Advantest Talks Semi

TechInsight Semiconductor Industry Outlook 2025 and Beyond

• Don Ong, Head of Innovation for the Advantest Field Service Business Group, Andrea Lati Director of Market Research at TechInsights Inc • Season 3 • Episode 1

Share episode

Join us for a captivating conversation with Andrea Lati from TechInsight as we unpack the transformative forces shaping the semiconductor industry. 

Andrea Lati is the Director of Market Research at TechInsights Inc. Since joining the company in 2001, Andrea has been managing and developing forecasting models as well as performing market analysis and research on electronics, semiconductor, and equipment markets. Andrea has played a key role driving detailed fundamental-based research and developed benchmark data and forecasts widely used by the industry. In addition, Andrea has performed many custom studies for various clients and is a co-author of The Chip Insider ®. He earned his bachelor’s degree in Mathematics and Economics from the University of California Santa Barbara.

With market growth reaching an impressive 23% and IC sales surging by 28% in 2024, Andrea offers his expert perspective on what’s fueling this remarkable performance. We'll explore why, despite climbing average selling prices, unit volume growth is still playing catch-up, particularly in the PC and smartphone sectors.

Our discussion also covers the substantial shifts in the landscape caused by government funding and export restrictions. As companies build fabrication facility "shells," we examine how they're aligning investments with genuine market demand to avoid overcapacity. The episode delves into the intricate dynamics of regional policies, highlighting the challenges of navigating export restrictions with China. We also explore TSMC's strategic expansion into advanced packaging and the burgeoning field of silicon photonics, driven by AI's insatiable demand for enhanced bandwidth, latency, and power efficiency.

Looking ahead, we discuss the optimistic projections for the semiconductor market with an eye on growth in chiplets, memory, and logic segments. Major players like Meta, Microsoft, Google, and Amazon are making hefty CapEx investments, fueling this bullish outlook. While traditional markets like PCs and smartphones show signs of recovery, the automotive industry's increasing semiconductor content, especially in electric vehicles, presents a compelling opportunity. We conclude with a look at macroeconomic factors and the critical role of innovation and strategic collaboration in sustaining growth amidst the industry's cyclical challenges.

Highlights include:

• Examination of the semiconductor market's remarkable growth potential
• Discussion of AI's pivotal role in driving semiconductor demand
• Analysis of supply chain challenges and geopolitical impacts
• Overview of advancements in foundries and semiconductor fabrication
• Insights on talent shortages and workforce challenges
• Examination of the impact of electrification and automotive trends
• Importance of collaboration and strategic partnerships for innovation
• Predictions for the market's future, focusing on AI and advanced technologies

Thanks for tuning in to "Advantest Talks Semi"!

If you enjoyed this episode, we'd love to hear from you! Please take a moment to leave a rating on Apple Podcast. Your feedback helps us improve and reach new listeners.

Don't forget to subscribe and share with your friends. We appreciate your support!

Don Ong: 0:00

Hello and welcome to another exciting episode of Advantest Talks Semi. Today we're diving deep into the trends, technologies and market forces shaping the future of semiconductor industry. I'm thrilled to have Andrea Lati with us. Andrea is not only a leading expert in semiconductor market analysis at TechInsight, but also someone who truly understands the intricate dynamics of this rapidly evolving landscape. Andrea, it's great to have you. Welcome to the show.

Andrea Lati: 0:27

Thank you so much. It's very good to be here.

Don Ong: 0:29

Andrea let's start with getting a snapshot of where we are today. The semiconductor industry is clearly undergoing some big changes, whether it's new technologies, evolving the supply chain or shifting market demands. Can you help us break down the key forces that are shaping the industry right now? What's the driving growth and where we are seeing the biggest challenges?

Andrea Lati: 0:51

This year actually has been quite an interesting year. If you look at the semiconductor market, it has performed much, much better than what we expected. Early in the year we actually went through many forecast revisions and if we look at the overall semiconductor market for 2024, we believe that it's going to increase around 23%. If you look at the IC market, which is a subset of the semiconductor market, the IC sales will grow about 28%, which is the fastest growth that we've seen in over a decade. So, this has been a really incredible year in terms of growth, and I've put together some slides here to show where we are in the cycle because, as you know, the semiconductor market is a cyclical market. It is a growing market, but we also have to deal with cycles that we see every two to three years and, as you can see in this chart here that I've provided, the top it shows the weekly IC sales growth on year over year basis and there you can actually see the yearly growth actually peaked in early October of this year. However, it's hovering close to over 30%. So, again, very, very strong growth rate and again, if we look at the peak, actually it's very close to what we saw in the previous peaks back in 2022 and also 2017. Now the question is how long can we expand? Because the semiconductor market has been expanding quite rapidly this year. The chart below is what you see is you see the nominal weekly IC sales started from 2007 to the present and what you see there is the nominal IC weekly IC sales is a blue line and the black line is a trend of the IC sales. What you can see here is that the IC market jumped above the trend line early in the year. So, after trending below trend line in 2022 and also 2023. So, what this chart actually shows us is that, even though we are hitting, you know, very, very growth high rates, we're still early in the cycle right, and again, if we look at an up cycle, the up cycle tends to last between two to three years. So, we moved above this trend line early this year. So, we feel pretty positive about 25 and 26 being growth years for the semiconductor industry, mainly because, we're just finally getting into this growth area following, the downturn that we had in 2023.

