Latest News

  • Industry Cooperation

  • Publish Date:2026-11-03
From Proof of Concept to Industry Collaboration: NYCU and Hon Hai Advance Silicon Photonics for AI Data Centers
Researchers from National Yang Ming Chiao Tung University and the Hon Hai Research Institute received the 2026 Future Tech Award at the Taiwan Innotech Expo for their terabit-scale next-generation silicon photonic communication technology for AI data centers.
Researchers from National Yang Ming Chiao Tung University and the Hon Hai Research Institute received the 2026 Future Tech Award at the Taiwan Innotech Expo for their terabit-scale next-generation silicon photonic communication technology for AI data centers.
 
Edited by Amanda Yang | Photo courtesy of Hon Hai Technology Group
______
As generative AI drives rapidly growing demand for data transmission, National Yang Ming Chiao Tung University (NYCU) and the Hon Hai Research Institute (HHRI) are advancing high-capacity silicon photonic communication technologies designed for next-generation AI data centers.

Following publication of the research in Optics Express and its selection as a post-deadline paper at the 2026 OptoElectronics and Communications Conference (OECC 2026), the team has now received a 2026 Future Tech Award at the Taiwan Innotech Expo for its “Terabit-Scale Next-Generation Silicon Photonic Optical Communication Technology for AI Data Centers.”  The recognition marks another step in moving the technology beyond proof-of-concept research toward industry engagement and future applications.

As AI Computing Grows, Data Transmission Emerges as the Next Bottleneck

As large language models and generative AI systems continue to scale, AI data centers must move enormous volumes of data among processors, servers, and networking equipment. With computing performance increasing rapidly, transmitting that data at higher bandwidth and lower power consumption has become a critical challenge for next-generation AI infrastructure.

To address the problem, NYCU and HHRI have jointly developed a silicon photonic communication system with high transmission capacity. The system combines an ultra-wideband quantum-dot comb laser, high-speed signal modulation, multicore fiber, and a high-frequency electrode design. Together, these technologies increase data transmission capacity while reducing the number of optical components and making more efficient use of fiber resources.

Traditional chips and data centers rely heavily on electrical signals traveling through metal interconnects. As AI accelerators and GPUs process increasingly large volumes of data, however, electrical interconnects can face limitations in bandwidth, power consumption, heat generation, and signal loss.

Silicon photonics offers another approach. By integrating optical components with silicon-based platforms, it enables high-speed optical interconnects that can move data more efficiently between chips, servers, and other computing systems. One of the key advances in the study is the use of a single quantum-dot comb laser to generate 23 wavelength channels, reducing the need for multiple discrete laser sources and simplifying the overall system architecture. Each wavelength channel is modulated using four-level pulse amplitude modulation (PAM4) at up to 212 Gbit/s.

The system further combines 23 wavelength channels with a seven-core fiber, allowing multiple data streams to be transmitted through a single fiber and significantly increasing overall transmission capacity. A high-frequency GSSG electrode configuration also helps reduce electromagnetic interference and maintain signal quality at high transmission speeds. Together, these design choices support higher bandwidth density while reducing system complexity, power consumption, and heat generation.
 
From International Research Validation to Industry Engagement

The technology was jointly developed by Hao-Chung Kuo, director of the Semiconductor Research Institute at HHRI and Chair Professor at NYCU; the research team led by NYCU Distinguished Professor Chi-Wai Chow; HHRI team leader Yu-Heng Hong; and HHRI researcher Yun-Han Chang. The work also received support from Taiwan’s National Science and Technology Council (NSTC), the Industrial Technology Research Institute (ITRI), and a team led by Distinguished Professor Mu-Hai Cheng of National Chung Hsing University.



AI illustration: Next-generation silicon photonics uses multi-wavelength optical transmission to support high-speed interconnects for AI data centers.
AI illustration: Next-generation silicon photonics uses multi-wavelength optical transmission to support high-speed interconnects for AI data centers.

The research was published in the international optics journal Optics Express and was also selected as a post-deadline paper at OECC 2026. Post-deadline sessions are generally reserved for timely research results emerging after the regular submission deadline. According to Hon Hai, only five papers were selected for the category at OECC 2026, highlighting the attention the work received within the international optical communications community.

Beyond academic recognition, receiving the 2026 Future Tech Award creates further opportunities for technology demonstration, industry matchmaking and international collaboration. The award program is designed not only to recognize forward-looking research, but also to connect research teams with industrial and international partners.

