A briefing session with Ralf Muenster, the Vice President at SiLC Technologies, Inc. Interviewed by Eva Sharma 07/06/2021
Frequency Modulation Continuous Wave (FMCW) LiDAR, or coherent sensing technology, is being considered to be the game changing technology. It is said to offer a revolutionary step change in perception capability of machines enabling them to attain true autonomous mobility. The technology is not entirely new but its cost and complexity has so far prevented its ubiquitous deployment. However, given its ability to supply instantaneous velocity, immunity to interference, and range and accuracy capabilities, it has recently become a popular preference amongst auto OEMs and Tier-1s.
The year 2021 started with several announcements revolving around FMCW LiDAR from renowned brands such as Mobileye (Intel), Aurora Innovations, and Denso. But the following questions arise:
Does FMCW technology fit well in the performance criteria of automakers?
Does this technology have the capability to overcome the cons of LiDAR sensors for self-driving applications?
Is it reliable, scalable, and cost-effective? And what advantages does enable over ToF technology?
Companies working on both ToF and FMCW technology have their own pointers to counter one another. To get a deeper understanding on this, M14 Intelligence has had the opportunity to discuss this topic with one of the recognized emerging players in the FMCW LiDAR space, SiLC Technologies, Inc.
The discussion was intent towards gaining in-depth understanding about SiLC’s recent technology breakthroughs in the autonomous vehicle industry, SiLC’s version of FMCW LiDAR, competition, business model, and perspective on the FMCW LiDAR market for ADAS and Autonomous Vehicles Application.
Below is just an excerpt of the extensive dialogue with VP of Business Development and Marketing at SiLC Technologies Mr. Ralf Muenster.
Why FMCW over ToF?
Ralf: There are many advantages of Frequency Modulation Continuous Wave (FMCW) technology over Time-of-flight (ToF) technology that contributes to ensuring the safety of self-driving vehicles. I have been talking with lots of OEMs and tier-1s and almost everyone agrees with the fact that FMCW is a must for perception.
GM in 2017 acquired Strobe, Volkswagen (VW) has invested around $120 million in AEVA, BMW and Toyota invested in Blackmore three years ago, Aurora bought Blackmore and recently acquired OURS Technology. At CES this year, even Mobileye has showcased its new FMCW silicon photonics LiDAR SoC. In addition, lots of OEMs and Tier-1s have expressed their desire for FMCW technology. The major reason is it’s easier to understand and much easier to detect and classify objects accurately.
Let us take an example of scanning out at 200-300m and compare the results from case 1-FMCW and case 2-ToF.
Case 1 (FMCW)- In the first frame and first measurement you detect a pixel return (if not multiples, due to long range capability of the technology). With that return you receive the distance and a velocity vector associated with that pixel. At this point you already know that there is something that is relevant, and you know where this object is going to be in the next frame (as you have the velocity information). Velocity information even helps in object identification by providing a match to velocity signatures of different objects. Hence, all you need to do is wait for another frame to confirm that assumption, and you are done. You have the object detected and your system can predict where it’s going to be in the next few seconds and is ready to plan its action.
Hence, due to the additional velocity information with FMCW, results are immediate without prior knowledge of the situation or time-consuming training of neural networks and it’s done with zero noise interference.
Further more, the availability of velocity information and the systems immunity to interference is a significant advantage to avoid false positives, where systems can take drastic (and potentially dangerous) action based on false information.
Case 2- In ToF, you might get a pixel between 200m to 300m, but you do not know if it is an object or just noise. Generally, in the absence of velocity, you need at least 10 pixels to identify the object. So, you must wait for one frame and another and another to confirm if there is an object and then compare the frames to determine if that object has moved and from there, derive a velocity vector. So basically, it is a very long and process-intensive procedure adding to the overall system latency and compute requirements.
Lots of people using ToF have told me that there is a good deal of pre-clustering needed. Meaning, if you get pixel returns, you create many hypothesis that there are objects and then predict their motions. You may end up with a stack of 20 or 30 pre-clustering vectors for every frame just due to noise and uncertainity. This involves lots of compute overhead to determine where real objects are.
However, on the contrary, in FMCW you get the pixel and then the associated velocity vector instantaneously and you’re pretty much done - with very low compute power and very low latency.
This is the key reason that there is a shift in preference toward FMCW. The only argument they make about FMCW is that it is far away from mass production, but this is where SiLC Technologies comes into play.
How does SiLC differentiate itself from other players in the industry? And what unique differentiation does SiLC offer in the autonomous vehicles (AV) space?
Ralf: What really differentiates SiLC in the FMCW LiDAR space is that our company’s founders are pioneers in the silicon photonics industry and they have been working on this technology for the past 25 years.
Our CEO, Dr. Asghari and Dr. Luff, the VP of engineering, have been pioneers in silicon photonic technology ever since they worked at Bookham Technology, England which resulted in a multibillion dollar successful IPO in early 2000. Bookham was the first company ever to develop silicon photonics for commercial applications and its original focus was on sensing which is an analog application. The foundation of technology they developed was therefore a great match to the needs of an FMCW system which is also analog in nature. The team then contributed to the repositioning of Kotura, Inc. focusing on datacenter optics. Kotura was later bought by Mellanox Technologies, and is now part of Nvidia. SiLC has a significant advantage over other players due to its unique brand of silicon photonics technology which is mature and developed grounds up for imaging. I can say that our platform is 10 to 100 times better across 10 or more relevant parameters that are essential for making a good FMCW LiDAR compared to what our competitors have access to, which is the 'Vanilla flavor' of the technology developed for datacom applications.
This allows us to provide a coherent imaging platform with the best performance and widest range of data compared to alternative solutions available today.
