Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 34 of the 34 references it cites were found in a registry or an index; 18 of the 72 sources its claims cite were checked for support.

Laser writing in glass for dense, fast and efficient archival data storage

Open access

Abstract

Long-term preservation of digital information is vital for safeguarding the knowledge of humanity for future generations. Existing archival storage solutions, such as magnetic tapes and hard disk drives, suffer from limited media lifespans that render them unsuitable for long-term data retention1–3. Optical storage approaches, particularly laser writing in robust media such as glass, have emerged as promising alternatives with the potential for increased longevity. Previous work4–16 has predominantly optimized individual aspects such as data density but has not demonstrated an end-to-end system, including writing, storing and retrieving information. Here we report an optical archival storage technology based on femtosecond laser direct writing in glass that addresses the practical demands of archival storage, which we call Silica. We achieve a data density of 1.59 Gbit mm−3 in 301 layers for a capacity of 4.8 TB in a 120 mm square, 2 mm thick piece of glass. The demonstrated write regimes enable a write throughput of 25.6 Mbit s−1 per beam, limited by the laser repetition rate, with an energy efficiency of 10.1 nJ per bit. Moreover, we extend the storage ability to borosilicate glass, offering a lower-cost medium and reduced writing and reading complexity. Accelerated ageing tests on written voxels in borosilicate suggest data lifetimes exceeding 10,000 years. An optical archival storage technology based on femtosecond laser direct writing in glass addresses the practical demands of archival storage.

Overview

CoChat Summary

## TL;DR

Silica is an end-to-end archival storage system based on femtosecond laser writing in glass that writes, reads, and decodes data with fully automated hardware and machine-learning-based decoding. The paper reports up to 1.59 Gbit mm−3 density, 25.6 Mbit s−1 write throughput per beam, 10.1 nJ per bit write efficiency, and an extrapolated data lifetime exceeding 10,000 years.

## Key findings

  • Using birefringent voxels in fused silica, the system achieves 1.59 Gbit mm−3 data density, 4.84 TB per platter usable capacity, 301 layers, 0.500 μm × 0.485 μm voxel pitch, 25.6 Mbit s−1 write throughput, and 10.1 nJ per bit write efficiency.
  • Using phase voxels in borosilicate, the system achieves 0.678 Gbit mm−3, 2.02 TB per platter, 258 layers, 0.5 μm × 0.7 μm voxel pitch, 18.4 Mbit s−1 throughput, and 8.85 nJ per bit write efficiency.
  • A four-beam multibeam writing setup reaches 65.9 Mbit s−1 without thermal damage; the authors state thermal simulations suggest 16 or more beams should be possible.
  • Accelerated ageing of phase voxels suggests modification lifetimes exceeding 10,000 years at 290 °C, implying even longer at room temperature.

## Methods

  • End-to-end archival system built around femtosecond laser direct writing in glass, with two voxel regimes: birefringent voxels in fused silica and phase voxels in borosilicate.
  • Writing uses a 10 MHz laser and polygon scanning; birefringent writing uses pseudo-single-pulse seed/data pulses and polarization modulation, while phase writing uses single-pulse amplitude modulation and a multibeam variant with four independently modulated beams.
  • Reading uses wide-field microscopy: polarization-resolved imaging for birefringent voxels and Zernike phase-contrast microscopy for phase voxels.
  • Decoding uses a CNN plus LDPC and erasure coding; the paper also optimizes symbol sets and modulation levels using mutual information and Pareto-front style trade-offs.

## Limitations

  • The lifetime test for phase voxels reflects thermal stability under isolated conditions and “does not account for external influences, such as mechanical stress or chemical corrosion” (stated in the paper).
  • For phase voxels, Zernike phase contrast has inherently poor optical sectioning, increasing axial cross-talk; the authors mitigate this with dual-image acquisition (stated in the paper).
  • The paper notes that read throughput was not considered because it is “not a significant cost component” of the system (text makes evident that read-speed performance was not fully evaluated here).

