The paper introduces an all-optical Diffractive Deep Neural Network (D²NN) architecture that can learn to implement various functions through the deep learning-based design of passive diffractive layers. The authors experimentally demonstrate the success of this framework by creating 3D-printed D²NNs that learned to classify handwritten digits and function as imaging lenses in the terahertz spectrum. The D²NNs are physically formed by multiple layers of diffractive surfaces, where each point on a given layer acts as a neuron with a complex-valued transmission or reflection coefficient. These coefficients are trained using deep learning to perform specific functions between the input and output planes of the network. After training, the D²NN design is fixed, and once fabricated or 3D-printed, it can perform the learned function at the speed of light. The paper highlights the potential of D²NNs for all-optical image analysis, feature detection, and object classification, as well as their applications in new camera and microscope designs. The authors also discuss the advantages of D²NNs, such as their scalability, power efficiency, and the ability to handle complex functions using passive optical components.The paper introduces an all-optical Diffractive Deep Neural Network (D²NN) architecture that can learn to implement various functions through the deep learning-based design of passive diffractive layers. The authors experimentally demonstrate the success of this framework by creating 3D-printed D²NNs that learned to classify handwritten digits and function as imaging lenses in the terahertz spectrum. The D²NNs are physically formed by multiple layers of diffractive surfaces, where each point on a given layer acts as a neuron with a complex-valued transmission or reflection coefficient. These coefficients are trained using deep learning to perform specific functions between the input and output planes of the network. After training, the D²NN design is fixed, and once fabricated or 3D-printed, it can perform the learned function at the speed of light. The paper highlights the potential of D²NNs for all-optical image analysis, feature detection, and object classification, as well as their applications in new camera and microscope designs. The authors also discuss the advantages of D²NNs, such as their scalability, power efficiency, and the ability to handle complex functions using passive optical components.