For the team, this means the work is beginning to move beyond the question of whether the technology can be demonstrated toward a broader question: how it can eventually be integrated into practical systems and industry applications.

Building the Optical Interconnect Foundation for Next-Generation CPO and AI Data Centers

For AI data centers, computing power alone is not enough. As AI models grow, moving massive amounts of data quickly and efficiently among chips and servers has become a critical challenge.

HHRI said the technology provides a foundation for future co-packaged optics (CPO), helping improve optical interconnect efficiency while reducing data-transfer bottlenecks. By combining silicon photonics, wavelength multiplexing and multicore fiber, the research points toward faster and more energy-efficient AI infrastructure.
 
文/楊茜雯  圖/鴻海科技集團

面對生成式AI快速發展帶來的資料傳輸需求,陽明交大與鴻海研究院持續推進高容量矽光子光通訊技術,以提高AI算力與量能。繼研究成果發表於《Optics Express》並獲選OECC 2026 期限後論文後,團隊再以「兆位元級次世代矽光子光通訊技術供AI資料中心使用」獲得2026年台灣創新技術博覽會「未來科技獎」的殊榮,讓研究從技術驗證進一步走向產業鏈結與未來應用。

AI算力持續成長,資料傳輸成為下一個瓶頸

隨著大型語言模型與生成式AI規模持續擴大,AI資料中心需要在大量處理器、伺服器與網路設備之間高速交換資料。當運算能力不斷提升,如何以更高頻寬、更低功耗傳輸大量資料,也成為下一代AI基礎設施的重要課題。為回應這項挑戰,國立陽明交通大學與鴻海科技集團旗下鴻海研究院(Hon Hai Research Institute, HHRI)共同投入矽光子光通訊研究。團隊整合超寬頻量子點梳狀雷射、高速訊號調變、多核心光纖及高頻電極設計,在有限的光學元件與光纖資源下,提高資料傳輸容量,同時降低系統複雜度。

在傳統晶片或資料中心裡,晶片主要靠金屬導線傳送電訊號。但當 AI 晶片、GPU 越來越快、資料量越來越大時,電訊號會面臨頻寬不夠、耗電高、發熱多、訊號衰減的問題。矽光子在這個過程則扮演高速資料互連的角色,將光學元件應用在矽晶片或矽基平台上,讓資料透過光來傳輸得更快、更省電。

本次研究最重要的突破,在於只利用一顆雷射,產生23個不同波長的光訊號,取代傳統需要多顆雷射的複雜架構,降低空間、耗能與散熱負擔。同時搭配PAM4訊號調變,每個通道傳輸速率可達212 Gbit/s,在相同頻寬下承載更多資訊。再透過單根光纖整合7個核心,最終實現超高速傳輸。此外,高頻GSSG電極設計可降低電磁干擾、維持高速訊號品質,也因為僅需單一光源,大幅降低了晶片的發熱量與耗電,實現資料中心綠能省電的目標。
 
 


從國際學術驗證到產業接軌

這項技術由鴻海研究院半導體所所長暨陽明交大講座教授郭浩中、陽明交大特聘教授鄒志偉團隊,與鴻海研究院半導體所洪瑜亨組長、張雲翰研究員共同研發,研究過程並獲國家科學及技術委員會(NSTC)、工業技術研究院(ITRI)及中興大學鄭木海講座教授團隊支持。

研究成果已發表於國際光學期刊《Optics Express》,並獲選為2026年光電子與通信國際會議(OptoElectronics and Communications Conference, OECC)的期限後論文(Post-deadline Paper)。期限後論文通常用於收錄會議截稿後出現、具時效性的重要研究成果;OECC 2026僅有5篇論文獲選,反映團隊研究受到國際光通訊社群關注。

除了學術上的肯定,獲得2026未來科技獎,也讓這項技術有更多機會進入技術展示、產業媒合與國際合作場域。換句話說,這項研究不再只是證明技術做得到,而是開始進一步思考未來可以怎麼用,以及如何與產業接軌。

為下一代CPO與AI資料中心建立光互連基礎

對AI資料中心而言,真正的挑戰不只是晶片算得夠不夠快,還包括大量資料能不能在晶片與伺服器之間快速流動。當AI模型越來越大,如果資料傳輸跟不上運算速度,再強的晶片也可能受到通訊瓶頸限制。鴻海研究院表示,此次高容量矽光子傳輸技術也為未來共同封裝光學(CPO)的發展奠定基礎,大幅提升AI算力。這項由台灣自主研發的綠能晶片,將成為守護地球的與加速AI發展的重大科技創新。

Related Image(s):