We are pleased to present to you this special issue dedicated to designing and implementing engineered photonic materials facilitated by artificial intelligence technology. In the past two decades, the use of rationally engineered photonic structures for the unconventional control of light represents one of the most exciting frontiers in photonics and materials science. These artificially structured optical media, including but not limited to photonic crystals, plasmonic components, and optical metamaterials, have led to transformative changes to the entire vistas of optics science. While most design processes for these materials have relied on the conventional trial-and error approach via iterative numerical modeling, the last few years have witnessed a growing trend of exploring artificial intelligence, such as machine learning and deep learning, to design a variety of photonic structures. Although still in its infancy, this emerging area of research has revealed an exciting potential as for how future photonic materials will be discovered and implemented. The continued advances in artificial intelligence-enabled design of engineered photonic materials open up an exciting opportunity to revolutionize the way we discover, design, and utilize photonic materials and devices beyond the conventional wisdom. The state of the art in this subject has been comprehensively covered in several very recent review articles [1–6].

This feature issue is intended to offer a glimpse of the evolving field that consolidates artificial intelligence and photonic design. Instead of compiling a copious volume, here we present a slim collection of 6 research articles that deal with representative branches of the field that cover a variety of geometries, material selections, operating frequencies, and functionalities. Starting from individual photonic components, Wu et al. leverage a deep neural network to design plasmonic nano-antennas with simultaneously optimized near- and far-field properties [7]. The developed approach can capture the highly nonlinear, complex relationship between the plasmonic geometry and the resultant responses in both the near- and the far-zones. Likewise, Li et al. utilize deep convolutional neural networks to accurately model the correlation between design parameters and the quality factors of photonic crystal nanocavities [8]. With open-source database and codes, the on-demand data-driven scheme exhibits decent prediction accuracy and convergence speed, useful for rapid design of such cavities for nanoscale lasers and photonic integrated circuits. Steering towards periodic structures, Lin et al. demonstrate the inverse design of resonant nanostructures for extraordinary optical transmission with periodic metallic gratings [9]. Exploiting a topology optimization approach, the designed structures are shown to possess prescribed resonant properties thanks to effective excitation and guidance of surface plasmon polaritons, enabling extraordinary optical transmission with enlarged bandwidth and reduced spectral sensitivity. Beyond the optical regime, Noh et al. present a powerful deep neural network for the expedited design of multi-layered transmissive metasurfaces for the microwave frequency range [10]. The developed framework elegantly handles the structural complexity as well as side effects such as couplings among adjacent meta-units. On the application side, Noureen et al. show how to design compact and efficient solar thermophotovoltaics that serve as perfect solar absorbers and selective emitters [11]. Using a hybrid deep-learning model that consolidates a deep convolutional network and a recurrent neural network, the authors are able to infer the absorption and emission of such thermal solar cells within a fraction of seconds with high accuracy. The efficacy of deep learning and neural networks in photonics is not limited to classical electromagnetics. As a telling example, Burgess et al. demonstrate the use of a recurrent framework to model the non-Markovian dynamics where two-level atoms interact with the radiation reservoir of photonic crystals [12]. The developed strategy captures precise details in the atomic evolution, including the fractional steady-state population inversion and spectral splitting of the electronic transition. We anticipate rapidly growing interest and research that utilize deep learning and neural networks for the prediction and representation of complex processes in quantum optics.

This feature issue is by no means an exhaustive display of all the exciting topics at the intersection of the fields of artificial intelligence and engineered photonic materials. Nevertheless, it reveals the ongoing paradigm shift in discovering and designing photonic materials, structures, and systems. We hope you enjoy this special issue. We are grateful to all the authors, reviewers and OSA staff members for their contributions and efforts to make this issue possible.