# How Chipmakers Are Using On-Chip Optical Clocks for Timing

An episode of [The Hardware Podcast with Fexingo: Chips, Devices, and Electronics Engineering Conversations](https://invisiblenetwork.ai/podcast/1896799982), 5 Jul 2026, 8:16.

Audio: https://audio.fexingo.com/business/the-hardware-podcast/episode-0093.mp3

Episode 93 of The Hardware Podcast explores a critical but often overlooked component in modern chips: the on-chip optical clock. Lucas and Luna discuss how traditional quartz oscillators are reaching their limits as chip speeds increase, and how integrated photonics are enabling tiny, stable optical clocks that can synchronize billions of transistors with picosecond precision. They dive into the work of researchers at UC Santa Barbara and imec, who have demonstrated a silicon photonics clock that consumes just 10 milliwatts while achieving timing jitter below 50 femtoseconds. The hosts explain why this matters for everything from 5G base stations to AI accelerators, and how it could reshape chip design by eliminating the need for external timing crystals. #OpticalClock #SiliconPhotonics #ChipTiming #IntegratedPhotonics #Semiconductor #Hardware #Technology #UCSantaBarbara #Imec #AIAccelerators #5G #ClockDistribution #Jitter #QuartzOscillator #Timing #FexingoBusiness #BusinessPodcast #TheHardwarePodcast Keep every episode free: buymeacoffee.com/fexingo