So, one interesting thing about this year's upturn is that it's really very lopsided. If we look at the growth, what's driving this year is really ASP driven. In fact, if you look at ICASPs, they are going to soar to a 20 year high this year and so, again, most of this upturn is ASP driven. And what's driving this ASP are two things. First of all, we have the recovery in the memory market in 2023 and 2022. The memory market went through a pretty steep downturn, so what we're seeing in the memory market, in both DRM and NAND, is really a rebound from that big downturn that we saw. And the other big driver that is fueling ASPs is the NVIDIA Impact.

As you know, NVIDIA and AI have been really growing tremendously this year and what's unique about that market is that, in terms of unit volumes, the unit volumes are very low compared to the rest of the IC market.

So, what we have ended up with is that if we look at the overall ASPs for ICs, as you can see in this chart here, asps are growing about 28% this year, which is the highest growth rate that we've seen the IC ASPs since 1995.

However, if you look at the semiconductor unit volumes, they're only going to grow in the low single digit range this year, about 2% or so. So, this shows that this upturn has been mainly AI driven. If we look at the rest of the market, some of the broad market like PCs, smartphones, automotive and so on, those markets actually have lagged and some of them are still in a correction territory. And the clear indication of that is the unit volumes, the unit volumes that we're running this year are actually below the levels that we saw back in 2021 and 2022. So, very uneven, an upturn that is primarily being driven by AI and the rest of the markets are still in this recovery phase. And this year, most of those markets, those broader markets, have really struggled to return to growth territory.

Don Ong: 5:58

Great, it's good to understand all that, but I'd like to understand what's your take on factors like supply chains and also like regional policies that's impacting the industry.

Andrea Lati: 6:08

There are many things at play here. If you look at, first of all, the capex that we are spending as an industry, the capital spending has been quite healthy, actually, relative to what the industry experienced in the last two years. And when we look at the capital spending in 2023, which was a down year for the semiconductor industry, the capital spending was barely down. This year capital spending is expected to be up, and so is in 2025. Now what we're seeing there is that most of the spending actually has been driven by China. If we look at the top 10 capex spenders in 2023, there were three Chinese companies in the top 10, which is the first time ever that in the you know three Chinese companies, namely CXMT, SMIC and also YMTC. So that's a huge change actually in the spending in the makeup of the semiconductor industry and what we have seen in the last three, four years is that this higher spending level that we've seen from these Chinese manufacturers has actually translated into much higher spend in the WFE (Wafer Fab Equipment) market. So, if we look at the wave of fabrication spending, most of the growth there has come out of China. Last year, China accounted for about 35% of total WFE spending. This year is probably going to be about 45%. So, these are very big numbers. So, they are the ones that have been driving most of this spending in WFE and, a lot of that is, government funded as well, because the governments around the world are spending a lot of money to support their semiconductor industry. We're counting about $200 billion of government funding initiatives in the pipeline right, whether it's in US, China, Japan, Europe, and China actually makes up about a quarter of the $200 billion in government incentives. And, of course, as a reminder, this $50 billion that they are spending in the pipeline is their third round of spending. China has already had their first round about 10 years ago and the second one several years back, which they spent a cumulative, I think, $50 billion dollars in semiconductor related initiatives. So, what has happened as a result because of this big spending surge in China is that we are seeing the capacity there increase quite a bit, especially in trailing edge technologies, and the concern there is now that we could see, because of the big spending levels that are taking place in China, an oversupply at the trailing edge devices, because, a lot of this spending actually has gone through those markets. So, there is a lot of concern, especially among the European and American manufacturers, that we could see some oversupply, in the distraining edge markets. And that's why you're seeing some of the governments, whether it's here in US or in Europe, start to implement some tariffs of this Chinese made semiconductors because of this risk associated with this big capacity surge that we're seeing in China. So that's one of the drivers that we've seen in terms of the manufacturing side.

Another big driver was AI, as you can imagine. But the impact of AI in terms of spending this year has been primarily in advanced logic essentially TSMC, whether it's in there five nanometer node or three nanometer nodes. Those nodes are running very, very strong at TSMC, very high utilization compared to the rest of their manufacturing node. And another has been in high bandwidth memory, because for AI to run you need a lot of memory as close as possible to those GPUs. So HBM has been very, very strong this year. Spending there has been increasing substantially because supply is really trailing demand. So, all the major direct manufacturers are expanding quite aggressively in HBM in order to catch up with demand. So those have been sort of the hot areas. Some of them again have been AI-driven, the other ones have been more regionally-driven.

