A DVD holds 4.7 GB; a single-layer Blu-ray, 25 GB. A paper published online on September 29 in *Nature Photonics* pushes the theoretical capacity of the same-size medium to the order of fifty terabytes (TB) — equivalent to about 2,000 single-layer Blu-ray discs stacked into one disc's thickness.
The work comes from the team of Academician Gu Min and Professor Zhang Qiming at the School of Intelligent Science and Technology, University of Shanghai for Science and Technology, titled "Giant phonon-assisted upconversion luminescence of in-situ photocarbonized polymer dots for three-dimensional high-definition optical data storage." It is optical storage: writing marks into a material with a laser, then reading the marks back with a laser.
Key numbers
- Each 3D storage point distinguishes 1,024 grey-scale states, carrying 10 bits of information.
- Writing uses a single monochromatic beam; reading uses a single monochromatic beam — high-speed scanning for write, wide-field parallel readout.
- Readout fidelity exceeds 99.99% under the reported experimental conditions.
- Blu-ray count: 50 TB ÷ 25 GB = 2,000 discs; 400 TB = 16,000 discs. The "equivalent to 2,000 single-layer Blu-rays" claim, consistent with 50 TB, appears in the Xinmin Evening News text.
- Order-of-magnitude claim: the other wire report says capacity exceeds Blu-ray "by more than three orders of magnitude." 2,000× ≈ 3.3 orders fits; 16,000× ≈ 4.2 orders would make the claim oddly conservative.
- Team convention: for the 2024 *Nature* work, CAS's news release converted 1.6 Pb to "~8,000 commercial discs or ~100 standard hard drives," i.e., 200 TB — showing the team's "Pb" means petabit, so 0.4 Pb should be 50 TB, not 400 TB.
- Single-beam read/write works, so no nanoscale co-alignment of two beams is needed — devices can be simpler and smaller.
- The 1,024-state discrimination depends on SNR; 99.99% fidelity means roughly one in ten thousand errors, with error-correction coding still needed on top.
- Optical media store data without power, targeting cold-data scenarios: AI training data, research archives, medical imaging.
For context: in conventional binary storage, one cell holds one bit (two states). An 8-bit grayscale image pixel has 256 shades and needs 8 binary cells. Here, one point stores 10 bits — enough for a full pixel with headroom. As Zhang Qiming put it: "Information that previously required multiple recording cells can now be carried by raising the capacity of a single point."
Two paths to higher capacity
Optical disc capacity has historically grown by shrinking the mark: shorter-wavelength lasers, stronger lenses, marks going from micron scale into the diffraction limit. A second path records in multiple layers inside the material.
Two years ago, Gu Min's team took the first path: dual-beam super-resolution (STED-style), marks of about 54 nm, 100 layers inside the material, 1.6 Pb theoretical capacity — published in *Nature* in February 2024. The cost soon showed: smaller marks and more layers demand extreme laser control and device precision, and point-by-point read/write limits throughput regardless of capacity.
The new work changes strategy: each point stores 10 bits directly, with one beam each for read and write — subtracting system complexity while adding throughput.
| Dimension | 2024 dual-beam (*Nature*) | New single-beam (*Nature Photonics*) | |---|---|---| | Read/write beams | Two coordinated beams | One monochromatic beam each | | Mark size | ~54 nm | Not disclosed on comparable terms | | Per-point information | Binary (1-bit class) | 1,024 states / 10 bits | | Capacity figure | 1.6 Pb (≈ 200 TB) | See below | | Readout | Point-by-point super-resolution | Wide-field parallel |
An eight-fold discrepancy in press coverage
Chinese coverage of the same result gave two capacity figures:
| Source | Statement | Conversion | |---|---|---| | Xinmin Evening News | ~400 terabits, i.e., 50 terabytes | 400 Tb ÷ 8 = 50 TB | | China News Service/Jingwei | ~0.4 Pb (noted as ~400 TB) | would be 400 TB |
The numbers differ by exactly 8 — the bit/byte conversion factor, a classic unit mix-up. Three consistency checks point to the 400 Tb = 50 TB side:
> Note: the original paper could not be directly accessed, so this judgment rests on cross-source inference.
Notably, 50 TB does not exceed the team's own 2024 figure of 200 TB. The paper's claim lies in the architecture: single beam instead of dual beams, 10 bits per point instead of 1, wide-field parallel instead of point-by-point readout — the three factors that matter most when moving from a lab-scale storage point to a whole-disc device.
Material and mechanism
The core is the material: in-situ photocarbonized polymer dots, exploiting "giant phonon-assisted upconversion luminescence" for writing and reading.
> Upconversion luminescence means the material emits higher-energy photons after absorbing lower-energy ones — the reverse of normal fluorescence. "Phonon-assisted" means lattice vibrations participate in this energy climb; "giant" describes the effect's strength. During read/write, light emitted by the illuminated point carries the grey-scale information; the 1,024 distinguishable states map onto spectral or intensity dimensions.
Engineering implications:
From storage point to whole disc: speed is the bottleneck
Gu Min was direct about the remaining bottleneck: read/write speed. "Once storage capacity increases, we still need to write and read large amounts of data within a time acceptable for real applications." The team is pursuing optical GPT (OGPT) research — using optical processing to improve data handling and I/O efficiency while preserving fidelity and stability — and industrialization efforts have begun.
| Remaining problem | Status | Direction | |---|---|---| | Write throughput | High-speed scanning write validated; whole-disc write undisclosed | Optical parallel processing (OGPT) | | Read throughput | Wide-field parallel read validated | Same | | State stability | 99.99% fidelity in lab conditions | Aging and whole-disc consistency untested | | Error correction | Not disclosed in coverage | 1,024 states need multi-dimensional ECC |
The 2024 *Nature* work proved petabit-class capacity is physically reachable; this *Nature Photonics* work answers "how many steps remain to engineering." What's left is throughput — and the team's bet is optical computing. There is no timeline yet for a shelf-ready product; "advancing industrialization efforts" is itself an honest progress marker.
---
Sources and caveats: Paper online in *Nature Photonics*, 2026-09-29, Gu Min and Zhang Qiming's team, University of Shanghai for Science and Technology. Details cross-checked across Xinmin Evening News, Shangguan News, and CNS/Jingwei; the capacity discrepancy was identified in this post's fact-check. Comparison work: "A 3D nanoscale optical disk memory with petabit capacity," *Nature*, Feb 2024; the 1.6 Pb = 200 TB conversion follows the CAS news release. The original *Nature Photonics* text was not accessible; the capacity judgment (400 Tb = 50 TB side) is inference-based.