Tuesday, November 17, 2026
Science
No Result
View All Result
  • Login
  • HOME
  • SCIENCE NEWS
  • CONTACT US
  • HOME
  • SCIENCE NEWS
  • CONTACT US
No Result
View All Result
Scienmag
No Result
View All Result
Home Science News

Liquid Crystal Photonic Processor Harnesses Light-Driven Molecules for In-Memory Computing

November 17, 2026
in Science News
Cassandra Pierce
By Cassandra Pierce Scienmag Editorial Profile - Systems Neuroscience
Reading Time: 6 mins read
0
Liquid Crystal Photonic Processor Harnesses Light-Driven Molecules for In-Memory Computing

Liquid Crystal Photonic Processor Harnesses Light-Driven Molecules for In-Memory Computing

65
SHARES
587
VIEWS
Share on FacebookShare on Twitter
ADVERTISEMENT

A research team has demonstrated a large-scale photonic in-memory computing architecture that exploits light-driven molecular reorientation in photo-rewritable liquid crystals to merge sensing, data storage, and optical processing. Published in eLight, the experimental platform implements a liquid-crystal-enabled vectorial diffractive neural network whose physical operations are mapped onto interconnected Poincaré spheres. By leveraging nonvolatile molecular alignments, the platform achieves megabyte-level optical memory that functions simultaneously as an array of reconfigurable artificial neurons. Across two liquid crystal layers, the prototype integrates up to 9.44 million reconfigurable neurons with eight-bit weight precision, providing an aggregate memory capacity of 151.04 megabits. This optical storage capacity represents a 100,000-fold expansion over previous photonic in-memory hardware implementations, offering an energy-efficient, massively parallel route toward physical neural networks and brain-inspired machine intelligence.

The development addresses physical bottlenecks inherent in conventional computing hardware based on the standard von Neumann architecture. In traditional electronic processors, arithmetic computation and memory storage are physically partitioned, requiring continuous, energy-draining data shuttling across bus lines that severely constrains processing throughput. Emerging electronic in-memory and in-sensor concepts attempt to mitigate these transfers, but they face operational penalties from analog-to-digital conversions, parasitic voltage drops in crossbar arrays, and device variability. Photonic computing offers a compelling alternative, leveraging the high carrier frequencies and intrinsic wave parallelism of light to execute computations with picosecond latencies. However, previous optical computing designs have operated predominantly as accelerators that still depend on external electronic memory and frequent electro-optic conversions for weight programming. By establishing nonvolatile optical storage directly within the diffractive medium, the liquid crystal architecture minimizes data-movement overheads across diverse computing tasks.

At the heart of the system is the reversible, light-guided realignment of anisotropic organic molecules that function as nonvolatile optical memory cells. The researchers fabricated thin-film devices using a dual-material system composed of a photosensitive sulfonic azo-dye known as SD1 and a nematic liquid crystal designated 5CB. When exposed to linearly polarized ultraviolet or blue illumination within the dye absorption band, SD1 molecules undergo photoisomerization and dichroic absorption cycles. These photodynamic cycles cause the absorption oscillators of the dye molecules to reorient until they settle perpendicular to the incident polarization vector. Through anisotropic intermolecular interactions at the substrate interface, the aligned SD1 layer forces adjacent 5CB liquid crystal molecules to adopt the exact same orientation. Because this molecular arrangement remains locked after illumination ceases, the device establishes a stable, nonvolatile optical pattern that persists without consuming electrical power during passive computation.

To exploit these molecular reorientations mathematically, the platform maps light modulation onto the geometric surface of the Poincaré sphere, which visualizes arbitrary optical polarization states. At the input and output planes of the diffractive network, local polarization states correspond to discrete coordinate points on the sphere. Inside the network hidden layers, space-variant polarization transformations induced by the liquid crystals are represented as continuous geometric trajectories traversing the sphere surface. Polarization modulation of each spatial neuron is governed by a two-by-two Jones matrix parameterized by in-plane molecular orientation angle alpha and out-of-plane molecular tilt angle theta. By tuning the optical phase retardation of the cell to satisfy the first-order half-wave condition, the Jones matrix depends strictly on the in-plane orientation angle alpha. Consequently, rewriting the molecular orientation directly rotates the transformation axis on the Poincaré sphere, altering the physical trajectory of propagating light waves.

