Microchip Market Dynamics – Automotive Microchips

Interview Transcript

Article | Microchip Market Dynamics – Automotive Microchips
4th May 2021 Atheneum Team

Expert Profile


Head of Global Automotive, Nexperia (part of NXP Semiconductors)




Juergen’s 38 year career in semiconductors culminated to his appointment as Head of Global Automotive for Nexperia, an NXP Semiconductors business. He previously served as VP Global Car OEM Business Development for NXP. He currently advises a fabless semi conductor high speed optical network start up.

Section 1: Microchip Design & Value Creation

1.1. What is your current design philosophy behind your microchips?

Microchips serve two purposes in the automotive industry. One is looking for customer needs. 20 years ago, Philips Semiconductor developed chips that are going to the 3 big vendors, like IBM. The start of the design process is to have discussions with customers about their solutions, we start with their challenge and then try to understand what they want to achieve.

How can we help the customer to solve their problem? We have to understand the entire situation the customer is in and understanding the capabilities of the company you’re working for. Second part is then to understand their objectives, which is sometimes difficult because some people are not able to formulate clear objectives. The next step is to ask if this is a market need? When is this needed? Is it needed in 1 or 2 years?

Once we started a project, we had to make a ROI calculation. In the regular semiconductor space, it is evaluated over two years. In automotive, it is evaluated over five years which is the equivalent of two generations of technology. This difference is very important.

When you’re looking into the consumer space, like the Apple iPhone, they take the latest available technology and the cost isn’t really taken into account in certain calculations. However, when you go into automotive, calculations are cost sensitive, but they do not need to be the latest technology. When you start designing a microchip in 2021 which is going into production in 2024-25, then you have to look at which types of technology you have today and in 3-4 years’ time. What does it cost in three years?

The final calculation in three years is when the production will start. This means two things. One thing, is that whilst you have an expensive technology, you know in three years it will be substituted by another more expensive technology that is the latest in development. And so, your current technology becomes a little cheaper. The point is to understand is that this is a depreciation. The depreciation of a technology in the next five years. Whereas, in the first two years there could be a depreciation of 50%.

You then have to select, what can I build in and what is the overhead? You can build in a certain overhead because the customer at the end does not notice and recognize this cost. On the other hand, it gives you much more flexibility to sell the microchip to customer A, but also customer B and customer C. Here the scaling effect helps to reduce the costs as well.

1.2. How do you design microchips to meet the demands for the automotive market?

You have to understand the market development. For example, the electrification of cars at the moment. 10 years ago, the transistors you used to, switch on and switch off the power of the batteries. They have a quite high resistance so would get very warm. So, it was clear many years ago that there must be another upcoming technology. This technology must have much lower resistance because the power consumption is the most important. The second part, the warming up of components means you have to reduce the heating.

Which means 10 years ago, the people started to invest in two technologies. One is SiC (Silicium Carbid), the other is GaN (Gallium Nitrid). Both technologies today are in the market. Both technologies are coming in at the right time and are fitting the automotive market requirements. This is also the point, where you have to look in the automotive market, what are the tendencies? What’s going to come? And to prepare, technology wise, yourself for this next step. Again, this requires a lot of research and talking to customers. You have to listen to your customers and to customers, who are perceived as leading edge.

The other point in this case is once you have this type of technology. There’s a demand for understanding. Then you have to look at whether you have the technology or not, which means then you have to start purchasing. GaN (Gallium Nitrid) for instance has been known for 50 years, but only very small companies who had special demand used it and it is produced at low volumes. So, then you have to see how you can industrialize this type of technology. What is the industrialization cost to you? You have to look really ahead in the market, what is needed in future.

1.3. What new technologies have you built into your microchips?

Currently, they have a CMOS technology because CMOS is one thing that is cheap and If you have the chance to get the dimensions, extremely small like 0.17. You have one big additional advantage when you go to CMOS, you just scale it down. It means if you have something which you’ll set by micron, and you say, “Okay, I want to reduce it by 30%.” Then you just scale down everything by 30%. You have to conduct some tests, but then you have your chip ready. I see some other technologies have built in certain applications and have needed analog digital converters. Because our world is speaking analog, but the processing is digital, and this is incompatible. It means there have been cases in earlier years, where people had put analog and digital in one chip. Whereas they could reduce the CMOS space but never the analog, which means cost reduction was quite low.