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41586-025-10042-w

Cites

34 of 34 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 The system of Silica. We compare Silica with other storage technologies in the Supplementary Information. a, Data are received from a user. These are prepared as a stream of bits, for example, using compression, encryption and FEC. b, The bits are encoded as symbols. One symbol corresponds to a modulator configuration. c,d, The glass sample is loaded into a write subsystem and the modulator settings are changed in time as the laser beam is moved relative to the glass. The symbols are written layer by layer, from the bottom up, to fill the full thickness of the glass. e, The data can be stored securely in the glass for more than 10,000 years. f, To read, we use an automated microscope with a camera to capture images of each 2D layer of voxels. g, Images are passed to a decoder to recover user data. Source Data
Fig. 1 The system of Silica. We compare Silica with other storage technologies in the Supplementary Information. a, Data are received from a user. These are prepared as a stream of bits, for example, using compression, encryption and FEC. b, The bits are encoded as symbols. One symbol corresponds to a modulator configuration. c,d, The glass sample is loaded into a write subsystem and the modulator settings are changed in time as the laser beam is moved relative to the glass. The symbols are written layer by layer, from the bottom up, to fill the full thickness of the glass. e, The data can be stored securely in the glass for more than 10,000 years. f, To read, we use an automated microscope with a camera to capture images of each 2D layer of voxels. g, Images are passed to a decoder to recover user data. Source DataCC BY-NC-ND · PubMed Central
Fig. 2 Schematics of high-throughput write. a,b, Setup (a) and modulation methods (b) for birefringent and phase voxels. c, Emissions monitoring to enable consistent data writing. d, Schematic of pseudo-single-pulse writing of birefringent voxels and scanning electron microscopy (SEM) top-view image of elongated nanovoids in fused silica, e,f, Schematics of single beam (e) and multibeam (f) single-pulse writing of phase voxels. In e, the SEM and phase-contrast microscopy (PCM) images show top-view and side-view profiles, respectively, of the different phase voxel symbols written in borosilicate glass. EOM, electro-optic modulator; AOM, acousto-optic modulator; RF, radiofrequency. a.u., arbitrary units. Scale bars, 200 nm (d, top); 2 μm (d, bottom; e, middle and bottom; f, bottom); 500 nm (e, top). Source Data
Fig. 2 Schematics of high-throughput write. a,b, Setup (a) and modulation methods (b) for birefringent and phase voxels. c, Emissions monitoring to enable consistent data writing. d, Schematic of pseudo-single-pulse writing of birefringent voxels and scanning electron microscopy (SEM) top-view image of elongated nanovoids in fused silica, e,f, Schematics of single beam (e) and multibeam (f) single-pulse writing of phase voxels. In e, the SEM and phase-contrast microscopy (PCM) images show top-view and side-view profiles, respectively, of the different phase voxel symbols written in borosilicate glass. EOM, electro-optic modulator; AOM, acousto-optic modulator; RF, radiofrequency. a.u., arbitrary units. Scale bars, 200 nm (d, top); 2 μm (d, bottom; e, middle and bottom; f, bottom); 500 nm (e, top). Source DataCC BY-NC-ND · PubMed Central
Fig. 3 Read and decode. a, Schematic of birefringent voxel readout using polarization-resolved microscopy. b, Schematic of phase voxels readout using phase-contrast microscopy. c, Decode pipeline showing images being decoded into recovered sectors. d, LDPC curves showing the trade-off between code rate (horizontal axis) and quality factor (vertical axis) for birefringent and phase voxels, written under state-of-the-art conditions detailed in section ‘Silica system analysis’. e, Example heatmap of code rates required to recover the original data encoded in phase voxels across layers (vertical axis) and tracks (horizontal axis). The data indicate the code rate is mostly dependent on depth in the glass rather than xy position. OL, objective lens. Source Data
Fig. 3 Read and decode. a, Schematic of birefringent voxel readout using polarization-resolved microscopy. b, Schematic of phase voxels readout using phase-contrast microscopy. c, Decode pipeline showing images being decoded into recovered sectors. d, LDPC curves showing the trade-off between code rate (horizontal axis) and quality factor (vertical axis) for birefringent and phase voxels, written under state-of-the-art conditions detailed in section ‘Silica system analysis’. e, Example heatmap of code rates required to recover the original data encoded in phase voxels across layers (vertical axis) and tracks (horizontal axis). The data indicate the code rate is mostly dependent on depth in the glass rather than xy position. OL, objective lens. Source DataCC BY-NC-ND · PubMed Central
