Making More Reliable And More Efficient Auto ICs

Sam Geha, executive vice president of memory solutions at Infineon Technologies, sat down with Semiconductor Engineering to talk about automotive chips, supply chain issues, and integration challenges. What follows are excerpts of that conversation.

SE: How do you build an automotive chip that will work in any environment?

Geha: The automotive market is, of course, one of the most demanding segments in terms of chip reliability. Products made for this segment require more rigorous verification at the design phase as well as rigorous testing of the chips at automotive-grade temperatures, which range from -40°C to 125°C in order to meet the quality standards defined by the automotive industry. Product requirements for the industrial and consumer markets are less rigorous and don’t need the safety margin required for automotive applications.

SE: How about aerospace?

Geha: The requirements are a little bit more stringent for aerospace and defense. The aerospace market requires a temperature range of -55°C to 125°C, so they have incremental requirements beyond automotive on the colder side.

SE: Are your aerospace chips being manufactured by a foundry?

Geha: Yes, we currently use two different foundries for the aerospace and defense market. SkyWater used to be our internal fab, and still runs about 75% of its overall volume for Infineon (formerly Cypress). We run the majority of our aerospace volume there. We are also qualified at UMC for some of these aerospace and defense products.

SE: So how do you ensure reliability across the supply chain?

Geha: Infineon products have the lowest PPB or PPM level in the industry. About 45% of Infineon’s revenue is automotive, which is almost €5 billion. The challenge there is to make sure you don’t get any returns. It all starts with a Zero “customer spill” mentality. Everyone is focused on making sure that ‘Our part is not the reason why that car is sitting on the side of the road.’ Or it could be, ‘This part worked fine and it tested fine, and I gave it to the customer and it failed. But was it our part, or was it another part on the customer board that caused the problem?’ You have to investigate all these cases. We control the front end for automotive. We look at what’s called the CPK (process capability index), which includes the parameters that are controlled in the factory for automotive. In some cases, we ‘cherry-pick’ lots. If a lot is outside of the distribution of when a certain tool is running, we say, “This lot is not for automotive.” We declassify it right away. We also perform all this back-end testing, at 125°C, room temperature 25°C, and -40°C, etc., and add additional tests to ensure the product is robust. There is a premium associated with running wafers for automotive, and the premium additional cost from the front end is about 10%. From the back end, it could be something similar, in order to ensure the product meets automotive quality. That’s why those chips are more reliable.

SE: But as you develop these chips, is there a recognition that not every piece of electronics is going to be perfect? At older nodes, it’s easier to prove that a chip is pretty darn good. It’s a lot harder when you get to some of these advanced chips.

Geha: Yes, we recognize that there could still be failures but we do our best to ensure perfection. This includes adding a built-in self-test (BiST) for advanced chips, which is used to validate the chips will really function. This is especially important after packaging because it may come out perfect from the fab, and then you have some issues with assembly. At the end of the day, there is no way to guarantee 100% perfection. You still take your car to the dealer for upgrades, improvements or to replace damaged boards caused by failures. For newer cars, the advanced chips have the enhanced ability to do more over-the-air updates, similar to what you do with a phone. This is the world that we’re heading toward. The new focus is shifting towards more safety and better security to ensure that people don’t hack into your car.

SE: Infineon is betting heavily on new compound materials like SiC and GaN. How is that going, and what kinds of issues are you encountering?

Geha: Silicon carbide is the main focus in the near-term and mid-term with major growth and expansion coming in the next 2-3 years. Infineon already built one fab in Austria for €1.6 billion, and we’re building another fab in Kulim, Malaysia, to more than double the capacity of silicon carbide. The short-term challenge is producing enough samples for customers now to get qualified in the next generation products. We have to select who we want to work with long-term because everything we’re making now is already booked or about to be booked. If a car company has a vision to grow to double-digit millions of electric cars, think about the amount of silicon carbide devices required. The questions are: where are you going to make it, which factory, and how much volume do you need — and, therefore, how soon can you ramp? The challenge in silicon carbide today is about ramp and about the material itself, but it’s also about whether you can scale it. Will it be akin to Moore’s Law, where there’s going to be next-generation chips where you can shrink the design and actually get more material in the same silicon area? We’re looking at all that. So, there will be a next generation of silicon carbide, as well, to serve the same purpose, same high breakdown voltage, low R, or R_DS(on) [drain-source on resistance]. The learning curve for GaN is similar, with challenges in ramping supply of material and yield that are being steadily addressed. The timeline for this ramp is probably a few more years. The long-term big benefit of GaN is that it is easier to ramp since it uses silicon-based substrates, which are easier to produce than silicon carbide substrates.

SE: One of the big problems with SiC has been wafer defectivity. Is that solved?

Geha: Silicon itself, even before silicon carbide, had similar types of problems. The good news is that while these are not the exact problems silicon had, they’re similar enough, and they are solvable to the point where yield will improve. It’s always a question of how rapidly yields can be improved. For example, cracks in the substrate are similar to silicon, which is why people started adding epi layers on top of the silicon. Whatever defects you had in the silicon, you protected it with epi, and then you built your device on top of the epi. Those are solvable manufacturing or technology development problems. But getting the substrate material itself for silicon carbide is expensive. It’s rapidly becoming a capacity problem as you fast forward.

