SourceStalecollected in 5h

Microsoft Stores 4.8TB in Glass for 10,000 Years

Microsoft Stores 4.8TB in Glass for 10,000 Years
PostLinkedIn
🧠Read original on 机器之心
#glass-storage#archival-data#optical-storageproject-silicamicrosoftproject-silicanature

💡Breakthrough glass storage holds 4.8TB eternally—vital for AI's exploding data needs (Nature pub)

⚡ 30-Second TL;DR

What Changed

Stores 301 layers of data in 120x120x2mm glass at 1.59 Gbit/mm³ density

Why It Matters

This enables ultra-durable, high-density archival storage for AI datasets and models, reducing long-term data center costs and environmental impact compared to tapes or disks.

What To Do Next

Download the Nature paper and simulate glass storage density for your AI dataset archival needs.

Who should care:Researchers & Academics

Key Points

  • Stores 301 layers of data in 120x120x2mm glass at 1.59 Gbit/mm³ density
  • Achieves 4.8TB capacity with 25.6 Mbit/s per beam write throughput
  • Borosilicate glass validated for >10,000-year data lifespan via aging tests
  • First glass-based storage meeting all production system requirements
  • Published in Nature journal

🧠 Deep Insight

Background and context from public sources — not the original article. 8 sources cited.

🔑 Enhanced Key Takeaways

  • Microsoft's shift from expensive fused silica to borosilicate glass reduces material costs and sourcing constraints, addressing a critical commercialization barrier that previously limited practical deployment[1][2]
  • The new 'phase voxel' encoding method requires only a single femtosecond laser pulse per data point versus multiple pulses in prior approaches, significantly reducing write complexity and manufacturing costs while enabling parallel high-speed writing[3][5]
  • Microsoft's reader hardware has been simplified from three or four cameras to a single camera, substantially reducing the physical footprint and cost of the read apparatus[5]
  • The company conducted accelerated aging tests using a novel nondestructive optical method to validate the 10,000-year lifespan claim, moving beyond theoretical projections to empirical validation[5]
  • Despite the technical breakthrough, Microsoft's official statement indicates the research phase is complete with no clear commercialization timeline, suggesting uncertainty about market viability or business prioritization[1]

🛠️ Technical Deep Dive

Storage Architecture

  • Data encoded as three-dimensional microscopic structures called voxels using femtosecond lasers
  • Phase voxel method modifies the refractive index of glass to create phase shifts in transmitted light, with four levels of amplitude modulation per voxel[6]
  • Voxel pitch: 0.500 µm × 0.485 µm with 6 µm layer spacing and 301 layers per 2mm-thick plate[6]
  • Achieves 1.59 Gbit/mm³ density, translating to 4.84 TB usable capacity in a 120 mm × 120 mm × 2 mm glass platter after overhead[6]

Write Performance

  • Write speed: 25.6 Mbit/s per beam (fused silica); 18.4 Mbit/s with borosilicate glass[3][6]
  • Energy efficiency: 10.1 nJ per bit[6]
  • Single laser pulse per phase voxel enables parallel writing across multiple beams[5]

Read Mechanism

  • Polarization-sensitive microscopy using regular light for data retrieval[4]
  • Simplified to single-camera reader versus previous three-to-four camera systems[5]
  • Read-only design guarantees true airgap security—insufficient power during reading to accidentally modify glass[4]

Material Properties

  • Borosilicate glass: low-cost, durable, electromagnetic field-proof, water and temperature resistant[4]
  • Write-once, read-many (WORM) media prevents accidental data overwriting[4]
  • Validated lifespan: >10,000 years via accelerated aging tests and nondestructive optical aging identification method[5]

🔮 Future ImplicationsAI analysis grounded in cited sources

Borosilicate glass adoption removes the primary cost barrier to commercialization, but Microsoft's ambiguous commitment suggests internal uncertainty about market demand or profitability.
The shift from fused silica to kitchen-grade borosilicate glass directly addresses the 'cost and availability' barriers Microsoft identified[1][5], yet the company's statement reads as a research wrap-up rather than a commercialization roadmap[1].
Glass storage could displace tape archival for long-term institutional data preservation if manufacturing scales and pricing becomes competitive with LTO tape.
The 10,000-year lifespan, electromagnetic immunity, and elimination of periodic media migration cycles address core pain points in enterprise archival workflows[4], but no pricing or production timeline has been disclosed.
Machine learning-based error correction becomes essential for phase voxel reliability as data density increases and inter-voxel interference grows.
The new phase voxel method trades single-pulse simplicity for increased inter-voxel interference, requiring ML models to decode data accurately[3]—a dependency that may complicate manufacturing quality control.

Timeline

2019-01
Project Silica research initiative launched by Microsoft Research
2024-06
Earlier experiments with fused silica achieved 4.84 TB per 2mm plate at 25.6 Mbps write speed
2026-02
Breakthrough published in Nature: borosilicate glass encoding with phase voxel method, 18.4 Mbps write speed, simplified reader hardware, and validated 10,000-year longevity
📰

Weekly AI Recap

Read this week's curated digest of top AI events →

👉Related Updates

AI-curated news aggregator. All content rights belong to original publishers.
Original source: 机器之心

This is a summary, not the original. Read the source, or get the weekly briefing.

The weekly digest

One email a week. Unsubscribe anytime.