At CDNLive in Taiwan in August, there was a record number of attendees, with about 880 customers and partners. There were almost 40 presentations on six different technical tracks, including a wide variety of user-authored presentations addressing all aspects of design and IP creation, integration, and verification. As with all the CDNLive events around the world, one of the major aims is for users of Cadence tools and IP to learn from each other. One session was an automotive panel discussion. Michael Shih, President of Cadence Asia Pacific, said when he introduced the panel, automotive is a "red-hot market" driven by the drive towards self-driving cars. The four panelists were: Kazuyoshi Yamada, Director, System Business Development, TSMC Masayasu Yoshida, Senior Director, Automotive Solution Business Unit, Renesas C.S. Ma, R&D Director, Phison Raja Tabet, Corporate Vice President of Emerging Technologies and Solutions, Cadence Michael started out with Yamada-san, who is obviously Japanese but lives in Hsinchu right now (although he was grateful for the panel being in English since he says his Mandarin is poor). He started by pointing out that TSMC is a foundry, so its main products are process and wafer manufacturing. Between consumer and automotive there is not much difference. The basic process is qualified to grade 1 at 125°C. There are two options for specialized automotive package. The first is the same regular manufacturing but with stricter manufacturing controls: more wafer inspection scans, stricter wafer scrap limit. So the product will be more tightly centered. This also comes with long-term documentation retention for 15 years, so if something happens after 15 years, TSMC can identify the lot and what process control was done. This is available for most processes. But the second option is that they have started to provide an automotive platform (at 16FFC initially). This has high-temperature SPICE, and also special IP that is ISO 26262 ready, as well as working with companies like Cadence to have their IP ready to meet all the functional safety compliance requirements. Next up was Yoshida-san of Renasas, also Japanese of course, although I think based in Japan. Renasas is the biggest MCU provider worldwide with 40% market share. He was asked a similar question, what is the difference between a consumer grade and an automotive grade MCU. Reneasis used to supply a lot of mobile and consumer MCUs but is now focused on MCUs for automotive and industrial. The biggest difference is quality, no matter what happens, the MCU shouldn't stop. Autonomous driving is demanding higher and higher performance. It is a challenge to deliver high performance with automotive quality and safety. The Chairman of Taiwan-based Phison was originally scheduled on the panel, but couldn't make it so he sent what Michael called "his best engineering head", C.S. Ma. Michael wanted to know what are the challenges to use solid-state memory modules in cars. C.S. pointed out that Phison is a total solution provider of all kinds of memory in mobile phones and storage, but, yes, it will be important in cars. According to Intel, a car will generate 4 terabytes of data in 1.5 hours. So the first challenge is capacity, if 4 terabytes can be filled in 90 minutes. Bandwidth, too, at 700 megabytes per second, but with low latency. You can’t take a long time to make decisions. Another challenge is robustness, which is very important to the car industry, as everyone else has said. A car on the road has to do with life so we have to be very careful. Another challenge, he said, is security, since a lot of data will be collected in real time, so the storage system needs to provide a very secure storage environment. Cadence's Raja was asked what automotive has to do with Cadence. Raja was hired to run this vertical, so what are we doing? Raja pointed out that many things are converging on cars. It is a connected device. But it also has to have an increasing amount of computer power at the edge as we move from level 2/3 to level 4/5. A lot of sensors and data acquisition has to be handled in the car without going all the way back to the cloud. Cars are clearly a big focus for future innovation in storage, process, and so on, as the earlier speakers said. But we need innovation on the architecture, on IP, tools, and flows. Cadence has been investing a lot in safety, security, reliability. We are creating a portfolio of IP, not just interface IP, but also Tensilica for automotive processing. For levels 4 and 5, you will not be able to deliver the performance required with traditional MCUs or even CPUs. Many people will need to move to application-specific acceleration technologies, and we have the Tensilica platform which is well suited for ADAS processing. There is also a lot of work going on in the area of reliability tools and flows for automotive. We believe automotive is on the technology forefront of innovation right now. We can also leverage what we develop for automotive into neighboring spaces like robotics and industrial. Michael