Lattice Semiconductor Corporation (LSCC) Earnings Call Transcript & Summary

Hello, and a very big welcome to our global audience today. Welcome to the OpenSystems Media Webcast for Autonomous Vehicles: We're Already There (Sort Of). I really like that title and you will understand exactly what it means very shortly. This webcast is sponsored by Cadence, DFI, ARM, Achronix and Lattice. My name is Rich Nass and I will be your moderator for this webcast. I'm also the Executive Vice President of OpenSystems Media. I lead the Embedded and IoT teams for Embedded Computing Design. If you are not familiar with that property, you will find that it is a great resource for anything related to the design of hardware and software for embedded systems. You can find it at www.embedded-computing.com. It is filled with design articles, whitepapers, blogs, videos, podcasts and a whole lot more. On this webcast, you'll be hearing from, in order of appearance: Robert Day, the Director of Autonomous Vehicles at ARM; Robert Schweiger, the Director of Automotive Solutions at Cadence; Bob Siller, the Director of Product Marketing at Achronix; then Jatinder Singh, the Marketing Manager for Automotive at Lattice Semiconductor; and Crystal Tseng, the Director of System Product Management at DFI. In the interest of time, I will not go through their bios. You can find them all on the console in one of the windows, but trust me when I say they are all more than qualified to speak on this topic. Also, keeping my own comments as brief as possible, so we can get right to the experts. Obviously, you know why you're here. You want to learn about the components that are needed, both the software and the hardware, to finally realize the dream of autonomous drive. Before I bring out the first speaker, let me very quickly go over some of the housekeeping items. This entire presentation will take roughly 50-ish minutes, and hopefully, we'll have some time at the end to answer your questions, and we will get to as many as time permits. We tend to answer them in the order that we receive them. So I encourage you to ask them as soon as you think of them rather than waiting until all of the presentations are over. You'll find on the console window an area where you can answer those questions in real time. And as I said, we will get to as many as we have time for. If we do not have a chance to answer your question within the webcast, someone will get back to you after the webcast is over with an answer. So every question gets answered. When you do ask your question, if you can tell us who it's being directed at, that will help us afterwards to get you an answer if we don't have time to answer it during the webcast. So we'll have that person or representative from that company get back to you. If you have a question pertaining to the operation of the webcast itself like if the slides aren't working right or if the audio isn't sounding right, put that question into the question box, and one of the tech people will get back to you with a solution. There's also a maximize button that allows you to see the slides in a larger window, and then you can download the slides from the handouts widget that's on your -- on the console. This will be available in an archived version when we're done. So if you missed something or if you want to go back and review or if you want to hand it off to a friend or a colleague, it's there for you. Okay. Enough for me. Let's bring out the first speaker, who is Robert Day from ARM. Robert, take it away.

Okay. All right. So what we're going to do is have a quick look at how close we are. So what we see right now is a lot of the driverless technology is really pretty being prototyped. So it's a whole ton of compute in the back of typically a traditional vehicle, and then there's usually a safety driver and a technician when these cars are going along. To get to true driverless really though, we need to get rid of the -- all of the compute -- all of the big compute in the trunk so that the vehicle is useful and actually have a vehicle that's fit for purpose fully with no steering wheel and no brakes and no driver. And that's the challenge, is how close are we to that right now. One way of really looking at this is to sort of see what's going on in the world and specifically other regulations that are going to either impede or help with truly driverless vehicles on our streets. What we've been seeing in California, which is where I'm based, is a lot of the autonomous technology companies are now obtaining permits from the California DMV to actually run autonomous vehicles with no safety drivers on certain streets and certain areas within California. And you can see here, we've got companies like Cruise, Nuro, AutoX and Waymo that are all starting basically testing without safety drivers. So one of the interesting things about autonomy is that you can see here we've got the different levels, which go all the way from Level 1 from the SAE International, which is really ADAS functionality, all the way to Level 5. If you think about Level 5, this is defined as an autonomous vehicle that can drive under all conditions pretty much at all speeds, and this is probably where we'll need to get to for our own vehicles, if we ever want them to be truly autonomous, because we probably want them to take us anywhere we want to go. What's happening in reality is a lot of these autonomous vehicles are actually within a restricted operational design domain. So what that says is this vehicle is going to operate in these conditions and that condition might be speed limitation, it might be geographic limitation or it might be environmental like it will only go out when it's daylight or sunny. The interesting thing with these restricted ODDs is that they actually then make the task of autonomy a little easier. So if you're operating in an area, for example, where there is no pedestrians, then you don't have to worry about calculating, is there a pedestrian, or what should I do about them. And that's what we're really seeing now as far as these initial truly autonomous vehicles, is they are operating within certain operational design domains. And let's have a little look at what some of those look like. So we are seeing the -- probably the most kind of famous robotaxis, which is basically taking your ride hailing and