Advantest Talks Semi

Why your next Smartphone will be AI-powered: A Look at Mobile Semiconductor Trends with Yole Group

• Keith Schaub vice president of technology and strategy at Advantest, Pierre Cambou Principal Analyst for Global Semiconductors at Yole Group • Season 2 • Episode 15

Share episode

How is the semiconductor industry evolving in the face of global market trends and technological advancements? 

Today, we are thrilled to welcome Pierre Cambou to our podcast. Pierre serves as the Principal Analyst for Global Semiconductors at Yole Group. With a rich background that includes founding a semiconductor startup and roles at notable companies like Thomson TCS and e2v Semiconductors, Pierre brings a wealth of industry knowledge. He is recognized as an expert in the semiconductor field, often cited by international media for his market and technology insights. Pierre holds an engineering degree, an MSc from Virginia Tech, and an MBA from Grenoble Ecole de Management. 

Together we'll explore the industry's impressive annual growth rate of 6-7% and delve into the mobile market's fascinating journey from the late 90s to today. Discover how AI chips are revolutionizing smartphones and potentially transforming the sector. 

We'll also uncover the cultural and practical reasons behind the persistence of feature phones in regions like India and Europe while they're nearly extinct in the US. We'll compare this trend to the automotive industry's ADAS level zero vehicles and assess how AI and edge technologies might shape future smartphone production. Moreover, we'll discuss the nuanced role of embedded AI in advancing mobile technology, predicting a slow but steady growth aligned with the global GDP.

Finally, we'll examine the global semiconductor manufacturing landscape, touching on government incentives, reshoring efforts, and strategic joint ventures like the TSMC and Sony collaboration in Japan. Pierre provides insights into the competitive dynamics among countries like Japan, Taiwan, and Korea, and the significant investments by China. We wrap up by discussing the industry's future, considering challenges such as managing energy consumption and the importance of green energy in mitigating environmental impacts. Tune in for a compelling dialogue packed with valuable perspectives and forward-looking predictions.

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!

Keith Schaub:

Welcome back to Advantest Talks Semi. Today we're thrilled to host the experts from Yole Group, the pioneers at the heart of the semiconductor market intelligence Renowned for their meticulous research and profound insights. Yole Group not only tracks but also predicts the pulse of semiconductor test consumable markets. Today, from Yole Group, we're joined by Pierre Cambou. Pierre serves as the Principal Analyst for Global Semiconductors at Yole Group. With a rich background that includes founding a semiconductor startup and roles at notable companies like Thomson TCS and E2B Semiconductors. Pierre brings a wealth of industry knowledge. He is recognized as an expert in the semiconductor field, often cited by international media for his market and technology insights. Join Pierre and I as we delve into the latest developments in the semiconductor industry, where data meets strategy and future technologies come into focus. Welcome, Pierre, to Advantest Talks Semi.

Pierre Cambou:

Hello, Keith glad to be here.

Keith Schaub:

So, on today's episode, what we'd like to do is really talk about the semiconductor industry, what's going on in the market and, in particular, what's going on in the mobile market. So you're from a company called Yole Group, which actually studies and provides reports on this. So how about we start with just a general overview of what you guys are seeing in the semiconductor industry?

Pierre Cambou:

Yeah, thank you, Keith. So the semiconductor industry is really, you know, our main topic at Yole Group. So we cover this for 25 years and it's not a very old industry. It's only 50 years that it really exists and from the humble beginnings in some lab in the US, it grew quite significantly in the 80s, mostly in Japan with radios and VCRs, and more recently it's really through the internet and mobile that the semiconductor industry grew to this 500 plus billion industry that we know today. This industry is growing really fast. I mean, it has the kind of growth rate that China is enjoying 6% to 7% growth rates annually. And now we see some cycles. So sometimes it's 20%, like we enjoyed in 2021. And sometimes it's minus 10% or minus 8%, exactly last year, in 2023. So, yeah, there are ups and downs, but globally it's a fun ride.

Keith Schaub:

Yeah. So it's very interesting how, over the past 30, 40 years, the average is 6% to 7%, but, like you said, we've had some extreme cycles with positive, double-digit positives of 20, sometimes even 30%, and conversely, we've had some deep declines also in the negatives, which are not the best years for the semiconductor industry. So of the 500 or so billion of the semiconductor industry, about how much of that is the mobile market?

Pierre Cambou:

So we consider mobile and consumer markets, which is 60%, and let's say 40% will be mobile. Another 15% will be personal computing and also consumer devices. So it's less than 20 years and even more since they are called smart. So our phone is really powering our lives, but also the semiconductor industry.