SE: How do you see that capacity problem playing out over the long term?

Geha: Infineon targets at least 25% share of the world’s market for power electronics with silicon MOSFETs, IGBTs, silicon carbide, and gallium nitride. We have plans and we’re investing enough to stay at that level even as more factories go online and various companies compete with us, which is a big deal. We have a lot of technical expertise in designing and manufacturing SiC as well as GaN and we will continue to be a leader in the power electronics market.

SE: Do you foresee it being used in markets other than automotive?

Geha: Absolutely, silicon carbide is being used in charging stations for electrification as well as the energy infrastructure. There are a lot of applications in industrial where SiC makes a lot of sense. But who do you want to serve first? Your industrial customers? Your automotive customers? Or your energy infrastructure customers? Also, every automotive company is now is looking at multiple suppliers because they know there’s a shortage.

SE: Within automotive there are a lot of unknowns. How do you see that evolving?

Geha: There are three trends that we focus on in automotive. Trend number one is what we call zero emissions become real, and that’s mostly about electrification. The second is what we call a driver becomes a passenger. This is about driver assist and it’s about driving safety, what we call Vision Zero, which is basically no accidents ever — zero fatalities. And, the third is, when a car becomes smarter, and we want the user experience to become a key differentiator. So, you’ve got safety, comfort, and connectivity features.

SE: We’re seeing a push toward zonal architectures in automotive, which is the third major architecture after ECUs and domains. Will this work, or will it be replaced by something else?

Geha: The zonal approach is the future. And that’s a big transition for the automotive companies and Infineon is participating in it.

SE: So how are automakers going to differentiate if everyone is basically using the same motor and transmission?

Geha: The best way to differentiate is through reliability. Another is self-driving and smarter cars. That requires secure connectivity and data integrity for things like smarter tires, pressure sensors, and magnetic switches, which is inductive sensing.

SE: A lot of those features have been around for a while. But what’s changed here is people are trying to put them together into some modules that can do some processing at or close to the source, right?

Geha: Absolutely. Mostly, you want it to self-adjust. Instead of a human pushing a button to make it happen, you want it to be automatic. That’s part of self-driving. You’ve got enough sensors that are telling you, ‘Hey, there’s a problem here, you have to swerve right or swerve left, or whatever.’ And yes, the sensors themselves have been around for a long time.

SE: So how do you deal with the challenge of making all the pieces work together? Who’s responsible for doing that?

Geha: It’s not us alone, we work closely with leading companies throughout the supply chain on integration and system issues. We call it product-to-system; we have application engineers who work closely with these companies to make sure the system works. We don’t necessarily sell directly to the original equipment manufacturers (OEMs), but we communicate with them about their requirements. We usually sell to the board makers (Tier 1) that make sure those boards will work.

SE: With automotive, we’re dealing with more complicated interactions and a lot of different kinds of chips. Where does this go next?

Geha: For microcontrollers, which is probably the technology that needs to continue to be the fastest and most advanced, you want smarter microcontrollers. They need to be able to handle a lot more transactions. In our case, people are going to the TriCore-based AURIX MCU for higher performance in real-time control applications. If a chip is manufactured at 22nm, 16nm, or even smaller geometries, versus an older chip at 40nm, it’s going to have a much faster response time. It will take less time to analyze the data. The trend is toward both faster processing and more data to process.

SE: So it’s all about data now?

Geha: Yes, and that includes safety. We’re heading toward Level 4 and Level 5 safety levels, so you have to make sure you meet ASIL-D (Automotive Safety & Integrity Level – D) requirements at board levels. Mostly, you’re enhancing ADAS. As ADAS improves, you get tons of information coming in from every sensor in your car, from every camera, and you’re trying to analyze it and do something with it. All of this is used to control the steering wheel to move the car left, right, or stay on current course. That continues to be the trend, and it’s going to be for many years. To get to Level 5, where you’re working to prevent all accidents, there are a lot more things you have to watch out for, because not all the roads are ready for that. You can make what we call a ‘smart city and have all kinds of sensors on the roadside. But today, you have to deal with images and visuals that are not necessarily perfect.

SE: As we move forward into the different levels, bringing all these different sensors into basically a system-wide view, what happens on the security side?

Geha: Security basically needs to be built into every chip. Every component needs some type of security, and a handshake with the other chips in the system, all the way down to memory. The microcontroller itself has to have that, but it also needs secure memory. It’s a secure node that combines functional safety and security. We’re protecting code and data from hackers, making sure that we are only talking to the exact microcontroller doing the controlling, saying, ‘Okay, give me information now, NOR chip, and let me handshake with you to make sure I’m really the right microcontroller and you’re really the right NOR chip.’ For each one, there’s a security level. So, there’s secure boot, and secure over-the-air updates through a secure gateway. Every interaction has to have that security connection.