Fig. 4 System analysis. a, Plot of density and voxel quality for different xyz phase voxel pitches (in μm). b, Example response of a phase voxel channel shown as predicted modulation compared with input modulation. The colour scale shows the model’s estimate of the likelihood of each output value given the input. c, Finding the best symbol count and write target emission for a given voxel pitch. d, Quality factor and density for repeated reads and writes showing the robustness of Silica results. Different colours correspond to different write repetitions of the same data. e, Arrhenius plot of decay time against temperature (inset actual, outer extrapolated) suggesting a voxel lifetime that exceeds 10,000 years at 290 °C. a.u., arbitrary units. Source Data
Fig. 4 System analysis. a, Plot of density and voxel quality for different xyz phase voxel pitches (in μm). b, Example response of a phase voxel channel shown as predicted modulation compared with input modulation. The colour scale shows the model’s estimate of the likelihood of each output value given the input. c, Finding the best symbol count and write target emission for a given voxel pitch. d, Quality factor and density for repeated reads and writes showing the robustness of Silica results. Different colours correspond to different write repetitions of the same data. e, Arrhenius plot of decay time against temperature (inset actual, outer extrapolated) suggesting a voxel lifetime that exceeds 10,000 years at 290 °C. a.u., arbitrary units. Source DataCC BY-NC-ND · PubMed Central
Extended Data Fig. 1 Continuous writing. (a) Schematic of voxel writing geometry. The pulse train from the laser is gated such that isolated voxels are formed from pristine material. (b) Azimuth (left) and retardance (right) images of experimentally written voxels. Voxels spacing is 2 μm. (c) Schematic of continuous line writing geometry. The pulse train from the laser is not gated and instead a continuous line is written. Here, previously modified material acts as a seed amplifying the retardance signal. (d) Azimuth (left) and retardance (right) images of experimentally written lines. Line spacing is 2 μm. Note that the birefringence signal is far stronger when compared with isolated voxels. (e) Experimental azimuth image (left) and a 1D slice through a section of scanline (right). Line spacing and voxel pitch is 1 μm. In this demonstration, the laser repetition rate is 1 MHz and there are 10 pulses per voxel. (Time per voxel is 10 μs with a stage speed of 100 mm/s). Sharp transitions between symbols (<300 nm) suggest that the material response is a relatively fast process. (f) Results of an x pitch sweep for a single layer. The voxel pitch in the y dimension is fixed at 0.8 μm, while the voxel pitch in the x dimension is swept from 100 nm to 300 nm. Calculation of retardance and azimuth from raw images shows the distinct azimuth levels. This allows us to estimate the number bits per voxel. (g) Decode analysis on a sample produced by continuous writing. A data density of 2.136 Gbit/mm3 (5.1 TB/platter usable capacity) was achieved. (h) (Left) Experimental setup showing 7-beam splitting, modulation, and recombination components used for parallelized continuous writing. (Right) Example of a 75 mm × 75 mm × 2 mm sample with 7 written data bands.
Extended Data Fig. 1 Continuous writing. (a) Schematic of voxel writing geometry. The pulse train from the laser is gated such that isolated voxels are formed from pristine material. (b) Azimuth (left) and retardance (right) images of experimentally written voxels. Voxels spacing is 2 μm. (c) Schematic of continuous line writing geometry. The pulse train from the laser is not gated and instead a continuous line is written. Here, previously modified material acts as a seed amplifying the retardance signal. (d) Azimuth (left) and retardance (right) images of experimentally written lines. Line spacing is 2 μm. Note that the birefringence signal is far stronger when compared with isolated voxels. (e) Experimental azimuth image (left) and a 1D slice through a section of scanline (right). Line spacing and voxel pitch is 1 μm. In this demonstration, the laser repetition rate is 1 MHz and there are 10 pulses per voxel. (Time per voxel is 10 μs with a stage speed of 100 mm/s). Sharp transitions between symbols (<300 nm) suggest that the material response is a relatively fast process. (f) Results of an x pitch sweep for a single layer. The voxel pitch in the y dimension is fixed at 0.8 μm, while the voxel pitch in the x dimension is swept from 100 nm to 300 nm. Calculation of retardance and azimuth from raw images shows the distinct azimuth levels. This allows us to estimate the number bits per voxel. (g) Decode analysis on a sample produced by continuous writing. A data density of 2.136 Gbit/mm3 (5.1 TB/platter usable capacity) was achieved. (h) (Left) Experimental setup showing 7-beam splitting, modulation, and recombination components used for parallelized continuous writing. (Right) Example of a 75 mm × 75 mm × 2 mm sample with 7 written data bands.CC BY-NC-ND · PubMed Central