SiLC is the only player today that offers depth, motion, and polarization intensity concurrently.
We offer a solution that integrates all needed optical functions such as laser light source and amplifier, monolithically integrated photo detectors, and all optical circuitry needed to mix the outgoing and incoming signal in a single low-cost silicon platform. Our solution is compatible with a wide range of 3rd party scanning solutions and we’re even developing an-inhouse solid-state scanning solution.
The exclusivity that SiLC has is the high-volume capacity, qualified, and already running manufacturing facility that provides us the ease to mass-produce our technology and drastically reduce the time-to-market.
Our solution utilizes shortwave infrared wavelength (SWIR), which is up to 100x more eye safe than traditional ToF NIR wavelengths. Our solution offers long range and high resolution which allows identification of objects quicker along with analysis of their form and shape.
We are the only company that offers polarization and wavelength information.
What is SiLC’s product roadmap and development schedule?
Ralf: Since our inception in 2018, we have raised a total of over $30 million in funding including the recent closure of a series A round of $17 million. Since mid-2020, we have been shipping development kits and planning a product launch later this year.
Also, we are going to have a vehicle on road equipped by one of our tier-1 customers this summer, so basically our A sample will be out later this year.
What short-term goals do you want to achieve with the investment you have received recently?
Ralf: We want to get our solution ready for volume production, expand our customer engagements and grow our engineering team and overall resources.
Which are SiLC Technologies’ target markets and focus areas?
Ralf: The entire mobility space is our prime target market including the robotic delivery vehicles, ADAS, autonomous trucks, shuttles, basically where velocity, long range, and interference-free operation is key.
Also, the other markets of interest are security, surveillance, consumer electronics, industrial robotics, and environmental monitoring where we see a lot of traction. In these segments, our solution with its high precision velocity, accuracy, long range, less interference, and eye safety is helpful in figuring out behavioral differences, monitoring crowds, and 3D face detection.
From which applications do you see maximum demand/traction coming from?
Ralf: More than half of our customers are from automotive and transportation sectors, where we see a wide opening for SiLC’s solution. Despite the popularity that other LiDAR companies have gained through SPACs and IPOs, we are getting requests for our samples from the majority of tier-1s, and many of them are ready to work with us. We have witnessed demand for our product amongst leading OEMs, AV companies, and robotic delivery vehicle companies.
We have good traction in other sectors as well, including the industrial robotics and surveillance industries. To expand our reach in these sectors, we are working with very large consumer conglomerates of leading industrial sensing and robotics players.
Then, the big market of course is the mobile and consumer electronics space where AR plays a critical role. In this segment, we are a couple of years away as we push ourselves to further reduce size and cost. In the mobile industry, we are confident that we can achieve millimeter level accuracy and about 30 meters in range with very low laser power. Also, we are working on a very steep cost reduction curve, which will take a couple of years and volume production is required to control the price parameter. We are working with very large consumer players in this industry as well.
What is the price range you are targeting for your automotive grade lidar solution?
Ralf: Later this year we will be the first FMCW chip integrator solution available for production.
For automotive ADAS applications a highly integrated multi-channel FMCW LiDAR chip solution is expected to enable sub $500 LiDARs in reasonable volumes, like a million units per year.
However, I do not expect to see it hit production until 2025 for the automotive ADAS segment.
What is SiLC’s business model?
Ralf: We are a fabless optical sub-system IC manufacturer, like Texas Instruments or Xilinx. We provide our own components, and create reference designs, and software solutions. We work with leading tier-1s, systems integrators, and OEMs to integrate our solution into their system products. Unlike other LiDAR players, we do not create our own FPGA and DSP, as they are readily available and there is no need to recreate. So, our business model is quite different from some of our LiDAR competitors as they try to do everything in-house.
We are directly connected to OEMs in the autonomous shuttle and robotic vehicle sector who are building their own hardware. However, in the automotive space, we work with OEMs through their traditional tier-1s.
Typically, the automotive design cycles are from 3 to 5 years, but as we are working with some renowned tier-1s for quite some time, we may be less than 3 years away from mass-production.
Ultimately, the automotive tier-1s want to retain their differentiation and are extremely good at using semiconductor components, putting them together in a housing, getting them qualified for automotive-grade, integrating software, and supplying the final product to OEMs. In this scenario, I think we are very well-positioned in the industry as we are working with a number of leading tier-1s on a number of programs. This is exactly our business model.
Who do you consider as your biggest competitor?
Ralf: There are several players working aggressively on FMCW technology. Intel (Mobileye) and AEVA are the major ones.
We believe Intel is likely to be a captive supplier to Mobileye and according to public statements, Intel is limited to below to a maximum of 100mW laser power on their chips and that is a big constraint. I think, at least 200mW for longer range LiDAR is essential, and at SiLC our technology can handles up to 900mW. Intel is not really a direct market competitor and according to their own comments won’t be ready for production until 2025.
AEVA also claims to work on silicon photonics technology; however, the company claims to do everything including perception software, DSP, ASIC, and scanner. Silicon photonics is not easy to do and needs dedication and experience in the field, which is a major core competence for us. In the end, if they don’t have the right technology, starting from scratch will take quite a long time. Also using an off-the shelf platform won’t work as it is not optimized for these applications. Last year at the CES show, SiLC demonstrated it’s silicon photonics FMCW chip under a microscope while scanning 350m out. We haven’t seen a similar demonstration from AEVA or anyone else and believe that we’re well ahead of the competition in terms of chip integration.
FMCW LiDAR Industry Dynamics and Market Trends
Read in-depth market analysis of FMCW LiDAR technology and its comparison with other LiDAR technologies for the application in ADAS and autonomous vehicles