removing the driver. These are typically prototyping right now with a safety driver that typically in certain areas that they know very well and they're geofenced. They're typically in areas where the weather conditions are good like California or in Arizona, and these are typically built using an existing vehicle and then basically a bunch of sensors and computer bolted on. And this is kind of like where a lot of the publicity is going around the Ubers and Waymos and people like that. We then get this other category, which is sort of similar, but they are basically first or last-mile shuttles. So they generally -- these are generally prototyping with no driver. They are often built for purpose. They are certainly restricted in where they go, often things like college campuses, retirement communities, transportation hubs, that sort of thing. And they typically run at relatively low speed. And then on the far right, you can see a last-mile delivery. This is kind of an interesting -- this is getting some good publicity within COVID time because they are potentially being able to deliver groceries to us without having us to interact with any humans to get those groceries. These always are with no driver -- pretty much always with no driver because they're -- again, they're built for purpose. They typically go really slowly because the groceries don't care how fast they go and they're typically on very fixed routes, let's say, from a grocery store to a residential area. So at ARM, we've been trying to figure out how do we get these to real deployment, how do we get rid of all that prototyping stuff and get to an actual deployment. So we actually commissioned Forrester to go and talk to people that is developing autonomous systems and really figure out what some of the challenges are, and this is -- it's actually available on our website, arm.com. So yes, go and take a download and you will get an idea of what a lot of people in the industry, a lot of the engineers, are kind of looking at autonomy, what they think about getting to actual production versus prototyping. We also had a look at what some of the companies are saying publicly now and really they -- people are kind of backing off a little bit when we get full autonomy or saying that there'll be some constraints, as you can see at the top there, implying a restricted ODD. So we dug a little deeper and what we've really looked at were some of the challenges that were kind of outlined in this. And I kind of liken it to a term in the avionics world, SWaP-C, which is size, weight, power and cost. And if you look at a lot of the things that are challenges to getting to production, it all goes around reducing the size, the thermals, the cost and everything else to try and actually make this deployable. The other thing is as you remove the safety driver, then things like safety and security become a real issue and that has to be built into the compute complex. So what we're doing at ARM is we're looking at, okay, what does this really mean. So we have to keep the level of compute up. It has to be high performance. It has to be scalable to go through these different types of ODD. But at the same time, we need to reduce the power. We need to reduce the costs and we need to introduce safety into the system. So we've really been looking at, well, what does an autonomous -- a deployable autonomous solution look like, both from a hardware perspective and also from a software perspective. And so I'm going to give you kind of like an example of the types of technology that really need to go into this autonomous solution, and we are really looking at this being system on chips that will actually go into real production vehicles. And you can see that we are having high-performance CPUs, high performance with safety built into them; having different accelerators, GPUs, MLAs, et cetera, et cetera; having things like ISPs to make sure that the vision -- the camera images are usable by the compute; building in safety islands and security enclaves to make sure that the system can't be hacked and to make sure that it will always work in a safe manner, especially when there's no safety driver to hand back to. And then the other thing is having connectivity. So basically being able to scale these systems up as the ODDs get more complex, and we can do that through things like C6 and basically PCI bridges between the systems. So that's a sort of hardware system architecture, which at ARM we typically think about because we're building that IP. But now we need to layer on the software. So there needs to be some standard architectures for software. There needs to be open-source development platforms and also commercial. So basically, people can start developing using open source, but then as they get closer to actual deployment, could move into commercial and safety critical software solutions. Having middleware is essential. Having things like ROS and DDS available as a way of communicating within the system and then finally, having the right set of standard based tools and development environment. And I think we'll see a lot of that as we go through today's webinar. And then on top of that, we need a whole vibrant ecosystem to make sure that the autonomous solutions can be taken to market. And again, we have people building SOCs. We have open source, we have commercial software, and we have people actually building the autonomous technologies. There are open-source autonomous stacks out there like Autoware and Apollo. And finally, we are now starting to see consortiums being formed, where people can work together to try and solve this challenge. And the last slide I'm going to show you today is a consortium that was founded last year, the Autonomous Vehicle Computing Consortium, which was really looking at the challenge of how to get deployable, high-performance, scalable, heterogeneous compute that basically has a standardized interface. And so you'll see a lot of the companies here, OEMs, Tier 1s, tech companies, silicon providers, they're all coming together to try and figure this out. And these are people that traditionally compete with each other, are actually sitting around in this consortium and trying to form what an autonomous platform could look like. And so I will end there and hand over to Robert Schweiger from Cadence to continue the story.

