June 2023
Spotlight Summary by Guangyuan Li
Self-design of arbitrary polarization-control waveplates via deep neural networks
Artificial intelligence algorithms enable high-performance polarization manipulation with arbitrary functionalities, which can be further combined with other dimensional manipulation of light, of the wavefront for instance. Conventionally, a specific structure is chosen a priori based on empirical knowledge or intuitions, leaving the structure's geometrical parameters to be optimized for achieving the desired performance. This, however, limits the degree of freedom, and sometimes fails for the real arbitrary tailoring of the polarization or/and the wavefront if it has never been explored in the literature. The work of Liu et al. reports a general method for the self-design of arbitrary functional waveplates based on Bayesian optimization and deep neural networks. Nanostructures with arbitrarily desired polarization conversion functionality and optimized performance are generated without any specific physical guideline, expanding the realm of nanophotonics. As a proof of concept, half- and quarter- waveplates are fabricated and a polarization conversion ratio up to 90% is achieved experimentally over broad bandwidths. By harnessing phase shifts covering the (0, 2π) range and group delays in the (0,10 fs) range introduced by nanostructures gathered in an optimized database, the authors also designed an achromatic metalens accompanied with polarization conversion function, showing the versatility of the proposed method. The proposed design strategy can be further expanded to other properties of light and beyond (e.g., sound), and thus will advance the metasurface design for multiplexed functionalities with optimized performance.
You must log in to add comments.
Add Comment
You must log in to add comments.
Article Information
Self-design of arbitrary polarization-control waveplates via deep neural networks
Zhengchang Liu, Zhibo Dang, Zhixin Liu, Yu Li, Xiao He, Yuchen Dai, Yuxiang Chen, Pu Peng, and Zheyu Fang
Photon. Res. 11(5) 695-711 (2023) View: Abstract | HTML | PDF