Extended Data Fig. 2 Birefringent voxel writing. (a) Schematics of the pseudo-single-pulse writing of birefringent voxels. The single column of data voxels is formed in the y direction. Data modifications are shown in different colours to represent different azimuths. (b) Birefringent voxel azimuth dependence on the polarization of the seed and data pulses demonstrated under cold writing conditions. (c) The effect of distance (walk-off) between the seed modification and data pulses on the quality of the birefringent voxel formation: (i) under “cold writing” conditions, when the seed modification is formed in advance and then the data pulses are delivered to the same location, (ii) under pseudo-single-pulse writing conditions. (d) Birefringent voxels recorded inside the silica glass media using a pseudo-single-pulse writing technique. (i) Retardance-azimuth image and (ii) extracted 8 azimuth values plotted in polar coordinates.
Extended Data Fig. 2 Birefringent voxel writing. (a) Schematics of the pseudo-single-pulse writing of birefringent voxels. The single column of data voxels is formed in the y direction. Data modifications are shown in different colours to represent different azimuths. (b) Birefringent voxel azimuth dependence on the polarization of the seed and data pulses demonstrated under cold writing conditions. (c) The effect of distance (walk-off) between the seed modification and data pulses on the quality of the birefringent voxel formation: (i) under “cold writing” conditions, when the seed modification is formed in advance and then the data pulses are delivered to the same location, (ii) under pseudo-single-pulse writing conditions. (d) Birefringent voxels recorded inside the silica glass media using a pseudo-single-pulse writing technique. (i) Retardance-azimuth image and (ii) extracted 8 azimuth values plotted in polar coordinates.CC BY-NC-ND · PubMed Central
Extended Data Fig. 3 Phase voxel writing. (a) Schematics of (i) a laser pulse focused inside the glass medium, and (ii) the laser-induced RIchange in the excitation region. (b) Schematics of the amplitude multiplexed data writing: different symbols are represented/written by different amplitudes of the laser pulses. (c) Example of phase-contrast microscopy images of 2-bit data (4 modification levels) written in borosilicate glass taken at two different depths of a single voxel. (i) Image of the positive contrast of a laser induced RImodification. (ii) Negative contrast image of the voxel obtained at a slightly different depth. (iii) Difference of positive and negative contrast images of the voxel with increased signal to noise ratio. (d) Example histogram of the normalized voxel intensity distribution showing the separation of four brightness levels. (e) Write (left) and read (right) research instruments.
Extended Data Fig. 3 Phase voxel writing. (a) Schematics of (i) a laser pulse focused inside the glass medium, and (ii) the laser-induced RIchange in the excitation region. (b) Schematics of the amplitude multiplexed data writing: different symbols are represented/written by different amplitudes of the laser pulses. (c) Example of phase-contrast microscopy images of 2-bit data (4 modification levels) written in borosilicate glass taken at two different depths of a single voxel. (i) Image of the positive contrast of a laser induced RImodification. (ii) Negative contrast image of the voxel obtained at a slightly different depth. (iii) Difference of positive and negative contrast images of the voxel with increased signal to noise ratio. (d) Example histogram of the normalized voxel intensity distribution showing the separation of four brightness levels. (e) Write (left) and read (right) research instruments.CC BY-NC-ND · PubMed Central
Extended Data Fig. 4 Multibeam writing. (a) Schematic method of parallel writing with four scanned beams: (i) side view and (ii) top view. (iii) Experimental setup of the parallel writing system. HWP: half-wave plate, PBS: polarization beam splitter, AOM: acousto-optic modulator. (iv) Schematic illustrates the mechanism of the beam position controller. (b) Phase contrast images of voxels written in glass by 10 MHz femtosecond laser pulses. (i), (ii), (iii) with 4 beams, and (iv) single beam. (c) (i) One histogram plot of voxel intensities from data in (b). The blue line with dot represents the histogram of the image intensity at the region where no voxel was formed (symbol 0), and the red line with dot represents those at the region where voxels were formed (symbol 1). (ii) The standard deviation of the voxel intensities plotted against pulse energy per beam. (iii) The energy window, evaluated from the plots in (ii), against the number of beams for writing. (d) Quality factor plotted against the code rate for (i) four-beam and (ii) single-beam writing. The emission target was set to 3000 counts per beam. The pulse energy to get the emission target was 17.6 nJ for four-beam and 19.5 nJ for single-beam.
