Adaptive and deterministic laser polishing enabled by in-process interferometric feedback

High-precision finishing of hard–brittle materials requires the simultaneous control of submicrometer form and nanometer-scale roughness, ideally with high yield and low consumable usage. Noncontact beam-based processes such as laser machining can avoid tool wear, reduce reliance on contact consumables, and relax tool-access constraints. However, deterministic finishing at the waviness and roughness scales remains difficult when the effective material-removal response varies with local absorption, near-surface state, and accumulated irradiation. Conventional laser finishing approaches are typically performed either in open loop or with between-cycle corrections based on offline surface metrology and an assumed stationary removal function, and are therefore vulnerable to such uncertainty. Here, this study examines whether deterministic surface-quality control in laser finishing can be achieved by using in-process surface metrology and feedback, without relying on a precisely calibrated removal function. To this end, we propose an on-machine interferometric feedback framework in which white-light interferometry (WLI) defines a virtual reference surface, a residual height map is computed after each iteration, and irradiation sites are replanned to selectively remove protrusions. Implemented with femtosecond-pulse ablation, the closed loop achieved convergent reductions in both the residual metric and the surface roughness during planarization of single-crystal diamond (SCD). In two experiments on chemical-vapor-deposition-grown SCD, the areal roughness S a decreased from 82.6 to 23.1 nm and from 129 to 44.3 nm, with the minimum-roughness states reached after seven to eight iterations. The roughness-improvement ratio Δ S a / Δ z reached 0.493 to 0.900, which is one to two orders of magnitude higher than values reported for representative open-loop laser-polishing processes. Loop-to-loop scatter in removed volume and single-point tests on rough and smooth surfaces showed that the effective material-removal response is topography dependent, explaining why iterative replanning from measured residuals is beneficial. Post-machining roughness measured by WLI agreed with measurements obtained by atomic force microscopy, supporting quantitative in-process evaluation under evolving surface-optics conditions. The attainable roughness was ultimately bounded by two coupled mechanisms: irradiation-induced loss of fringe visibility in WLI and residual laser-induced periodic surface structures. These results establish adaptive replanning as a measurement-driven route to deterministic, high-yield, noncontact laser polishing under non-stationary material removal.