# How Chips Use On-Chip Lidar for Autonomous Navigation

An episode of [The Hardware Podcast with Fexingo: Chips, Devices, and Electronics Engineering Conversations](https://invisiblenetwork.ai/podcast/1896799982), 18 Jul 2026, 8:55.

Audio: https://audio.fexingo.com/business/the-hardware-podcast/episode-0118.mp3

This episode dives into the integration of lidar directly onto silicon chips, a breakthrough that could make autonomous navigation cheaper and smaller. Lucas explains how researchers at MIT and the University of California, Berkeley have fabricated solid-state beam-steering arrays on CMOS chips, replacing the bulky rotating lasers used in self-driving cars. Luna asks about the key technical challenge: achieving a wide field of view without moving parts. They discuss the use of optical phased arrays and the current limitations in range and resolution. The conversation also touches on potential applications beyond cars—drones, robotics, and even smartphone depth sensing. Lucas shares a specific milestone: a prototype chip that achieved a 70-degree field of view with centimeter-level accuracy at 10 meters. The episode closes on the question of when this technology might hit consumer products, with Lucas estimating three to five years for specialized industrial use, and longer for automotive. #OnChipLidar #AutonomousNavigation #SiliconPhotonics #OpticalPhasedArray #SolidStateLidar #CMOS #SelfDrivingCars #MIT #UCBerkeley #LidarTechnology #AutonomousVehicles #DroneNavigation #Robotics #DepthSensing #Technology #HardwarePodcast #FexingoBusiness #BusinessPodcast Keep every episode free: buymeacoffee.com/fexingo