was worried how we will find enough engineers to have these consumer-type products with automotive reliability. Worse, a car is a big ecosystem. OS, applications, mechanical, and more. How do we solve this ecosystem issue so we don’t have to start from scratch? The panel felt that electric vehicles are part of the solution. Engine control with internal combustion is very complicated, but motor control is much simpler with electric traction. So engineers with software experience in other areas will be able to participate. Germany, Japan, and US have history of vehicle production, but the market is moving to China, and so is a lot of engineering. But we need to simplify the architecture. Today a high-end vehicle has 50-80 ECUs. That is already difficult to manage, so the OEMs want to reduce the number of units. So SoCs and networks will be the key components of this new architecture. Yoshida-san pointed out that autonomous driving has to connect into the cloud for traffic information, and also for software update so the latest algorithms for autonomous driving are used. Renasas has provided MCUs and also SoCs but they intend to provide a total platform. Future vehicles will be more like Knight Rider. Renasas has created R-car, which is an ecosystem. Phison is also a member. CS said that Phison has traditionally been in all types of consumer electronics, so they can focus their attention on controller design, which is the core of storage devices. They already have a lot of experience in infotainment systems in cars, which obviously don't have the same reliability requirements. The key is to make storage devices very robust and reliable for automotive. Phison builds everything in-house, including the firmware, software, etc. They see it as a big business opportunity. Michael turned to Yamada-san of TSMC. "You have an automotive line-of-business. I don't know if this is secret, but is it growing? How fast? And is it more about advanced node or mature node?" Yamada-san said that it is growing. But traditionally automotive semiconductor was made by IDMs like Renasas, not in foundry. In foundry, it is small, but growing rapidly from that small base, In the beginning, 10 years ago, growth rate was about 30%. Today we have a good sized business and we are seeing growth of 18-20%, compared to TSMC as a whole growing 5-10% a year. But it is still not that big, total volume-wise it is 3-4% of TSMC’s business. But due to ADAS, the high end, 16nm, is taking off. We expect a lot of growth there but in the meantime we have a lot of business at 0.13um, 0.18um, and 90nm with eFlash. Automotive is really spread around all the processes TSMC offers. Michael asked Raja about software. A Boeing 787 apparently has 10M lines of code (LoC). It seems the most advanced cars today require 100M LoC. And we are not up to level 5 (self-driving) yet. Raja agreed, saying that was what makes cars such an interesting platform for innovation. But he emphasized that the 100M lines of code is not all ADAS, there is infotainment, and third-party providers of services. He told us that we shouldn't forget that ISO 26262 contains requirements for software, too, not just hardware. But those requirements depend on the application. Your infotainment is different from your braking system, obviously. Automotive is no longer truly the vertical with the structure that we’ve known and loved for years, OEMs, tier-1s, tier-2s. Now people from outside of automotive are creating technology, many in software. Two big things are that we need to care about safety and reliability, and second we need to embrace standards. We will need to build redundancy in software as well as hardware. Also, there is a third element: machine learning and deep learning. We are moving away from traditional languages like C++, but deep learning will need the same level of rigor, and that comes from the amount of data that is used for training. Data is the new oil. Time was running out, so Michael's final question for each panelists (with a 1-minute time limit) was this: Today, when people see Tesla cars in Taiwan they go "wow" because they are not everywhere, commonplace. When will self-driving cars no longer be a "wow" in at least somewhere in the world. Some say 2020 but that seems too fast, some say 2040 but that seems too slow. Yamada-san of TSMC: I think by 2025, by the time we have true level 3. No wow-factor by 2025. By 2030, all cars will be somewhat autonomous. Yoshida-san of Renasas: By 2025, it will be available. But full autonomy won't come until much later. C.S. Ma of Phison: I don’t know. But from storage system perspective, Phison is ready. Raja of Cadence: Level 2/3 is imminent. I think McKinsey has an aggressive prediction that by 2030 up to 50% with level 4. Their non-aggressive estimate is 5% or less. Deployment could be geography and urban dependent. It will probably come first for specific use cases such as freeway, taxis/Uber. Maybe 2030. So, thumbs up for our self-driving cars in 2025-2030. Sign up for Sunday Brunch, the weekly Breakfast Bytes email![]()
↧