چكيده لاتين
An optical needle beam, as a Bessel-like beam, features a bright central core and concentric rings of alternating brightness in its transverse profile. Owing to its diffraction-free propagation, the transverse profile remains unchanged during propagation. Furthermore, due to its self-healing capability, after encountering an obstruction, the beam is able to reconstruct its intensity distribution within a short distance. These characteristics make needle beams more effective than conventional Gaussian beams in certain applications, such as optical communications, particle trapping, and biological imaging. In this study, we aim to theoretically analyze and simulate the intensity pattern of a needle beam, and then experimentally generate and investigate it using a spatial light modulator. To achieve this, we need to design a diffractive element corresponding to the needle beam. In traditional approaches, this design is carried out using mathematical relations; however, such methods impose limitations on engineering the propagation properties of the beam. For example, the mathematical function used in direct design has a singularity at the origin, necessitating the central region of the diffractive element to be set to zero within a certain radius to mitigate this issue, which in turn reduces the power transfer efficiency. Other limitations of this approach include the inability to independently adjust the beam’s propagation length and width, or to control the axial position of the needle, as well as the presence of intensity oscillations along the propagation axis, which decreases energy localization. These limitations led us to replace the direct mathematical design with inverse design of the diffractive element. Among the available methods for this purpose, we chose to perform the inverse design using artificial neural networks. In recent years, the superiority of artificial intelligence (AI) over traditional methods in solving certain problems has led to its application in various areas of photonics, including the design of waveguides, integrated optical circuits, metasurfaces, and the inverse design of diffractive elements for structured light generation. In our research, we employed a specific class of artificial neural networks known as physics-informed neural networks. These networks offer advantages such as not requiring large datasets obtained from simulation or experimental results for training, reducing computation time to conventional AI-based approaches.