Interconnections between successive liquid crystal layers are formed through free-space diffraction of the vectorial optical field across amplitude, phase, and polarization dimensions. The authors modeled this propagation by treating orthogonal horizontal and vertical polarization components as independent wave fields governed by Rayleigh-Sommerfeld diffraction theory and the angular spectrum method. In the forward propagation model, each neuron acts as a secondary wave source transmitting across an axial diffraction distance of ten centimeters. To train the network, the research team used gradient-based error backpropagation implemented in Python and TensorFlow on a computing server equipped with a graphics processing unit. Optimization was conducted over individual diffractive units measuring 256 by 256 pixels across an aperture of 4.19 square millimeters. By iteratively calculating gradients against task-specific loss functions, the training algorithm updated the spatial alpha distributions, establishing the optical weights required for complex inference.

The team validated the computational capability of the Poincaré-sphere-connected diffractive architecture through an intelligent polarization encoding task using handwritten digits from the MNIST dataset. Images of handwritten digits representing numerals one through four were assigned to four distinct target polarization coordinates on the Poincaré sphere. When horizontally polarized light corresponding to an input digit traversed the trained diffractive liquid crystal layer, diffraction guided the beam toward a centralized detection zone while rotating its polarization vector along the planned trajectory. Testing across 200 blind handwritten digit samples revealed clear separation in polarization space, yielding a single-model classification accuracy of 94.5 percent. To boost classification robustness, the researchers reconfigured the same physical device into a complementary secondary distribution called alpha-2. Through joint inference and optical ensemble learning combining predictions from alpha-1 and alpha-2, the network corrected individual misclassifications, elevating overall experimental accuracy to 96.5 percent without adding physical layers.

Demonstrating functional reconfigurability, the liquid crystal device was subsequently reprogrammed into a third configuration, designated alpha-3, to execute Poincaré-sphere-perceptive computing that simultaneously senses and classifies polarized input data. In this operational mode, nine image categories spanning three polarization states were mapped into three spatial detection zones monitored through a fixed horizontal analyzer. The network achieved polarization-multiplexed execution of three distinct machine learning problems—classifying letters from E-MNIST, fashion articles from Fashion-MNIST, and digits from MNIST—yielding standalone experimental accuracies of 88 percent, 92.5 percent, and 93 percent, respectively. To compensate for minor experimental aberrations, the researchers introduced a shallow hybrid optoelectronic architecture containing a single fully connected layer with twelve learnable parameters and a Softmax activation. Although this electronic stage contributed only 0.05 percent of total trainable parameters, it raised the final experimental classification accuracies across the three multiplexed tasks to 91 percent, 95 percent, and 95.5 percent.

The team extended this perceptive capability to the automated identification of polarized physical objects by assessing their composite polarimetric signatures. In an experimental evaluation involving 600 test polarization features across three object classes, the diffractive neural network converged matching features into designated detection areas, achieving a classification accuracy of 91.2 percent. Building on this principle, the authors designed a specialized convolutional vectorial diffractive neural network tailored for all-optical biological sample analysis. The system featured a 4-f optical arrangement integrating a static liquid crystal convolutional layer containing a 768-by-768 spiral-phase pattern at the Fourier plane and a trainable 256-by-256 diffractive classifier at the output plane. When imaging real-world botanical specimens, the spiral-phase plate converted edge orientations into distinct polarization states, allowing the system to differentiate onion epidermal cells exhibiting dominant vertical cell walls from cauliflower cells displaying omnidirectional boundaries directly in the optical domain without electronic digitization.

Device fabrication and testing were carried out using a unified, dual-functional optical apparatus that eliminated physical alignment steps during memory rewriting and computational inference. Indium-tin-oxide glass substrates were UV-ozone treated and spin-coated with a 0.35 weight percent solution of SD1 dissolved in dimethylformamide, cured at 100 degrees Celsius for ten minutes, and assembled with seven-micrometer spacers before capillary filling with nematic 5CB at 55 degrees Celsius. In the shared experimental setup, a 450-nanometer blue laser and a 632.8-nanometer red laser were co-aligned and expanded. An arbitrary linear polarization profile was generated by sandwiching a reflective spatial light modulator with eight-micrometer pixel pitch between two achromatic quarter-wave plates. The generated pattern was projected onto the liquid crystal cell via a 4-f lens relay. By defining the first diffractive propagation distance as zero, optical reconfiguration and red-laser computing were conducted in situ on the exact same device, completely bypassing mechanical realignments.

