Quantum Transformer Applied to Materials Generation

Designing novel crystalline materials remains a central challenge in condensed matter physics and materials science. Here, we present a hybrid quantum–classical Transformer framework for inverse materials design that combines quantum self-attention with the SLICES crystal representation. The model employs parametrized quantum circuits to perform query–key attention operations within a Transformer decoder, while the SLICES representation provides an invertible and symmetry-invariant tokenization of crystal lattices. Trained on Materials Project datasets, the hybrid model seeks to generate physically valid, novel crystal structures and achieve enhanced diversity and reconstruction accuracy compared to fully classical Transformers. We demonstrate conditional generation targeting specific electronic properties such as band gap and perform simulations using NVIDIA's CUDA-Q platform, which is designed for efficient GPU scalability. This work explores how quantum-enhanced attention mechanisms can improve the expressivity of generative models for materials discovery and highlight a path toward integrating quantum computing with large-scale materials informatics.