Don Ong: 11:06

So, with all these current supply chain challenges, do you think they are temporary or are they indicative of a more systematic shift that will affect the whole industry for years to come?

Andrea Lati: 11:12

I think, if we look at the level of investments that are taking place, whether it's on this bottleneck that we're seeing on AI, I do think they are temporary because, companies are investing quite aggressively and we're already seeing some signs of some, let's say, ease in the supply chain in some of the areas that have been some really big bottlenecks, like CoWoS has been a very critical area that TSMC has been capacity constrained to really meet NVIDIA's demand. CoWoS capacity is doubling, will double again next year. So, the investments are being made and I believe that you know, in ‘25 and ‘26, we are going to see a more balanced market in these areas that we've had shortages this year, which have been primarily in the AI area. On the other hand, if you look at other areas, the market is oversupplied, and one of the headwinds that we've had this year, and also it's a little bit of a headwind moving into next year, is inventories, if we look at the overall industry inventories, like, without including AI right, where things are still tight.

The rest of the industry actually is struggling with high level of inventories. If you look at segments like mobile, in the PC area, automotive, industrial, the inventory levels are still concerning and there is still some work that the industry has to do in order to bring those inventory levels to reasonable levels. So, for those markets actually, inventories have been an overhang. There are no supply constraints. If anything, there is too much supply. So, those segments actually are working very hard to work down their inventories. If we look at the market, it has been really a tale of two markets: The AI market, which is doing very well, and then the non-AI market which has been struggling with software, demand, high inventories and it's really beginning to sort of work those inventories down and resume growth.

Don Ong: 13:22

I'd like to dive a little bit deeper into that and see the role that geopolitical situation, the whole tension is shaping the whole current semiconductor market. So, what impact do you think the CHIP Act and also other similar initiatives in different countries are having on the whole semiconductor production?

Andrea Lati: 13:39

Well, there are two things. First of all, there have been a lot of government initiatives to support their own semiconductor industry, and a lot of this money actually is finally being released, so we are going to see more fabs being built to essentially support these government initiatives. So that's part one. What we are seeing, though, is that, with all this money coming into the market, there is always this risk of overcapacity. As you know, our industry is very, very sensitive to this overbuilt, because anytime we end up with overcapacity, prices decline. When prices decline, profits fall, losses mount, so we've seen this many, many times. So, this money that is moving in the industry does increase the risk of oversupply. The good thing is that a lot of this money is going to be used, to build shells. And companies will gauge when to move in equipment, because the moment that you move in equipment, you have to start with equipment, because the depreciation starts, because the depreciation is huge. If you look at leading-edge fabs now, they cost $30 billion plus. So, once those tools move in, you want to make sure that they are utilized properly. So, what we believe will happen is that companies will gauge how much tools they will put into these shells and these clean rooms that they are building, based on market demand. Because, if you think about it, the risk of building a shell it's not as high because the depreciation for those shells is much, much longer compared to tools, which is only five years typically. And it's a very good insurance company, because, if the market is up and you need more capacity, you can easily add more tools. However, if you don't have the shell ready, because you're waiting, then you can miss on the whole upturn because it takes one or two years to build the shell bringing the tools. So, for chip makers, a very good let's call it insurance policy to have these shells ready and fill them based on market demand. And I think that's what we're probably going to see in terms of how this money is spent.

Now, the other factor that we have to keep an eye and again creates a lot of uncertainty in the market is these export restrictions. And that's happening a lot with China, and part of that is because these export restrictions have basically forced China, in a way, to hoard as many tools as they can, especially critical tools, because they feel that they might be shut out of that market. So, if you look at tools in 2023, they were up 40%, mainly driven by China. Of course, China didn't really need all those tools right away because they were not building as much capacity to be used, but the fears of being locked out of the market forced them to buy a lot of tools. So, these export restrictions always cause a lot of pulling in buying, because companies are worried, they might not have the tools to expand their capacity.

So, we don't know how, how far these export restrictions will come moving into 2025. But it's a pretty safe assumption that the ones that are already in place will stay. We could see higher, especially with the new administration coming here to the US. But it adds another layer of complexity in the industry and also in forecasting because all these purchases are not really being made because of market fundamentals, but they are more made because of these security reasons, which makes forecasting the industry and planning much trickier than it would be otherwise.

Don Ong: 17:44

So, you know about all the money that the government is spending right now is mainly on fabrications and we see that the foundries are evolving. So, Intel has been off their foundry. We know TSMC is leading the world in all the foundry, taking over 50, 60% of capacity, especially the whole AI and global foundries shifting their focus a little bit towards silicon photonics. What's your take on this? How do you see the role of foundries evolving in the current market?

