Deep Photon Network Platform, Empowering Any Functional Photon Integrated Circuit

The widespread application in the fields of optical communication, computing, and sensing continues to drive the growing demand for high-performance integrated photonic components. Recently, Ali Najjar Amiri of Kochi University in Türkiye and other scholars proposed a highly scalable and highly flexible deep photonic network platform, which is used to realize optical systems on chip with arbitrary functions. Traditional devices based on forward or reverse design often have limited functionality, limiting the type, complexity, and bandwidth of optical operations. The deep photonic network platform proposed in this study breaks through these limitations, allowing for the design of integrated photonic devices with arbitrary broadband capabilities, bringing unprecedented flexibility and performance improvement to the next generation of photonic integrated circuits (PICs). The relevant research results have been published in Nature Communications. (DOI: 10.1038/s41467-024-45846-3)

More efficient, flexible, and complex ways to manipulate light

The deep photon network architecture consists of an input layer, a series of MZI layers, and an output layer, as shown in Figure 1. The advantage of this MZI network-based architecture lies in its ability to implement the functions of any optical system and excellent signal processing capabilities. The input optical signal is processed through a multi-layer customized MZI interferometer, and each MZI is equivalent to a matrix function. The modular transmission matrix constructs the entire network framework. Specifically, each MZI consists of two pairs of waveguide cones with customized geometric shapes and two directional couplers. These unique waveguide cones are determined through iterative optimization algorithms to achieve a unique spectral phase distribution different from straight waveguides, while also having higher design degrees of freedom. Customizing photon network paths according to requirements, combined with the fast search function of machine learning algorithms, can quickly and accurately regulate the matrix of MZI, allowing for the rapid design of integrated photon devices with any function.

Researchers utilized the deep photon network platform to demonstrate ultra wideband power splitters and spectrum duplexers, each design completed within 2 minutes. This platform provides an easy to handle path for systematic and large-scale photon system design, providing customized power, phase, and dispersion distributions for high-throughput communication, quantum information processing, and medical/biosensing applications.

Figure 1 Deep Photon Network Architecture and Components

Quick design and efficient implementation of any function
The core of deep photonic network architecture lies in its high scalability and flexibility, and the ability to design photonic devices with any spectral specification. In this article, researchers demonstrate how to use this deep photon network to achieve arbitrary optical functions. As proof of principle, three independent devices were selected for explanation: two broadband power dividers with spectral ratios of 50:50 and 75:25 operating in the range of 1400-1600nm, and a spectral duplexer operating between 1450nm and 1630nm. According to the complexity of the required functions, an appropriate interference layer and the number of parameters in each waveguide taper can be selected. For example, both power dividers are designed with three layers of grid, while duplexers are designed with six layers of grid; For the waveguide taper in the MZI interferometer arm, five trainable widths and one trainable length were used, providing a total of 24 optimization parameters for each MZI in the photon network. The 50:50 and 75:25 power dividers have 72 trainable parameters and a device length of 240 μ M; The spectral duplexer has 144 trainable parameters and a device length of 480 μ M. The optimization iteration process of the three devices is shown in Figure 2, and fast convergence can be achieved within 1-2 minutes. This new method combining simulation and optimization significantly shortens the development cycle of optical equipment from concept to manufacturing. The use of computing power and cutting-edge algorithms not only simplifies the construction process of optical systems, but also brings more possibilities for photon technology innovation.

The researchers also conducted experimental verification on two power dividers and spectral duplexers. The test results show that the insertion loss of both power dividers is less than 0.61dB, and the experimentally measured 1dB bandwidth is as high as 120nm, which is consistent with the simulation results. The manufactured duplexer also has excellent performance, with an insertion loss of less than 0.66dB and a cutoff wavelength shift of only 5nm. These demonstrations validate the practicality and effectiveness of the deep photon network platform in handling complex optical tasks.

Figure 2 Optimization and final simulation results of power splitter and spectral duplexer deep photon network

Summary and Outlook

In this work, researchers propose a computationally efficient, physically accurate, and systematic deep photon network platform for creating and implementing on-chip optical functions. This platform can expand its functions according to specific requirements, with high practicality and effectiveness, providing scalable and robust solutions for designing and manufacturing optoelectronic devices and systems with new functions. The multifunctional demonstration of deep photon networks not only improves the performance of devices, but also opens up new paths for customized optical system solutions, which is expected to bring new technological changes in the fields of communication, computing, and sensing.

Source: Sohu