Thanks, Rob, for the introduction. So hello, everybody. I will focus a little bit more on the hardware side of things, where we are right now with autonomous driving. And so today, automated and autonomous driving is redefining the role of the automobile, and many functions need to be seamlessly integrated into a high-performance compute platform. However, 2020 is almost over and we still do not even have Level 3 automated driving vehicles in mass production. So let's have a look at the challenges in the current status of automated driving platform. So if you look on the right hand side of the slide, and Robert has already mentioned a couple of things, computing performance is a major concern in vehicle networking, E/E architecture, which depends on the networking speed grades of 5 that become available over time and reliability, safety and security. And I will focus on a couple of these challenges. So right now, OEMs are transitioning from a distributed network to a domain-based network and in the future to a zonal network architecture. Domain controllers based on multi-core SoCs enable a domain-centralized E/E architecture, while the domain-based architecture looks a lot more structured. It is not optimized for the physical layout. However, the next step in vehicle -- next step in a vehicle-centralized E/E architecture, also known as zonal architecture, will rely on a central, eventually redundant high-performance compute unit. So different E/E architectures will also have an impact on the sensor architecture, going from a smart sensor via a smart sensor front end to a dumb sensor. However a dumb sensor requires high bandwidth, low latency network architecture for raw sensor data fusion. As shown in the table, for a 4K, 60 frames per second, 16-bit camera produces already a data rate of 8 gigabit per second, which require a 10 gigabit per second Ethernet PHY to transmit the data from the camera to the central processing unit. And so this requires a high bandwidth network. So another interesting area is the evolution of the automated driving platform. For ADAS, Mobileye was the market leader in smart camera for many years. Audi was probably the first OEM that developed their own proprietary ADAS platform based on standard components. Afterwards, companies like NVIDIA, Renesas, NXP and others created various more or less open systems that OEMs and Tier 1 suppliers could use. Tesla was the first OEM that also developed their own proprietary SoC software and machine learning environment, which was tailored to exactly their system requirements. Tesla's full self-driving computer is still today the industry benchmark in terms of TOPS per watt. So let's do a quick comparison of the Audi's zFAS platform and the Tesla's full self-driving computer based on Hardware 3.0. So Audi used for their zFAS system available ICs from companies like NVIDIA, Intel, Mobileye and Infineon using a kind of a bottom-up design approach. By using a top-down design approach, Tesla developed their own dedicated SoC tailored to their overall system requirements. By doing so, Tesla achieved with the same processor footprint a quite substantial performance improvement of about 150x. So as already mentioned, the zonal architecture is optimized for a physical layout in a car. Multi-functional zonal controllers are physically placed on various locations in the car, which helps to reduce the cable length to react to ADAS and the sensors and could also improve the bandwidth. The zonal controllers are connected via high-speed Ethernet backbone to transfer the data to a redundant central compute unit. As a result, a high-resolution raw sensor data processing is now possible. So this is really the next step. And so here you can see the various components. So we will have intelligent gateways, redundant ECUs like in the Tesla platform. So for the bidirectional communication, we will use an Ethernet backbone architecture and we've got high speed endpoint connections. And for that, we're going to use most likely either a MIPI type of protocol or it could be also like mentioned by Robert, a PCI Express type protocol. And of course, everything needs to be scalable from a small, eventually electric vehicle up to a premium type of car. So you need a scalable architecture. So since 2 years, we've been collaborating with the University of Nagoya, who developed a full autonomous driving software called Autoware. We are operating a test vehicle with a hardware platform that is based on a high-performance 10-core SoC. And out of those 10 cores, a quad-core Tensilica Vision P6 processor cluster is doing all the sensor processing and the sensor fusion of the camera and the LiDAR data. The driving platform that is shown on this slide consumes less than 10 watts, thanks to the Tensilica low power consumption. So if we come back to the initial slides on those challenges, Cadence is really a broad-based partner for automotive intelligence system design, and as you can see, we do provide technologies that can be leveraged to address most of these challenges. However, the capability of designing a purpose-built SoC is key for autonomous driving. And with that, I'm done with my presentation and I hand over to the next speaker.