To demonstrate memory scalability and cascade operation, the researchers scaled the architecture into a multi-core processor array and recorded optical computational outputs into secondary liquid crystal memory cells. The team integrated a 12-by-12 array containing 144 identical vectorial diffractive units, comprising 9.44 million reconfigurable neurons operating across 576 parallel polarization encoding channels. Using a three-by-two subarray programmed under configuration alpha-5, the platform processed the six-letter text sequence “bad cab” in parallel, mapping each letter to a four-spot intensity array where the correct letter corresponded to vertical polarization. These blue-light output patterns were directly projected onto a second liquid crystal cell, physically recording inference results into localized horizontal molecular alignments. The recorded data exhibited a natural sigmoid-shaped exposure curve advantageous for optical activation functions, and the stored results were nondestructively reconstructed at the speed of light under crossed polarizers, providing a viable pathway for feeding outputs into downstream optical computing stages.

Comprehensive performance assessments highlight significant operational strengths alongside specific technical trade-offs inherent to molecular liquid crystal processors. Because the molecular reorientations are nonvolatile, the system draws negligible static electrical power during operation, requiring only an initial low-voltage one-kilohertz signal to sustain the half-wave retardance condition. The optical device demonstrated a high optical transmission efficiency of 92.5 percent, corresponding to an insertion loss of just 0.34 decibels. Based on its parallel execution of approximately 2.72 billion real-valued multiply-accumulate operations per cycle, the platform attains an effective computational throughput of 5.44 x 10^16 floating-point operations per second when coupled with ten-megahertz photodetectors. Memory reconfiguration speed remains bounded by optical exposure kinetics: writing required 60 seconds using a low-power 2.05-milliwatt-per-square-centimeter blue laser combined with ultraviolet light-emitting diode erasing and a two-megahertz radio-frequency signal. However, tests confirmed that higher-intensity 405-nanometer continuous illumination shortened writing times to 100 milliseconds, while ultrahigh-peak-power femtosecond laser pulses reduced rewriting to three nanoseconds.

Subject of Research: Liquid-crystal-based photonic in-memory computing using light-driven molecular reorientation

Article Title: Light-driven molecular reorientation for large-scale photonic in-memory computing

Article References: Liu, S., Hu, X., Chen, P., Fang, X., Chen, F., Li, K., Sun, R., Ding, S., Zhang, Y., Li, K., Li, B., Ge, S.-J., Wang, F., Gu, M., & Lu, Y.-Q. (2026). Light-driven molecular reorientation for large-scale photonic in-memory computing. eLight, 6(1), Article 26. https://doi.org/10.1186/s43593-026-00141-0

Image Credits: AI Generated

DOI: 10.1186/s43593-026-00141-0

Keywords: photonic in-memory computing, liquid crystals, molecular reorientation, diffractive neural networks, Poincaré sphere, vectorial light fields, optical neural networks, nonvolatile photonic memory, azo-dye photoalignment, parallel optical computing, Jones matrix method, optical convolution, polarization multiplexing, 5CB nematic liquid crystal, spatial light modulators, electrothermal effect, optical pattern recognition

Cite Scienmag News

Cassandra Pierce. (November 17, 2026). Liquid Crystal Photonic Processor Harnesses Light-Driven Molecules for In-Memory Computing. Scienmag. https://scienmag.com/liquid-crystal-photonic-processor-harnesses-light-driven-molecules-for-in-memory-computing/

Cassandra Pierce. "Liquid Crystal Photonic Processor Harnesses Light-Driven Molecules for In-Memory Computing." Scienmag, 17 November 2026, https://scienmag.com/liquid-crystal-photonic-processor-harnesses-light-driven-molecules-for-in-memory-computing/. Accessed 17 November 2026.

