Microsoft's glass-storage researchers turned a sample upside down to work out why some of its data was harder to read.
The weaker region moved with the reading direction. That pointed to the reader, rather than the way those layers had been written. The observation appears in the team's February 18 Nature paper on Project Silica, which stores data as tiny laser-made changes inside glass.
Reading the archive starts with microscope images. A neural network estimates which stored symbol each tiny region represents, then the system converts those estimates into bits and applies error correction. Neighboring marks can interfere with one another in the images, so surviving glass alone isn't enough to recover a file.
The upside-down test made that distinction visible. The researchers found that the reading difficulty varied with depth. Reversing the glass reversed the pattern. More demanding parts of the readout needed more redundancy to recover the original data.
There is a material change in the same work: the technique now extends to borosilicate glass, the family used in kitchen cookware, rather than relying only on specialized fused silica. Microsoft says the new reader needs one camera instead of three or four.
The project is aimed at long-term archives. Its widely quoted 10,000-year lifetime comes from accelerated aging tests, not ten millennia of observation. The less spectacular upside-down experiment addresses the other half of that ambition: being able to get the original bits back out of the glass.
Sources:
https://www.nature.com/articles/s41586-025-10042-w
https://www.microsoft.com/en-us/research/blog/project-silicas-advances-in-glass-storage-technology/
Image: AI-generated editorial illustration of glass under an optical instrument, not Project Silica equipment or a recorded sample.
The weaker region moved with the reading direction. That pointed to the reader, rather than the way those layers had been written. The observation appears in the team's February 18 Nature paper on Project Silica, which stores data as tiny laser-made changes inside glass.
Reading the archive starts with microscope images. A neural network estimates which stored symbol each tiny region represents, then the system converts those estimates into bits and applies error correction. Neighboring marks can interfere with one another in the images, so surviving glass alone isn't enough to recover a file.
The upside-down test made that distinction visible. The researchers found that the reading difficulty varied with depth. Reversing the glass reversed the pattern. More demanding parts of the readout needed more redundancy to recover the original data.
There is a material change in the same work: the technique now extends to borosilicate glass, the family used in kitchen cookware, rather than relying only on specialized fused silica. Microsoft says the new reader needs one camera instead of three or four.
The project is aimed at long-term archives. Its widely quoted 10,000-year lifetime comes from accelerated aging tests, not ten millennia of observation. The less spectacular upside-down experiment addresses the other half of that ambition: being able to get the original bits back out of the glass.
Sources:
https://www.nature.com/articles/s41586-025-10042-w
https://www.microsoft.com/en-us/research/blog/project-silicas-advances-in-glass-storage-technology/
Image: AI-generated editorial illustration of glass under an optical instrument, not Project Silica equipment or a recorded sample.