The answer to the question, “What will it take to make LiDAR scalable for mass automotive deployment,” is the technology that is currently being developed by Insight LiDAR.
There are three different types of technologies that can be used to provide automotive vehicles (cars, trucks, tractor-trailers, etc.) with the ability to perceive what is going on around them in the outside world: camera-based, radar- (radio detecting and ranging) based, and LiDAR- (light detection and ranging) based. Such a machine perception ability is important for advanced driver assistance systems (ADAS) in general, and it’s critical in the case of the fully autonomous (self-driving) vehicles that are to come.
Why do we need to use all three of these technologies? Well, each has different capabilities. When it comes to detecting and understanding road markings and signs, for example, camera-based systems are the way to go because radar and LiDAR systems are unsuited to this task. Cameras combined with artificial intelligence (AI) are also good when it comes to object detection and recognition in general, so why do we need radar and LiDAR?
The answer is that the biggest challenge of autonomous driving is reliability. Having different sensors at different wavelengths with different capabilities goes a long way in providing this reliability. Many of their failure modes don’t correlate — where one fails, another succeeds. Cameras struggle where there’s low color contrast, for example, while LiDAR doesn’t care about color contrast. Both cameras and LiDAR are affected by environmental conditions like fog, which is not a significant issue with radar, and so it goes.
Given that context, there are problems equipping vehicles with all three types of sensors. These problems boil down to size and cost. Today’s cameras and radar systems are relatively small and cheap. By comparison, today’s early-generation LiDAR systems are currently large and expensive. Many people are asking the question: “What will it take to make LiDAR scalable for mass automotive deployment?” The answer is simple – we know that LiDAR systems will achieve scalability when they are smaller and cheaper, so what’s the roadmap for achieving this?”
Select the Right Wavelength
The first step is for the industry to settle on selecting the wavelength of light that is most appropriate for use with automotive LiDAR applications. As we discussed in a previous blog (see Which Wavelength is Better for Automotive LiDAR: 900, 1550, or 1310 nm), the light from lasers with wavelengths around 900 nm can damage the human eye. While some LiDAR developers have responded by simply reducing their power, this has not proven to be a viable solution because that also reduces their effective detection range.
At the other end of the scale, lasers with wavelengths of 1550 nm are safer and can be powered sufficiently to see farther, but supplying this power requires large and expensive fiber amplifiers. Also, 1550 nm lasers are created using indium phosphide, which is itself expensive because it can be fabricated only on small wafers due to problems with yield.
There’s also the fact that current LiDAR sensor systems use a detection technique called time-of-flight (TOF), while next-generation LiDARs employ a more sophisticated detection technique known as frequency-modulated continuous wave (FMCW), which provides much more useful information (900 nm lasers don’t support FMCW).
The sweet spot for automotive LiDAR is found at the 1310 nm wavelength. In addition to being safe to the human eye, sufficient power for long distance detection can be generated using a small, inexpensive, on-chip semiconductor amplifier. Furthermore, 1310 nm lasers are fabricated using gallium arsenide. The key points here are (a) there is already a huge infrastructure in place for working with gallium arsenide (many LEDs are made using this material), and (b) the devices can be created on much larger wafers, which drives down costs, especially at scale.
Reduce the Component Count
In the context of these discussions, the term “scalability” embraces a multi-dimensional solution space that includes being reducible and producible: reducible in terms of size and number of components, and producible in terms of creating tens or hundreds of millions of units at a low cost.
To put this another way, the overriding driver for scalability is low cost in high volume. The fewer the number of parts there are in a system, the less expensive it is to build and the easier it is to assemble and maintain. There’s also the fact that smaller is better because space inside automobiles for sensors is at a premium. Things are already very crowded, so getting smaller is important. The solution to all of these factors is to get as many of the components as possible onto the same semiconductor chip.
As we’ve already discussed, a LiDAR based on a 1550 nm wavelength requires a lot of components to support the laser, including a large, expensive fiber amplifier that is necessary to boost the power to achieve detection over distance. In addition to costing thousands of dollars (thereby placing such a unit outside the cost envelope of consumer cars), the resulting unit is physically ungainly and large, being approximately the size of a flattened shoe box (about 1/3 the height of a regular shoe box).
By comparison, Insight LiDAR’s best-in-class 1310 nm FMCW LiDAR has most of its components, including the laser and amplifier, constructed on the same semiconductor device. Furthermore, in addition to being physically small, our state-of-the-art scanning technique provides 1,000+ points wide by 300+ points deep, which is much higher than competitive solutions. The result is a powerful FMCW LiDAR with a large detection range and high resolution that’s only about the size of a pack of playing cards.
So the answer to the question, “What will it take to make LiDAR scalable for mass automotive deployment?” The right wavelength, coupled with the right scalable semiconductor technology, is the clearest most visible path to wide-scale affordable deployment at a mass-level consumer scale the automotive industry requires.