Enhancing Adjoint Optimization-Based Photonic Inverse Design with Explainable Machine Learning

Abstract

A fundamental challenge in the design of photonic devices, and electromagnetic structures more generally, is the optimization of their overall architecture to achieve a desired response. To this end, topology or shape optimizers based on the adjoint variable method have been widely adopted due to their high computational efficiency and ability to create complex freeform geometries. However, the functional understanding of such freeform structures remains a black box. Moreover, unless a design space of high-performance devices is known in advance, such gradient-based optimizers can get trapped in local minima valleys or saddle points, which limits performance achievable through this inverse design process. To elucidate the relationships between device performance and nanoscale structuring while mitigating the effects of local minima trapping, we present an inverse design framework that combines adjoint optimization, automated machine learning, and explainable artificial intelligence. Integrated with a numerical electromagnetic simulation method, our framework reveals structural contributions toward a figure-of-merit (FOM) of interest. Through an explanation-based reoptimization process, this information is then leveraged to minimize the FOM further than that obtained through adjoint optimization alone, thus overcoming the optimization's local minima. We demonstrate our framework in the context of waveguide splitter design and achieve between 39 and 74% increases in device performance relative to state-of-theart adjoint optimization-based inverse design across a range of telecom wavelengths. Our results highlight machine learning strategies that can substantially extend and enhance the capabilities of a conventional, optimization-based inverse design algorithm while revealing deeper insights into the algorithm's designs.

Recent progress on inverse design for integrated photonic devices: methodology and applications

“…Furthermore, some researchers combine TO algorithm with machine learning to push the optimization out of its local optimal. 140 In addition, since the TO algorithm generates very small features and their sizes are different, how to improve its robustness to manufacturing errors has been extensively studied which is discussed in Sec. 6.…”

Recent developments in Chalcogenide phase change material-based nanophotonics

“…Emphasis has been put on smart sampling strategies [160,161], active learning [162,163] and transfer learning [164,165] to reduce the burden of training data generation. The use of auto-encoders for design space's dimensionality reduction [159] and explainable AI to generate meaningful training data [166] can be instrumental in this direction. Robust training regimens for parametric [143] and pixellated [167] shapes' inverse design can alleviate the challenge of response variance related to fabrication imperfections.…”

Predicting Broadband Resonator-Waveguide Coupling for Microresonator Frequency Combs through Fully Connected and Recurrent Neural Networks and Attention Mechanism

“…In particular, neural networks have already shown promise in a variety of photonics applications such as inverse photonic design, − material and device characterization, ,,− optical sensing, image processing and classification, and optical communication . In inverse photonic design, − the goal is to design optical components or devices with specific desired properties such as the desired transmission spectrum, scattering properties, bandwidth, or quantum efficiency. This goal can be achieved using numerical optimization algorithms that iteratively adjust the design parameters until the desired performance is achieved.…”

“…However, there are also some other techniques that follow completely different strategies to do photonic inverse design. For example, in the adjoint method, − the gradient of an objective function with respect to the design parameters of a device is calculated first. Then, this gradient is used to optimize the device design through a process of iterative refinement.…”

“…The field of photonics also has seen a surge of interest in the use of artificial intelligence, particularly neural networks, to enhance the understanding and application of light–matter interactions. In particular, neural networks have already shown promise in a variety of photonics applications such as inverse photonic design, − material and device characterization, ,,− optical sensing, image processing and classification, and optical communication . In inverse photonic design, − the goal is to design optical components or devices with specific desired properties such as the desired transmission spectrum, scattering properties, bandwidth, or quantum efficiency.…”