Phonon-assisted photon upconversion (UPC) is an anti-Stokes process in which incident photons achieve higher energy emission by absorbing phonons. This letter studies phonon-assisted UPC in twisted 2D semiconductors, in which an inverted contrast between UPC and conventional photoluminescence (PL) of WSe2 twisted bilayer is emergent. A 4-fold UPC enhancement is achieved in 5.5° twisted bilayer while PL weakens by half. Reduced interlayer exciton conversion efficiency driven by lattice relaxation, along with enhanced pump efficiency resulting from spectral redshift, lead to the rotation-angle-dependent UPC enhancement. The counterintuitive phenomenon provides a novel insight into a unique way that twisted angle affects UPC and light-matter interactions in 2D semiconductors. Furthermore, the UPC enhancement platform with various superimposable means offers an effective method for lighting bilayers and expanding the application prospect of 2D stacked van der Waals devices.

Chirality plays an important role in biological processes, and enantiomers often possess similar physical properties and different physiologic functions. In recent years, chiral detection of enantiomers become a popular topic. Plasmonic metasurfaces enhance weak inherent chiral effects of biomolecules, so they are used in chiral detection. Artificial intelligence algorithm makes a lot of contribution to many aspects of nanophotonics. Here, we propose a nanostructure design method based on reinforcement learning and devise chiral nanostructures to distinguish enantiomers. The algorithm finds out the metallic nanostructures with a sharp peak in circular dichroism spectra and emphasizes the frequency shifts caused by nearfield interaction of nanostructures and biomolecules. Our work inspires universal and efficient machinelearning methods for nanophotonic design.

The manipulation of polarization states beyond the optical limit presents advantages in various applications. Considerable progress has been made in the design of meta-waveplates for on-demand polarization transformation, realized by numerical simulations and parameter sweep methodologies. However, due to the limited freedom in these classical strategies, particular challenges arise from the emerging requirement for multiplex optical devices and multidimensional manipulation of light, which urge for a large number of different nanostructures with great polarization control capability. Here, we demonstrate a set of self-designed arbitrary wave plates with a high polarization conversion efficiency. We combine Bayesian optimization and deep neural networks to design perfect half- and quarter-waveplates based on metallic nanostructures, which experimentally demonstrate excellent polarization control functionalities with the conversion ratios of 85% and 90%. More broadly, we develop a comprehensive wave plate database consisting of various metallic nanostructures with high polarization conversion efficiency, accompanying a flexible tuning of phase shifts (0–2π) and group delays (0–10 fs), and construct an achromatic metalens based on this database. Owing to the versatility and excellent performance, our self-designed wave plates can promote the performance of multiplexed broadband metasurfaces and find potential applications in compact optical devices and polarization division multiplexing optical communications.

Structural design is an important driving force for technological development in nanophotonics. To achieve better performance of nanophotonic devices, the freedom of structural design space needs to be expanded. Unsupervised learning algorithm in deep learning provides a great platform to expand design freedom of structures and avoid the “curse of dimensionality” effectively. It performs well on extracting important features from high‐dimensional data and excavating potential rules. In this work, an unsupervised convolutional neural network is built to inversely design nanostructures with unidirectional transmission. Near‐field information with high dimensions is recognized and extracted into a 2D feature space which maintains high physical continuity and maps to far‐field transmittance effectively. The feature space is further expanded to the whole space by optimistic Bayesian multisampling, from which nanostructures with transmittance over 95% forward while less than 40% backward are inversely designed. Moreover, the relation between near‐field information and far‐field transmittance is explored. A feasible design method of nanostructures is proposed based on unsupervised learning with design space expanded. This design mentality exhibits a way of extracting near‐field features to analyze far‐field spectra with deep learning algorithms, which is suitable for more abundant physical design and can be extended to other similar systems.