## Transcript

Lucas: You ever think about how a chip knows when to do what it's supposed to do?
Luna: I mean, there's a clock, right? Every transistor waits for the tick.
Lucas: Right. But where does that tick come from? Traditionally, it's a quartz crystal oscillator sitting off-chip, connected by a trace. That works fine at a few gigahertz, but as chips push toward 10, 20 gigahertz and beyond, the clock distribution becomes a nightmare.
Luna: So the solution is to put the clock on the chip itself?
Lucas: Exactly. And not just any clock — an optical clock. Using light instead of electrical signals to generate and distribute the timing reference. This is one of those quiet revolutions in chip design that doesn't get a lot of attention, but it's absolutely critical for the next generation of processors.
Luna: Okay, so what does an on-chip optical clock actually look like? Is it a tiny laser?
Lucas: It's a laser, yes — but a very specific type. Typically, it's a mode-locked laser diode integrated directly into the silicon photonics layer of the chip. That laser produces a train of extremely short optical pulses at a precise repetition rate. Think of it as a metronome made of light.
Luna: And the key advantage over a quartz oscillator is stability, right? Less jitter?
Lucas: Exactly. Timing jitter — the tiny random variations in the clock edge — is the enemy of high-speed digital design. A good quartz oscillator might have jitter around one picosecond. These on-chip optical clocks are demonstrating jitter below 50 femtoseconds. That's twenty times better.
Luna: Fifty femtoseconds. That's hard to even wrap your head around. Light itself only travels about 15 micrometers in that time.
Lucas: Right. And that stability translates directly into performance. You can either run the same circuits faster with fewer errors, or you can reduce the voltage margins and save power. For data centers, that's a huge win.
Luna: Who's actually building these things? Is it still lab-scale, or are we seeing commercial prototypes?
Lucas: Both. Researchers at UC Santa Barbara have been doing pioneering work on heterogeneous integration — bonding iii v laser material onto silicon photonics. And imec in Belgium has demonstrated a fully integrated optical clock in their silicon photonics platform. They published a paper last year showing a 10-milliwatt optical clock that can drive an entire chip's timing network.
Luna: Ten milliwatts is actually pretty low. Some of those clock distribution networks can eat up 30, 40 percent of a chip's power budget.
Lucas: Yeah, power is a huge driver. The electrical clock tree on a modern high-end CPU can consume tens of watts just to push the clock signal across the die. With optical, you distribute the clock via a waveguide — essentially a tiny pipe for light — and you don't have the same RC delay or power loss. You can cover a much larger area with less energy.
Luna: So does this mean we'll see optical clocks in consumer chips soon? Like, the next iPhone?
Lucas: Probably not in the next generation or two. The biggest challenge is integration. You need to combine a laser, modulators, photodetectors, and waveguides all on the same chip, and do it in a way that's manufacturable at scale. That's hard. But for high-end server chips and AI accelerators, I could see it happening in the next five years.
Luna: And for applications where timing is absolutely critical — like 5G base stations, or high-frequency trading — the femtosecond stability might be worth the extra cost.
Lucas: Exactly. In fact, that's where the early adopters will be. Any system that relies on precise synchronization across multiple channels benefits enormously. And there's another interesting angle: optical clocks can also serve as a reference for on-chip analog to digital converters, improving signal fidelity in radar and communications.
Luna: So it's not just about making computers faster, it's about making them more accurate.
Lucas: Right. And that brings me to something I wanted to flag, because it's actually how this show stays on the air. We don't run ads, we don't have sponsors — the whole thing is supported by listeners who find value in these deep dives. If today's conversation about optical clocks and chip design gave you something you can use, or just satisfied your curiosity, the best way to keep these episodes coming is to toss a few bucks into the hat at buy me a coffee dot com slash fexingo.
Luna: Yeah, and it really does make a difference. We hear from people who say these episodes help them understand what's actually happening in hardware, beyond the press releases.
Lucas: Totally. So, back to the tech — one more thing I think is cool. These optical clocks can also be used to generate multiple synchronized clock domains on a single chip. You can take that pulse train and split it optically, sending the same ultra-stable timing to different parts of the chip with virtually no skew.
Luna: That's huge for heterogeneous integration, where you have chiplets from different fabs running at different speeds.
Lucas: Exactly. If you can synchronize a bunch of chiplets with femtosecond precision, you can build much larger, more complex systems without worrying about clock domain crossing issues. It's like having a single metronome for the entire orchestra, no matter how far apart they're sitting.
Luna: What about reliability? Lasers on a chip — do they wear out?
Lucas: That's actually one of the big open questions. Laser diodes do degrade over time, especially if they get hot. But the mode-locked lasers used for optical clocks are typically run at lower power, and researchers are seeing lifetimes in the tens of thousands of hours. For server chips that might be replaced every three to five years, that's probably acceptable.
Luna: And if one laser fails, could you have redundancies? Like, put two or three on the chip?
Lucas: You could. The waveguide network could be designed with redundant sources, and the chip could switch to a backup if the primary clock drifts. That kind of fault tolerance is already common in aerospace and telecom.
Luna: So what's the main barrier right now — cost, yield, or ecosystem?
Lucas: I'd say all three, but ecosystem is the biggest. Foundries are set up for CMOS or for photonics, but not both in the same process flow. You need a cmos compatible photonics platform that can also handle iii v materials. That's starting to emerge — companies like Tower Semiconductor and GlobalFoundries are offering silicon photonics processes — but it's not yet a standard offering.
Luna: And the design tools? I imagine EDA tools for optical circuits are less mature than for electrical.
Lucas: Definitely. You need to simulate both the optical and electrical domains together. That's an area where companies like Synopsys and Cadence are investing, but it's still early. However, there's a lot of momentum. The IEEE recently formed a study group on optical clock distribution, which is a sign that the industry sees it as a real future technology.
Luna: So if someone wanted to follow this space, where should they look?
Lucas: I'd keep an eye on the Optical Fiber Communication Conference — OFC — and the IEEE International Solid-State Circuits Conference. That's where the latest results appear. And definitely check out the work from imec and UC Santa Barbara's John Bowers group. They've been pushing the envelope on heterogeneous laser integration for years.
Luna: Alright, I'm going to start paying more attention to the clock. It's not just the tick, it's the whole mechanism behind it.
Lucas: Yeah, and that mechanism is about to get a lot faster and a lot more precise. It's one of those quiet transformations that will show up in everything from your phone to the cloud.

Shows are listed from Apple Podcasts' charts and directory, and Podcast Index where it is used. Every show and episode belongs to its publisher and plays from the publisher's own feed; this site hosts none of them.