Source: Ars Technica

Clear as Glass

Femtosecond lasers etch data into a very stable medium.

Note: Right now, Silica hardware isn’t quite ready for commercialization.

Credit: Microsoft Research

---

Archival storage poses many challenges. We need media that is extremely dense, stable for centuries (or longer), and ideally consumes no energy when not being accessed. Numerous concepts have been explored—even DNA has been considered—but one of the simplest approaches is to etch data into glass. Many types of glass are physically and chemically stable, and they can be relatively easy to pattern.

There has been substantial preliminary work demonstrating various aspects of a glass‑based storage system. In the latest issue of Nature, Microsoft Research announced Project Silica, a working demonstration that can read and write data into small glass slabs with a density exceeding 1 Gb/mm³.

Writing on Glass

We tend to think of glass as fragile, prone to shattering, and capable of flowing downward over centuries—although the latter claim is a myth. Glass is a category of material, and a variety of chemicals can form glasses. With the right starting chemistry, it’s possible to make a glass that is, as the researchers put it, “thermally and chemically stable and resistant to moisture ingress, temperature fluctuations, and electromagnetic interference.”

While such glass still needs to be handled carefully to minimize damage, it provides the stability we want for long‑term storage.

Writing Data

Putting data into glass is conceptually simple: etch the information onto the material. The challenge has been that traditional etching is slow. The development of femtosecond lasers—lasers that emit pulses lasting only 10–15 femtoseconds and can fire millions of pulses per second—dramatically reduces write times and allows the beam to be focused on a very small area, increasing potential data density.

Reading Data

To read the data back, several options exist:

• Laser reading – similar to how we read optical disks, though traditionally slower.
• Microscopic inspection – any technique capable of detecting the tiny etched features could work.

Project Silica

With these considerations in place, the theoretical foundation for Project Silica was solid. The remaining question was how to integrate the writing and reading technologies into a functional system. Microsoft chose to answer that question twice, proceeding cautiously while exploring multiple implementation paths.

A Real‑World System

The difference between these two answers comes down to how an individual unit of data (called a voxel) is written to the glass.

Birefringence‑based voxels

One type of voxel they tried was based on birefringence – a phenomenon where the refraction of photons depends on their polarization.
It’s possible to etch voxels into glass to create birefringence using polarized laser light, producing features smaller than the diffraction limit. In practice this involved:

1. A first laser pulse that creates an oval‑shaped void.
2. A second, polarized pulse that induces birefringence in that void.

The identity of a voxel is defined by the orientation of the oval. Because multiple orientations can be resolved, more than one bit can be stored in each voxel.

Refractive‑index‑magnitude voxels

The alternative approach varies the magnitude of the refractive effect by changing the energy of the laser pulse.
Again, more than two states can be distinguished, allowing multiple data bits per voxel.

---

Example: Silica Storage Medium

The map data from Microsoft Flight Simulator etched onto the Silica storage medium.
Credit: Microsoft Research

---

Reading the Data

Reading these voxels in Silica uses a microscope that detects differences in refractive index (i.e., phase‑contrast microscopy). The microscope determines how many layers of voxels can be placed in a single piece of glass:

• The layers are spaced far enough apart that only one layer is in the microscope’s focal plane at a time.
• Etching incorporates alignment symbols that let the automated system position the lens over specific points on the glass.
• The system then slowly changes its focal plane, moving through the stack and capturing images of each voxel layer.

Decoding the Images

Microsoft employed a convolutional neural network (CNN) that combines data from images both in and near the focal plane of a given layer. This works because nearby voxels subtly influence how a voxel appears; the CNN can learn these nuances given sufficient training data.

Data Encoding

The Silica system encodes data as follows:

1. Raw bitstream → add error‑correction using a low‑density parity‑check (LDPC) code (the same code used in 5G networks).
2. Neighboring bits are combined into symbols that exploit the voxel’s multi‑bit capacity.
3. The resulting symbol stream is written to the glass.

Performance

Writing remains a bottleneck in the system, so Microsoft developed hardware that can write a single glass slab with four lasers simultaneously without generating too much heat. This enables a write speed of 66 megabits / s. The team believes it could be possible to add up to a dozen additional lasers, which may be needed because a single slab can store up to 4.84 TB (dimensions 12 cm × 12 cm × 0.2 cm). At the current speed, it would take over 150 hours to fully write a slab.

Data‑density trade‑offs

Longevity

Borosilicate glass offers extreme stability. Microsoft’s accelerated‑aging experiments suggest the data would remain stable for over 10 000 years at room temperature. The company therefore declared:

“Our results demonstrate that Silica could become the archival storage solution for the digital age.”

Real‑world scalability

The Square Kilometre Array (SKA) telescope is expected to generate ≈ 700 PB of data per year. Storing that amount would require:

• ≈ 140 000 glass slabs (700 PB ÷ 4.84 TB per slab)
• Even with many more lasers, ≈ 600 Silica machines would need to operate in parallel to keep up with the write rate.

The SKA is just one of many data‑intensive projects, highlighting the challenge of scaling this technology.

Advantages

• Zero energy consumption for data preservation.
• Rapid retrieval, unlike DNA storage which can take days.
• A futuristic look that feels straight out of science‑fiction (see the video here).

Nature, 2026. DOI: 10.1038/s41586-025-10042-w – About DOIs

---

Author

John Timmer – Science editor at Ars Technica.

• B.A. in Biochemistry, Columbia University
• Ph.D. in Molecular and Cell Biology, University of California, Berkeley

When he’s not at his keyboard, John can be found on a bike or hiking in scenic locations.

---

80 Comments