Disorder unlocks the strength-toughness trade-off in metamaterials

Disorder is ubiquitous in nature, found in both soft biological materials like leaves and strong, tough structures such as diatoms. However, its effect on mechanical properties – whether enhancing or degrading – remains poorly understood. To explore this, we generated 50,000 Voronoi network architectures with varying degrees of disorder and evaluated their mechanical response under uniaxial tensile stress using high-throughput finite-element simulations. Our analysis revealed two distinct failure mechanisms, with some disordered networks outperforming regular hexagonal honeycombs by up to 20% in strength and 100% in toughness, effectively overcoming the conventional strength-toughness trade-off. Remarkably, optimal architectures emerged across all disorder levels, challenging prior assumptions that such performance is achievable only with quasi-order. The mechanical impact of disorder is driven by local geometric features that determine whether the disorder has a positive or negative effect. By training Convolutional Neural Networks (CNNs) on this dataset, we accurately predicted mechanical properties, quickly identifying configurations that exceed traditional limits. This approach offers a pathway for designing lightweight, strong, and tough metamaterials by utilizing disorder to enhance mechanical performance.