Extended Data Fig. 4 Multibeam writing. (a) Schematic method of parallel writing with four scanned beams: (i) side view and (ii) top view. (iii) Experimental setup of the parallel writing system. HWP: half-wave plate, PBS: polarization beam splitter, AOM: acousto-optic modulator. (iv) Schematic illustrates the mechanism of the beam position controller. (b) Phase contrast images of voxels written in glass by 10 MHz femtosecond laser pulses. (i), (ii), (iii) with 4 beams, and (iv) single beam. (c) (i) One histogram plot of voxel intensities from data in (b). The blue line with dot represents the histogram of the image intensity at the region where no voxel was formed (symbol 0), and the red line with dot represents those at the region where voxels were formed (symbol 1). (ii) The standard deviation of the voxel intensities plotted against pulse energy per beam. (iii) The energy window, evaluated from the plots in (ii), against the number of beams for writing. (d) Quality factor plotted against the code rate for (i) four-beam and (ii) single-beam writing. The emission target was set to 3000 counts per beam. The pulse energy to get the emission target was 17.6 nJ for four-beam and 19.5 nJ for single-beam.CC BY-NC-ND · PubMed Central
Extended Data Fig. 5 Multibeam writing simulations. (a) (i) Heat maps of the temperature at the depth of the laser focus. The laser pulses are scanned from the bottom to the top. (ii) The peak temperatures of each beam plotted against the number of photoexcitations. The red dotted line indicates the 15th photoexcitation, where the peak temperature saturates. (b) Temperature history at the centre of the 15th photoexcited region by each beam of (i) single beam, (ii) double beams and (iii) four beams. Time zero is defined at the point corresponding to the 15th photoexcitation event. The temperature was normalized by the temperature rise by a single photoexcitation, ΔTex. (c) (i) Normalized simulated temperature rise by pre-heating plotted against the number of beams. (ii) The energy window plotted against the simulated temperature rise by pre-heating.
Extended Data Fig. 5 Multibeam writing simulations. (a) (i) Heat maps of the temperature at the depth of the laser focus. The laser pulses are scanned from the bottom to the top. (ii) The peak temperatures of each beam plotted against the number of photoexcitations. The red dotted line indicates the 15th photoexcitation, where the peak temperature saturates. (b) Temperature history at the centre of the 15th photoexcited region by each beam of (i) single beam, (ii) double beams and (iii) four beams. Time zero is defined at the point corresponding to the 15th photoexcitation event. The temperature was normalized by the temperature rise by a single photoexcitation, ΔTex. (c) (i) Normalized simulated temperature rise by pre-heating plotted against the number of beams. (ii) The energy window plotted against the simulated temperature rise by pre-heating.CC BY-NC-ND · PubMed Central
Extended Data Fig. 6 Emissions-based control of voxel writing. (a) Detection of the emission line to get the laser energy modulation. (i) An example image, and associated amplitude profile, of the emission line generated on the image sensor by scanning a laser beam inside the test material. The spatial inhomogeneity of the scanner and beam delivery causes the inhomogeneous emission profile along the scanline. (ii) Emission profiles along the scanline obtained with different laser energies. (iii) Laser energy along the scanline to get the emission of the target amplitude. (iv) Emission amplitude along the scanline with modulated laser energy of (iii). (b) Example of the variation of the voxel amplitudes written without offline flattening. The laser beam was scanned in y by a polygon scanner while the glass was translated in x at a constant velocity. (i) An example sector written with all the facets (24 in total) of the polygon scanner. The x variation in voxel amplitude indicates the inter-facet variation, resulting from the variation in reflectivity across polygon facets. (ii) An example sector written with the same single facet per rotation. The vertical variation indicates the intra-facet variation, resulting from the variation in reflectivity along each facet and different beam delivery at different scan angles. (c) Map of the emission amplitude along the scanline (y axis) of all the 24 facets (x axis) of the polygon scanner. (i) Emission amplitude map without energy modulation. (ii) Emission amplitude map with energy modulation to flatten the emission amplitude in the central region. (iii) A sector of voxels written with the energy modulation, which gives the intended emission amplitude within the written area. (d) Example of the closed-loop control to compensate for the temporal variations. (i) Emission profiles during feedback. (ii) Average emission amplitude plotted against the count of feedback.