Therefore, in NXP, 20 years ago, we put a lot of research into replacing analog by digital. Which means you have to make a huge effort to invest into research, but on the other hand you have the scaling capability for this one. So that means demand, today, is going whenever possible into CMOS. Which is in my judgment, 90% of the demand. Other things are made with other technologies, but this is time for dedicated special production where there is no other technology available for the moment. It means in this case; the customer has a need for this technology pay out the price for this one. But this was not the core business of NXP at that time and even today at the time. It means, whatever we have it has to go on volume because we have to fill our FAB.

1.4. What is the future of microchip design in the next 5-10 years?

If I’m looking 10 to 15 years in advance, I see one clear tendency. At the moment when you design the chip, you have a lot of computer power behind it. But there’s still many, many mistakes you can make. Artificial intelligent, AI, is one of the tools which is already being heavily used in the design of some microchips, for compressors of course not for transistors. The tendency is to define it as a black box and say, “Okay, here are my input parameters and here is the description what I want out of its output.” The design of the chip has to be done by the system itself. You can say that we have a robot today. The robot is then building the next robot by itself. This is something which could come in 10 to 15 years because the complexity of design is constantly increasing.

You simply don’t have enough brain power to understand all the criticalities that a chip can have. When you look in the AI activities in the moment, there’s a lot of possibilities where you can go through everything and can make a test in the software area about everything, so this is one direction that I see growth in, at least from a design point of view.

The other part is concerning the technology. Technology of course in semiconductors is getting smaller and the next level of technology will be where you have even only one atom. That means it’s just a function. It’s not any longer that you switch something. That means that you’re just moving a few atom electrons from one stage to the other stages for this.

Because in quantum physics at the moment. The quantum computer today will be in cars in 20 years because the processing need is so huge. It’s getting more and more necessary. So that means you can’t reduce costs any longer within the current dimension of semiconductors. You have to go to other technologies, where you can build up an affordable type of product microchip, which is affordable from the production costs, the development costs and so on.

Whatever will be in cars in five years, is already under development. Likely it is a semiconductor. I would say you can optimize additional development to it. The limitation is only the hardware elements. That means, that the TSMC who is leading in the scale down the technologies in the production and this is a limiting factor at the moment. Technology wise, it could go further down. But at the end, this is a question about if you have a chip aligned, which is below 0.17 micro. You can imagine this still needs to be manufactured.

1.4.1. Fiber Optic

The next technology going to come out is glass fiber used for the ethernet connection in the car. Speed and the data bandwidth in fiber ( e.g. glass) allows for a much higher frequency than copper. There are connectors Phy`s and switches in the line. The optical Phy has the capability to send the signals optically over the fiber. This is a technology which the first companies are starting to have huge advances in development.
What is also really impressive are the Korean and Chinese car makers. The European and the American car makers are still discussing the concerns, whilst in Korea and China they’re discussing the benefits.
The next technology in the car will be optic and will come in 2-3 years. This is so important as some parts of a car’s computer are failing over a number of things. It needs security liability. Or when you have autonomous driving, it is extremely important that you have no failures.

Section 2: Microchip Manufacturing & Production Cost Cutting

2.1. How as the Covid-19 pandemic impact your manufacturing process?

From the market point of view, huge. In the latest report I’ve seen, the overall semiconductor demand of the automotive industry is between 5-9%. So, the remaining 90% is the consumer electronics market and anything else. But the automotive industry, likes to think they are number one. So, when they give a sign, the rest of the industry has to invest. Which last year did not happen. When COVID-19 popped up in spring 2020, was the car industry saw their demand dropping by 50%-80%. This is a type of flatness for commercial stuff because the car industry never gave any commitments. Which means you have a contract to deliver for five years, but the quantities are varied based on their success in the market.

They want to have delivery more or less on time when it’s needed. No stock, nothing. Now it’s coming to a point when you look for new capacity because the demand is huge. Tesla has a 2000 Euro or dollar demand in semiconductor. Even 10 years ago, it was about maybe $200. Investments into a new Wafer Fab, are about $2-3 billion. It takes up to 2 years to bring this additional capacity into market. A wafer fab is only efficient when it’s running 85%, which means you must have a load of 85% as a minimum. Ideally it 100%. If you go below 75%, you make losses and then you just make adjustments. So, you have to make this investment, but have no security of supply.

