Rapid inverse design of super-resolution metalenses via a differentiable vectorial diffraction solver

Metalenses based on optical superoscillation principles are widely applied in super-resolution imaging and can be achieved using inverse design methods. However, traditional approaches predominantly rely on heuristic algorithms, which perform random searches within the solution space. This results in low design efficiency and challenges in obtaining optimal solutions, particularly when dealing with high-dimensional continuous variables, such as multiple nanobricks rotation angle. In this work, we propose an end-to-end inverse design framework for super-resolution metalenses. This framework integrates a differentiable vectorial diffraction solver with a gradient descent algorithm to accurately compute the optical field and efficiently optimize super-resolution metalenses, including single focus, multifocal, and optical needle types. Compared with conventional methods, our approach reduces the optimization time by about 30 times and diminishes the reconstruction loss by more than 1-2 orders of magnitude. This work establishes a new paradigm, to the best of our knowledge, for efficient super-resolution metalens design with broad application potential.