Extended Data Fig. 6 Emissions-based control of voxel writing. (a) Detection of the emission line to get the laser energy modulation. (i) An example image, and associated amplitude profile, of the emission line generated on the image sensor by scanning a laser beam inside the test material. The spatial inhomogeneity of the scanner and beam delivery causes the inhomogeneous emission profile along the scanline. (ii) Emission profiles along the scanline obtained with different laser energies. (iii) Laser energy along the scanline to get the emission of the target amplitude. (iv) Emission amplitude along the scanline with modulated laser energy of (iii). (b) Example of the variation of the voxel amplitudes written without offline flattening. The laser beam was scanned in y by a polygon scanner while the glass was translated in x at a constant velocity. (i) An example sector written with all the facets (24 in total) of the polygon scanner. The x variation in voxel amplitude indicates the inter-facet variation, resulting from the variation in reflectivity across polygon facets. (ii) An example sector written with the same single facet per rotation. The vertical variation indicates the intra-facet variation, resulting from the variation in reflectivity along each facet and different beam delivery at different scan angles. (c) Map of the emission amplitude along the scanline (y axis) of all the 24 facets (x axis) of the polygon scanner. (i) Emission amplitude map without energy modulation. (ii) Emission amplitude map with energy modulation to flatten the emission amplitude in the central region. (iii) A sector of voxels written with the energy modulation, which gives the intended emission amplitude within the written area. (d) Example of the closed-loop control to compensate for the temporal variations. (i) Emission profiles during feedback. (ii) Average emission amplitude plotted against the count of feedback.CC BY-NC-ND · PubMed Central
Extended Data Fig. 7 Preambles and CNN model architecture. (a) Schematics of preambles along the bottom of a sector. The pattern is designed to be easy to detect so that we can unambiguously find the sector boundary. (b) Schematic of the write geometry. Each layer is filled with data sectors. For non-polygon writers, the fast axis is the x axis. For polygon writers, the fast axis is the y axis. (c) Architecture of the neural network used for decoding. Our network architecture consists of an image embedding stem, which extracts high resolution features. Then, three consecutive stages, each operating on successively downsampled inputs, extract features at coarser resolutions. Additional projection layers balance information flow between different stages directly to the final classification layers. The three different stages combine blocks of 2D convolutional layers, 2D batch-normalisations and rectified linear unit (ReLU) or Gaussian-error linear unit (GELU) activations. (d) Impact of number of context images on decode quality. Context images are additional images provided to the neural network to give it more information about the 3D structure of the data (see section ‘Machine learning model’).
Extended Data Fig. 7 Preambles and CNN model architecture. (a) Schematics of preambles along the bottom of a sector. The pattern is designed to be easy to detect so that we can unambiguously find the sector boundary. (b) Schematic of the write geometry. Each layer is filled with data sectors. For non-polygon writers, the fast axis is the x axis. For polygon writers, the fast axis is the y axis. (c) Architecture of the neural network used for decoding. Our network architecture consists of an image embedding stem, which extracts high resolution features. Then, three consecutive stages, each operating on successively downsampled inputs, extract features at coarser resolutions. Additional projection layers balance information flow between different stages directly to the final classification layers. The three different stages combine blocks of 2D convolutional layers, 2D batch-normalisations and rectified linear unit (ReLU) or Gaussian-error linear unit (GELU) activations. (d) Impact of number of context images on decode quality. Context images are additional images provided to the neural network to give it more information about the 3D structure of the data (see section ‘Machine learning model’).CC BY-NC-ND · PubMed Central