All right. Thank you, Robert. Hi. My name is Bob Siller and I'm the Director of Product Marketing at Achronix. Achronix is a fabless semiconductor company, which has been producing high-performance FPGAs and eFPGA IP since 2004. We're uniquely positioned as the only FPGA company that has both standalone high-performance FPGA devices as well as embedded FPGA or eFPGA IP in high-volume production. And today, I'm going to talk about Achronix' eFPGA solutions, which will enable the next generation of autonomous vehicles. FPGAs have been used in automotive applications for well over a decade. There are over 200 million FPGAs in cars today and 75 million in production ADAS applications. Why FPGAs? Well, FPGAs provide a significant value in applications for performance and latency are critical. The programmable nature of FPGAs allows them to address complex algorithms that require high-speed parallel processing capabilities. Designers can develop custom hardware accelerators that outperform CPU-based systems. In addition, in space and power-constrained environments, FPGAs offer the best performance per watt compared to more general-purpose GPU and CPU-based systems. Now we're seeing automotive customers design further integration and the eFPGA IP from Achronix off this opportunity. As mentioned by the previous speakers, this really allows for a significant cost, power and space savings compared to using solely standalone FPGA devices, and this value proposition makes it attractive to replace some of these standalone FPGA devices with custom ASICs that incorporate eFPGA IP. Let's go into some examples of applications that are using standalone FPGA devices in production today. Daimler, the MBUX Interior Assistant, this is an AI-powered system which brings deep learning capabilities inside the vehicle. One example is the search light function, which during nighttime detects when a passenger is looking for something and can turn on the interior lights with just an extension of the driver's arm. The FPGA device runs the AI algorithms and was selected due to its ability to meet the thermal and performance per watt requirements for this application. Next, Continental has developed a highly flexible FPGA-based control unit for automated driving system. It's available for developers to create custom algorithms, which connect to external sensors to perform sensor fusion on the incoming data streams. It's offered in different configurations to allow for scalability to perform basic to advanced ADAS functionality. As mentioned previously, Audi has also used FPGAs for their Level 3 ADAS platform called zFAS. zFAS performs a range of functions such as adaptive driving assistance, active suspension, parking assistance and traffic congestion assistance. As mentioned previously as well, zFAS uses multiple high-performance devices, including the Cyclone V SoC FPGAs, NVIDIA Tegra K1 GPU, Mobileye EyeQ3 devices and really demonstrates that the amount of processing requirement that is required for Level 3 autonomy and that really no single off-the-shelf component is powerful enough to meet these application requirements. And for small form factor applications, BYD selected FPGAs for its front camera system. The core functions include lane departure warning, forward collision warning, pedestrian collision warning, and all performed in an SoC FPGA, and this is done in a very low-power envelope. Finally, Subaru leverages FPGAs for their EyeSight stereo vision camera systems. In this application, the FPGA generates a 3D point cloud from the stereo cameras, which enable the system to perform ADAS functions such as pedestrian and object detection. As you see in each of these applications, the FPGA is being used for different functions, which demonstrates the wide range of applications that FPGA technology powers. However, as these systems evolve, there's going to be even a greater need for cost and power reduction with higher performance. This is why embedded FPGA IP has started to gain traction in the automotive market. So you might ask, what exactly is eFPGA IP and how does it work? So an embedded FPGA is a licensable ASIC IP core that can be embedded just like any other IP core into an ASIC device. The eFPGA block contains a user-defined amount of FPGA logic, embedded memory and DSP or machine learning processor blocks, which provide high-performance mathematical operations and support various different number formats, such as floating point, block floating point and integer. And unlike standalone FPGAs, designers are not limited to a predefined configuration of the FPGA resources and instead are able to define the exact amount of resources required for their applications. Achronix' eFPGA IP is called Speedcore, and it's available on multiple different process technologies, and the Speedcore eFPGA IP can be designed and configured with Achronix' standard development tools and design flow. So if you already have RTO for an existing standalone FPGA, it can be easily ported to the Achronix' architecture. Once