# How Chip-Scale Switch Arrays Are Shrinking Data Centers

An episode of [The Hardware Podcast with Fexingo: Chips, Devices, and Electronics Engineering Conversations](https://invisiblenetwork.ai/podcast/1896799982), 28 Aug 2026, 9:56.

Audio: https://audio.fexingo.com/business/the-hardware-podcast/episode-0169.mp3

In this episode, Lucas and Luna explore the quiet revolution in data center networking: chip-scale optical switch arrays. These tiny silicon photonic devices can reroute light beams in nanoseconds, replacing bulky, power-hungry electrical switches. The hosts break down a 2025 demonstration from a startup that packed 64 channels onto a single die, slashing switching energy by 80 percent. They discuss how these arrays could cut data center energy use, enable fully optical data centers, and reshape network architecture. With a nod to hyperscalers like Google and Microsoft, they consider implications for cloud costs, latency, and the future of copper. The conversation is grounded in specifics but accessible to anyone curious about the tech behind the internet. Tune in for a clear-eyed look at a chip that might quietly change everything. #ChipScale #OpticalSwitching #DataCenter #SiliconPhotonics #MEMS #CloudComputing #NetworkHardware #Semiconductor #EnergyEfficiency #LaserTech #Photonics #Hyperscaler #Google #Microsoft #TechInnovation #ElectronicsEngineering #FexingoBusiness #BusinessPodcast Keep every episode free: buymeacoffee.com/fexingo

