Microsoft this week detailed new research aimed at preserving data in borosilicate glass plates for thousands of years longer than conventional media like hard drives or magnetic tape, without needing to worry about bit rot.
The boffins at Redmond have been exploring this concept as part of Project Silica since 2019, using femtosecond lasers to encode data as voxels (3D pixels) inside glass.
Glass is resilient to water, heat, and dust. It also breaks down more slowly than the plastics or mechanical components found in hard disks or magnetic tape drives, both still widely deployed for archive storage media.
Up until recently, Microsoft’s experiments into glass-based storage have required special fused silica glass, which is both challenging and expensive to manufacture. In a paper published in the journal Nature this week, Microsoft researchers now say these long-term storage qualities can be achieved using the same kind of borosilicate glass found in oven doors and Pyrex glassware.
In their testing, they were able to etch 258 layers of data totaling roughly 2.02 TB onto a 2 mm thick borosilicate glass plate while achieving write speeds of between 18.4 and 65.9 Mbps depending on the number of laser beams used. That top speed is significantly faster than the 25.6 Mbps achieved using fused silica glass, but with less than half the density, at 2.02 TB vs 4.84 TB per platter. Additionally, they were able to reduce the amount of equipment required to read back data from the plates from three or four cameras to just one.
The way Microsoft is etching the voxel data into the glass hasn’t changed — it’s still using femtosecond lasers — but its method for doing so has. Early attempts at glass-based storage, including Microsoft’s, used “birefringent” voxels, which means they refract light differently depending on their polarization. According to Microsoft, this required multiple laser pulses to encode the data, which they were eventually able to reduce to two. They’ve now developed a different phase-based voxel, which requires just one laser pulse.
In addition to reducing the number of laser pulses required to write data to the plates, Microsoft has also managed to increase the number of voxels written in parallel, significantly boosting transfer rates.
As we noted last year, this is just what Microsoft has managed to achieve in the lab and isn’t necessarily reflective of the technology’s storage density or transfer rate if it’s ever productized. For example, the researchers suggest that 16 or more beams writing in parallel could dramatically increase write speeds over the one to four beams used in its borosilicate glass trials.
Also, while more efficient, the approach isn’t perfect. The phase-based voxels showed a greater propensity for interference, but the researchers note that this can be effectively mitigated by machine learning-based classification models.
Finally, Microsoft conducted an accelerated aging test to extrapolate the media’s viable storage life. Despite our best efforts, existing storage media have a set life span before they begin to break down and are subject to bit rot. This can happen relatively quickly in the case of flash media, particularly when left unpowered in warm climates.
This doesn’t appear to be an issue with glass, with Microsoft’s trials supporting a viable storage life of more than 10,000 years.
Despite Microsoft’s ongoing progress into glass-based storage, its future remains somewhat vague. In a blog post published alongside the paper, the company notes that the “research phase is now complete, and we are continuing to consider learnings from Project Silica as we explore the ongoing need for long-term preservation of digital information.”
That’s not exactly a development road map. “Microsoft, like other companies, takes many considerations into account when determining productization and investment priorities. Microsoft continues to value the intellectual property in Project Silica and is exploring options for how to apply the research learnings,” a Microsoft spokesperson told El Reg ®.