UC Irvine

Artificial intelligence is reshaping the discovery and engineering of photonic materials. Here, we introduce a data-driven framework that employs a generative deep neural network to create ultrathin, multilayer epsilon-near-zero absorbers with record-breaking broadband performance. Within this framework, the neural network learns a generative distribution over high-performing multilayer designs, optimizing the refractive index, layer arrangement, and thickness for maximal broadband absorption. Guided by AI, we experimentally demonstrate an optimized photonic structure that achieves near-perfect light absorption (exceeding 98%) across more than 1000 nm bandwidth in the near-infrared range, while remaining under 160 nm thick. Our approach demonstrates the power of integrating advanced computational algorithms with state-of-the-art fabrication technology, offering a scalable route to high-efficiency nanophotonic devices for target applications. Altogether, this work establishes a generalizable strategy for algorithm-driven photonic design with implications across energy, sensing, and integrated optics.