the eFPGA IP resource mix and IO interfaces are defined, we'll simply add this as a target device to our standard software tools, so it appears as a new FPGA design option. Once again, this is a standard FPGA design flow. So, it's really easy for FPGA designers to program eFPGA IP. So over the past several years, there has been increased desire for automotive manufacturers to develop custom checks to differentiate their ADAS offering. Tesla has taken this approach and developed their own full self-driving ASIC, and this allowed them to significantly differentiate their solution by creating the ASIC which has exactly the right amount of processing capability needed. However, this wasn't without challenges. With any ASIC design, you're forced to define the specification years in advance of deployment, and therefore, there is a risk that things may change by the time the vehicle is in production. In the case of AI and ML applications, the algorithms are changing at such a rapid pace that there could be a much better one available when the vehicle is ready for mass production. The ever-changing nature of AI/ML is one reason why an FPGA would be needed. By using eFPGA IP, not only can you adapt to these changing algorithms like you would using a standalone FPGA, but also significantly reduce the unit cost and power of standalone FPGAs. The eFPGA integration provides our customers up to a 90% cost savings compared to standalone FPGA devices, much faster time to market and extend their product life cycles. In addition, eFPGA IP allows up to a 75% reduction of the FPGA power by eliminating the unused resources and the peripheral IO, that's typical in a standalone FPGA. And finally, by integrating the ASIC and eFPGA monolithically, designers can achieve about a 10x increase in the interface bandwidth between the eFPGA and other system components, and this also increases the reliability by simply removing the number of -- reducing the number of components on the PCB. So let's take a look at eFPGA IP and how it maximizes your ASIC's time to market. The black line in this curve demonstrates a standard ASIC product life cycle, and the green line shows that by including eFPGA IP, you can extend the ASIC time to market by enabling earlier tape-out, supporting more product variants and extending the product life cycle, which translates into increased revenue and market opportunity. Designers can tape out earlier by partitioning the parts of the design that might change into the eFPGA, and this allows designers to incorporate last-minute changes without having to respin the ASIC. Simply, the designer just have to reprogram the FPGA to support the latest changes after the ASIC is in production. Additionally, using eFPGA IP inside, your ASIC designers can support more product variants. For example, you can support the entry, mid-range and premium class vehicles with the same ASIC in a different eFPGA program. Finally, eFPGA IP allows designers to extend their product life cycle by developing new features and capabilities and deploying these through remote updates over the time. So let's take a look at some applications of eFPGA IP in ADAS systems. The eFPGA IP can be used for custom in-image and radar sensor interfaces. It can perform tailored sensor fusion algorithms, formatting and aligning disparate data streams and video preprocessing for functions like edge detection or feature extraction. Designers can also create custom data accelerators inside the eFPGA, which are reprogrammable depending upon the application. For example, when the vehicle is moving forward, it can do traffic sign recognition, and while on reverse, they can be reprogramed to do cross-traffic alert. Additionally, in V2X applications, the eFPGA IP can incorporate custom crypto algorithms to increase vehicle security and prevent unauthorized access to the ADAS ECU. Finally, in the case of a camera-based application, where an external image is being displayed on the infotainment console for the driver, the eFPGA IP can be used for different resizing algorithms, compression, formatting the image to be displayed on various different-sized displays. This is just a few of the applications of how you can use eFPGA IP in your automotive design. As you can see, the flexible nature of an FPGA allows for multiple deployment options and provides the ability to differentiate your system versus those that are built solely off -- using off-the-shelf components. So as you can see, FPGAs are already being used in mainstream automotive applications today. They found great success in ADAS application due to the rapidly evolving nature of these systems and the need for more and more processing and performance. The next evolution of ADAS design will be focused on migrating many of these discrete FPGA functions into embedded eFPGA IP, which allows for much lower cost, power and increased performance compared to standalone solutions. So thank you for your time today and I'll kick it over to the next speaker, Jatinder.

