CEA-Leti has been working on optical phased arrays (OPAs) to produce cost-effective and compact LiDARs.
The LiDAR sets the scene in 3D, which makes it a valuable technology for future sensing and vision systems. Its high cost, however, remains a major obstacle to mass integration. Leveraging silicon photonics, CEA-Leti has been working on Optical Phased Arrays (OPAs) to produce cost-effective and compact LiDARs.
In an interview with EE Times Europe, CEA-Leti’s researchers Sylvain Guerber and Daivid Fowler explained the development, calibration and characterization steps of silicon photonics OPAs for LiDARs. CEA-Leti has indeed developed genetic algorithms to calibrate high-channel-count OPAs and a measurement setup compatible with wafer-scale OPA characterization.
A LiDAR is a remote sensing method that calculates the time it takes for a beam of light to strike an object or surface and reflect back to the laser scanner. Traditional LiDAR systems are made of discrete components whose mechanical assembly makes them fragile, expensive, and cumbersome. Using a photonic chip, CEA-Leti claims the current optical beam scanning system (moving mirrors) can be replaced by an integrated OPA (no moving mechanical parts). The cost and size of the LiDAR are significantly reduced while its performance (e.g. scanning speed, power efficiency, resolution…) is improved thanks to solid-state beam steering.
An OPA is an array of closely spaced (around 1µm) optical antennas with adjustable phase. It performs beam steering by controlling the phase of light radiated by each antenna. If the phase gradient between the antennas is linear, a directional beam will be formed. By changing the slope of the linear gradient, the direction of the beam can be controlled, which enables solid-state beam steering, the research institute explained.
CEA-Leti said it is using advanced microelectronics manufacturing processes.
Calibration and characterization
Developing an OPA is the first part, but it is not enough to have a system that can scan a beam and steer it in a given direction. “When we make these OPAs, even though the microelectronic technology is precise and allows us to control all the dimensions of the circuits, there can be variations,” said Guerber. “The problem, if you want a beam to be formed vertically, above the OPA, you need all the phases of all the antennas, so hundreds or even thousands of antennas, to be exactly the same.”
Due to small manufacturing imperfections, an OPA will not form a very clean and usable beam to make a LiDAR. A calibration phase is then necessary. The CEA-Leti researchers have placed a camera at the exit of the OPA and observed the spatial distribution of the outgoing beam. “If you turn on the OPA, without doing anything special, you get a more or less random distribution of power, so the light comes out in all directions.” To avoid this problem, the team implemented a genetic algorithm so as to analyze the camera image and draw conclusions about what adjustment to make in the OPA to then be able to form a clean, usable beam.
Genetic algorithms are one of the evolutionary algorithms that use techniques inspired by Darwin’s theory of evolution. They have been developed to quickly (up to 1000x faster than previously used algorithms) calibrate high-channel-count OPAs. Genetic algorithms have another advantage: They are relatively robust to the starting conditions, said Guerber. “This means that if we take several OPAs that do not have exactly the same beams because they have been manufactured slightly differently, the genetic algorithm is free of this variation and always manages, more or less, to find a solution close to the optimal solution. Other algorithms are more dependent on the initial conditions and will be less efficient in certain contexts.”
Accelerating or even automating the calibration process is essential to enable the industrialization of OPA-based LiDARs. Consequently, the CEA-Leti researchers have developed an advanced measurement setup enabling wafer-scale OPA characterization. “It takes a large testing capacity to produce these components by the thousands,” said Fowler. “On a wafer, there can be 100 OPAs and, for each OPA, thousands of calibrations must be done. It must be automatic. We worked on the calibration, but also on the automation of this test process.”
When asked if CEA-Leti was the only one to have a wafer-scale OPA characterization setup, Fowler said, “Personally, I have not yet seen any research group doing wafer-scale OPA characterization.” Usually, they dice the wafer into chips, do the wirebonding and proceed with the characterization, he added. “We avoid this step and put the whole wafer on the prober, which allows us to test an OPA in a standard microelectronics environment and to move towards industrialization.”
A LiDAR comprises many elements, including a laser, an electronic driver, an OPA steering system, a detector, and data-treatment capability. All of them must work together, and the OPA is only a part of the system.
About four years ago, CEA-Leti started with small OPA circuits with about 16 optical channels. “Today, we’re up to about 200, but we’re not yet at the complexity needed to operate in a real system,” said Fowler. Thousands of channels are actually needed.
“The OPA is one thing, but the integration into the system is a more complex task,” said Fowler, specifying that CEA-Leti is far from having a complete system. “At CEA-Leti, we have a silicon photonics lab, but we have cross-functional working groups to integrate our OPAs into a system.” The research groups will move forward in stages, said Fowler. “For instance, we will start by integrating an OPA, a laser, and a detector. We will integrate one component and try to make the system work. Then two, three and four. The goal is to integrate them all in a small box, and it will certainly take years.”
Applications down the road
LiDAR innovation has moved fast, and automotive applications are expected to be the main drivers in the next five years, according to market research firm Yole Développement (Lyon, France). The LiDAR market for ADAS is set to achieve an annual growth of 114%, from US$19 million in 2019 to US$1.7 billion in 2025.
“Today, automotive applications are driving the development of OPAs, but the specifications are so aggressive in terms of beam quality, temperature resistance, and safety that it is difficult to have an OPA-based LiDAR that meets these specifications,” said Guerber. Medical imaging applications are also concerned, but the level of specifications is just as high. In the meantime, OPAs should come out, but with lower performance and for applications like robotic arms and drones.
This article was originally published on EE Times Europe.