Phase voxel glass archival storage: TB-scale in borosilicate with 10,000 year longevity

Microsoft Research’s Project Silica has moved from laboratory curiosity to a demonstrable archival platform: researchers report in Nature (published 18 February 2026) that they can encode terabytes of data inside millimetre‑thin pieces of glass — including ordinary borosilicate — using femtosecond lasers, and that accelerated ageing tests indicate those inscriptions can survive for at least 10,000 years under typical storage conditions.

Background / Overview​

Project Silica began as an exploration of durable, long‑term digital archival media by encoding information as three‑dimensional microstructures — voxels — inside transparent materials using tightly focused femtosecond laser pulses. Early work concentrated on high‑purity fused silica because its optical and chemical stability made it naturally suited to ultra‑long‑term retention. The new Nature paper reports two major advances that materially change the product and commercialization calculus:

• The team developed a phase‑voxel writing regime that works reliably in borosilicate glass (the same family used in many household ovenware products), dramatically widening the choice of media and lowering material cost.
• The researchers scaled write throughput through parallel multi‑beam writing and simplified the readout pipeline (a single‑camera reader plus machine‑learning decoding), bringing the system closer to practical archival workflows.

Taken together, these developments answer two of the most stubborn barriers to archival optical glass storage: expensive media and low throughput. The study documents a highest‑density demonstration of 4.84 terabytes stored in a 120 mm × 120 mm × 2 mm fused silica platter (301 layers), and an accessible borosilicate demonstration showing robust phase‑voxel storage with multi‑layer inscription and accelerated ageing that supports claims of 10,000‑year retention.

---

How the technology works — the science in plain terms​

Writing data in glass with femtosecond lasers is an exercise in controlled micro‑damage. A femtosecond pulse — a pulse lasting a few quadrillionths of a second — creates a micro‑plasma at the laser focus inside the glass. That event locally changes optical properties (refractive index, birefringence or the creation of nanoscale voids), and those tiny changes become the physical representation of bits.

Two voxel regimes: birefringent vs. phase voxels​

• Birefringent voxels (fused silica)
  These are anisotropic micro‑modifications whose optical response depends on the orientation of polarization. The Project Silica team developed pseudo‑single‑pulse techniques that reduce the writing energy and the number of pulses required per voxel, enabling very high density (reported 1.59 Gbit / mm³) and very fine pitch (≈0.5 µm lateral). The fused silica demonstration reached 301 layers and a usable capacity of ~4.84 TB in a 12 cm square, 2 mm thick piece.
• Phase voxels (borosilicate)
  Phase voxels change the local refractive index isotropically. Crucially, Project Silica showed phase voxels can be written with a single femtosecond pulse per voxel in borosilicate glass — a simpler physical mechanism that makes the approach work with cheaper, widely available glass. The phase‑voxel regime gives lower peak density than the birefringent regime (reported ~0.678 Gbit / mm³ and ~2.02 TB usable per platter in the demonstration), but it greatly simplifies hardware and media supply.

Parallel writing and the path to throughput​

To address the long time required to write terabytes with a single laser beam, the researchers parallelized the process: they split the laser into multiple independently modulated beams and scanned them through synchronized optics to write many voxels simultaneously. The paper reports a practical multibeam throughput of tens of megabits per second and shows that models of heat diffusion and pulse timing allow writing many voxels close together without thermally damaging the glass.

Reading and decoding​

Reading the encoded data uses high‑resolution microscopy and optical modalities suited to the voxel type:

• Birefringent voxels are read with polarization‑resolved imaging.
• Phase voxels are read with Zernike phase‑contrast microscopy.

A major engineering win reported in the paper is the simplification of the reader: where previous approaches required multiple imaging channels and complex optics, the new approach can use a single camera combined with machine‑learning decoders to mitigate inter‑symbol interference and reliably classify symbol states.

---

The headline technical metrics​

The Nature paper spells out concrete, reproducible numbers that matter for engineers and archivists:

• Demonstrated capacity (fused silica, birefringent voxels): 4.84 TB in a 120 mm × 120 mm × 2 mm platter (301 layers), equivalent to 1.59 Gbit / mm³.
• Demonstrated capacity (borosilicate, phase voxels): ~2.02 TB per platter with 258 layers, 0.678 Gbit / mm³.
• Voxel pitch examples: ~0.5 µm × 0.485 µm (birefringent) and ~0.5 µm × 0.7 µm (phase).
• Layer spacing: 6 µm (birefringent) and 7 µm (phase) in the demonstrations.
• Per‑beam write throughput (single beam): ~25.6 Mbit/s (birefringent) and ~18.4 Mbit/s (phase).
  With a four‑beam parallel writer, reported combined throughput reached ~65.9 Mbit/s; thermal simulations indicate the technique scales with even more beams.
• Write energy efficiency: ~10.1 nJ / bit (birefringent), ~8.85 nJ / bit (phase).
• Longevity: accelerated ageing and nondestructive optical analysis support data retention on the order of 10,000 years when stored under benign conditions; results suggest even greater stability at ambient temperatures.

