Certify Alexametrics gif

    Europe's Semi Share Loss and Most Advanced Memory Maker

      Sep 9, 2021

    VLSIinsiders’ cloudside chat — September 09, 2021

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

    • Has Europe really lost so much share of semiconductor manufacturing, going from 44% in 1990 to 9% in 2020?
    • Semiconductor forecast of 26% for ICs and current market status.
    • The DRAM disconnect of spot prices and contract prices.
    • Memory.
      • What’s Micron up to as it locks down a leadership position.
        • DRAM and NAND.
      • Embedded memory: TSMC takes a big leadership position.
        • Move to STT cell gives 1-2x performance and lower power.
        • MRAM at 22nm is real in an IoT Fitbit.

    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

    Transcript

    VLSIinsiders' Cloudside Chat — September 09, 2021

    DAN HUTCHESON: Welcome to this week's feel VLSIinsiders. This is Dan Hutcheson.

    ANDREA LATI: This is Andrea Lati.

    JASON ABT: And this is Jason Abt.

    DAN HUTCHESON: So, it's been kind of a quiet couple of weeks. But we did finish our forecast, and, Andrea, you have a pretty spectacular forecast for the equipment market this year.

    ANDREA LATI: Yes, we did upgrade the equipment to about 42%, which is something that we did, actually, almost last month. And the IC’s forecast, it's pretty much coming along where we expected. We're still forecasting at 26% growth year over year. All the data that we've been getting lately is pretty much matching our numbers. The Taiwanese data for July, for August actually, were still very, very strong, especially if you look at the memory area, actually. Memory prices, especially on DRAM, if you look at sales, they're actually there. They are accelerating. I think revenues on yearly basis in July were about 50%, and then in August about 60. So, again, pretty healthy numbers. The only thing that I'm a little bit worried is that if we look at spot price, especially for DRAM, is continues to be very weak, right? So there has always been this lag between spot prices and contract prices, with spot prices leading contract prices. So, if we look at the spot market, actually, prices are falling at 2 to 3% a week, which is quite high. And it's something that we'll have to worry, especially about next year, because if that continues, that means that at some point and contract prices will have to reset and match whatever is happening at spot level. So, and I think that's one of the reasons why memory stocks have been quite weak in the last month or so. It’s because of this fear that perhaps next year, we're going to see a rollover in pricing, especially in the DRAM side. But again, it remains to be seen, because you know, not all the cycles are the same, as you know, so we'll have to see once we get into next year.

    DAN HUTCHESON: Yeah, so that's been kind of interesting. What about, you know, last time we talked about DRAM prices and spot prices in the CPPI? Because the CPPI has gotten pretty scary lately. But that's not showing up in the contract or the WSTS numbers, right?

    ANDREA LATI: No, not at all. And actually, if anything, we're still seeing sort of an acceleration of strength on the contract level, right? So, you know, this gap actually is quite big. So, it's more, and I think there is still enough momentum this year for the contract prices to be pretty healthy. It's more what happens next year. It's these things will need to be in phase and if they are in phase, then that would mean that contract prices have to roll over. And if that happens, I think we'll probably see the sales begin to solve them. Because again, pricing is really what drives most of the sales momentum. So again, it remains to be seen, but there is, yeah, quite a bit of a disconnect between these two points.

    DAN HUTCHESON: I think something has to happen with contract prices, too, just because of the fact that notebook prices and PC prices are down and also much.

    ANDREA LATI: Yes, and that's the other part. I still think that there has to be a correction along the way because it will look at sort of the end market, right? We are seeing plenty of availability there. And also, you know, we don't see that tightness that we saw earlier in the year or even last year. So, I think at some point, we will probably have to see this pricing pressures roll at the contract level as well.

    DAN HUTCHESON: There is a Boston Consulting Group study that's been circulating around that it says that 44% of the world semiconductor manufacturing was in Europe in 1990. I think it's a backward extrapolation, but I don't know how they arrived at it? But there's no way it can be 44% because I remember Japan was a huge percentage back then. And the battle was mainly between Japan and the US and people were talking about Europe's loss of manufacturing chair back then. You were telling me earlier you got a couple numbers on that.

    ANDREA LATI: Yes. I mean, they're still rough numbers, but I think they're fairly accurate. If we look at the early 2000s, right? Japan actually was about 30% of total manufacturing. North America was about 20%. Asia Pacific, which includes primarily Korea and Taiwan, was 35% and Europe was only 15%. So, Europe actually was the smallest back in the early 2000s.

    DAN HUTCHESON: So, in ten years, that would mean they'd lost, you know, they’ve gone from 44 to 15 they lost 30% and that doesn't make sense to me either.

    ANDREA LATI: No, I doubt it. Unless, again, that talking about a specific part within that capacity, but yeah, Europe it's, you know, close to 15% and it has lost some share since then. But I will be very surprised, you know, to know why that number is so high, you know, what the number is being quoted 44% because it's really high numbers.

    DAN HUTCHESON: So, Jason, you've been, you guys been doing a lot of work in memory lately. And I want to know who's on first in memory?

