UST030

83 certain point, their calibration is not always easy or simple, and their reliability – their mean time between failure, or MTBF – is often in question. Production cost and size are the biggest disadvantages of fully mechanical Lidar systems. While many system developers are working on flash architectures that are becoming available to system designers, there is still significant development in the design and manufacturing of mechanical and MEMS systems. Mirror lasers The first Lidar sensors had a spinning mirror to deflect the various laser beams, with 128, 64 and now 16 beams. Reducing the number of beams but keeping the same range and resolution allows the sensor subsystem to be smaller and use less power. Mirror-based Lidar design has relied heavily on micro-machining technology. This is based on the same process that is used to build silicon chips, creating single mirrors or arrays of mirrors on a chip that can be digitally controlled. The implementation of a moving-mirror Lidar has two approaches. The dominant technology is a single rotating mirror, but that is evolving into a single flexed micro-machined mirror that can be digitally controlled. Then there is an array of micro- mirrors, each tens of microns in size that act as pixels for the laser. This ‘digital micro-mirror device’ has had issues with reliability in the past, with individual mirrors sticking in place, but the manufacturing process, yield and reliability testing has improved over the past 10 years to address those issues. The initial design of a 64-channel sensor with a rotating mirror used a biaxial design with the sensor and receiver on two different axes, requiring two lens systems. A 128-channel rotating mirror with biaxial laser and sensor is currently the best performing Lidar sensor for autonomous vehicles, providing the ability to detect an object with 5% reflectivity at a range of 200 m. The sensor can detect a rabbit for example hopping across the road at 60 m, and the image processing on the point cloud, reflectivity information and distance data means the sensor can routinely read road signs. This has evolved into a coaxial design, with the laser and receiver on the same axis with flexed mirrors, so there are no moving parts to wear out. Using manufacturing techniques from the silicon chip industry required new factory equipment to be designed and built but this equipment reduces the cost of the sensor dramatically. There are plans to move the 128-channel, high-resolution, long range Lidar sensor to a coaxial architecture. However, this architecture is currently aimed at less complex sensors. The latest single-channel sensor using a single- mirror coaxial architecture and 905 nm laser with bulk avalanche photodiode (APD) receivers costs less than $100. For a range of 100 m the power consumption is expected to be 2-3 W with a 10 º field of view (FoV) that generates 12 to 16 lines of resolution. This is intended for detecting objects when changing lanes or reversing. The low power consumption is important as it means less cooling is required, reducing the sensor’s weight to less than 250 g. It would be possible to use a carbon fibre casing to reduce the weight further for UAV applications. Optical waveguides Using an optical waveguide to steer a beam instead of a mirror is an increasingly popular approach for a Lidar sensor. This is a truly solid-state implementation as there are no moving parts, but the various techniques used for optical beam steering introduce thermal challenges in the packaging of the devices. The optical phased arrays consist of an arrangement of optical waveguides where the phase of the light passing through can be changed, either by heating the waveguide or by using periodic nanostructures similar in size to the wavelength of the light. The waveguides are typically built from silicon nitride on a silicon substrate using the same etching processes as for a silicon chip. The waveguides, with hundreds or thousands in an array, do not absorb any light; they just adjust the phase, so they can provide a long range. The thousands of waveguides are separated by 1-2 microns to create an array of static beams with a phase difference. This creates areas of constructive and destructive interference to produce a pattern of spots for sensing in a horizontal scan. The wavelength of the light, typically 1550 nm, can also be varied, with different lasers producing 1551 or 1552 nm light to get a different vertical Lidar sense & avoid | Focus Unmanned Systems Technology | February/March 2020 High-speed processing is key to all Lidar sensor architectures (Courtesy of Seamless Microsystems)