Never miss a story from Breakfast Bytes. Subscribe for in-depth analysis and articles.
At this year's Automobil Elektronik Kongress, there was a panel session titled Semiconductors: The Base for the Software-Defined Car. Given the troubles of the automotive industry over the last couple of years with semiconductor shortages, maybe just "the base for cars" is all that is needed.
The panel comprised (from left to right in the above picture):
As usual, when I cover panel sessions, to make things easier to follow, I will precede Alfred's questions with "Q." Anything in [brackets] is my commentary, not something the panelists said. Also, this is not a full transcript of everything said.
Q: A car is a semiconductor product on wheels, and without hardware, there is no software. In this panel, we will discuss semiconductor availability issues, strategy, and architectural trends. But let's start with politics. I’ve never seen so many non-technical media writing about semiconductors until the last couple of years. So political bodies finally took action. Was this long overdue? Why?
Magnus: It was long overdue that we got some attention, and people at OEMs understood how the semiconductor supply chain works, and long overdue that we paid attention to the details.
Dipti: One thing the people on this panel recognize but the general population does not is how dependent we are now that semiconductors are part of every moment of our lives. I can't get my car repaired right now because I can't get a headlight. Governments started to recognize how dependent we are too.
Calista: It's not just the dependencies we've had for a long time; it's the thirst for more. We continue to want everything at our fingertips, the immediate experience.
Lars: The semiconductor supply chain is truly global. We have aligned in the last 30 years to a very efficient supply chain. But this supply chain has local singularities. Supply chain resilience is a big topic. It's not that the whole semiconductor supply chain crashed. NXP is firing on all cylinders. We are delivering 45% more than pre-covid. But automotive has a semiconductor content growth of 20% per year, so up 60%. It was a big surprise that it came so fast.
Jens: I'd rather add to the competency. In Europe, we have the #1 in MEMS [Philips, I think], the #1 in power semiconductors [Infineon], we are really great in mixed-signal, we are really great in automotive, and in microcontrollers. That's a real competence we have but we forgot about the value chain which became globalized and became a bit out of balance. I'm glad that politics is paying attention to this issue.
Q: Let’s dig into the depths. The automotive industry competes with the consumer industry for capacity in the factories. How do we guarantee we get our fair share?
Jens: That's an interesting question in the sense of 'when do you recognize that you want to invest in something?'. I'm just remembering when we announced the Dresden fab in 2018 [Bosch has 3 of its own fabs] it was long before we were short on parts. It was obviously good timing, but at that point we didn't see it. We need good planning since we have to develop capacity in advance or we won’t make it. Lars already talked about supply chain resilience, and you can create that by strategy. We play with two partners, one our own manufacturing, one a partner, and that creates resilience.
Lars: The first thing is careful planning. We have treated semiconductors as just-in-time goods, and we switch on and off the ordering behavior, and then we are surprised at the delays because for semiconductors, it takes six months. These are long cycle times. Another thing is the big SoC, and this determines everything in the car. Usually, it is below 20nm, where mobile phones and PCs are. Nevertheless, 98% of automotive semiconductors in the next 15 years will be above 20nm. But in the last three years, 80% of all factory investments have been below 20nm, so consumer. The old factories above 20nm are all written off, so nobody with a financial background wants to build a new factory and write it off over five years. So, what we are seeing is that we have overcapacity below 20nm but shortages above 20nm. I spend my time educating politicians on this. We need to go one level deeper than just "we need more semiconductors."
Jens: I promise we will invest more in these 20nm and above nodes because that's the interesting stuff for automotive.
Lars: Guys in Brussels and Germany are wondering what should be done. We need these AI SoCs but only a few. So we coined the phrase "let's build the NATO of semiconductors," to have friendly nations with the need and the customer base. So put 5nm factories in the US, but 28-40nm factories in Europe and help each other. And then you have supply chain resilience. This is what we are discussing in politics at the moment.
Q: There are voices calling for automotive to join forces and bundle to a large purchasing power. What do you think of the idea?
Magnus: We are bundling within the company already. If we should do it on an industry basis, we need to see if that is financially the right thing to do for the industry.
Q: But often, this is not differentiation. You all buy the same components from the same suppliers, and if all the companies team up and say we need ten million of this component, then the manufacturer will know, and we won't have the shortages.
Lars: On the one hand, our anti-trust agencies will get goosebumps if we are not careful. We saw in the US with the CHIPS act and in Europe with the CHIPS act [see my post China, US, Europe: Everybody's Got a CHIPS Act] the politicians are saying that in case of crisis, they want to manage our supply chain. And I'm at the end of my imagination as to what the lawyers will allow. Take the NXP example: I will have to disclose to the market what I am shipping, and I'm forbidden by law to tell Bosch what I'm shipping to Conti [Continental] and to tell Conti what I'm shipping to Bosch. Even if the politicians know that they then need to decide how much Bosch is using for power tools, white goods, BMW, and the same for Conti. I can't see this as a politician-managed ecosystem. We need other mechanisms because that one doesn't work in my opinion.