Andrea Lati: 18:13

Foundries have become extremely important, and if we look at leading edge foundry, it's pretty much dominated by TSMC. At this point, as you said, they have over 60% market share, and leading-edge nodes is even higher, right, but what we're really seeing is that the foundries are really expanding their offerings not just to wafer production and wafers, but also in assembly and test services. And you can see, TSMC is a very good example, they are making a huge push in advanced packaging, whether it's on interposers, chip on wafer, SoCs, huge, huge push, because chiplets are becoming a very important market and TSMC is trying to capitalize on that. But other companies are doing the same thing. Intel has a huge program in advanced packaging as well. They are making a big push in leading edge technologies. Samsung is pretty much going the same way.

We are seeing these new technologies, or emerging technologies, rather like silicon photonics, which are also becoming more, let's say, prevalent at this point, and part of this is because of AI. If you think about AI, the bandwidth, latency, power requirements are increasing quite a bit. So, if we look at, for example, silicon photonics, that's one of the options that the industry has to explore in order to essentially reduce those huge power consumptions that these AI models need. As you know, copper is running out of these physical limits, so going through silicon photonics can help quite a bit. Latency is an issue. Bandwidth is also a huge issue. So, we are seeing quite a bit of innovation there. There are a lot of light upstarts, companies like AR Labs, Light Matter, which are coming up with really interesting in-package optical IOs to improve this AI proliferation that we're seeing. Because if we look at it just from the power perspective, these data centers are consuming a lot of power, they're putting a huge strain on the grid, so these new technologies actually can help essentially deal with this. For AI to expand, we have to take care and deal with all these power issues, bandwidth issues, and I think silicon photonics offers some solutions, especially at the package level, right. So, those are the things that our industry has to do in order to make AI viable. We're seeing that Global Foundries is expanding there, TSMC is expanding there, Intel has been doing a lot of work there. So, there is a lot of work in these new technologies.

And also, when we look at them, you know, let's say, silicon technology, even though the slower is slowing from scaling perspective, companies are still pushing hard.

You have, you know, TSMC is now ramping two nanometer, three nanometer node. Next year they're moving to two nanometers. So, a different technology which promises improvements. And then you have, after that, in 2027, TSMC will go to 16 angstroms technology which will have backside power delivery, more improvements. Intel is following similar steps in terms of technology and their roadmap. They are moving to their 18 angstroms next year, in a couple of years they are probably going to hit their 14-angstrom version, Samsung the same way. So, all the market leaders are pushing very, very hard on the technology front. Even though we're seeing a diminishing return there, that's not stopping them from moving to these lower nodes, because the industry needs them. Even though those density gains are slowing, they still need them, and that's why these companies are pushing really, really hard on their roadmaps.

Don Ong: 22:53

I'd like to explore a little bit, because we talk about AI and machine learning. That's helping us drive the growth in semiconductor right now. One of the applications of that could be back into semiconductor in terms of design and manufacturing to help us get more efficiency, get more yield. What's your take on this? Are we doing enough? Can we do more? What's your read about the situation?

Andrea Lati: 23:15

Well, this has been a very important area for a while now. It started with DFM back in the day, and then we had DTCO, right? So, there is already a very strong link between fabless companies, foundries and EDA. But I think what we are going to see in the coming years is that collaboration expanding the chiplet market, because that's an up-and-coming market where there has to be very strong collaboration between these.

You know three parties, Fabless Companies, Foundries, EDAs, and also the OSATs. So, that collaboration has to take place. And because you cannot just take the design and throw it off to your foundry or your OSAT, especially if you're dealing with chiplet. Before you have to design the chiplet, you have to think about every single piece that goes in it and make sure that all the parties involve optimize and figure out what has to happen before the chiplets comes to life. So, there has to be far greater collaboration within the semiconductor industry, even much more than what we've seen in the last few years in order to make this chiplet market or advanced node markets, a reality.

Because things are getting more complicated, far more expensive. Especially when you look at these leading-edge nodes. Yes, chiplets can reduce costs, but they are fairly complex packages, so it will require a lot of collaboration, whether it's reduced costs, whether it's on the EDA side, fables, OSATs, foundries, to make sure that progress is being made there.

Don Ong: 25:09

So, Andrea, you talk a lot about chiplets. So that's something that's coming and you know I would like to, at this point, talk about looking forward to 2025. I'm really curious to hear your take on what's coming. We know the whole semiconductor industry is set for both incredible opportunities and some tough challenges. So, what sector do you think are going to take off and what emerging technologies are going to make the biggest difference, like what you mentioned about chiplets? So, let's dig a little bit into your projections for the coming year.

Andrea Lati: 25:38

We are actually quite bullish about next year. We are expecting the semiconductor market to grow by about 20%. And if we just look at the major segments, we do think that memory will still have a pretty good year next year, especially DRM, because HBM and AI this AI craze is not going to go away anytime soon. We still think that even next year the AI demand will be very, very strong, and part of that is because, if you look at the Capex budgets of the hyperscalers, they are accelerating. 