## Transcript

Lucas: So last week I was reading through the latest from a startup out of Berkeley called LightWave Switch, and they announced a chip-scale optical switch array that fits sixty-four channels on a single die. The switching energy? About eighty percent lower than a comparable electrical switch.
Luna: Okay, that sounds like a big deal, but let me play dumb for a second. Why do we need optical switches in data centers anyway? Aren't we already using fiber for long-haul?
Lucas: Right, so the long haul is fiber, sure. But inside a data center, when a packet hops from one server to another, it often goes through electrical switches that convert light to electricity, switch it, and then back to light. That conversion costs power and generates heat. And as data centers scale, that switching energy is becoming a real bottleneck.
Luna: I've read that networking can be something like twenty percent of a data center's energy budget. Is that in the right ballpark?
Lucas: It's in the zone. Some hyperscalers report even higher. And that's why the idea of all-optical switching has been the holy grail for a while, but the challenge has been making the switches small and fast enough. Traditional optical switches, like the ones using liquid crystal or MEMS mirrors, are bulky, and they can take milliseconds to reconfigure. But this new chip-scale array using silicon photonics can switch in nanoseconds.
Luna: Nanoseconds. That's the speed of electrical switching, but without the conversion. So what's the catch? Why hasn't this been done before?
Lucas: The catch has always been loss and crosstalk. When you route light on a chip, you lose some signal, and light can leak between channels. But LightWave says they've gotten insertion loss below one decibel and crosstalk down to minus thirty decibels. Those numbers are good enough for commercial use.
Luna: So how does the switch actually work? Is it a tiny mirror or some kind of interferometer?
Lucas: It's based on what's called a Mach-Zehnder interferometer, which is a classic photonic building block. You split a light beam into two paths, then apply a voltage to change the phase in one arm, and when you recombine them, you get constructive or destructive interference, effectively steering the light. The whole thing is fabricated in a standard CMOS foundry, which means it's cheap to produce at scale.
Luna: So they're using the same manufacturing process as chips in your phone. That's got to be a game-changer for cost.
Lucas: Exactly. And that's the punchline. Because it's cmos compatible, you can integrate the control electronics right on the same die, which reduces packaging cost and improves performance. The startup says their switch can handle forty gigabits per second per port, and they're aiming to scale that up.
Luna: Who's actually going to buy these? I mean, is this a niche research thing, or are hyperscalers lining up?
Lucas: The big cloud providers are definitely interested. Google has been experimenting with optical circuit switching for years, and Microsoft has been funding research in this space. The promise is to reduce power consumption and latency in their data centers, which directly affects their bottom line. If they can cut the energy bill by twenty percent, that's billions of dollars in savings across their fleet.
Luna: That's a huge incentive. But what about the existing infrastructure? Are they going to rip out all their electrical switches overnight?
Lucas: No, not overnight. The transition would be gradual, starting with the top of rack switches that handle the most traffic. You'd replace the electrical switch with an optical one, and then you can eliminate some of the transceivers that are needed to convert between light and electricity. Over time, you might move to a fully optical data center, but that's years away.
Luna: Let's talk about the practical gains. If you switch optically, how much faster is the actual data transfer from one server to another?
Lucas: The latency savings come mainly from cutting out the conversion steps. Each conversion adds maybe a few hundred nanoseconds, but when you're doing millions of packets per second, that adds up. More importantly, the power savings allow you to pack more switches into the same footprint, which means you can increase bandwidth without expanding the data center.
Luna: So it's not just about speed, it's about density and energy. I can see why cloud providers would be all over this.
Lucas: And it's not just the big guys. Smaller colocation providers are watching too, because they can differentiate on energy efficiency. Plus, there's a sustainability angle. Data centers account for something like one to two percent of global electricity use, so any reduction is meaningful.
Luna: You mentioned this is based on research from Berkeley. Is there a story behind how they got from lab to startup?
Lucas: Yeah, it spun out of a group at UC Berkeley that was working on silicon photonics for years. They had been developing the interferometer design and realized they could push the performance far enough for commercial use. With some seed funding from venture capital, they licensed the IP and set up shop in a small office near campus.
Luna: That's the classic Silicon Valley story. But what are the actual specs? You said sixty-four channels, insertion loss under a decibel. What's the footprint?
Lucas: The chip is about five millimeters by five millimeters, so smaller than a postage stamp. And it's designed to be compatible with standard fiber connectors, which makes it easy to integrate into existing systems. They're also working on a version with integrated drivers, so the control electronics are on the same chip.
Luna: So you could plug this into a standard switch chassis and just swap out the old module?
Lucas: That's the idea. They're aiming to make it a drop-in replacement for the current optical modules, at least in the initial phase. But long-term, they want to redesign the whole switch architecture around these chips, which would be more radical.
Luna: What about reliability? Optical components sometimes have issues with temperature drift and vibration. How do they handle that?
Lucas: Good question. They've integrated heaters to control the temperature of the interferometer arms, which keeps them stable. And the chip is fabricated in a standard CMOS process, which is inherently robust. They've tested it across a range of temperatures and it holds up.
Luna: So they've thought through the practical stuff. What's the timeline for shipping? Are they in production now?
Lucas: They're in pilot production with a handful of customers, and they expect volume shipments by early next year. The pricing is competitive with existing optical switches, which is key. But the real value is the energy savings over the lifetime of the equipment.
Luna: So if I'm a data center operator, the upfront cost might be similar, but I save on power and cooling. That's a pretty easy sell.
Lucas: Exactly. And that's why the industry is paying attention. The same technology could also be used in everything from autonomous vehicles to medical imaging, where you need fast, compact optical switching.
Luna: What about the competition? Are there other companies working on this?
Lucas: There are a few, but most are using micro-electromechanical mirrors, which are slower and more expensive to package. LightWave's approach is unique because it's all solid-state and cmos compatible. That gives them a cost and speed advantage that could be hard to beat.
Luna: So this could be a case where the right engineering at the right time changes the industry.
Lucas: It has that potential. But there are still hurdles, like getting the big cloud providers to adopt new hardware in their massive data centers. That takes time and trust.
Luna: Speaking of trust, I know our listeners are curious about the broader field. What are the other chip-scale technologies you're keeping an eye on?
Lucas: There's a lot of work in chip-scale gyroscopes for navigation, and also chip-scale spectrometers for environmental sensing. We've covered some of those in previous episodes. But the switch array is particularly exciting because it's addressing a critical bottleneck in the digital economy.
Luna: It makes you wonder what else is lurking in the labs that could suddenly become mainstream.
Lucas: Speaking of that, we should note that this kind of deep-dive is exactly what our listeners support. We don't run ads, and that's only possible because a few hundred of you chip in a few bucks a month through buy me a coffee dot com slash fexingo. It's a small gesture that keeps these conversations ad-free and independent.
Luna: Yeah, we really do appreciate it. And if you're new to the show and find this stuff fascinating, that's the best way to keep it going.
Lucas: So back to the future of data centers. I think the next big step is integrating these switches with the existing network protocols. Right now, they're circuit switches, which means they route a whole wavelength. But the next generation might switch individual packets, which would be a bigger leap.
Luna: That's a tougher challenge, but if they pull it off, we're looking at a fundamental shift in how data centers are built.
Lucas: Absolutely. And for now, I'm just excited to see the first deployments happen. It's one of those rare moments where a chip-scale component could genuinely reshape the infrastructure we all rely on.
Luna: Well, I'm ready to see it in action. Thanks for breaking it down.
Lucas: Of course. And if you want to dive deeper into the specs, we'll have links in the show notes, but the short version is: small chip, big energy savings, coming to a data center near you.

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.