Fig. 1: Overview of the customization process for digital material (DM) design with multiple-target properties.
From: Multi-target digital material design via a conditional denoising diffusion probability model
a The core process starts with random noises and progresses through coarse and fine structure formations to generate the desired DM designs. Target properties, such as resonant frequencies of specific vibrational modes, guide the denoising process to customize DM structures. The example shows a disk-shaped resonator with a labyrinth-like pattern. b The detailed architecture consists of a U-Net model, where the denoising process begins with a Gaussian noise input, \({x}_{N}\), representing a quarter of the labyrinth-like pattern with the symmetric constraint. The U-Net progressively refines this pattern through 800 denoising steps with fully connected layers that encode target properties and time representations. The process concludes at \(t=0\), for the final design, \({x}_{0}\).