If you look at the, the capex of the hyperscalers, of the top eight, they are spending about 270 billion dollars’ worth in capex in this a primarily in this AI initiatives and the top four, which is essentially Meta, Microsoft, Google and Amazon they make up about 90% of that spending. So, the capex for these companies is going to go up about 56% this year. So huge numbers. I mean think about $270 billion in capex. I mean that's a huge number and I have a chart here so you can see how the capex has increased over time. So clearly, we're at an all-time high. And also, if we look at 2025, all the major hyperscalers pointed to continued expansion in terms of data centers and spending for next year as well. And our estimates are between 15 to 20 percent growth next year as well. Yes, it's a little bit slower than this year, but it's still growing. I mean 20% growth. It's a high growth. So, in terms of you know, hyperscale capex, we're going to look at well over $300 billion next year. So, where is all this money going? Most of it is going to be AI initiatives, right? And clearly, the big favorites there will be all these GPUs NVIDIA, primarily HBM, of course, GPUs run on HBM. There's going to be plenty of ASICs. We expect to continue growing at quite rapid rates in the next year, all these custom ASICs and probably a lot of things that are AI related to the networking side. So those will be very, very high growth areas.

When I look at the logic market, next year we expect the logic market to grow about 19%. But the highflyers, there will be GPUs for sure, because NVIDIA is expected to grow between 40 to 50% next year. They have a lot of demand that they can't keep up with this point with. So, that will be a big driver next year. But we also think that other markets within the logic segment, let's say the broader market, like PCs, smartphones, should do better next year because again, we do see a recovery taking place in those markets, which is why we think that next year, on the logic side, at 19% growth,  we should see a very, very strong market.

When we look at the memory market, we are also quite optimistic. DRAM will be riding the AI wave in terms of HBM. When we look at other areas of the memory market, the improvement in this broader market, like PCs and smartphones, later in the year we should probably see improvements in consumer, automotive and industrial, which are lagging right now. That should start giving the memory market, especially the NAND market, a nice boost. So, even when we look at the memory market, we expect double-digit growth next year. So very, very strong growth for both memory and logic.

The only segments where we expect some softness will be automotive and also industrial, and part of this is because end demand has been weaker, inventories have been higher. But we do think that in the second half of next year market fundamentals there will improve. We do think that there will be more inventories, let's say reasonable as we head into the second half of next year. And end demand should also improve. Now that you know interest rates are coming down and also now that we're seeing some signs of recovery in those markets.  

Don Ong: 30:22

Yeah, I was going to ask you about automotive, and you mentioned that as well, so I'd like to just dive a little bit deeper into that. So, with electric vehicles on the rise, how will this impact semiconductor demand?

Andrea Lati: 30:33

Well, what we've seen actually in the auto market is that auto units, in terms of units, they're not growing very fast because it's a function of population growth and all that. So, they are growing actually in the low single digits, like a couple percentage points. But where the growth there is happening is the content increases because, as cars become smarter than what we're seeing is that the content per car, the IC content per car, is growing substantially right. So, in the end, what you get is that you have this really strong growth. In the end, what you get is that you have this really strong growth in the semiconductor side, or IC side of automotive applications and part of that is the electrification trend. Another one is the increased safety features that we're seeing in cars, especially in these higher levels of autonomy, and also, better infotainment systems and so on. So those are really the big drivers.

So, in the automotive market it is a content-driven increase that we are seeing as it relates to the auto IC growth and I think our projections are by 2029, the auto ICs will be over $100 billion. And it's primarily contact driven and we're still again in the early phases of this electrical vehicle penetration. And this is going to be happening for several more years before it starts sort of rolling over and maturing. But during those years we are going to continue to see a pretty healthy growth in semiconductor content per car and that's going to be the primary driver in that market and it's one of the fastest growing areas in the semiconductor market. Fastest growing areas in the semiconductor market outside of AI of course.

Don Ong: 32:40

So now I'd like to shift focus a little bit and talk about the advanced packaging. So, as we know the whole, we're moving to advanced nodes three nanometer, two nanometers. Things are going to get smaller. So, it will be a lot of challenges in advanced packaging technology, such as 3D stacking, and also you talk about chiplets. So how do you think that is going to transform the industry and what opportunities do we see over there?

