Inverse design method for large-scale metalenses in reflection employing innovative deep neural networks envisioning future virtual reality applications

Metalenses are the next generation of high numerical aperture lenses. They provide very high efficiencies with minimal aberrations. Nevertheless, there are still challenges remaining to be surpassed in order for such devices to become commercially available. Firstly, the elements of the metasurface have dimensions close to the wavelength of the light, demanding high precision lithography methods to conceive such devices. Techniques such as nanoimprint lithography help reducing the cost with limitations on the possible shapes of the elements. Secondly, the optimization of certain devices remains a challenge, specially ones bound by the limitations of nanoimprint lithography. In this work, we demonstrate the inverse design of a large-scale metalens in reflection using an efficient inverse design method based on deep learning.