youtube-transcript.ai

@HPCpodcast-104: Silicon Photonics, w Keren Bergman (2)

Watch with subtitles, summary & AI chat
Add the free Subkun extension — works directly on YouTube.
  • Watch
  • Subtitles
  • Summary
  • Ask AI
Try free →

This video is for professionals and enthusiasts interested in the future of high-performance computing and AI hardware.

TL;DR

This episode of the ATH HPC podcast features Keren Bergman discussing the advancements in silicon photonics for AI and HPC systems. Key topics include the maturation of the manufacturing ecosystem, the transition of optical interconnects into the data path, and the projected commercial readiness around 2028.

Key Takeaways

In This Video

  1. 00:00Introduction to HPC Podcast

    Welcome to the ATH HPC podcast with hosts Doug Black and Shaheen Khan, discussing supercomputing technologies.

  2. 01:30Guest: Keren Bergman, Optical IO Expert

    Introducing Keren Bergman, an expert in optical IO interconnect technology and co-founder of Escape Photonix.

  3. 02:51Changes in Optical IO Technology

    Discussing advancements in optical IO technology over the past two years, focusing on ecosystem maturation.

  4. 03:39Photonics in the Data Path

    Explaining the concept of bringing photonics into the data path for AI and HPC systems.

  5. 04:25Co-Packaged Optics vs. Copper

    Clarifying that optical IO is for chip-to-chip communication within servers, not within chips.

  6. 05:10Advantages of Optical Interconnects

    Highlighting the low loss and high bandwidth advantages of optical communication over electrical.

  7. 07:07Commercial Readiness and Cost

    Addressing the long-standing question of commercial readiness and the primary barrier of cost.

  8. 08:51Manufacturing Ecosystem Maturation

    The manufacturing ecosystem for photonics is maturing, with major players investing in its future.

  9. 09:28Projected Deployment Timeline

    Estimating 2028 as the year for the first significant deployment of co-packaged photonics in systems.

Questions & Answers

What is silicon photonics and its role in AI and HPC systems?
Silicon photonics is a key technology for photonic interconnects, enabling AI and HPC systems to run faster and cooler by moving data optically within the data path, rather than relying solely on electrical interconnects.
What is co-packaged optics or embedded photonics?
Co-packaged optics, or embedded photonics, involves bringing the photonic interface closer to the chip, allowing data to be transmitted optically over short distances within a server or motherboard.
What are the advantages of optical interconnects over copper for HPC and AI?
Optical interconnects offer significantly lower energy consumption per bit and higher bandwidth density over distance compared to copper, which experiences more loss and requires amplification over longer runs.
When will co-packaged photonics be widely deployed in systems?
While research has been ongoing for decades, widespread commercial deployment of co-packaged photonics in systems is anticipated around 2028, driven by the maturation of the manufacturing ecosystem and significant industry investment.
Why hasn't silicon photonics been fully implemented sooner?
The primary reason for delayed implementation has been cost. While photonics can outperform electrical interconnects, the cost of manufacturing and the ecosystem's development have been significant factors.
What is the main change in silicon photonics over the last two years?
The most significant change has been the maturation of the silicon photonics ecosystem, with major companies making strategic investments and commitments to bring these technologies to the commercial market.

Key Terms

Download or copy the punctuated YouTube transcript (Markdown)

Full Transcript

Loading transcript…

Source

YouTube video. Original: https://www.youtube.com/watch?v=hLjeY2_wuiY
Transcript captured and processed by youtube-transcript.ai on 2026-06-04.