Breaking
ITIran war, Trump: "We will strike hard if Tehran is involved in the Flydubai air attack"TRNew presidential candidates are on the agenda after the detention of MHK President Ferhat GündoğduINHuge demonstration in Delhi-Mumbai demanding resignation of Chief Election Commissioner Gyanesh KumarCRYPTO-FRUS Treasury sanctions underground banking network A7 NetworkTRInflation in the Eurozone is at its peak in three yearsESPP and Vox reject the Government's housing decrees in CongressITSummer 2026: record marine heat waves in Italian seasAUManchester City Files Appeal Against Premier League Financial RulingGLOBALOil prices fall on reports of potential diesel and crude stock releasesARTensions have risen between Europe and Washington over diesel supplies and global inflation challengesITIran war, Trump: "We will strike hard if Tehran is involved in the Flydubai air attack"TRNew presidential candidates are on the agenda after the detention of MHK President Ferhat GündoğduINHuge demonstration in Delhi-Mumbai demanding resignation of Chief Election Commissioner Gyanesh KumarCRYPTO-FRUS Treasury sanctions underground banking network A7 NetworkTRInflation in the Eurozone is at its peak in three yearsESPP and Vox reject the Government's housing decrees in CongressITSummer 2026: record marine heat waves in Italian seasAUManchester City Files Appeal Against Premier League Financial RulingGLOBALOil prices fall on reports of potential diesel and crude stock releasesARTensions have risen between Europe and Washington over diesel supplies and global inflation challenges
Back光電大廠Coherent前CEO:AI帶動基礎設施革命 1.6T光學產品正進入放量階段
光電大廠Coherent前CEO:AI帶動基礎設施革命 1.6T光學產品正進入放量階段
Tech
自由时报1 hour agoTech4 min readChinaView translation

光電大廠Coherent前CEO:AI帶動基礎設施革命 1.6T光學產品正進入放量階段

AI不只是軟體轉型,更是一場基礎設施革命,隨著運算規模擴大,資料移動已成為新的限制因素,推動光學技術持續向運算核心靠近。

Quick Look

光電大廠Coherent前董事長暨執行長Chuck Mattera表示,AI基礎設施革命推動運算規模擴大,資料移動成新限制因素,1.6T光學產品正進入放量階段,3.2T收發器已完成展示,光學正持續向運算核心靠近。

AI-generated summary

Why It Matters

AI基礎設施持續擴大,運算規模從晶片擴展至整棟建築與園區,對資料傳輸與光學技術的需求顯著提升。

Font size

光電大廠Coherent前董事長暨執行長Chuck Mattera指出,在光學持續深入AI運算架構下,產業下一階段仍有待業者共同推進。(記者方韋傑攝)

〔記者方韋傑/台北報導〕人工智慧(AI)基礎建設持續擴大,光電大廠Coherent前董事長暨執行長Chuck Mattera指出,AI不只是軟體轉型,更是一場基礎設施革命,隨著運算規模從晶片、伺服器、機櫃一路擴大至整棟建築甚至園區,資料移動已成為新的限制因素,推動光學技術持續向運算核心靠近,1.6T光學產品正進入放量階段,3.2T收發器也已完成展示,共同封裝光學(CPO)則已走出實驗室。

Mattera曾任Coherent董事長暨執行長,日前以「光子的世紀」為題發表演講。他表示,AI背後其實是一部龐大的實體機器,運算單位已不再只是晶片、伺服器或機櫃,而逐漸擴大至建築,甚至整個園區,並需要十萬級加速器、液冷系統及數以百萬計的光學連結;在此規模下,運算、記憶體、電力、散熱與連接已成為5個同等重要的系統,當運算達到建築規模時,互連本身也成為架構的一部分。

Mattera指出,每個運算時代都有不同的限制因素,1990年代產業追求時脈速度,直到散熱形成限制;2000年代持續提高電晶體密度,2010年代則面臨記憶體頻寬壓力,之後封裝與資料移動逐漸承壓,如今瓶頸再次轉移至互連,功耗、頻寬、延遲與傳輸距離必須同時解決,「移動資料現在已成為新的限制因素」。

不過,Mattera強調,這並非電子與光子的競爭,「電子負責運算,光子負責傳輸」,未來將是兩者的工程化融合。過去50年,光纖解決跨越海洋、大陸、城市及資料中心的長距離傳輸問題,如今光學傳輸距離一路縮短至1公里、10公尺、1公尺、1公分,甚至毫米尺度;隨著頻寬提高,銅線面臨功耗、散熱與延遲壓力,促使光學持續向運算端靠近。

在技術路線方面,Mattera表示,可插拔光模組(Pluggable)、近封裝光學(Near-Package Optics)、共同封裝光學及光學輸入輸出(Optical I/O)正朝相同方向演進,但不同架構並非單純互相取代;可插拔方案仍將持續升級,近封裝光學扮演銜接角色,共同封裝光學進一步整合,而光學輸入輸出則讓光更接近運算核心,各自回應更高頻寬、更低功耗、更低熱量與更短延遲的需求。

Mattera並提出產業轉型正在加速的4項訊號,包括1.6T光學產品正進入放量階段,採用400G光通道的3.2T收發器已完成展示;可插拔光學持續擴張;共同封裝光學已離開實驗室,相關平台開始出貨;此外,將光學元件移至特殊應用積體電路(ASIC)旁,相較傳統可插拔方案,光互連功耗可降低約70%。他直言,「這場轉型已經不再只是預測,而是正在發生。」

Open Questions

  • 1.6T光學產品的具體市佔率與營收貢獻為何?
  • 共同封裝光學(CPO)的大規模商業化時間點為何?

Related Topics

This article was originally published by 自由时报.

Related Stories

The advanced deeds of Fujian science and technology commissioners group are released: writing papers in the fields and countryside
Tech·

The advanced deeds of Fujian science and technology commissioners group are released: writing papers in the fields and countryside

Fujian Province held a ceremony to release the advanced deeds of the "Eight Fujian Model" science and technology commissioners group. Over the years, more than 10,000 scientists have taken root in the fields, promoting new technologies and products, and promoting the enrichment of the people and agriculture. Juncao and other technologies have spread to more than 100 countries and regions around the world.

中国新闻网
3 min read
More on this topiccoherent