Shrining lidar down to a chip offers big rewards

Elon Musk’s disdain for lidar used to be well-established. “Lidar is a fool’s errand. Anyone relying on it is doomed. Doomed! They’re expensive sensors that are unnecessary,” Tesla’s Technoking proclaimed during a discussion about self-driving vehicle technology in 2019. He appears to have softened his stance a bit, as earlier this year a Tesla Y test vehicle fitted with rooftop lidar sensors was spotted in Florida. Subsequently, Bloomberg uncovered that Tesla has a testing and development contract with California lidar company Luminar Technologies. 

With that change of mind, Tesla joins the consensus among automotive companies that lidar is an essential technology for making advanced driver-assistance systems (ADAS) and self-driving vehicles a reality. Radar has a great range in a wide range of conditions but limited resolution. Cameras have excellent resolution, but they don’t work well in the dark or bad weather. Lidar works fine at night and is also capable of providing high resolution. The three, therefore, complement each other nicely. As an added, possibly crucial bonus, some types of lidar can not only gauge the distance to objects but also determine their speed. 

On the one hand, one can sympathize with Musk’s initial skepticism. By automotive standards, lidar used to be a clunky and costly instrument relying on moving parts to scan the environment. On the other hand, the global automotive market is a big prize, and Musk could have known that a technological arms race would ensue to develop affordable lidar that meets the requirements of the car industry. Indeed, dozens of research groups and companies around the world have set out to make that happen. 

The strategy to get there is all too familiar: integration and miniaturization. The ‘lidar-on-chip’ is compact, lightweight, contains no moving parts and can be mass-produced at a greatly reduced cost. As a light-based technology, solid-state automotive lidar, therefore, presents a great opportunity for integrated photonics, according to the Integrated Photonics for Automotive Roadmap recently released by Photondelta, an independent industry accelerator for the integrated-photonics sector. 

It’s not just automotive that would benefit from an integrated photonic lidar, however. “Lidar is used in a wide range of applications, ranging from land surveying to robotics and industrial metrology to archaeology. The latest Iphone models have lidar on board, to map rooms for example. Automotive will be without a doubt the biggest market, but I can’t think of a lidar application that wouldn’t benefit from reductions in cost and/or footprint. There’s a long road ahead, but the rewards will be great,” says Carol de Vries, program manager at Photondelta and co-author of the roadmap. 

Bottleneck 

Participating in the EU program New Control, aimed at developing subsystems required for highly automated vehicles, integrated-photonics researchers at Eindhoven University of Technology (TUE) are working on lidar-on-chip technologies. “We believe that monolithic lidar is ultimately the best solution, though it’s a huge challenge to develop it. That’s why current commercial approaches typically focus on co-integration of different technologies, for example combining indium phosphide with silicon photonics, MEMS and others. That’s the shortest path to a working chip-based lidar, but not necessarily the most optimal solution technologically, nor the most scalable, economically speaking,” says Victor Dolores Calzadilla, senior researcher at TUE’s Eindhoven Hendrik Casimir Institute (EHCI). 

Dolores Calzadilla and co-workers chose indium phosphide (InP) as the platform for their monolithically integrated photonic circuit (PIC). This comes as no surprise, given the fact that InP-based integrated-photonics technology has its roots at TUE. But it’s also the logical choice since InP is the only platform that allows for the integration of active components such as lasers, amplifiers, photodetectors and modulators alongside passive elements. 

“Other platforms can’t generate light, so they have to ‘import’ it. That’s always going to be less efficient than being able to generate, amplify and manipulate light on the same chip. InP provides more flexibility to go to higher power levels, which means extending range. This is particularly important in automotive: self-driving on highways will require a range of 200 meters or more. That’s quite challenging,” says Dolores Calzadilla. 

Another reason to pick InP is the ability to implement coherent sensing. This technique greatly improves sensitivity, even at a relatively low power level, by mixing reflected signals with a reference signal to remove out-of-phase noise. “Basically, you’re using the unique and predictable nature of the laser source to block out most light from other sources, minimizing interference from other lidars. It’s possible to implement coherent sensing on different platforms, but it’s easiest in InP.” 

These days, a coherent sensing technique called frequency-modulated continuous wave (FMCW) is drawing a lot of attention. FMCW involves continuously emitting light while periodically modulating its frequency and comparing the incoming frequency with what was sent out. The distance follows from how much the frequency changed while the reflected light made its round trip. Further processing extracts the Doppler shift to yield the object’s velocity. 

“FMCW lidar is an excellent candidate for fully integrated automotive lidar,” says Photondelta’s De Vries. “Pulsed techniques, such as time-of-flight systems, require more power than can be generated by a chip. Flash lidar, which acquires an image much like a camera does, can only be realized through heterogeneous integration, though that would result in very complex systems. And neither of these alternatives can apply coherent sensing or measure velocity.” 

Still, getting automotive FMCW lidar on the road will take years, De Vries predicts, and reaching full integration will most likely be a step-by-step process. “Putting a modulating laser and a coherent detector on a chip is relatively easy. The biggest hurdle will be the optical phased array, which is used to steer the light beam. The first automotive lidar implementations will likely take advantage of other, more mature means of controlling the beam, such as MEMS mirrors.” 

Optical phased arrays (OPAs) consist of multiple antennas that produce a single beam of light. The beam is steered by adjusting the phases of light at each antenna such that the output waves interfere constructively in one direction and destructively in all other directions. 

“The optical phased array is the biggest bottleneck for achieving a truly solid-state lidar,” Dolores Calzadilla agrees. “Many lidar building blocks, such as photodetectors and optical amplifiers, were developed years ago for other applications, like telecommunication. Even though they’re generally not yet optimized for lidar, they are available in principle. OPAs were not needed in telecom, so work on them started much later. This component is the least mature.” 

Dolores Calzadilla’s research team focuses on OPAs, among other things. “The performance of an OPA is highly dependent on the efficiency of the optical components as well as the density of the array. The higher the number of antennas, the better the control over the light beam. Currently, however, we’re limited in how densely we can pack components on a chip.” 

“The goal in the coming years is to go from tens to hundreds of antennas in a single array. We’re working together with the Eindhoven-based foundry Smart Photonics to increase integration density by upgrading fabrication technologies, allowing us to shrink components. This mission isn’t unique to lidar, of course. Almost any application in the integrated-photonics community would benefit from increased density and energy-efficiency.”