These are not hand‑wavey claims — the paper includes methodological detail for pulse energy, objective numerical aperture, repetition rates (example: 10 MHz lasers), and the tools used to verify and decode symbols.

---

Strengths and why this matters now​

Project Silica’s latest results deliver several real and consequential benefits for long‑term archival storage:

• Media availability and cost: Demonstrating storage in borosilicate (a commodity glass) is a practical breakthrough. It reduces dependence on specialty fused silica, easing supply constraints and lowering media cost.
• Longevity without power: Once written, glass requires no power to preserve the encoded information — an attractive property for cold archives, cultural institutions, and disaster‑resilient preservation.
• Physical durability: Glass is naturally resistant to many degradation modes (water, dust, many chemicals). Properly handled, it can survive centuries to millennia far better than magnetic or organic media.
• High information density: The capacity figures shown put glass storage well above many tape and optical archival options per unit volume for immutable, long‑lived copies.
• Improved write throughput: Multi‑beam writing and single‑pulse regimes substantially reduce the time to inscribe large datasets, moving the idea from a lab trick toward something that could be integrated into archival workflows.
• Simplified read hardware: Reducing reader complexity to a single camera plus ML decoding lowers cost and improves the chance of fielding practical, maintainable readout systems.
• Energy efficiency in writing: The reported nanojoule/bit regime shows the approach is not frivolously power hungry during write operations — an important consideration if many units are deployed.

For archivists, museums, scientific facilities (e.g., particle physics, astronomy) and government records offices, these properties map directly to real problems: how to keep verifiable, accessible, and immutable copies of data for centuries without continuous migration cycles.

---

Practical limits, risks, and unanswered questions​

No technology is without constraints. Project Silica’s results are encouraging, but there are practical, logistical, and systemic hurdles to solve before glass archives become a mainstream component of digital stewardship.

Scaling to exabytes: throughput and factory economics​

• Writing terabytes on a single platter is impressive; writing exabytes is a different scale. Even with multibeam writers, the throughput numbers imply many writer units would be required to keep pace with the annual output of large scientific instruments or hyperscaler cold tiers.
• The cost of high‑precision femtosecond lasers, precision optics, synchronized scanners and robotic handlers is nontrivial. Industrial cost engineering and volume manufacturing will need to reduce per‑bit capital and OPEX dramatically for mass adoption.

Reader and decoder dependency​

• The technique depends on specific microscope hardware and machine‑learning decoders to interpret subtle optical signatures. The long‑term viability of archived data relies not only on the glass but on preserving the means to decode it: file formats, decoding models, and calibration routines.
• Without an open, well‑documented specification and archived reader software (and ideally reference hardware designs), future readers could be hard to build or license.

Software and format obsolescence​

• Even if the bits survive a thousand generations, meaning is lost if file formats, compression schemes, or metadata are not preserved in a durable and documented way. Standardization and embedding human‑readable metadata will be essential.

Physical risks and access patterns​

• Glass is durable but brittle. Large‑scale physical storage implies handling, packaging, and environmental control protocols to prevent mechanical damage.
• Retrieval latency is not negligible: reading a platter requires microscope scans and ML decoding — acceptable for deep cold archives but not for high‑frequency retrieval.

Trust, verification, and independent audits​

• Extrapolating accelerated ageing to 10,000 years has scientific precedent but is inherently inferential. Independent replication, long‑term monitoring experiments, and public data will be needed to build institutional trust.

Security and confidentiality​

• Glass archives are effectively immutable once written; errors or leaked data are hard to remove. Policies and processes for encryption, access control, and secure destruction (where legally required) must be addressed.

---

Use cases where glass archival shines​

Project Silica should be judged not as a drop‑in replacement for all storage but as a purpose‑built solution for specific needs.

• Cultural heritage preservation (libraries, museums): storing canonical copies of images, audio and textual artifacts for millennia.
• Government and legal records: tamper‑resistant, long‑lived retention of high‑value records.
• Scientific data vaults: permanent, power‑free archives of raw data that must be preserved intact for reproducibility decades or centuries hence.
• Strategic backups for critical infrastructure (e.g., genomic databases, climate records): where nonvolatile longevity and physical security matter more than retrieval speed.

These are all cold storage scenarios where the cost of write hardware is balanced against the value of maintaining a single immutable copy for many generations.