Verification

Claims the paper makes

11 Valid · 7 Not valid · 17 Not checked · 65 Own results and interpretations

45 more claims

Create a free account to see all 100 claims and to verify the 1 reference still open.

See all claims — free account
  • It may state two separate claims as one. 34
  • Probably not a substantial claim. 21

Create a free account to see these claims.

Citations

Cites

Show all 33 references

Research works printed but not linked: 4

  1. Ashby, M. & Jones, D. Engineering Materials 1: An Introduction to Properties, Applications and Design 3rd edn (Elsevier, 2005).
  2. Anderson, P. et al. Glass: a new media for a new era? In Proc. 10th USENIX Conference on Hot Topics in Storage and File Systems 13 (USENIX Association, 2018).
  3. MacKay, D. J. C. Information Theory, Inference and Learning Algorithms (Cambridge Univ. Press, 2003).
  4. Kingma, D. P. & Ba, J. Adam: a method for stochastic optimization. In Proc. 3rd International Conference on Learning Representations (ICLR) (2015).

Cited by

See all 25

Metadata

Authors
Microsoft Research Project Silica Team, Allison, J., Anderson, P., Aranas, E., Assaf, Y., Black, R., Caballero, M., Canakci, B., Chattaway, J. A., Chatzieleftheriou, A., Clegg, J., Cletheroe, D., Cooper, B., Deegan, T., Donnelly, A., Drevinskas, R., Feng, Z., Gkantsidis, C., Gomez Diaz, A., Haller, I., Hong, F., Ilieva, T., Joyce, R., Kapitany, V., Kunkel, M., Lara, D., Lawson, T., Legtchenko, S., Liu, F., Liu, X., Magalhaes, B., Nowozin, S., Overweg, H., Rowstron, A., Sakakura, M., Schreiner, N., Smith, A., Snowdon, O., Stefanovici, I., Sweeney, D., Verkes, G., Wainman, P., Whittaker, C., Wilke Berenguer, P., Williams, H., Winkler, T., & Winzeck, S.
Affiliations
Microsoft Research (United Kingdom) (GB)
Corresponding author
  • Black, Richard
  • Donnelly, Austin
  • Stefanovici, Ioan
Venue
Nature
Published
2026-02-18
Volume
650
Issue
8102
Pages
606-612
ISSN
  • 0028-0836 (ISSN-L)
  • 1476-4687
OpenAlex
W7130374112
Cited by
25
References
33
Citations by year
2026: 25
FWCI
70.28
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by-nc-nd)
Topics
  • Physical Sciences › Engineering › Industrial and Manufacturing Engineering › Laser Material Processing Techniques
  • Physical Sciences › Materials Science › Materials Chemistry › Phase-change materials and chalcogenides
  • Physical Sciences › Computer Science › Computer Networks and Communications › Advanced Data Storage Technologies
Keywords
femtosecond laser direct writing, data density, borosilicate glass, data lifetime, accelerated aging, write throughput, energy efficiency, long-term digital preservation, silica

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Laser writing in glass for dense, fast and efficient archival data storage. https://app.cochat.ai/papers/10.1038/s41586-025-10042-w
The paper
Microsoft Research Project Silica Team, Allison, J., Anderson, P., Aranas, E., Assaf, Y., Black, R., Caballero, M., Canakci, B., Chattaway, J. A., Chatzieleftheriou, A., Clegg, J., Cletheroe, D., Cooper, B., Deegan, T., Donnelly, A., Drevinskas, R., Feng, Z., Gkantsidis, C., Gomez Diaz, A., . . . Winzeck, S. (2026). Laser writing in glass for dense, fast and efficient archival data storage. Nature. https://doi.org/10.1038/s41586-025-10042-w

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.