Certify Alexametrics gif

    TSMC's Gift to the Industry

      Sep 22, 2021

    VLSIinsiders' cloudside chat — September 22, 2021

    This week Andrea Lati, Dan Hutcheson, Jason Abt, and Risto Puhakka discuss what insiders are saying about …

    • TSMC’s gift to the industry.
    • 2021 forecast updates: Few notable changes.
    • 2022 the big question of much higher growth:
      • Is spending out of line?
    • The next big thing in image sensors:
      • Dynamic vision sensors bring more dynamic range and are extremely fast.
      • Hybrid bonding of pixel-to-pixel may be possible.

    About VLSIinsiders: Every week our analysts have a cloudside* chat to discuss current events and key issues of concern while sharing what they’ve heard over the past week from the semiconductor industry insiders. 
    * cloudside chat: A fireside chat without the fire and is across the clouds of the internet

    Other VLSIinsiders


    VLSIinsiders' Cloudside Chat — September 22, 2021

    DAN HUTCHESON: So, welcome to this week's VLSIinsiders. This is Dan Hutcheson.

    ANDREA LATI: This is Andrea Lati.

    JASON ABT: This is Jason Abt.

    RISTO PUHAKKA: And this is Risto Puhakka.

    DAN HUTCHESON: So, there's not a lot going on these days, in the last week or two other than, you know. I know, Risto, you just did a new 2020 forecasts and it was a big yawn.

    RISTO PUHAKKA: Yeah, it was spelled basically 21, there was basically known about 40% increase on equipment. You know, WFE is going to be north of north of 90 billion this year, so it's a huge number in multiple levels there. The interesting thing really is what came out of the critical subsystem side was, you know, the forecast is about 30% growth with the year and equipment on 40%. So that kind of confirms one way also the 10% price increase or ASB increase going in equipment side, where are the subsystems are in the long term, supply contracts with more fixed pricing. So that's kind of an interesting and it's kind of confirms what we already know and have been seen indicators, but that's just one more data point to confirm that. So that's the thing. But then Andrea, as soon as I release a forecast, Andrea starts to start to tinker around it. So, what about 2022? You got that?

    ANDREA LATI: Yeah, we just have a monthly update. And yeah, there were no changes to the 2021 forecasts, they were the same as what we had in August. However, we did upgrade the forecast for equipment for 2022 from six and a half percent to about 10 and a half percent, so basically double digits. And the reason was that we're seeing a lot of activity in the equipment space, there is a lot of unfulfilled demand that is throwing in 2022 and the orders are just too strong at this point. So, I think a lot of that will roll into next year and that was the reason for going in double digit growth. It could even be higher than that, but at this point, we're basically moving that double digit territory.

    DAN HUTCHESON: So, everybody's concerned about all this spending. And Andrea, when you look at it in terms of the capacity that's planned to be put in place, but it's unrealized yet, it's not actually making wafers in the next year or two. How much does that account for would add capacity to today's production capacity which is maxed out?

    ANDREA LATI: Well, yeah, I'm not going to go about talking about MSI here, but it's not a significant jump there. And part of that is because there is some inflationary pressure, that like Risto mentioned, that at least 10% is real price increases. It's not really a physical increase in the fab, in terms of output. Yeah, there are conversions that are happening in some capacity that's being added. But I believe if we look at the overall capacity increase for this year, it's only about 11%, which is not a big number actually, considering that there wasn't much than the year before and the previous year, right?


    ANDREA LATI: So, and the industry has been growing quite a bit.

    DAN HUTCHESON: Right. And…

    ANDREA LATI: So, they need to start catching up actually. And so, if you look at from that perspective, actually, we are not seeing any huge increases in capacity. And most of these, you know, increases will sit at the production level. It's because, you know, companies are running very, very high utilization rates and whatever capacity comes online gets converted quickly and then around at a very high utilization rate. So, but it is really driven through, you know, getting the most out of, you know, the existing fabs.

    DAN HUTCHESON: Okay. But you know, I remember seeing in your TCI graphics yesterday that you had an unrealized capacity at about 15% of…

    ANDREA LATI: Correct.

    DAN HUTCHESON:capacity. Yeah, so.

    ANDREA LATI: Yeah, but this spread out. It doesn't, even next year, is going to be probably…


    ANDREA LATI:grow throughout the year. It won’t all come in January, right? Because there always this, so…

    DAN HUTCHESON: So Risto, what about this $206 billion that Samsung is talking about? How does that square with just his 15%?

    RISTO PUHAKKA: Oh, okay, I mean. You know, the chip makers have been giving out these big numbers $206 billion or $300 billion and you really, and then they give some timeframe, you know, three years, five years, ten years. And you really need to look at that relative to their historical spending and you need to really look at what's on that number, that $206 billion and basically. I mean, Andrea, you did the math on that.

    ANDREA LATI: Yeah, so it was basically it was $206 that they are planning this next year versus $185 that they did in the previous period, so the increase is only 14% if you look at an…


    ANDREA LATI: …aggregate level.

    DAN HUTCHESON: So, it is just not outlined at all is it?

    ANDREA LATI: No, it's not.

    RISTO PUHAKKA: No, and it's just a, you know, but it also the message on those. It's a PR message, “Hey, we will have capacity. We will build more. As a customer, don't you worry, we will have it.”

    DAN HUTCHESON: Yeah. Okay.

    RISTO PUHAKKA: So that's kind of the messaging out there and it's kind of important not to take that number out of the context and look at it within the, within this stream of data. And that's why it's important to look at that way also.

    DAN HUTCHESON: So Risto, I really want to know what TSMC’s gift to the industry is? You keep talking about that, what the hell are you talking about?