Thank you, Bob, and hello, everyone. My name is JP Singh, and I'm the Automotive Marketing Manager at Lattice Semiconductor Corporation. For those who are not familiar with Lattice, we are the largest supplier of low-power FPGAs in the world. We have shipped more than 1 billion devices during the last 4 years. I will be sharing my perspective on the state of autonomous vehicle in the next few minutes and providing a bit of a hardware level view with respect to autonomy. First, let's look at who needs vehicle autonomy. Vehicle autonomy is more than just a self-driving car on our roads. Granted the technology is the same, there are other areas that need this autonomy. One of these areas is logistics and delivery droids perform the mundane tasks of, say, delivering packages to our home, as Bob mentioned, the last-mile delivery. During the current pandemic, we see an increase in investment in these kind of technologies. Another area that can greatly benefit from autonomy is long-haul autonomous trucks that tread thousands of miles to connect our supply chains. Currently, these are human-operated systems. With autonomy, we can reduce the supply chain delivery times and add safety and reduce the overall cost for our systems. Industries like farming, mining, construction can also benefit from the efficiency that these autonomous systems bring. Unlike humans, these systems can operate around the clock to reduce cost, improve the efficiency and add safety to these industries. The next set of services on the left you can see that can greatly benefit are the commuting services. These could be systems like autonomous ride-sharing services, robotaxis, runabouts within our campuses and senior living spaces and of course, the self-driving shuttles in our cities and the downtowns. So let us look at where we are in terms of autonomy and where are we going, how are we going to progress with autonomy. If we look at today's time frame, farming, mining, construction, some autonomous robots on a factory floor in a warehouse have already reached autonomy. Unlike our roads, these are very controlled environments. There are no pedestrians walking around. There is no unexpected events that are happening, and hence, the parameters of autonomy are definite and autonomy is achievable. If we shift our focus to the next decade, today many universities around the world have pilot programs with runabouts as well as some of the senior living spaces to provide mobility and independence to our seniors. A lot of the big-name companies are making investments in logistics and delivery services, too. So I believe that within the next 5 years to 10 years, these vehicles will achieve autonomy, too. Right around a decade is when we will also start to see technology mature enough to be deployed in shuttle services in our cities and long-haul trucks. That is when we will also have autonomous self-driving cars, which I believe are the precursors to the robotaxis. Fully self-driving, humanless robotaxis are at least 1.5 decades to 2 decades out. This time line is actually driven by maturing of technologies that will make it all happen. We have the technologies, but they need to be improved and matured over time. Let us look at these technologies. Autonomy for a vehicle is essentially making our vehicles smarter. What it means is that these vehicles are able to sense, receive and process the information and make intelligent decisions, a very self-thinking machine if you may. So there are sensors needed, the systems that can process data in real time without latency, for example, and AI that can make smarter decisions and all of this in a very safe and secure system. During the next few years, our cars -- sorry, during the last few years, we have seen more and more features that are getting added in our cars. They are making our cars safer. They are making them intelligent, features like surround view, backup cameras, parking assistance, driver monitoring system, distance maintenance between the car in front of you and behind you, collision avoidance, these are becoming more and more standard features in a modern car. On your screen, you can see the various levels of autonomy and the features in our cars that are needed to make it happen. As we increase the autonomy level of a car, we are adding functionalities that are allowing humans to rely more and more on the electronic systems. With the increasing level of autonomy, the cars need a lot of surrounding information and spatial information, and hence, a need for a system that supports more sensors. But if you notice, right around Level 3 onwards, you can see the number of sensors in a vehicle will not increase dramatically. However, the performance and the intelligence embedded in these sensors and in these systems using these sensors will exponentially improve. Some of these improvements include higher resolution and faster sensors or edge sensors with some edge capabilities, processing units with extremely low latency, higher levels of AI and algorithms and more secure systems. All these technologies