---

Engineering and policy recommendations for stakeholders​

If you’re an IT manager, archivist, or CTO evaluating whether to pilot Project Silica‑style glass archival, consider this prioritized checklist:

• Establish a preservation package:
• Store human‑readable metadata and decoding instructions on the media itself and in independent paper or microform copies.
• Demand open standards:
• Insist that any vendor provide an open, documented format and reference reader/decoder software that can be compiled and run on community platforms.
• Pilot on high‑value cold data:
• Start with datasets where write‑once immutability and energy‑free longevity are worth the capital investment (cultural collections, long‑tail scientific data).
• Architect for layered preservation:
• Use glass as one component in an N‑version preservation strategy (e.g., a glass platter plus geographically distributed tape/filestore and cloud copies).
• Plan for hardware obsolescence:
• Archive a means to build or emulate readers: CAD files, firmware images, and trained ML models, plus instructions for calibration.
• Require independent validation:
• Work with third‑party labs to verify accelerated ageing claims and perform periodic readback audits.

---

What commercialization will need — technical and ecosystem bets​

Several engineering and industry moves would accelerate adoption:

• Commodity femtosecond lasers and optics: volume manufacturing would lower per‑unit costs and increase throughput through parallelization.
• Standardized golden‑master formats for glass archives, including canonical encodings and human‑readable manifests.
• Open‑source reader software and trained ML weights distributed together with media.
• Packaging and handling standards: shock‑resistant housings, barcodes and robotic interfaces for large‑scale libraries.
• Certification programs (NIST or equivalents) for long‑term data fidelity, backed by continuous readback testing.

If those pieces converge, glass archival could move from bespoke high‑value proofs to a reliable archival tier for specialized workloads.

---

Ethical, legal and governance considerations​

• Irreversibility: once data is written, it’s effectively permanent. Laws and policies about the right to be forgotten, record deletion, or data correction create legal friction that archivists and legal teams must manage.
• Access equity: who gets to read these platters centuries from now? Public institutions should consider widely accessible, nonproprietary decoding paths to avoid gatekeeping.
• Cultural stewardship: selection decisions for what to encode permanently are inherently editorial. Transparent governance frameworks are essential.

---

How to read the longevity claims (a cautious scientist’s take)​

Accelerated ageing tests are standard practice for extrapolating long‑term degradation from elevated temperatures and stressors. The Project Silica team used nondestructive optical metrics and accelerated protocols to estimate that the glass inscriptions would remain readable for 10,000 years under typical storage conditions. That is a strong indicator of exceptional stability, but it is not a literal time‑travel guarantee:

• Extrapolation over millennia carries uncertainty. Environmental extremes, unexpected chemical interactions, mechanical shocks, or unforeseen failure modes cannot be fully simulated.
• Independent replication and multi‑decade monitoring will be necessary to build institutional confidence.
• The media (glass) is only one half of the story — the decoding ecosystem must also be preserved. Without it, unreadable bits are as good as gone.

In short: the physics looks solid and the engineering is sound, but long‑term archival practice must plan conservatively and redundantly.

---

Final assessment — opportunity and caution in equal measure​

Project Silica is a rare combination of rigorous materials science and systems‑level thinking about archival workflows. By demonstrating:

• high densities in fused silica,
• practical phase‑voxel inscription in commodity borosilicate,
• multi‑beam writing and simplified readers,
• and careful accelerated ageing analyses,

the team has moved glass optical storage from a laboratory oddity to a credible archival technology with a distinct niche: write‑once, power‑free, millennial durability.
That said, the path to widespread deployment is not automatic. Major challenges remain in scaling throughput and lowering hardware cost, formalizing open standards for encoding and decoding, preserving decoding tools, and designing physical archives that balance durability with access. For institutions holding irreplaceable cultural, legal, or scientific records, Project Silica offers a compelling new option — provided deployments are accompanied by layered preservation strategies, open standards, and an explicit plan for reader and format preservation spanning generations.
For the storage industry, the project is both an invitation and a warning: invest in rugged, standards‑based archival design or risk creating beautiful, durable artifacts that are effectively unreadable when the specialized readers are gone. For archivists and technologists, the sensible next step is careful pilot programs, joint work on standards and open readers, and well‑documented preservation packages that keep both the bits and the means to read them alive for the centuries ahead.

---

Conclusion
Project Silica’s Nature paper marks a pivotal moment for archival storage: glass, once exotic as a data medium, now demonstrates practical writing, decoding, and longevity characteristics that make it a serious candidate for permanent archives. The innovation is technological and cultural — a call to build preservation ecosystems that ensure the encoded knowledge remains both physically intact and meaningfully accessible to the generations that follow. The research is no longer merely proof‑of‑concept: it is a roadmap for how institutions might lock away selected parts of our digital civilization in a medium designed to outlast current storage paradigms — if and only if we do the engineering, governance, and standardization work that future readers will need.

---

Source: News9live Microsoft introduces Project Silica for storing data in glass