Andrea Lati: 33:05

Advanced packaging has been one of the fastest growing areas in the last few years. It's expected to continue to grow very fast. Because, in the last few years, it's expected to continue to grow very fast. As we move into this chiplet world we need more capacity there and you can see that all the major manufacturers, whether it's Intel, TSMC, or OSATs, are expanding very, very aggressively in the advanced packaging field. And it's quite interesting because especially when it comes to AI, there is a little bit of a collision course there because typically what our industry did for many, many years was rely on the Moore’s law. Moore’s law worked great for the industry, because not only were able to drive the transistor cost down, but we got very nice density gains as we went from one node to another. We have made a lot of improvements in performance, low power. So, there were a lot of nice added benefits from the Moore’s law that our industry enjoyed for many, many decades. Clearly, that's essentially slowing down quite a bit, as we've seen now that we've moved to FinFET and we're moving to this gate all around transistor architectures and I have actually a couple of slides that I can show there. But what we're seeing is that, as we, in the last five years or so, the transistor density growth has slowed right. And also, when we do projections, as we move to horizontal nanosheets and FinFET, we expect actually pretty sort of flattish growth in this transistor density increases and that's a sign actually that Moore's law is hitting a ceiling, and this is where we are seeing to see the impact of that right.

When we look at, for example, the transistor counts, let's say, in Apple devices, they're not growing as fast anymore. Apple is bound by a certain die size. Apple cannot go in die sizes that are 400, 500 millimeters squared. They typically tend to want to keep the die size around 100 millimeters squared, in order to keep yields high and keep costs low. Because at the end of the day, apple sells $1,000 product so yields are super critical. But if we look at the transistor growth in application processors over years, the transistor growth there actually is slowing because Apple has these limits in terms of the die size.

If we look at the NVIDIA products and their GPUs over time, what we're seeing there is that the transistor growth actually has been faster than Apple's. However, NVIDIA has been able to achieve that by going to a bigger die size and clearly NVIDIA at this point in time doesn't have to worry very much about yields because they sell $80,000 products. So, they can take a little bit of a yield hit. But even with NVIDIA, what we're seeing is that the die size is getting very close to the radical size limit. So, if we look at the Blackwell chip, what NVIDIA had to do in order to cram those 208 billion transistors was really stitch two GPU dies, because they essentially ran out of radical size. Now, radical size fill limits. So, this is where we come into the chiplet world, because for the industry to continue to improve at least at the system level, chiplets are the next alternative that we have and of course it doesn't mean that we have to stop shrinking as an industry, but because shrinking is slowing, advanced packaging can help to continue improving the system performance to essentially enable AI, those automotive applications that we talked about earlier. We still need to get better system performance in order to get to this, to expand this market. And that's where chiplets actually help at the end of the day, they do improve system performance.

And it also depends on which way you're looking at it. Because clearly, if you're looking from the system level and you are putting everything into a chiplet. So, you're shrinking the board and taking all those devices and putting them into a chiplet, your performance would increase substantially. Going from the motherboard to a single chiplet. However, if you're looking from the chip level, basically from the bottom up, right, and you're desegregating that die, there's going to be some performance loss. Because you're moving away from this monolithic approach to a chiplet. So, depending which way you're looking at it, whether you're looking from a system level, from a chip level there could be either gains or losses, but essentially the industry needs chiplets in order to improve the overall system performance.

Don Ong: 38:43

Right. So, as you mentioned, with chiplets, because we're nearing the probably nearing the end of the Moore’s Law, I also just like to get your insights on are there any market segment that you think will shrink because of that, and is there any market segment that you think will shrink in 2025?

Andrea Lati: 39:02

Not really. I mean, if we look at the markets, the markets are still growing. And if you look at how our industry is built, even if we look at like legacy markets like PCs or even mobile if I want to throw them there now because mobile has clearly matured, now those markets maybe are not growing as fast anymore in terms of units but there are still improvements that are being made there in terms of higher chip content. So, essentially, even though those markets are not growing as fast as AI now, they are still very, very critical, important for the industry because they create this very nice base for our industry essentially maintain and have other markets basically contribute for future growth. So, when we look at, for example, PCs in terms of units, we don't expect much growth there because maybe 300 million is probably sort of the max there. But what we're seeing there is that the content there is continuously increasing, whether it's in terms of memory, processing power, the content is continues to increase. We are going to see more AI-enabled PCs, which will require more compute requirements. So, essentially, there is going to be improvement from that perspective.

Even though units are not increasing, the features are getting better. Same thing with smartphones, right? Yes, they're getting more expensive, the replacement cycle has been stretched, but on the other hand, features are also improving right. And also, with AI becoming a thing, we should see pretty interesting features coming down the road, whether it's in PCs or smartphone, in the coming years, which will drive higher chip content. So those markets are not going away, they're still going to support our industry. But we also want more growth. And when we look at growth, growth will primarily come from AI and also the automotive market in the coming years. 

Don Ong: 41:14

We're looking for great 2025, but I'd like to shift the focus a little bit and get your take on. What about talent? We talk about global talent shortages. How do you see that impacting the development of new semiconductor technologies and do you see any shift in how R&D investments are being allocated?