that will be used have some requirements to be suitable for use in a vehicle. A car -- a vehicle is a fairly complex system and a very harsher environment. So if we look at the perspective of an automotive vendor, what are these requirements? The very first requirement is that these should consume less power. They should be low-power systems. And this is true for all of our electronic systems from our cell phones all the way to the cars that we are driving. Within a vehicle, an electronic system needs to operate in extreme temperatures that in these -- the temperature ranges increases the power consumption of these systems, and hence, we need components that are low power. In autonomous vehicles, less power the electronic systems consume, longer the battery can last. Or in case of electric vehicles, we can get more miles per charge. The next is high-reliability environment for these systems and they need to be secure. These systems have to last the lifetime of a vehicle without any derating in performance. Hence, the need for reliability in an extreme temperature for a system and all the components which are there. Bob talked about the eFPGAs and reprogrammability and how FPGA is becoming more and more popular in automotive. And there is a need for this reprogrammability. Technologies and intelligence algorithms are evolving faster and there is a need to be flexible in implementing these changes quicker and have field upgradability. FPGAs -- the programmer devices in general are the right devices for this. And they enable parallel processing and they add the flexibility that the vendors need to help reduce time to market and overall upgradability to the system. And lastly, small form factor. As we deploy more and more electronic systems, we are fighting with the real estate within a vehicle as well as on systems, boards. So the components need to get smaller and more compact. You can see how our LiDAR systems are becoming smaller and smaller. These are some of the major requirements that any automotive vendor will ask for from a system or a component supplier. You can take a look at how Lattice meets these requirements for FPGAs designed for automotive systems. So back to what I call the eternal question, are we there yet? So we are in many areas as we saw. We are on the road to get self-driving cars and vehicles. We will get there by continuously testing and improving these technologies and intelligence algorithms, develop with a strong regard for safety and security. That is where I would like to end my section, that we are on the road to autonomous vehicles and we will get there soon enough. Now I would like to hand over the reins to Crystal Tseng from DFI. Thank you.

Thank you, Jatinder. Okay. Let's go to the first page. Hello, everyone. This is Crystal from DFI. I'm Director of our System Product Management. I'm very delighted to be here to share with you what DFI can contribute in the autonomous vehicle chain. First of all, I would like to have a short introduction of DFI. We were founded in 1981 in Taipei, Taiwan. It's about 40 years. We are going to celebrate our birthday next year. Along with ascending business outcome, DFI went public and launched its IPO in 2000. We offer versatile industrial-grade computing solutions, including industrial motherboards, system-on-module, industrial computers and panel PC, also displays in our DFI brand. In addition, we provide professional ODM and OEM services by about 200 experienced R&D and [ quality-approved veterans ]. To strengthen DFI's competence and provide the customer with higher added values. DFI has joined the Qisda/BenQ Group in 2017. Qisda/BenQ Group was recognized a Top 100 Global Technology Leader by Thomson Reuters at the same year 2017. Okay. Let's go to our main course. Along with the technology evolution, we have seen 4 major changes or say disruptions in transportation. Firstly, human driving will move to autonomous driving. Secondly, to achieve autonomous driving, the high-speed connectivity will be needed to reduce communication latency. The third, for the environmental protection concern, the EV car is definitely the way to go. Lastly, when the above elements are integrated to the vehicle, of course, the price of the vehicle will be much higher than today. So users may prefer to use shared cars instead of affording a private-owned car. This is the sales forecast for autonomous vehicle from Taiwan MIC report. As you know, SAE defines autonomous vehicles to 6 levels. Levels 0 and 1 is fully human controlled. Starting from Level 2, the vehicles are partially automated. As the previous speaker mentioned, the vehicles across ADAS belong to Level 2 and 3. From the diagram, you can see the Level 3 and Level 4 vehicles will increase year by year until 2025. There will be over 3 million L3 and even vehicles L4 in the market. To fully -- the fully automated car may not be popular in year 2030, but it definitely will get popular back in the next decade. This is the top 20 countries for the Autonomous Vehicle Readiness by KPMG surveyed in 2020. The