SE: One of the big knobs you can turn here is designing the software in conjunction with the hardware, whether you call it software-defined hardware or hardware-software co-design. What does that do in terms of the ability to respond to security threats?

Geha: We’re not to the point where the response shuts something down or responds to the threat, but we do have teams of software and firmware people who make sure they’re really enhancing and improving the signals and things like that. At the end of the day, it’s about protecting the data as opposed to responding to hacks. Trying to find ways to prevent the hacks from happening probably is the best way to think about it. That’s why we do this private handshake, which means this NOR chip only talks to this MCU, and you have to have the handshake to make sure it really happens. One challenge we have to address is, ‘Okay, I want to start my car and turn on the radio, and I am going into reverse and I want my camera to turn on right away.’ I don’t have time to wait for your handshake to happen before I turn on the car, because otherwise you’re losing features or making the response time slower. Our innovation is more about making sure boot-up time is super-fast while still supporting the appropriate level of security.

SE: What are the top check-off items for an automotive OEM or Tier 1 buying chips? Is it reliability? Is it software? Is it security?

Geha: Reliability has to be a given. Then, the strongest selling point these days is the longevity of the supply. They want to be able to buy it 15 and 20 years from now. It used to be 10 years, but they’re all extending it toward 20 and 25 years. With that come questions like, ‘Do you make it in-house? Do you use foundries? And what kind of contract do you have with the foundries, because lately, some of these foundries have been less committal to long term contracts?’ From a technical standpoint, it’s the PPM level or PPB level. Quality is important. And of course, features are important. But we keep noticing that not everybody, especially in cars, wants to venture out and go for all the most advanced features all the time, specifically Japanese and European carmakers. They’re innovating, but with products that already work. They’re spending more time getting the firmware and the software to work with it, but they don’t necessarily want a sensor that is significantly more sensitive. That’s planned for the next generation. The U.S. carmakers are not that much different. The biggest shift is that the priority on electrification has jumped ahead of the drive towards autonomy. Everyone went from talking about Level 3, 4 and 5 to, ‘The world will get at least to three, and then ultimately 4 and 5.’ Everybody’s focusing on zero emissions and electrification. That’s their priority.

SE: And now the chargers are all suddenly being used in parking lots.

Geha: Yes, and the next big challenge is about the on-board or off-board chargers and how fast they charge.

SE: That’s a whole separate market for a SiC, right?

Geha: Yes, and it’s a big one. You need an infrastructure that supports an electric fleet. Otherwise, you can’t get from Silicon Valley to Los Angeles or San Diego without having to stop and charge for several hours on the way.

SE: Where does Infineon play in memory these days?

Geha: We are in different sectors. With flash memory, we continue to be the leader in automotive NOR flash, which are the chips used for booting up the cars, the cameras in the cars, and the starters and things like that. That continues to be one of our bread-and-butter markets. We are expanding in this area significantly. Obviously, we’re adding security. We’ve added functional safety already. And as microcontrollers move to smaller process geometries, when they go below 22nm and move to finFETs, there is no embedded, non-volatile memory technology that’s available for them. The solution is to have a stand-alone memory chip that is going to have a super-fast external interface and communicates with the micro-controller. So instead of having your memory on-chip, your memory is going to be outside or you can package a two-chip solution using the external memory and the microcontroller under one package. While the microcontroller is doing the fast transactions, it can communicate fast with the memory. That’s a trend that is going to expand big time and we are the world’s leader in this area and well ahead of any competition.

SE: That sort of blows up the old definition of what’s an MCU, though, right? It was always on-chip memory.

Geha: People are evaluating RRAMs and MRAMs and other non-volatile memory technologies to be used on chip, but those technologies currently don’t work at the advanced nodes. And even if they end up working, they’re not going to be reliable at automotive temperatures. You need reliable memory. So, the answer is SiP (system-in-package) using a reliable non-volatile memory with the fast interface that communicates with the micro-controller. We’ve been working with some of the top automotive guys, and we’re actually making it work on a chip that’s a 16nm MCU, with external memory built-in to the package, for a top OEM. That’s only the beginning.

SE: Does flash have the same kind of problems with heat as DRAM does?

Geha: No, but not all flash is created equal. You still have to worry about it because it always behaves differently at high temperatures. That’s where we have a differentiated advantage with our technology over competitors. And that’s why we maintain our leadership position with automotive NOR. In the other markets for memory, we’re seeing a lot of interest in data logging applications, especially for medical. If you think of your pacemaker or other defibrillator, or products like that, you need to always continuously data log to track how things are going. We make the memory that does the data logging. And that requires what we call infinite endurance. In reality, it is something like 100 trillion cycles of endurance. You can always log to it, and it needs to be ultra-low power because you’re not going to change the battery inside the human body anytime soon. Now we’re looking at how do we do it even for higher density. Those are big vectors and big trends that are expanding.

SE: Nothing makes people angrier than when their technology does not work. Is technology becoming more reliable?

Geha: Technology is certainly focused on better reliability. Infineon has one of the best reputations in the industry for making extremely reliable products. We make and sell products only when they meet full reliability specifications. It is and continue to be our differentiated advantage.