## Transcript

Lucas: So you know how most self-driving cars today have that spinning cylinder on the roof — the lidar unit that costs tens of thousands of dollars and looks like a bucket of tennis balls?
Luna: Right, the Velodyne-style unit. I've seen them on test vehicles. They're huge.
Lucas: Exactly. What if you could shrink that entire system — laser, scanner, receiver — onto a single silicon chip smaller than your thumbnail? That's what a bunch of research groups are now doing with on-chip lidar.
Luna: And we're not talking some far-off lab demo. I've seen papers from MIT and UC Berkeley in the last year showing working prototypes.
Lucas: Yeah, that's the key thing — this is real silicon. The MIT group, led by Professor Jelena Vuckovic, fabricated a chip that uses an optical phased array to steer a laser beam electronically. No moving parts.
Luna: No spinning motor. So it's solid-state, which means it's more reliable, cheaper to manufacture, and smaller.
Lucas: Right. The phased array works by having dozens or hundreds of tiny optical antennas on the chip. By adjusting the phase of the light going to each antenna, you can steer the beam in different directions — kind of like how a phased-array radar works.
Luna: What kind of field of view are we talking about?
Lucas: The MIT prototype demonstrated a 70-degree field of view in one dimension. To get full 3D coverage, you'd need to either have multiple chips or use a second axis of steering. But 70 degrees is already useful for forward-facing obstacle detection.
Luna: And range? Because for autonomous driving, you need to see a couple hundred meters out.
Lucas: That's the current bottleneck. The MIT chip achieved centimeter-level accuracy at about 10 meters. For industrial robots or drones flying low and slow, that's fine. But for highway-speed cars, you need an order of magnitude more range.
Luna: So the physics — you're limited by the power you can put through a silicon waveguide without damaging the chip, and by the sensitivity of the on-chip detector.
Lucas: Exactly. The laser power on chip is constrained by the material's damage threshold, and the return signal drops off with the square of the distance. Groups are working on using more sensitive detectors, like single-photon avalanche diodes, or SPADs, integrated on the same chip.
Luna: There's also the question of eye safety. If you boost the power too much, you could damage someone's retina.
Lucas: Right, that's a regulatory constraint. But the beauty of the phased array is that you can spread the beam over a larger area, reducing the intensity at any point. So there's a design trade-off.
Luna: I've also seen a different approach from UC Berkeley — they used a frequency-modulated continuous wave, or FMCW, lidar on a chip. That gives you both range and velocity directly, like radar.
Lucas: Yeah, FMCW is promising because it's more robust against interference from other lidar units and sunlight. The Berkeley chip, led by Professor Ming Wu, demonstrated a solid-state FMCW lidar with a 20-degree field of view at 10 meters. They used a different beam-steering mechanism called a grating coupler.
Luna: So there's not one clear winner yet. Multiple architectures competing.
Lucas: Right, and that's healthy. The market for solid-state lidar is expected to grow from about $500 million this year to several billion by 2030, driven by autonomous vehicles, but also by drones, robotics, and even smartphones for depth sensing.
Luna: Smartphones — like Apple's Face ID, but with lidar instead of structured light. That could enable much better AR experiences.
Lucas: Exactly. And if the chip can be made cheap enough — say under $10 per unit — it becomes a commodity sensor. You could put it on a Roomba, on a delivery drone, on a factory robot.
Luna: Let's talk about the manufacturing challenge. Making a phased array with thousands of antennas requires nanoscale fabrication precision. Is that something that can scale with standard CMOS processes?
Lucas: That's actually the exciting part. Most of these prototypes are built in standard CMOS silicon photonics foundries — the same ones used for data center optical transceivers. So there's a clear path to volume production. The MIT chip was fabricated at AIM Photonics in New York, which is a multi-project wafer service.
Luna: So the foundry infrastructure already exists. It's not like we need a new fab.
Lucas: Exactly. The challenge is yield and uniformity. If even a few antennas have slightly different phase responses, the beam quality degrades. Researchers are developing calibration algorithms that can measure and compensate for these errors in real time.
Luna: That's reminiscent of how phased-array antennas in radar are calibrated — you inject a test signal and tune each element.
Lucas: Right. And in the optical domain, you can use on-chip heaters to adjust the phase of each antenna. So the chip has built-in self-calibration.
Luna: One thing I'm curious about — how does the on-chip lidar handle weather? Rain, fog, snow can mess with lidar.
Lucas: That's a challenge for all optical sensors. On-chip lidar uses near-infrared wavelengths, typically around 1550 nanometers — the same as fiber optics. That wavelength is relatively eye-safe and can penetrate fog a bit better than the 905-nanometer lasers used in some automotive lidars. But you still lose performance in heavy rain.
Luna: Some companies are combining lidar with radar and cameras for redundancy.
Lucas: Right. You'll never have a single sensor that works in all conditions. But the advantage of on-chip lidar is that it's small and cheap enough to put multiple units on a vehicle — say, one facing forward, one on each corner — giving you 360-degree coverage without a spinning tower.
Luna: You know, this kind of technology — where a complex system gets reduced to a chip — is exactly the kind of thing that makes me think about the importance of independent tech journalism. Shows like this that dig into the details without a corporate agenda.
Lucas: Absolutely. And we can keep doing that because of listeners who support the show. If today's discussion gave you something useful, and you'd like to help us stay ad-free and independent, you can buy us a coffee — literally at buy me a coffee dot com slash fexingo.
Luna: It's a simple way to say thanks, and it really does make a difference. We put that support right back into research and production.
Lucas: Alright, back to the chips. So we've got these lidar-on-chip prototypes working in the lab. When do you think we'll see them in a commercial product?
Luna: I'd say for specialized industrial robots — like automated guided vehicles in warehouses — maybe two to three years. For automotive, more like five to seven, given the safety validation required.
Lucas: That sounds right. There are startups like SiLC Technologies and Ocelot that are already sampling chip-scale lidar to early customers. And some big automotive suppliers are investing heavily.
Luna: What about consumer drones? DJI's latest models use vision-based obstacle avoidance, but lidar could give them night vision.
Lucas: Exactly. A chip-scale lidar could be small and light enough for a commercial drone. That would open up applications like autonomous inspection of bridges and power lines at night. The market is huge.
Luna: One more thing — the cost. When Velodyne first came out with the HDL-64, it was $75,000. Now you can get a solid-state lidar from Luminar for about $1,000. Where does the chip version come in?
Lucas: For a single-chip lidar, the bill of materials could be under $100 in volume. That's a game changer. At that price, you could put lidar on every car, even economy models.
Luna: And on robots, on lawnmowers, on delivery pods. It democratizes sensing.
Lucas: Exactly. So we're at an inflection point — the technology is proven in principle, the manufacturing path exists, and the applications are waiting. The next few years will be about engineering the system integration and driving down cost.
Luna: It's one of those rare moments where you can see a whole new capability set about to become ubiquitous.
Lucas: And the chip — not much bigger than a grain of rice — will be the thing that makes it happen. That's the power of on-chip integration.

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.