Can Silicon Photonics Solve the Bandwidth Bottleneck in AI Data Centers?

Moore's Lobby: Where engineers talk all about circuits

Steve Klinger, Vice President of Product, joins the Moore’s Lobby podcast to discuss how LightMatter is using silicon photonics to improve speed and reduce power consumption in AI data centers. With two previous $1B+ startups under his belt, Klinger knows a thing or two about identifying successful technology solutions to current industry challenges. While compute performance continues to grow rapidly, interconnect has not been able to keep pace. In this episode, Klinger explains how LightMatter’s flagship product, Passage, creates a programmable optical fabric for the efficient interconnect of chiplets and other silicon ICs.

Klinger explains that they are trying to solve the problem of efficiently accessing all of the bandwidth on one chip and sharing it with another chip. If they can improve the interconnect bandwidth density, it will allow performance scaling to continue increasing at the workload level. Klinger emphasizes, “There are data centers with hundreds of millions of dollars of GPUs sitting idle, waiting for the network topology or the interconnects to catch up.”

So, join our host, Daniel Bogdanoff, in this deep dive into silicon photonics with Klinger. In this discussion, they address many fascinating topics, including:

- What makes silicon photonics unique from traditional photonics?

- The common traits shared by Klinger’s previous $1B+ startups.

- The many job openings available at LightMatter.