Is TSMC-Sony A Sign Of Things To Come?

One wonders if TSMC’s plan to build a 28nm 40k wpm fab in Japan to make image sensors for Sony is a sign of things to come.

One normally thinks of a new TSMC fab as being a bleeding edge, industry-leading facility but TSMC has been building trailing edge plants in China for years.

This is however, the first time it has built a  fab for one customer and one product.

Sony has a $7 billion+  revenue business in image sensors which makes the  $2.5 billion cost of such a fab a reasonable proposition,

When you look at the car-makers with their $40 billion annual spend on chips made with equivalent technology you’d think it surprising that none of them have seen a captive fab as a way to secure their growing dependence on chips.

For the phone makers – who spend about the same on chips as the car-makers – Samsung has its own fabs but Apple doesn’t and nor do Huawei, Xiaomi or BBK which makes the No. 5 and No.6 best selling brands Oppo and Vivo.

So there’s a reluctance to run your own fab, which is very understandable, but if, like Sony, you can get TSMC to set up and run a fab just for you – why that’s a proposition which companies might go for and which TSMC might find a profitable way to go.


Comments

15 comments

  1. Dave,
    the question you raise re carmakers is, IMHO, easy to explain: car makers, in spite of the skyrocketing ampunt of semiconductors in the car, have little or no culture in that sector. The Ferrari’s CEO move is a paradigm of the attempt some are making to overcome that big issue

  2. If you counted all the different semiconductor devices in a car, it would amount to hundreds, probably thousands. It would be impractical surely to build a fab that could make all those devices? On top of that, they would be a mixture of different technologies – high power, microcontrollers, analogue, drivers. They would need dozens of fabs surely?

    Then they would need to design most of the devices themselves, as I can’t see the existing IP owners giving away their designs. That would require a whole army of designers in the various different technologies.

    The same argument could be expanded to metals – steel is currently in short supply too – plastics, etc. Why just concentrate on wafer fabs?

    It’s really quite understandable why they don’t have their own fabs.

  3. The problem the auto industry has is that it thinks it’s big and important — which it is, but not for semiconductors. The number and especially value of semiconductors used by the auto industry is *way* smaller than many other market sectors (computing, mobile, networking, consumer) and it’s split over a large number of different small-volume designs.

    So from the foundry point of view, when there’s a capacity crunch (or a factory fire, or an earthquake…) the auto industry goes to the back of the queue and their deliveries get pushed out. Which would be OK if they kept big stocks, but thanks to the JIT beancounters they don’t. So the lack of a $10 chip stops them making a $30000 car and the production lines grind to a halt.

    The way to avoid this is to either pay a price premium to make your business more attractive to the foundries, or to keep enough stock to smooth out supply fluctuations. But both these cost money, and the beancounters don’t like doing that…

  4. What surprises me is that automotive ICs rarely make use of the latest bleeding edge nodes, but mostly legacy processes (for analog, high voltage, etc). Hence, I don’t see the relation between increased volumes of CPUs, phones, XBox, etc and the automotive shortage. (except if it is due to the increased volume in peripherals like display drivers, PMIC, …)

    • You’re right johan the car guys are using different processes to the phone and gamer guys so they shouldn’t be competing for the same capacity. I suppose TSMC doesn’t keep a huge amount of trailing edge foundry capacity available because it becomes less profitable than bleeding edge and, in the current crisis, TSMC told some auto customers it could give them their chips if it moved them onto to more advanced processes where it had some spare capacity. It makes it all the more sensible for the car makers to ask TSMC to run trailing edge fabs just for them.

      • My understanding is that the design cycle for a car is significantly longer than for a consumer gadget. Also in specifying the components, auto manufacturers are inherently conservative. So by the time the volume manufacturing starts the semiconductors are going to be several generations behind the bleeding edge.

        Trailing edge fabs are essentially well written down financially so any production has to cover only the costs of manufacture. This suggests that running product at the trailing edge could be profitable for the fab owner.

  5. Although Apple does not have it’s own fab’s, they are VERY involved in the capacity planning and technology transfer / capability at the foundries they use (including on-site inspections). So they are basically doing what you propose, “get your foundry to set up a fab for you”. Important difference is lifetime, Apple changes the chips they need fast (<2 years) so a dedicated fab doesn't make sense, while imaging sensors have a much longer product(ion) lifetime.

    • Very good point martijn, Obviously Apple has a great relationship with TSMC and even if Apple could fill a 100k wpm gigafab it wouldn’t make sense to have a dedicated fab if was going to be obsolete in two years. Sony sensors, as you say, are a different beast.

  6. Bosch just started production in a new fab (at Dresden)

    • Very true johan but is all the output from that fab going to Bosch internal demand or will some of it be sold on the open market? After all Bosch is the 6th largest supplier of ICs to the auto industry so I expect it intends to sell to multiple customers in the merchant chip market.

    • That’s all MEMS and sensors. Huge market but totally different to Apple as it doesn’t need constant scaling to stay competitive.

Leave a Reply

Your email address will not be published. Required fields are marked *

*