Design of Experiments-Based Optimization of Crystallographic and Physical Characteristics of Atomic Layer Deposition-Grown TiO ₂ Thin Films

In ATomic Layer Deposition (ALD) processes, one-factor-at-a-time tuning remains standard, inflating experimental cost and time, and masking factor interactions. Here, a Taguchi L9 (3 3 ) design is employed to map how cycle number (A: 100/250/400), temperature (B: 170/180/190 °C), and pulse time (C: 10/15/20 ms) influence phase composition and thickness/roughness of ALD-grown titanium dioxide (TiO 2 ) on Si. Reflectivity was modeled using Parratt recursion with Nevot-Croce roughness in Advanced Material Analysis and Simulation Software (AMASS), and signal-to-noise ( S / N ) optimization applied the “smaller is better” criterion. Across the matrix, thickness is primarily set by A, rising from 4 to 5 nm at 100 cycles to 16–17 nm at 400 cycles, while the roughness percentage decreases as films thicken (14–20 to 3.7–4.0%); B and C tune the rutile content. Interaction plots revealed B × C coupling for rutile content and roughness, while Factor A showed additive effects. Analysis of variance (ANOVA) attributed 99.7 and 97.1% of variance in thickness and roughness to A, and rutile variance mainly to B (41.5%) and C (40.9%). Confirmation experiments validated predictions: A 1 B 3 C 2 minimized rutile (26.2 vs 25.1% predicted), A 1 B 1 C 1 minimized thickness (4.20 nm), and A 3 B 1 C 2 yielded the lowest roughness (3.63%). These results deliver statistically supported process windows for ultrathin, low-roughness, anatase-rich TiO 2 and outline a minimal-experiment strategy transferable to other ALD systems for controlling crystallography and film characteristics.