SetupAI
EN
來源較早收集於 5h

微軟將4.8TB資料刻入玻璃保存一萬年

微軟將4.8TB資料刻入玻璃保存一萬年
PostLinkedIn
🧠閱讀原文： 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 秒速覽

有什麼變化

在120x120x2mm玻璃中儲存301層資料，密度達1.59 Gbit/mm³

為什麼重要

此技術為AI資料集與模型提供超耐用、高密度檔案儲存，降低資料中心長期成本與環境影響，優於磁帶或磁碟。

下一步行動

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

誰應關注：Researchers & Academics

關鍵要點

  • •在120x120x2mm玻璃中儲存301層資料，密度達1.59 Gbit/mm³
  • •每光束寫入吞吐量25.6 Mbit/s，容量達4.8TB
  • •硼矽酸鹽玻璃經老化測試，資料壽命超過一萬年
  • •首個滿足生產系統所有要求的玻璃儲存技術
  • •已發表於Nature期刊

🧠 深度解析

背景與延伸：來自公開資料，非原文內容。引用 8 個來源。

🔑 增強重點摘要

  • •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]

🛠️ 技術深入

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]

🔮 前景展望基於引用來源的 AI 分析

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.

⏳ 時間線

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
🧠閱讀原文： 机器之心
📰

AI 週報

閱讀本週精選 AI 大事摘要 →

👉相關動態

同主題

探索 #glass-storage

同產品

更多關於 project-silica

同來源

最新來自 机器之心

AI 策展新聞聚合。所有內容版權歸原始發布者所有。
原始來源： 机器之心 ↗

這是摘要，不是原文。去看原站，或訂閱每週簡報。

每週電子報

每週一封，可隨時退訂。