Performing optical logic operations by a diffractive neural network

Chao Qian, Xiao Lin, Jian Xu, Yang Sun, Er‐Ping Li, Baile Zhang, Hongsheng ChenLight 2020 · DOI

Abstract

Optical logic operations lie at the heart of optical computing, and they enable many applications such as ultrahigh-speed information processing. However, the reported optical logic gates rely heavily on the precise control of input light signals, including their phase difference, polarization, and intensity and the size of the incident beams. Due to the complexity and difficulty in these precise controls, the two output optical logic states may suffer from an inherent instability and a low contrast ratio of intensity. Moreover, the miniaturization of optical logic gates becomes difficult if the extra bulky apparatus for these controls is considered. As such, it is desirable to get rid of these complicated controls and to achieve full logic functionality in a compact photonic system. Such a goal remains challenging. Here, we introduce a simple yet universal design strategy, capable of using plane waves as the incident signal, to perform optical logic operations via a diffractive neural network. Physically, the incident plane wave is first spatially encoded by a specific logic operation at the input layer and further decoded through the hidden layers, namely, a compound Huygens' metasurface. That is, the judiciously designed metasurface scatters the encoded light into one of two small designated areas at the output layer, which provides the information of output logic states. Importantly, after training of the diffractive neural network, all seven basic types of optical logic operations can be realized by the same metasurface. As a conceptual illustration, three logic operations (NOT, OR, and AND) are experimentally demonstrated at microwave frequencies.

Introduction: The Challenge of Optical Logic

The quest for ultra-high-speed information processing has positioned optical computing as a promising successor to traditional electronic systems. Central to this paradigm shift is the development of robust optical logic gates. However, traditional implementations have faced significant hurdles. Most current optical logic gates rely on the precise interference of multiple light beams. This necessitates exacting control over phase differences, polarization, intensity, and beam size.

Such sensitivity often leads to instability in output states and low contrast ratios between logic '0' and '1'. Furthermore, the bulky apparatus required for this level of control impedes the miniaturization necessary for integrated photonic circuits. To overcome these barriers, researchers have proposed a novel design strategy that utilizes diffractive neural networks (D2NNs) to perform logic operations using simple plane waves.

Methodology: Logic as a Spatial Classification Task

The core innovation of this work is the reframing of logic operations as a spatial classification problem within a physical neural network. Instead of relying on complex wave interference, the system uses a Diffractive Neural Network (D2NN) architecture implemented via Huygens’ metasurfaces.

Comparison between a conventional artificial neural network (a) and the proposed diffractive neural network (b) for optical logic operations.

Spatial Encoding and Hidden Layers

The process begins at the input layer, where the specific logic operation (e.g., AND, OR) and the binary input states are spatially encoded into a mask. This mask modulates an incident plane wave, which then propagates through "hidden layers" consisting of dielectric metasurfaces. These metasurfaces are composed of subwavelength "meta-atoms" that act as individual neurons, scattering light according to a pre-trained phase profile.

Training the Network

The metasurfaces are designed using deep learning back-propagation algorithms. By optimizing the phase shift of each meta-atom, the network is trained to steer light into one of two designated regions at the output plane, representing the binary states '0' and '1'. This approach allows a single, fixed metasurface design to potentially handle all seven basic logic operations (NOT, AND, OR, NAND, NOR, XOR, and XNOR) simply by changing the spatial encoding at the input.

Key Results and Experimental Validation

The researchers demonstrated the versatility of this approach through both numerical simulations and physical experiments.

  • Universal Functionality: Numerical results confirmed that the D2NN could successfully perform all seven basic logic operations with high fidelity.
  • Microwave Proof-of-Concept: To validate the design, a prototype was fabricated and tested at microwave frequencies (17 GHz). The experiments successfully demonstrated NOT, OR, and AND operations.
  • High Contrast and Compactness: The system achieved a contrast ratio exceeding 9.6 dB, ensuring a clear distinction between logic states. Remarkably, the entire computation occurs within a compact distance of approximately 17 wavelengths.
  • Passive Computing: Once fabricated, the metasurface performs calculations entirely passively at the speed of light, requiring no external power for the logic operation itself.

Discussion and Future Outlook

By moving away from interference-based logic, this D2NN approach offers a robust and scalable path for optical computing. The use of spatial encoding makes the system significantly less sensitive to the precise characteristics of the incident light compared to traditional gates.

However, several challenges remain for practical deployment. The current experimental setup uses static input masks; future iterations would require dynamic components like Spatial Light Modulators (SLMs) or tunable metasurfaces for real-time processing. Additionally, while the physics scales to optical frequencies, the nanometer-scale fabrication required for visible light presents a manufacturing hurdle. Future research into nonlinear materials could also enable more complex, multi-layered "deep" optical processing, further expanding the capabilities of this compact photonic platform.

Summary updated

Abstract

Optical logic operations lie at the heart of optical computing, and they enable many applications such as ultrahigh-speed information processing. However, the reported optical logic gates rely heavily on the precise control of input light signals, including their phase difference, polarization, and intensity and the size of the incident beams. Due to the complexity and difficulty in these precise controls, the two output optical logic states may suffer from an inherent instability and a low contrast ratio of intensity. Moreover, the miniaturization of optical logic gates becomes difficult if the extra bulky apparatus for these controls is considered. As such, it is desirable to get rid of these complicated controls and to achieve full logic functionality in a compact photonic system. Such a goal remains challenging. Here, we introduce a simple yet universal design strategy, capable of using plane waves as the incident signal, to perform optical logic operations via a diffractive neural network. Physically, the incident plane wave is first spatially encoded by a specific logic operation at the input layer and further decoded through the hidden layers, namely, a compound Huygens' metasurface. That is, the judiciously designed metasurface scatters the encoded light into one of two small designated areas at the output layer, which provides the information of output logic states. Importantly, after training of the diffractive neural network, <i>all seven</i> basic types of optical logic operations can be realized by the same metasurface. As a conceptual illustration, three logic operations (NOT, OR, and AND) are experimentally demonstrated at microwave frequencies.
BibTeX
@inproceedings{qian2020performing,
  author = {Chao Qian and Xiao Lin and Xiaobin Lin and Jian Xu and Yang Sun and Er‐Ping Li and Baile Zhang and Hongsheng Chen},
  title = {{Performing optical logic operations by a diffractive neural network}},
  booktitle = {Light},
  year = {2020},
  doi = {10.1038/s41377-020-0303-2},
}
Figure