assessment used 28 different measures organized into 4 pillars: policy and legislation, technology and innovation, infrastructure and consumer acceptance. Singapore is the top ranking of autonomous vehicles readiness and leading on both the customer acceptance and policy and legislation pillars. Singapore has expanded AV testing to cover all public roads in Western Singapore and aims to serve 3 areas with driverless bus from 2022. The Netherlands takes the second place of readiness but gets top ranking on the infrastructure pillar because of high-density of EV charging stations. From a technological point of view, Israel is leading on both AV-related companies and investment scaled by population. United States is second only to Israel on technology and innovation. American technology companies including Apple and Google's Waymo unit and vehicle makers such as General Motors and Ford continue to dominate AV development. This is the network architecture from the vehicle to the cloud. To make autonomous driving come true, everything should be connected and talk to each other when the vehicle can talk to other vehicle, pedestrian and infrastructure and then it doesn't need too much driver's judgment and dramatically increase the road safety and transportation efficiency in the driverless mode. The vehicle itself has to support multi-mode connectivity like 4G, 5G, DRSC or WiFi to connect to other vehicles and pedestrians and also infrastructures like traffic lights and road sensors to reduce the latency between the vehicle to the cloud. The roadside unit, we call it the RSU, plays the edge computing role. It's generally equipped with AI-based cameras to detect hazards and alerts and need to get -- generally needs to get a precise positioning and 3D HD map update to help the autonomous driving. In the autonomous vehicle ecosystem, DFI offers a powerful AI-based computing platform for video analysis by deep learning technology. As you can see the data flow on the slide, there are many sensors like camera, LiDAR, radars equipped on the vehicle. The sensors play the video streaming and then transmit to the AI edge computing platform via telematics inside the car. Through the [ chain new ] networks, the results are sent to the ECUs and activate the ADAS or autopilot system to take corresponding actions. What's DFI's solutions for the autonomous driving? DFI's VC300-CS is the fanless platform with the Intel 9th Generation Coffee Lake CPU and integrates NVIDIA MXM graphics module. The MXM graphics card is widely used in notebook and mission-critical platforms since its compact, rugged and low-power consumption features. Unlike PCI Express type card, the MXM is installed horizontally to the PCB and physically can contact the larger heat sink, thus resulting in better thermal dissipation and vibration assistance. Our VC300-CS can operate under minus 20 to 60 without active fan, even the total system power achieves around 200 watts. The VC300-CS also supports high-speed, low-latency 5G in our communication by option. The 3 mini PCI Express slots and 2 M.2 slots gives the flexibility to support more functionalities like CAN bus, LTE, WiFi and GPS. The VC300-CS will be E-Mark certified and support a wide wattage power input from 9 watts to 36 watts. Here is the list we support -- the MXM list we support on the VC300-CS. You can see our MXM supports 5 years life cycle and we support the GeForce and Quadro technology and the maximum CUDA core we can support is up to 2944 RTX 2080. DFI also has another AI edge computing platform option named EC533-KD-AI supporting PCI Express type card in up to 300 watts. It also supports 4 PoE ports for connecting IP cameras with power. As we are still in Level 2 and Level 3 ADAS stage, how to use AI technology to help on safety driving is the urgent and a realistic topic. DFI is based on NVIDIA Jetson Xavier platform, developing a ready solution which can monitor drivers' behavior like fatigue detection and distraction detection. This solution is widely used in fleet management that can effectively prevent the car accidents due to mainly driver behavior. Okay. The key takeaway. Autonomous vehicle is not far from us, as everybody knows. And second, DFI is a very reliable in-vehicle embedded platform provider. And third, the VC300-CS is the most powerful and robust video analysis platform designed for in-vehicle application. And last, if you have more interest on our platform, please visit our website. Thank you. That's all my presentation. I am going to hand over the -- to our moderator, Rich.

Thank you very much, Crystal, and thank you to all the presenters. We have lots of questions. We don't have lots of time. So hopefully, we can do this in a little bit of rapid-fire. The first one, I will ask this of Robert Day since he had the most time to rest. It has to do with security, Robert, and I'm going to paraphrase the question, but we didn't get into much detail on security in this presentation. What needs to be done from ARM's perspective to make sure these autonomous vehicles are secure?