AI+SLM: The Intelligent Revolution of Spatial Light Modulators

The Spatial Light Modulator (SLM) is an optical device that dynamically controls the wavefront distribution of light through electrical or optical signals. By altering the orientation of Liquid Crystal molecules with an external electric field, it can modulate the amplitude, phase, or polarization of light, enabling real-time programmable control of the optical field. Liquid crystal-based SLMs, with advantages such as simple fabrication, low cost, low power consumption, and ease of control, have shown remarkable performance in fields like optical communication, optical computing, and quantum information processing. In recent years, the integration of artificial intelligence (AI) with optical technologies has sparked a revolution. As a core device for optical field modulation, SLMs, empowered by deep learning and neural networks, are demonstrating unprecedented application potential.

The Power of Integration: When Deep Learning Meets Spatial Light Field Modulation

Deep learning (DL), neural networks (NN), and machine learning (ML) optimize complex tasks through data-driven models. Optical systems, characterized by high parallelism, low latency, and low power consumption, can overcome the bottlenecks of traditional electronic computing when combined with these technologies. The rise of artificial intelligence, particularly the integration of deep learning with spatial light modulator (SLM) technology, is bringing revolutionary changes to this field.

Deep learning, an extension of neural networks, extracts high-level features through multiple layers of nonlinear transformations (such as convolutional layers and recurrent layers). The main foundational models of deep learning include Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Generative Adversarial Networks (GANs). In recent years, advancements in machine learning, particularly driven by deep learning, have led to widespread applications in Optical Imaging detection and optical computing.

The core function of a spatial light modulator (SLM) is to control the alignment of liquid crystal molecules through voltage, thereby modulating the amplitude or phase of incident light in a spatially distributed manner. Phase-type SLMs can achieve phase modulation in the range of 0-2π, enabling precise control of the wavefront for beam transformation. Transmissive SLMs, on the other hand, are primarily amplitude-type, utilizing the optical rotation effect of liquid crystal light valves in combination with polarizers and analyzers to alter the intensity of the output light. By leveraging deep learning methods, SLMs can achieve goals such as parameter-free diffraction focusing and enhanced Resolution, leading to significant advancements in applications like iterative phase imaging, super-resolution imaging, and ghost imaging.

Technological Breakthroughs: Typical Application Scenarios of Intelligent SLMs

Traditional methods for computing holograms often rely on iterative algorithms such as Gerchberg-Saxton (GS). While these algorithms are well-established, they suffer from low computational efficiency, particularly when generating complex light fields or requiring real-time control, where their limitations become increasingly evident. The introduction of deep learning offers a novel approach to addressing this bottleneck.

Flowchart of the GS-CNN Method

In 2021, a domestic research team pioneered the GS-CNN hybrid algorithm, combining the traditional Gerchberg-Saxton (GS) algorithm with convolutional neural networks. This method maintains the stability of the GS algorithm while leveraging the powerful nonlinear mapping capabilities of neural networks, significantly improving computational efficiency. The Bessel vortex beams generated using the GS-CNN method exhibit notably lower root mean square error (RMSE) and higher diffraction efficiency (DE) compared to the results obtained with the traditional GS algorithm.

Simulation Experiment Results of the GS-CNN Method

(a)～(c) Target intensity distributions of Bessel beams; (d)～(f) Output intensity distributions of the generated vortex beams; (g)～(i) LC-SLM phase holograms of the generated vortex beams

Using the GS-CNN method, researchers successfully generated Bessel beams with different topological charges, significantly reducing computation time while maintaining high beam quality. The incorporation of deep learning enables the system to learn the complex physical processes of light field propagation, thereby rapidly producing high-quality holograms. Compared to traditional methods, a greater proportion of the input light field energy is effectively diffracted, and the difference between the intensity distribution and the target distribution is notably reduced, providing a more reliable light source for applications such as optical tweezers and microparticle manipulation.

Holographic Optical Tweezers: The holographic optical tweezers system developed by our company, based on SLMs, enables real-time loading and conversion of holographic images via a computer. This generates the desired light field at the focal plane of the objective lens, allowing for independent dynamic manipulation of microparticles in any optical trap within an array.

Holographic Optical Tweezers System

Holographic optical tweezers technology can generate optical traps with special modes, such as Laguerre-Gaussian beams, Bessel beams, and Airy beams, by modulating the wavefront of incident light. This enables functionalities such as rotation, transport, and sorting of microparticles.

Real-Time 3D Holographic Imaging: Beyond the Limits of Traditional Rendering

In July 2025, Stanford University and Meta showcased a VR headset with a thickness of only 3 millimeters to the world. This device utilizes spatial light modulators (SLMs) and waveguide technology, combined with artificial intelligence algorithms, to project real holographic images directly into the user's eyes, creating a visual experience that feels incredibly realistic in both perception and touch.

Holographic imaging technology is moving from the laboratory to commercial applications at an astonishing pace, with the core driving force being the deep integration of deep learning and spatial light modulators. The Shi Liang team at MIT has leveraged convolutional neural networks (CNNs) to accelerate 3D holographic imaging computations for AR/VR, with the optimized tensor convolutional neural network operating several orders of magnitude faster than physics-based calculations. Similarly, the Suyeon Choi team at Stanford University employed CNNs within the PyTorch framework, achieving unprecedented image quality in 3D holographic imaging using spatial light modulators. Our company's self-developed SLM products, featuring high refresh rates and high phase accuracy, provide an ideal hardware platform for such AI-driven holographic algorithms, making real-time holographic image generation possible.

Color Holographic System

Our company's self-developed color holographic system is based on a phase-type liquid crystal spatial light modulator (SLM). It achieves color holographic reproduction through the principle of time-division multiplexing, where the R, G, and B laser sources, along with the corresponding three-color holograms on the SLM, modulate the light field at the same sequential rate. Observation of the color holographic image can be achieved using a white screen or a CCD receiver.

Test Results:

Written in the End

Machine learning, deep learning, and neural networks form an evolutionary closed loop of "tool-architecture-paradigm": ML provides the methodological framework, NN establishes the computational foundation, and DL unleashes the ability to process complex data through deep architectures. With the deepening integration of AI and optical technologies, spatial light modulators are stepping into a broader development space. As the "dynamic brush" for light field programming, SLMs, when combined with the "intelligent algorithms" of deep learning, are reshaping the design paradigm of modern optical systems. In the future, with the deep collaboration of algorithms and hardware, SLM-AI integrated systems are poised to become the core engine of optical computing and intelligent perception.

References:

[1] Wenqi Ma, Huimin Lu, Jianping Wang, et al. Vortex Beam Generation Based on Spatial Light Modulator and Deep Learning [J]. Optics Journal, 2021, 41(11): 79-85.