    RISTO PUHAKKA: Oh my God. As a business guy, you know, we, it's a tremendous gift. You know, it's basically if, you know, TSMC announced, you know, and I like to think the number the higher number 20% price increase, you know, to be implemented. That's a gift to the whole industry because now everybody can go with the 19% price increase and tell their customers, “Aren't you lucky, you didn't get 20% price increase.” So, but in the sense, it's a recognition that we going to have a basically a higher cost, higher expenses, basically inflation, inflation working through the whole supply chain in the industry. And then of course, there's somebody who's going to have to figure it out, whether it's the consumer, electronic suppliers, or Apple. It's this role in how the device pricing at the end of the day will go, and that will define it, and how it works in the end markets. But in the general thing, it’s a, I like to joke it's a tremendous gift by the TSMC to the industry.

    DAN HUTCHESON: You think it could be a major inflection point, like what we saw in autos in the 20s, where all of a sudden, we just went from prices always going down to prices incrementally following inflation.

    RISTO PUHAKKA: Huge possibility for that, because, you know, it's basically the expectation through the industries that the prices to decline 2 to 3% per year because of Moore's law and efficiencies and everything else. And we already known and seen it, it's not holding all places. And in now it's like, okay, is this going to work through all of it, you know, all the all types of devices, it's going to be really interesting and see if that and that has an opportunity to become an inflection point.

    DAN HUTCHESON: So, I want to turn to Jason now because, you know, when we start to look at what's going to drive all this growth, right? A lot of it, what we're seeing is more and more AI. But the AI needs sensors and you've been seeing some really interesting things happening on the sensor side of the world.

    JASON ABT: Yeah, absolutely. I mean, image sensors are one of the most interesting parts of our industry, at least to me. I mean, so many different applications and what I've been looking at over the last probably few weeks or so are sort of non-traditional image sensors. They're called dynamic image sensors, or maybe more accurately, a dynamic vision sensor. So, if you think about a normal image sensor, like you'd have in your phone, it is still sort of based on capturing a frame, kind of imagine a shutter coming down and capturing a still image. The difference the dynamic vision sensor is that, it isn't capturing a frame. In fact, what it's doing is every pixel is always on the lookout for a change, probably change in brightness. And it can do this really, really fast. Because you're not trying to capture an entire frame, whether it's 60 times a second or 100 times a second. All you're doing is you're looking for changes in individual pixels, you can get speeds 10 times or more faster. It also has a far better dynamic range. For those of you who sort of have the math bent, it's actually a logarithmic detection, rather than a linear one. And so, when you combine those two pieces together for industrial applications where high speed is important, dynamic range is important. It's pretty clear all kinds of industrial applications. When you think about the automotive market, where, in fact, the whole reason of having these sensors around is to look at fast moving objects relatively and detect very small variations so that it can head back to the processor to figure out is this important, is this not? And so, these sensors are really interesting. But there's a downside, the pixels themselves, or you can include the peripheral circuitry are huge 10 microns by 10 microns versus maybe what you'd see out of Samsung. We're getting closer to that half micron by half micron size for like a single frame in a normal image sensor and it's because it's so complex. Now, one of the ways that the industry is moving and we can actually take advantage of other technologies. I think about hybrid bonding. And so, there are a number of companies right now that are looking at using hybrid bonding techniques, so very high density again wafer on wafer to ideally actually get to a single pixel, and you're actually pumping right out to your processor, which is sitting on the back side. So, you get the density improvement, you get incredibly high bandwidth, high speed, right to your processor. So, again, you know, we're talking about hybrid bonding, and as we talked about a couple of weeks ago, massive amounts of cash against the process, but there's so many more applications in the hybrid bonding space, including things like this dynamic vision sensor. So, you know, that the hybrid bonding, the 3D packaging, more and more importance in our industry,

    RISTO PUHAKKA: Yeah, hybrid bonding is coming up on different applications, there is actually a lot more than meets the eye right now in just with Skype. I would kind of, I don't want to say conventional packaging, but the natural thinking is, hey, this is going to be creating your packaging, but we see it on NAND manufacturing. Now you're bringing the vision sensor technologies, you know, and probably, we just got to scratching the applications based on that with even those.

    JASON ABT: Absolutely.

    DAN HUTCHESON: Now, the interesting thing is coming to along that same line is that, you know, we've been talking about hybrid, you know, heterogeneous stuff for 20-30 years now. But now today, what's happening is the test technologies come along and the design technologies come along. You know, where before you, the design technology really couldn't design a chip that was broken up into multiple sub components. And so, if you can't design it, you can't make it.

    JASON ABT: Absolutely.

    DAN HUTCHESON: Those barriers are broken now and you can do a lot.

    JASON ABT: You can, and I mean the complexity, like you said, actually increased already think about trying to pump your image data to a processor, not on the same chip, but actually to a different chip through the hybrid bonding. I mean, now that we have the manufacturing technology, fantastic, but you still need to be able to design the architecture, basically, to be able to accommodate that.

    DAN HUTCHESON: Yeah, the signal path becomes really complex. And, you know, it takes a lot longer, a lot more energy to get it out of the chip to the other chip. So now you start looking at things like, you know, because in the old days, they were just using wire bonding, right? And that's a really power-hungry way to move data around.

    JASON ABT: Yeah, yeah. Absolutely.

    DAN HUTCHESON: Now you can do so much with through hole via and all that stuff. And plus, you can test die to the point where they're as good as if they weren't a package. So, it's a really exciting time going forward.

    JASON ABT: I love this industry.

    DAN HUTCHESON: Okay, so thanks everyone for taking the time. I appreciate it.

    RISTO PUHAKKA: Thank you Dan.

    ANDREA LATI: Thank you Dan.

    JASON ABT: Thank you Dan.