Andrea Lati: 41:38

Well, talent has been a concern for a while now, because we have these really nice projections. I mean, what we're forecasting is that semiconductors should hit a trillion dollars by 2030, and I think we're on track to do that. However, to go from, let's say, 650 billion or whatever we are right now, to a trillion in a matter of a few years, you have to have the infrastructure in place to drive that growth. You have to have the fabs, you have to have the people to run those fabs, you have to have the design teams. The whole ecosystem has to scale in order to support that kind of you know market size that we are projecting. So, everything has to scale up to support that. It's not like, okay, we're saying it's going to hit a trillion with this. Everything has to scale up to support that and, of course, human talent. It's part of that. It has to scale so, and of course, this has been a concern for many, many years now because, especially here in the US, we are trying to get more people to go into the semiconductor space.

Clearly, there is a lot of competition because we have to compete with other industries which are pretty, let's call it rich in terms of how much profits and everything they generate. So, there is always going to be a competition there for talent, whether it's here in the US or other parts of the world and it's something that, of course, it's an obstacle for our industry. But I think there's been many initiatives, whether it's from SEMI or other industry organizations, to essentially attract more talent, because we do need more people in order to support and get us to those trillion-dollar market levels that we are anticipating. Clearly, AI can help keep up with those requirements that are needed. But talent has always been an issue and will continue to be an issue because not only we're not producing as many engineers, but we also have competition from our other industries for those talents.

Don Ong: 43:57

So, Andrea, the whole semiconductor industry is evolving so fast, and that means both established players and newcomers have some big decisions to make. I'd like to hear your thoughts on how companies can stay ahead, whether it's through strategic partnership, new technologies or just smart decision making. So, what are some of the key strategies that will help them navigate this whole complex and changing landscape?

Andrea Lati: 44:21

Technology is always evolving and can change very rapidly. And this is always a big concern, especially for the big players because they are not as nimble sometimes to adjust to changes in the market. And of course, every company has to innovate, that's a number one, sort of a must, innovation is a must, but also, they have to be very nimble because the market can change very quickly. Our industry tends to be quite cyclical. Sometimes you want to make sure that you are always investing in R&D, in capacity, in talent. Whether you're in an upturn or downturn. So, you have to manage those cycles very, very carefully and of course, that's very challenging sometimes, especially if there are huge swings and our industry can go through very significant swings sometimes. So, staying disciplined and staying focused on those investments, especially on R&D and talent, it's very, very critical.

Collaboration is another way, because you cannot always do everything on your own. So, you want to work with your suppliers as closely as possible because first of all, things are getting very, very expensive. Everything related to the semiconductor industry, whether it's designs at the leading edge, fabs at the leading edge, they are very, very expensive. So, you want to make sure that those decisions pay off because if those decisions turn out to miss the market or come to market late, you could actually lose significant share or could be even completely out of the market.

Andy Grove had this: “Only the paranoid survived”. You know what kind of saying and that's actually very true. Because in technology things change very rapidly. You always have to be very alert on what's changing, stay hungry, be innovative, collaborate and also try to manage these cycles right, whether it's technology cycle or silicon cycle, as best as you can. Because one thing that is for sure in our industry, that is always changing, that's a constant thing, always changing. Make sure that you're always adapting to these changes, because they will always be part of our industry.

Don Ong: 46:48

Right, as you mentioned, there's rapid changes. We change very quickly. Technology moves really fast. Do you think there's going to be an increase in MAA activities as a strategic move for growth or do you think because, with all the AI implementation right now, the industry is better served by internal innovation at this point?

Andrea Lati: 47:05

I think, let's say, at the lower level, probably, there's going to be plenty of those acquisitions, because companies, especially the bigger companies, will try to acquire as much as they can to make sure that they are in the right markets and segments. Now, when it comes to the high-profile acquisitions, I think, as you have seen in the last few years, they've become more difficult to go through the regulators because there's been far more scrutiny there. So, I don't think that's going to change, especially with the geopolitical tensions that we see right now. Even if some countries might allow some of these M&As to go through, maybe some other ones might not be right. I think that's probably going to be the sort of the norm moving forward, I think at the smaller scale, there's going to be plenty right At the bigger scale, M&A, I think there's going to be far more scrutiny mainly because you know of these geopolitical tensions, which I think that they're going to be far harder to pass.

Don Ong: 48:12

Do you think there are any market segment right now or market opportunities that big companies established names are underestimating, whereby we see some new startups starting to come in and working on the new technologies?

Andrea Lati: 48:26

Yeah, and I think there could be, and those are hard to detect because it all comes down to how these smaller startups or smaller companies innovate and how disruptive they are. And there's always going to be a disruption at some point. I'm pretty sure that you know all the big companies are looking for those disruptions because that's the nature of our industry. And the chiplet, I think, it's a pretty good example. If you think of the chiplet market Dan Hutchison was writing this article that was comparing the chiplet market to what Foundry was to IDMs 20, 30 years ago.

