Nanophotonics, which explores significant light–matter interactions at the nanoscale, has facilitated significant advancements across numerous research fields. A key objective in this area is the design of ultra‐compact, high‐performance nanophotonic devices to pave the way for next‐generation photonics. While conventional brute‐force, intuition‐based forward design methods have produced successful nanophotonic solutions over the past several decades, recent developments in optimization methods and artificial intelligence offer new potential to expand these capabilities. In this review, we delve into the latest progress in the inverse design of nanophotonic devices, where AI and optimization methods are leveraged to automate and enhance the design process. We discuss representative methods commonly employed in nanophotonic design, including various meta‐heuristic algorithms such as trajectory‐based, evolutionary, and swarm‐based approaches, in addition to adjoint‐based optimization. Furthermore, we explore state‐of‐the‐art deep learning techniques, involving discriminative models, generative models, and reinforcement learning. We also introduce and categorize several notable inverse‐designed nanophotonic devices and their respective design methodologies. Additionally, we summarize the open‐source inverse design tools and commercial foundries. Finally, we provide our perspectives on the current challenges of inverse design, while offering insights into future directions that could further advance this rapidly evolving field.
The on‐chip optical power splitter is a common and important device in photonic integrated circuits (PICs). To achieve a low insertion loss and high uniformity while splitting the guided light, multi‐mode interferometer‐based structures utilizing a self‐imaging principle are widely used mainly in the form of a 1 × 2 configuration. Recently, an inverse design method for nanophotonic devices has emerged to overcome the limited capability of the conventional design methods and make it possible to explore the vast number of design parameters. Because of the non‐intuitive shape of inverse‐designed structures, they allow us to discover interesting and complex optical responses which are almost impossible to find with conventional design methods. Here, we report two kinds of inverse‐designed 1 × 4 optical power splitters composed of silicon bars of different lengths, which are fabricated with a standard CMOS‐compatible process. The particle swarm optimization method was used to minimize the insertion loss and divide the power evenly into each output port with finite‐difference time‐domain method simulation. The first optical power splitter has a compact size of 8.14 × 12 μm 2 and the second optical power splitter has an even more compact size of 6.0 × 7.2 μm 2 . With the inverse designed structures, we fabricated the chip with a CMOS‐compatible fabrication process. Experimental verification of the structures is provided and good agreement with the numerical results is obtained. The first 1 × 4 optical power splitter has a low insertion loss of less than 0.76 dB and uniformity of less than 0.84 dB, and the second more compact optical power splitter has a low insertion loss of less than 1.08 dB and uniformity of less than 0.81 dB. As the complexity of on‐chip photonic systems has steadily increased, the inverse design of photonic structures holds great potential to be an essential part of advanced design tools.
Efficient extraction of light from a high refractive index silicon waveguide out of a chip is difficult to achieve. An inverse design approach was employed using the particle swarm optimization method to attain a vertical emitting meta-grating coupler with high coupling efficiency in a 220-nm-thick silicon-on-insulator platform. By carefully selecting the figure of merit and appropriately defining parameter space, unique L-shape and U-shape grating elements that boosted the out-of-plane radiation of light were obtained. In addition, a 65.7% (−1.82 dB) outcoupling efficiency and a 60.2% (−2.2 dB) fiber-to-chip vertical coupling efficiency with an 88 nm 3 dB bandwidth were demonstrated by numerical simulation. Considering fabrication constraints, the optimized complex meta-grating coupler was modified to correspond to two etching steps and was then fabricated with a complementary metal-oxide-semiconductor-compatible process. The modified meta-grating coupler exhibited a simulated coupling efficiency of 57.5% (−2.4 dB) with a 74 nm 3-dB bandwidth in the C-band and an experimentally measured coupling efficiency of 38% (−4.2 dB).
Precise imaging in three-dimension (3D) is an essential technique for solid-state light detection and ranging (LiDAR). Among various solid-state LiDAR technologies, silicon (Si) optical phased array (OPA)-based LiDAR has the significant advantage of robust 3D imaging due to its high scanning speed, low power consumption, and compactness. Numerous techniques employing a Si OPA have utilized two-dimensional arrays or wavelength tuning for longitudinal scanning but the operation of those systems is restricted by additional requirements. Here, we demonstrate high-accuracy 3D imaging using a Si OPA with a tunable radiator. As we adapted a time-of-flight approach for distance measurement, we have developed an optical pulse modulator that allows a ranging accuracy of less than 2 cm. The implemented Si OPA is composed of an input grating coupler, multimode interferometers, electro-optic p-i-n phase shifters, and thermo-optic n-i-n tunable radiators. With this system, it is possible to attain a wide beam steering range of 45° in a transversal angle with a 0.7° divergence angle, and 10° in a longitudinal angle with a 0.6° divergence angle can be achieved using Si OPA. The character toy model was successfully imaged in three dimensions with a range resolution of 2 cm using the Si OPA. The further improvement of each component of the Si OPA will allow even more accurate 3D imaging over a longer distance.
We proposed inverse-designed nanophotonic waveguide devices which have the desired optical responses in the wide band of 1450-1650 nm. The proposed devices have an ultra-compact size of just 1.5 µm × 3.0 µm and are designed on a silicon-on-insulator (SOI) waveguide platform. Individual nano-pixels with dimensions of 150 nm × 150 nm were made of either silicon or silicon dioxide, and the materials for the 200 total cells were determined using a trained deep neural network. While training the two networks, the hyperparameter optimization method was applied to make the training process efficient. We then fabricated the proposed devices using a CMOS-compatible fabrication process, and experimentally verified the fabricated device performance.
Albeit the recent successful demonstrations of nanophotonic device designs leveraging data‐driven supervised learning methods, several challenges still remain. One of the primary constraints of these methods is their computational cost because they require a massive amount of highly time‐consuming full‐wave electromagnetic simulations. In this study, semi‐supervised learning methods are implemented to avoid this issue. Photonic crystal waveguide devices, which offer interesting light‐matter interactions such as the slow‐light effect and the filtering effect, are employed to validate the methodologies. To utilize unlabeled data within this framework, a novel deep generative model, the denoising diffusion probabilistic model is introduced. Next, a pseudo‐labeling process is introduced to assign synthetic labels to the unlabeled data. The performance of a semi‐supervised learning method is compared with a supervised learning scheme, demonstrating that the performance of the neural network can be enhanced without additional numerical calculations. This method holds extensive potential to elevate the efficiency of designing and characterizing future nanophotonic devices.