Cassandra Pierce. "Liquid Crystal Photonic Processor Harnesses Light-Driven Molecules for In-Memory Computing." Scienmag. November 17, 2026. https://scienmag.com/liquid-crystal-photonic-processor-harnesses-light-driven-molecules-for-in-memory-computing/

Tags: 5CB nematic liquid crystalazo-dye photoalignmentdiffractive neural networkselectrothermal effectJones matrix methodliquid crystalsmolecular reorientationnonvolatile photonic memoryoptical convolutionOptical Neural Networksoptical pattern recognitionparallel optical computingphotonic in-memory computingPoincaré spherepolarization multiplexingSpatial Light Modulatorsvectorial light fields
Share26Tweet16
Previous Post

Study Uncovers Severe Social and Economic Strains Among Displaced Bangladesh River Communities

Next Post

Gut Microbiota and Metabolites Influence Immunotherapy Response in Liver Cancer

Related Posts

Researchers Validate Odia Version of Cancer Financial Toxicity Assessment Tool
Science News

Researchers Validate Odia Version of Cancer Financial Toxicity Assessment Tool

November 17, 2026
Fusobacterium nucleatum Extracellular Vesicles Trigger Neuroinflammation and Anxiety-Like Behaviors via IL-1β
Medicine

Fusobacterium nucleatum Extracellular Vesicles Trigger Neuroinflammation and Anxiety-Like Behaviors via IL-1β

November 17, 2026
Parametric Voronoi Models Generate Synthetic Defect Geometries for Cast Metal
Mathematics

Parametric Voronoi Models Generate Synthetic Defect Geometries for Cast Metal

November 17, 2026
Experiential Workshop Expands Mexican Students’ Understanding of Bat Diets and Ecological Roles
Biology

Experiential Workshop Expands Mexican Students’ Understanding of Bat Diets and Ecological Roles

November 17, 2026
New Framework Aligns Research Questions With Randomized Implementation Trial Designs
Science News

New Framework Aligns Research Questions With Randomized Implementation Trial Designs

November 17, 2026
Immune Cell Exhaustion May Drive Neurodegeneration in Alzheimer’s and Parkinson’s
Medicine

Immune Cell Exhaustion May Drive Neurodegeneration in Alzheimer’s and Parkinson’s

November 17, 2026
Next Post
Gut Microbiota and Metabolites Influence Immunotherapy Response in Liver Cancer

Gut Microbiota and Metabolites Influence Immunotherapy Response in Liver Cancer

  • Mothers who receive childcare support from maternal grandparents show more optimized

    Mothers who receive childcare support from maternal grandparents show more parental warmth, finds NTU Singapore study

    27656 shares
    Share 11059 Tweet 6912
  • University of Seville Breaks 120-Year-Old Mystery, Revises a Key Einstein Concept

    1061 shares
    Share 424 Tweet 265
  • Bee body mass, pathogens and local climate influence heat tolerance

    682 shares
    Share 273 Tweet 171
  • Researchers record first-ever images and data of a shark experiencing a boat strike

    546 shares
    Share 218 Tweet 137
  • Groundbreaking Clinical Trial Reveals Lubiprostone Enhances Kidney Function

    531 shares
    Share 212 Tweet 133
Science

Embark on a thrilling journey of discovery with Scienmag.com—your ultimate source for cutting-edge breakthroughs. Immerse yourself in a world where curiosity knows no limits and tomorrow’s possibilities become today’s reality!

RECENT NEWS

  • Researchers Validate Odia Version of Cancer Financial Toxicity Assessment Tool
  • Fusobacterium nucleatum Extracellular Vesicles Trigger Neuroinflammation and Anxiety-Like Behaviors via IL-1β
  • Parametric Voronoi Models Generate Synthetic Defect Geometries for Cast Metal
  • Experiential Workshop Expands Mexican Students’ Understanding of Bat Diets and Ecological Roles

Categories

  • Agriculture
  • Anthropology
  • Archaeology
  • Athmospheric
  • Biology
  • Biotechnology
  • Blog
  • Bussines
  • Cancer
  • Chemistry
  • Climate
  • Earth Science
  • Editorial Policy
  • Marine
  • Mathematics
  • Medicine
  • Pediatry
  • Policy
  • Psychology & Psychiatry
  • Science Education
  • Science News
  • Social Science
  • Space
  • Technology and Engineering

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 5,150 other subscribers

© 2026 Scienmag - Science Magazine

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • HOME
  • SCIENCE NEWS
  • CONTACT US

© 2026 Scienmag - Science Magazine

Discover more from Science

Subscribe now to keep reading and get access to the full archive.

Continue reading