Who takes the risk? This was under the industry of the semiconductor. And they said, “Okay, fine. Where is the demand?” This was of course in the mobile world; it was in the consumer world. The new PlayStation, whatever else they have eaten up the capacity.

Quite simply the capacity is fully booked. However automotive OEMs shared a forecast with producers, but it is non-binding because the condition says it is clearly a non-binding forecast. So, with a non-binding a microchip producer cannot plan with that. Which in this case from a financial controller perspective of a CarOEM, it was extremely good and efficient. “We don’t put something in stock because it’s just money we have to pay for, and we would be distributing our cashflow. As an unlikely effect of this strategy, the car industry has to shut down factories. On the other hand, if you are not the number one in the market, where you can dictate the market, you’re now running behind. Then you have a problem.

All the savings the car industry made in the last 10 years, they have paid this in one year, and paid back into the market. Clearly, the COVID-19 effects are running full steam. But there is a market differentiation. It is a small percent of the automotive players that are crying loudest, but the rest are smart.

In addition, the latest technology is TSMC in Taiwan. When you see how China says Taiwan is part of greater China or Mainland China. Then you see another problem.

2.2. How do you scale your production process?
2.2.1. Productivity

There are two messages. One message is going to come from the semiconductor area itself. The factory has to deliver each year between 5 to 7% productivity as a request from the car industry. The point is how to improve, how fast and how efficient are you? Then you have to look for the investments you have to take because at the end it has to pay off. Which means when you make some improvements and the improvement does not payback within six months or 12 months, or even in two years for this one. Then you have to look very carefully, whether you’re doing this or whether you look for the next technology because each one after two years has a new technology. So, it’s one thing for the investment manager.

The other point is how to optimize operations. For instance, how to optimize the tooling work you have. How to optimize the molding. How to optimize the final test program. And if you go for maybe 96-98%, this is almost a 100% or 50% less time you need for testing. And testing is something which can take several minutes for complex devices.

Whenever you move from one product to the other product you have to install the new program you have to run tests and so on and so forth, which means installing the new product takes time. It can sometimes take two days; it can be two hours, but it depends on the complexity.

2.2.2. Moldings

Next is when you see the moldings. That means the size of the chip and the plastic around the chip. When I started in semiconductor, a mold component was really a big chip. It was like, a normal razor. If you see today, the mold component for the same functionality it looks a little part on a table. The mold component has been reduced by 90%, with the same functionality and the same accuracy of the device. This is a huge improvement and huge cost savings in this case.

2.3. How can producers cut their production costs?
2.3.1. Reducing testing time

It’s difficult. Simply to the point you have to reduce the testing time.

Just as one example: You have to use the rate of each single piece of microchip because the less weight it has the faster it can be processed. If you have one kilogram, if you take it from A to B it takes a certain amount of time and energy. If you have a hundred grams to take from A to B it would use much less time and energy. If you are talking about micrograms, it is small but if you have millions of them, it can add up to something.

2.3.2. Less expensive parts

The bonding to connect the silicone, the chip inside, to the legs outside, there was a time where they used gold wires and gold was quite expensive. So, then they started using palladium, then we have moved to aluminum. It’s reducing the cost of the material and guaranteeing the same quality and technical stuff. There’s also cost saving with soldering as it depends on which material, you’re using so the soldering costs can also heavily increase or decrease your costs.

2.3.3. Additional software

Ten years ago, VW in Germany had 28 different car radio systems. They then reduced it down to a few systems, which means you still have different options when you look at the brochure. But the point is that the hardware was always the same. What you’re doing is you’re just getting a software key where they switch on the functionality which you want to add on.

Hardware costs are getting lower each year. 10 years ago, was 80% of cost was hardware, software was 20%. At the moment, hardware is maybe 15-20% and the software 80-85%. So, the point is when you put a lot of additional functionalities in the chip it costs less.

You buy a car for £30,000. Say in a few months you come back to seller and you ask for additional features. For this, you might have to pay a monthly fee which means the car manufacturer has a monthly income. 10 years ago, you just got your car and that would be it. If you translate what I’ve said, to the silicon status. You can also reduce the cost, which means to put more functionalities in the silicone itself, so the costs are irrelevant in comparison to the savings you can achieve.