Dipti: It is already difficult for each company individually to predict, so there is already variability in the forecast, which is how we got where we are today. We pulled back forecasting, we pulled back orders at the beginning of Covid, and now we are trying not only to recover but make up the upside.
Q: So we need to look for other solutions. We keep hearing OEMs want to develop their own chips. What does that mean for the classic semiconductor manufacturers and the tier-1s? What sort of chips are we talking about?
Magnus: I can only speak to Mercedes here. Our #1 is to make sure we have a good balance between power and compute, which has a direct impact on the range of an electric vehicle. So that is important, and there we are increasing our know-how about what we want to achieve. But we are not going to become a silicon producer. We are evaluating how deep we can go and get into the details of the supply chain.
Jens: Whenever we do this, as soon as volumes come together, prices and costs have implications, even if there is innovation going on. Closer to the computing area, we will see more software/hardware co-design so that the user experience. A key question is whether an OEM will be able to have a dedicated expensive device just for one brand. I'm not sure how long it will live, even if it gives a short-term advantage.
Lars: Look at Tesla. They came forward very early and said we are doing our own chip. But this is one specific class of chip. Tesla sources 99% of its chips from companies like NXP. They just do the AI chip, and it is all digital, no crazy RF woo-woo or anything. Elon thinks he has a pretty clear advantage presently, and I see similar things from other car OEMs to get better level 2,3,4 systems if they want to move fast. For the rest, scale is the name of the game, and robustness and qualification.
Dipti: I see this all the time, where the OEMs are asking whether to make a custom chip or buy from the industry. I think it is harder than people think it is. There is quite a lot of infrastructure required that needs to created and then needs to be loaded, and how do you not over-scale. Sometimes. this makes sense, but often it is more of a discussion about how to influence the entire value chain.
[At this point there was a discussion about Arm vs RISC-V architectures but this post is going to be too long, plus very little of the issues are related to the automotive semiconductor content specifically, so I'll skip that.]
Q: Going back to where we started, we had a lot of turmoil. Fire in fabs, flooding, Covid, supply chain. What have we learned for the future?
Jens: First planning. Second resilience, not just relying on a single fab somewhere in the world. Third, taking care of the right investments in the right nodes and rechecking that we are preparing for the future in the right setup. It's not one single activity; it is a combination of them.
Dipti: It is easy to get depressed with all the things that have happened over the last few years. But the innovation has not gone. The levels and pace are unprecedented, especially in automotive. We worked from home and figured it out.
Calista: We’ve had unprecedented growth over the last few years, now with 2,800 members. Maybe it was that we sat down, stopped getting on planes, and got to work!
Lars: We have found that these value chains are enormously deep. For instance, a year ago, we had flooding in Germany, and a copper mill got washed away. Three weeks later, we found out that this specific super-pure copper goes to Japan, and over 5 or 6 stages goes to leadframes for our products and back into our customers. We managed to avoid having another supply crisis on top of the chip crisis. For these things, we don't know where the next earthquake or disaster might hit and what effect it might have, geopolitics aside, which is even more complicated. The only reasonable answer is to sit together and do inventory planning...but inventories cost money, and you tie up capital. Toyota in Japan did much better than the European OEMs because after Fukushima, they had asked all their suppliers to up their inventory levels, and they had not reduced that yet. So they were OK, at least during the first year of the pandemic. If, over the next three years, we manage the inventory levels down again, just like we read in our MBA textbooks, then good luck in the next crisis. A clever and smart inventory system would be a great lesson to learn. It would be a big step towards a resilient supply chain.
Q: Do the OEMs need to suffer to get things to change?
Magnus: One big thing is that we have learned to be much more professional when dealing with risks. What are the risks? What investment do we need to do to mitigate those risks? Plus, every company needs to be comfortable at the end of the day since a risk-free environment doesn’t exist.
Q: Did it also change the just-in-time process?
Magnus: I'm probably the wrong person to be quoted on this, but my observation is "yes."
Lars: I recall early in the crisis that we were short at one foundry and asked to switch to a different one. But it is a 3-year exercise to move a 28nm product between manufacturers, qualify it again, and bring it back into the car industry. It's like you have a sausage factory, and you have a marmalade factory. Both provide you with something for breakfast. But try and swap recipes. The biggest problem in the semiconductor chain is that reaction times are so long. If we want to bring up a fab: 4 years. Swap products: 3 years. This is much better understood now.
Q: I'm glad this has percolated all through the industry...and talking of percolation, time for the coffee break.
Sign up for Sunday Brunch, the weekly Breakfast Bytes email.