Researchers Demonstrate 1,024-Mode Polarization Modulation for High-Dimensional Optical Computing
Updated
Updated · Nature.com · Aug 12
Researchers Demonstrate 1,024-Mode Polarization Modulation for High-Dimensional Optical Computing
2 articles · Updated · Nature.com · Aug 12
Summary
Using a phase-only spatial light modulator, researchers spatially programmed both degree and state of polarization across more than 1,024 optical modes, adding a previously unavailable controllable dimension of light.
The method engineers micrometre-scale polarization statistics so each mode can be placed anywhere inside the Poincaré sphere; experiments spanned the full 0-to-1 polarization range with about 4% DOP error.
That control let the team encode 32×32 RGB images into a single 532-nm laser beam, mapping color data to polarization rather than wavelength and reconstructing hidden grayscale and color patterns from Stokes measurements.
In a photonic neural network, the approach lifted CIFAR-10 test accuracy to 50.4% from 43.3% for phase-only grayscale encoding, while a related optical encryption system used a 12.2-Gbit decryption key for multidimensional image security.
Because the platform relies on mature spatial light modulator technology, the authors say it could scale to more than 10,000 macromodes and extend photonic computing, cryptography and sensing.
Will manipulating light's polarization at a microscopic scale render traditional electronic neural networks obsolete?
Could programming the hidden statistical properties of light finally make unhackable optical encryption a reality?
How does reconstructing full-color images from a single-wavelength laser redefine high-dimensional data encoding?
1,024-Mode Polarization Modulation: Transforming Optical Computing, Telecom, and AI Acceleration
Overview
In 2026, researchers achieved a major breakthrough by introducing a 1,024-mode polarization modulation technique, overcoming the limits of traditional scalar optical processing. By unlocking the full vectorial and statistical properties of light, this method enables highly scalable and efficient optical computing, including encoding high-resolution RGB images in a single laser beam and boosting optical neural network performance. However, scaling this technology faces challenges such as fabrication imperfections that cause unwanted mode coupling and signal loss in photonic chips. Despite these hurdles, the technology shows strong potential for next-generation communications, AI hardware, and secure optical systems.