Nanofabrication stops returning one shape: a conditional diffusion model learns the whole spread of possible outcomes

Nanofabrication outcomes are predicted as a distribution rather than a single shape. Existing models of the milling process typically predict a single deterministic fabrication result and neglect variability. The predictions reproduce edge morphology, linewidth variations, and defect probabilities observed in fabricated samples.

The model treats the input design as a condition and learns the statistics of many possible outcomes for that design. A conditional diffusion model learns the statistical distribution of process outcomes conditioned on input design. Its distributional predictions match the edge morphology, linewidth variations, and defect probabilities observed in fabricated samples.

The work addresses fabrication variability analysis for focused ion beam milling, for researchers and engineers working on micro- and nanofabrication modelling and fabrication-aware design. The material states it establishes a probabilistic modelling framework for fabrication variability analysis and fabrication-aware design optimization.

The material states that predictions reproduce edge morphology, linewidth variations, and defect probabilities observed in fabricated samples, but does not state the number of validation samples or a quantitative error range. Milling deviation sources are described as systematic and stochastic effects such as ablation-rate fluctuations and redeposition.