When Foundry started, nobody really made much of what they were saying. But look at Foundry market now. There could be many opportunities now that we're getting into this chiplet world, and these disaggregated designs that could create new opportunities in the market which I don't know all of them. If I did, I'd be very rich, but they would probably be out there. It opens up quite a few markets, actually that people can take opportunities and do something. But I'll bring up chiplets as a very good example, because it's a very disruptive market and we could see new companies there, out of the blue that could come up with something that are very, very innovative, very disruptive, that could change and shake things quite a bit in the semiconductor world.

Don Ong: 50:11

Andrea, what's your take on the whole ATE forecast in the whole grand scheme of things, with chip leads and advanced packaging, what's your take on it?

Andrea Lati: 50:20

If you look at the ATE market, this year turned out to be a pretty good year. It was a recovery year following two down years, 22 and 23. In 2022 and 2023. And a big part of it, of course, is AI and HBM. The test of demand has been very, very strong this year, but what we saw also was SOC compute, especially as it relates to AI, also was very, very strong this year and that's going to continue to be the case also in the coming years.

And what we're seeing in the ATE space is that these GPUs that are proliferating now and also this ASICs that the hyperscalers are building, these are very high-value chips, they have very high ASPs and ultimately will benefit because essentially, test assures the value that's being created there. I have this slide that shows the data center processor shipments right Units are not growing very fast, about 11% if you look at the two peaks. But if you look at, for example, the sales, sales are growing by almost 400% if you compare the previous peak to 2025, a peak that we're seeing here. So, there is a lot of value and a lot of profits that are being generated in this space and, of course, test assures that value. So, I do think that it's going to be a pretty important driver for SOC tests and also memory tests in the coming years.

And also, as we move into this chiplet world the testing requirements there are going to increase and these chiplets become more sophisticated and you have to test more. And what we believe is that these chiplets proliferate, we are going to see, over time, a shift in value from WFE to test when it comes to these expenditures. And this is one of the reasons when we look at our WFE, our test forecast, ATE forecast, it's pretty much going to be growing at pretty much the same rate as WFE, where in the past, if you look at it, test lost share relative to WFE. We don't expect that to happen in the next five to 10 years, mainly because of these changes that we are seeing in the market as it relates to the chiplets and their higher need for testing.

Don Ong: 53:01

Yes, we're seeing some startup in the chiplet area like silicon box, and also new technology like silicon photonics. That's coming in a lot of startups in that area, so we're seeing a lot of that.

Andrea Lati: 53:12

Yeah, and the other thing is that when we look at some of the challenges that we have as an industry, the transistor density gains are slowing. They are becoming more difficult because we are running into physics problems, whether it's at the transistor level, whether it's at the contacts, and resistivity is increasing quite a bit. So, we're not getting those big performance gains that we're used to, those are issues. But those issues actually are opportunities for companies to innovate, whether it's in new materials, new manufacturing technologies, packaging technologies. So, those can become new markets down the road.

These issues and technical challenges that people can figure out better ways to get around. So, sometimes it's not good to see them as challenges but more as an opportunity and as a market that later on can grow and make things you know better and more innovative. Those are the things that we have to keep an eye on because there are a lot of challenges our industry faces, but I view them as opportunities down the road for companies to find and improve on and continue to innovate.

Don Ong: 54:37

Fantastic, Andrea. Thank you so much for sharing your insights today. It's been incredibly enlightening to hear a perspective on where the semiconductor industry is headed, both the exciting opportunities and the challenges that lie ahead. I think our listeners will walk away with a much clearer picture of what to expect in the coming years. Before we wrap up, if there is one key takeaway or message that you'd like to leave with our audience, what would it be?

Andrea Lati: 55:03

My main takeaway would be that the future for semiconductor market is very bright. I think there are huge, huge opportunities here and we are actually in the early stages of AI. And of course, it's not going to be a straight line. I'm pretty sure there's going to be some up and downs there as well, just with anything else. But I think you know AI is here to stay and of course, our industry will make that happen, we are the enablers. I do think that we have a very, very promising future and AI preparation will be a huge driver. But also, as I said earlier, the other broader markets will continue to expand as well, maybe not as fast as they did in the past, but they're not going to go away anytime soon. They'll provide a very solid base for our industry going forward.

I think the future is quite incredible, especially if we see this whole AI revolution taking place, which is probably as profound as what we saw with the industrial revolution. It's just a different sort of settings that is taking place. But I think the implications and also the benefits that we are going to see are going to be tremendous. And, of course, at the end of it, the building blocks and what supports all that is semiconductors. So that's why I feel very, very optimistic about our future and our prospects in the next decade.

Don Ong: 56:33

Excellent. Thank you again, Andrea, for joining us, and thank you to all our listeners for tuning in. Stay tuned for more episodes of Advantest Talk Semi, where we continue to explore the trends, technologies and strategies shaping the semiconductor industry. Thank you, Andrea. 

Andrea Lati: 57:01

Thank you.