    JASON ABT: Oh, wow. So, it's been a wild ride. I mean, last few years have seen such amazing innovation scaling like we've never seen before and what we've seen in the last couple of months is Micron continuing with a very commanding lead in that space. And you know, if we take a look at where they are in DRAM right now, they came up with D1-alpha (1α). And what does that mean? I mean, we're all kind of moving away from talking about nanometers. But if you were to sort of use the old metrics, it’s a little, it'll sort of be in the 14 nanometers range. So, it really, it's the most advanced DRAM node out there. And that continued with their latest flash offering. We're seeing 176 layers, 3D NAND flash and it's well ahead of where even Samsung is. It's fantastic to watch. So that's, I mean, all good and, you know, we're hearing a lot about what's happening there and everything that we don't, doesn't get talked about very much, but I think is going to be much more important in the coming couple of years is something that we actually saw out of TSMC just in the last month or so. We were having a look at a Fitbit product and there's a processor inside and what we found is not only MRAM, so nonvolatile memory embedded in the processor, but in fact, the most advanced node that we've seen by a long shot down to 22 nanometers. Which means that the density is getting to the point where you know, it goes beyond just sort of niche applications in industry and actually used for consumer goods like the Fitbit and we're just going to see that expanding. The other interesting part is that technology we're seeing in MRAM.

    DAN HUTCHESON: It’s also. One thing is…

    JASON ABT: Yeah, of course. Go ahead.

    DAN HUTCHESON: On the Fitbit, that the cost of 22 nanometer is getting to the point that design costs, the, you know, that you can squeeze it into a market size of a Fitbit, right? So, it says that the leading-edge stuff is technology is really getting cost effective at that level even…

    JASON ABT: Absolutely.

    DAN HUTCHESON: …even though today 22 nanometers is what? Three nodes behind or four nodes behind.

    JASON ABT: Yeah, sure. Yeah, absolutely. And you know, if you take a look at the bulk of consumer goods, they're not at the leading nodes. We spend a lot of time in the media talking about the latest Apple’s processor, Samsung’s processor. But the reality is, this is a very advanced node. Yes, it's three generations behind, but it's very, very capable. And now with this MRAM, not only do they did they get the density and the shrink, but they've moved over to something called spin torque transfer. That's something that used to be something called toggle, but the spin torque transfer technology has at least the possibility, the capability of being like an order of magnitude or better in terms of power consumption on write cycles. It used to be fairly power hungry when you wanted to write to an MRAM now it is really power sipping and so beyond the sort of larger scale applications we're going to see this popping up in IoT applications where there's remote sensors and you need this nonvolatile memory. The other advantage, of course, the MRAM over say flash is its write tolerance. You can do 1000s of times more write operations to an MRAM than you could for a flash. The flash would break down the MRAM will not, so again, for a long-lasting high rewrite application. We're going to see this a lot more in the future and it'll become that much more important.

    DAN HUTCHESON: Plus, it doesn't really require a lot of extra costs in terms of mask layers and things like that.

    JASON ABT: It really doesn't. It's implemented at sort of the slightly upper layers within a processor or controller chip and I mean relatively inexpensive to implement and so you know. As you said, you know, the Fitbit is by no means an ultra-premium product like the latest iPhone or Samsung Galaxy phone and it's cost effective. So, this is a technology for us to watch.

    DAN HUTCHESON: Something interesting is the technology to manufacturer was developed, you know, with Applied Materials and their tool sets. And but a lot of the foundries have been talking about implementing MRAM for a while now. I mean, it's been on their roadmaps. To me, it's fascinating that it took this long for you to find one.

    JASON ABT: We've seen MRAM before. Again, it tends to be a niche applications and lower density applications that it's not sort of the leading edge, if you will. It's really now, I think we're just at that turning point where it's passing that threshold. I mean, Dan, you and I were talking a little bit beforehand about the adventures of phase change RAM and some of the challenges there. And, you know, some difficulty in getting traction in the market and seeing it in a lot of products. I think this is an indication that we're sort of beyond the challenges and we're ready for mainstream and MRAM.

    DAN HUTCHESON: Yeah, what's also interesting, but as you think about when you go off chip to another chip you, it cost you about a couple orders of magnitude, maybe three orders of magnitude and power. And if you're putting something out there, like a remote sensor, that you want to stay out there and just go for forever with your battery.

    JASON ABT: Absolutely. And from a performance standpoint, of course, the closer you can put your memory to the processing engine, the much quicker it is to get that data transferred back and forth. So, you actually get a performance advantage as well. Yep.

    DAN HUTCHESON: Okay, that's cool. So, I know we didn't talk about this before, but anymore cool things that are coming you guys are working on that we should be…

    JASON ABT: Oh, wow. Yeah, so there's actually quite a bit going on in the power transistor domain and we might have touched on this a couple of weeks ago. But there's this really fascinating war going on between silicon carbide and gallium nitride.

    DAN HUTCHESON: Right.

    JASON ABT: And for every week, what we're seeing are these sort of continued indications of one sort of stepping over the other. But you know, silicon carbide is really, you know, it's quite mainstream, but gallium nitride is taking off like mad. So, you know, if this is something that we're watching very closely. And we're honestly, we're given, we're challenged, we take a look at things like data sheets on these kinds of devices, they're all selling slightly different variations. It's hard to kind of normalize and so something that we're working on right now is actually a normalization to do a real one for one comparison from a cost versus performance standpoint. And I think that's something that would be worth us talking about in the coming weeks.

    DAN HUTCHESON: Any particular companies in the lead there?

    JASON ABT: So, um, there are so many, many players that are in this space. One of the ones that we're actively watching, actually, is a small company, called GaN Systems, and they’re actually based in Ottawa, Canada where I'm located. Now of course, they're working with a foundry partner, but we're seeing some really interesting performance out of what they're producing. So, keep, stay tuned.

    DAN HUTCHESON: Watch GaN Systems, huh? Thanks a lot. I appreciate you taking the time today. And Andrea, same.

    ANDREA LATI: You’re welcome.