Metalenses offer a promising path toward ultra-compact achromatic optical systems. However, conventional achromatic metalens designs based on hyperbolic phase profiles face limitations, as practical meta-atoms cannot fully satisfy ideal phase requirements. This reduces broadband focusing efficiency and weakens achromatic performance. Here, we propose an inverse design framework for broadband achromatic metalenses using the optical diffraction neural network. This framework integrates scalar diffraction propagation with gradient-based optimization and is constrained by focusing performance and meta-atom library responses. It enables end-to-end co-optimization of wavelength-dependent phase profiles and discrete meta-atom selection, alleviating the phase-matching problem of conventional designs. Therefore, the devices designed by this approach achieve high achromatic performance while maintaining full compatibility with practical fabrication. Using this framework, we simulated long-wave infrared (9–12 µm) achromatic metalenses with different apertures at numerical apertures of 0.1 and 0.4. These designs achieved a higher average focusing efficiency and a substantial reduction in focal length shift compared with conventional designs based on the identical library. Furthermore, we fabricated large aperture metalenses and experimentally validated their exceptional long-wave infrared imaging performance. When integrated with object detection algorithms, these devices enabled real-time and accurate detection of pipeline leaks even under partial obscuration. Our framework facilitates the development of broadband achromatic metalenses and holds significant potential for integrated, lightweight imaging systems.
Di et al. (Fri,) studied this question.