Fujian Normal University Team Encodes Data in Three Dimensions of Light, Boosting Holographic Storage Capacity With AI-Powered Decoding
Researchers combine amplitude, phase, and polarization in a single holographic data page, using a neural network to decode all three channels from intensity-only measurements.
Overview
A research team at Fujian Normal University in China has demonstrated a holographic data storage technique that encodes information using three properties of light simultaneously — amplitude, phase, and polarization — rather than the one or two dimensions used by conventional approaches. The method, published on March 26 in the journal Optica, relies on a convolutional neural network to reconstruct all three data channels from simple intensity measurements, eliminating the need for specialized detection hardware.
The work addresses a growing bottleneck in global data infrastructure. As data generation continues to outpace available storage capacity, researchers have long viewed holographic storage as a promising alternative to magnetic and solid-state media because it can record information throughout the volume of a material rather than on its surface alone. The challenge has been extracting enough usable data channels from a single holographic page to justify the added complexity.
What We Know
The team, led by professor Xiaodi Tan with co-authors Ruixian Chen, Jinyu Wang, Hao Wu, Minghui Song, Yi Yang, and Dakui Lin, refined an approach called tensor-based polarization holography. Traditional holographic storage typically encodes data using only the amplitude or phase of a light wave. By adding polarization — the orientation of the wave — as a third independent channel, the researchers substantially increased the information density of each holographic data page, according to ScienceDaily.
The encoding process uses a double-phase hologram technique with a single phase-only spatial light modulator. By adjusting the intensity and phase of two perpendicular polarization states, the system writes amplitude, phase, and polarization data into the same holographic volume, as described by EurekAlert.
Decoding posed the harder problem. Standard optical sensors can measure only intensity, not phase or polarization directly. To solve this, the researchers trained a convolutional neural network on pairs of complementary diffraction images — one captured through a vertical polarizer and one without. The neural network learned to simultaneously extract amplitude, phase, and polarization data from these intensity-only measurements, according to Interesting Engineering.
“Our results showed that multidimensional joint encoding substantially increased the information carried by a single holographic data page, thereby improving storage capacity,” Tan stated in the Optica Publishing Group press release.
What We Don’t Know
The research remains at an early laboratory stage. The team has not published specific storage density figures comparing the three-dimensional approach to existing commercial storage technologies, making it difficult to quantify the practical advantage. The researchers acknowledge in the paper that further work is needed on gray-level coding expansion, media stability, and volumetric multiplexing before the technique could be commercialized, as noted by ScienceDaily.
It is also unclear how the AI-based decoding performs at scale. Convolutional neural networks require training data that closely matches the conditions of the target system, and whether the model generalizes across different recording media, environmental conditions, and data densities has not been publicly demonstrated.
No timeline for integration into working storage devices has been offered, and the gap between laboratory proof-of-concept and a product that competes with established storage technologies — magnetic tape for archival use, SSDs for speed — typically spans years or decades.
Analysis
Holographic data storage has attracted periodic waves of research interest since the 1960s, with companies such as InPhase Technologies and Akonia Holographics attempting commercial products that ultimately failed to reach market viability. What distinguishes the Fujian Normal University work is the combination of a third encoding dimension with a machine learning decoding pipeline that sidesteps the expensive optical hardware traditionally required to read phase and polarization.
The use of a neural network as the decoder is significant because it lowers the barrier to readout. If a standard image sensor paired with a trained model can replace interferometric or polarimetric detection equipment, the cost and complexity of holographic storage systems could drop substantially. That said, neural network inference adds its own computational overhead, and the tradeoff between optical simplicity and computational cost will determine whether the approach proves practical.
Tan suggested the technology could eventually “enable smaller data centers and more efficient large-scale archival storage,” according to Interesting Engineering. If the technique scales as hoped, it would be most competitive in cold storage and archival applications where data density matters more than read/write speed — a segment currently dominated by magnetic tape, which faces its own physical density limits.