying the complete realised functional interface."

Improvements to AI Systems:

1. Multi-variable surface–function prediction models

Replace single-geometry (e.g., period-only) predictors with models that jointly encode period, depth, aspect ratio, hierarchy (LIPSS, particles), and surface state (oxide, adsorption, ageing). The AI should learn that identical period–depth pairs can yield different functions depending on interfacial chemistry and operating context.

2. Domain-aware transfer learning with explicit coupling variables

Build AI systems that automatically identify which variables are “coupling” (geometry-driven) vs. “expression” (interfacial-state-driven) for each application domain. For optics and controlled mechanical contacts, prioritize geometric transfer; for wetting, anti-icing, antibacterial, and biomedical, weight surface state and history more heavily.

3. Causal inference over correlation

Implement AI that designs experiments to perturb a single “gate” variable (e.g., depth, oxide thickness, storage time) while holding others constant, and uses the resulting functional endpoint (friction coefficient, contact angle, CFU reduction) to infer causal mechanisms—not just morphological correlations.

4. Uncertainty-aware design recommendations

Train AI to output not just a predicted performance but a confidence interval that reflects the completeness of the realised interface description. If surface state or hierarchy data are missing, the system should flag low confidence and request additional characterization before recommending a process parameter.

5. Process–state–function co-optimization

Extend AI optimization loops to include manufacturing constraints: high-throughput DLIP is only valid if depth distribution and surface state are verified at production rate. The AI should co-optimize laser parameters (period, fluence, pulse overlap) with post-processing steps (cleaning, coating, ageing control) to reproduce the mechanism of function, not just the written geometry.

6. Dynamic surface-state tracking and prediction

Develop AI that models temporal evolution of the realised interface (e.g., hydrocarbon adsorption, oxide growth, ageing) and predicts how functional performance drifts over time. This enables design for long-term stability, especially for anti-icing, biomedical, and electrochemical applications.

7. Cross-domain analogy engine

Use the period–depth–state–function framework to map overlapping parameter spaces (e.g., 2.6–2.7 µm anti-icing vs. 0.85–3 µm antibacterial) and propose transferable design rules, while automatically flagging when a function is state-dependent and thus not transferable across materials or environments.

8. Direct-endpoint evaluation module

Replace proxy metrics (e.g., contact angle for anti-icing, CFU for antibacterial) with direct functional endpoints (ice adhesion force, biofilm mass) in AI training data and evaluation. The AI should learn to reject designs that only improve proxies without improving the actual application-relevant outcome.

What the Improved AI System Can Do:

• Given a target function (e.g., low-friction lubricated contact, anti-icing, antibacterial), it can recommend a complete DLIP parameter set—including period, depth, hierarchy, and surface-state control steps—with predicted performance and confidence.

• It can distinguish between “geometry-transferable” and “state-dependent” functions, avoiding false generalizations across materials or storage conditions.

• It can design and interpret experiments that isolate causal variables, accelerating the creation of a predictive DLIP Design Atlas.

• It can monitor and predict performance degradation over time, enabling robust industrial deployment.

• It can flag when a reported “same geometry” result is actually due to different interfacial states, preventing misleading literature comparisons.

• It can optimize manufacturing processes to ensure that the realised interface, not just the written pattern, meets functional specifications at production throughput.

Direct laser interference patterning (DLIP) generates periodic micro- and nanoscale structures with increasing precision and throughput, yet similar geometries can produce fundamentally different functional responses. This review examines why morphology alone cannot predict friction, wetting and ice adhesion, bacterial response, cell behaviour, optical performance or electrochemical and photovoltaic function. DLIP is treated as a model system in which the optically prescribed geometry can be distinguished from the interface realised during processing. The written period is separated from relief depth, aspect ratio, hierarchical topography, surface chemistry, ageing and process history. Functional response is interpreted as a two-stage process: geometry creates the opportunity for interaction with an external agent, while the realised interface determines how that interaction becomes measurable performance under a specific interfacial state. A curated, metals-centred evidence base spanning tribology, wetting and anti-icing, antibacterial and biomedical surfaces, optics, electrochemistry, energy devices and manufacturing shows that period-depth coordinates alone rarely predict function across domains. Geometry remains highly transferable in optical systems and controlled mechanical contacts, whereas biological, state-dependent and device-level functions also depend on surface state, operating conditions and system architecture. The framework reconciles conflicting observations and identifies which variables are transferable and which remain context-dependent. Predictive DLIP surface engineering therefore requires identification of the interacting agent, separation of written geometry from the realised interface, isolation of variables governing interfacial coupling and evaluation through direct, mechanism-specific endpoints.