Keith Schaub:

So you mentioned that the last 20 years, which is interesting because that's actually when I came into the semiconductor industry, the mobile really took off sort of in the late 90s, and then in the 2000s it was just on fire and it's been an interesting ride to watch it progress from 2G, to 3G, to 4G, to 5G, I don't know how many Gs we're going to have, but it has been growing. And I remember when we eventually got to a billion or over a billion phones per year, which is a phenomenal number, and now we've reached I don't know 1.2, 1.3 billion. You probably know better than I do, but for the past two, three years or so we've been sort of sitting at that number. So could you sort of give us how you guys look at this market, what has happened to have it mature, and then what are we going to see, let's say, in the next six to 12 months?

Pierre Cambou:

Yeah, I mean we should first decide what we want to look at. Is it feature phones or smartphones? So we made a clear distinction in the past. Now I think we should not anymore. So, basically, you have 1.2 billion, which is the smartphones, and you have an extra 400 million, which is the feature phones. So we are on the range of 1.6 billion phones being manufactured every year. The peak was around 2017, with a peak of 2 billion phones, whatever kind being manufactured, and then we came down to kind of the situation of today 1.6, which is an extraordinary number of1. 1.6 billion phones. So it's difficult to have more than one phone per inhabitant on the planet. And this is basically what we have, more or less, you know. Now we are into a cycle of renewal and so we know that we will have a peak of the average age of phones around 2.5, which means that five-year-old phones are replaced every year.

Keith Schaub:

Now the AI chips are starting to be embedded into the phones and it looks like that's going to have a positive impact, or a positive boost, to the market in general. Can you say a few words on that?

Pierre Cambou:

I think this is the whole story of mobile. Mobile from the start revolutionized the industry because of the addition from memory and compute. That was the existing states of the semiconductor industry. We added all the analog RF chips and then all the autonomy, battery-powered aspects and also all the sensors. So this is what we call the more than Moore. So when looking at the kind of chips that are inside the phone or powering the mobile industry, there is a diversity between memory, compute and then also the more and more chips. So it's super diverse. And now we have the GAN AI that is invading the space. We saw Apple, for example, embedding neural engines a few years back, so this was a major event and I think we are really on that trend when we want to embed even more AI capabilities on the edge. So, yes, there is an ongoing technology shift within mobile and it's a world drive series. It's a renewal market, but there's always new things, new features to be added, and this is part of the game.

Keith Schaub:

Yeah, I agree. I want to go back to something you said a bit earlier, with about 1.2 billion being smartphones and around 400 million being the feature phones. I remember when that number was reversed, when most of the phones were feature phones and smartphones just sort of came into being and they were slowly increasing in volume. So my question is are the feature phones a percent of the overall market? Does that continue to reduce and the smartphones are sort of taking over as just the de facto default? And then how does that play out in the different regions? So, for example, in the US, smartphone feature phones versus India versus China?

Pierre Cambou:

There is almost no sales of feature phones in the US, but it's significant in India, yes, but very significant too in Europe and especially in Germany. So there is a trend and it's going down. We see the trend is going down, but not as fast as many have thought. I think it has become a niche. Some people understand the value of this kind of phone and to use it. I think we are also trying to mitigate our addictions to screens, and this can be the answer. So, yeah, it has become a market. That is interesting. It does not have much value for industry because the average selling price of a feature phone is $50. When the average smartphone is about $500. So, okay, it's a 10 times ratio, so you can do the compute.

Keith Schaub:

Do you have a theory or a reason behind why the feature phones are more popular in India and Europe versus they're basically non-existent in the US? What's driving that? Is it just cost?

Pierre Cambou:

Cost is part of the reason. As I mentioned the idea that some people are fighting against invasions of screens and they want to lower their screen time, and so they find a feature phone quite interesting to have. I also think there is a kind of culture and a status symbol. You know, in some societies like the US it's very difficult to go by your everyday routine without a smartphone. Otherwise, you could be quickly in trouble if you don't have a smartphone, and I think in China it's probably the same. Now In some places I know in Europe you can still go by with only a feature phone. So you know it's interesting to notice this and there is a significant number of components like cameras that are already on those phones. So it's good to know that it still exists, and as the South East Asia countries also rise in purchasing power, we know those markets for feature phones will maintain in the future.

Pierre Cambou:

I would like to make another parallel, maybe with the feature phones. It's with automotive. We were asked recently one of our customers what was ADAS level zero? So I believe you know ADAS advanced features for cars. Is there any market for no ADAS? Adas level zero yes, there will be probably, similarly to feature phones. So maybe 10% of the market will maintain without any electronics and mobile is the same.

Keith Schaub:

That's interesting. I hadn't thought of that. So where would that 10% be geographically?

Pierre Cambou:

Again, it's not just a geography. The key learning we have right now is it's a niche market. So even with advanced countries like Germany or Europe at large, you will find customers that want to stay clear from smartphones or, in some occasion, they want to have a very simple phone for their kids and they don't want access to screen, and so they will acquire this very simple phone. So it's pure marketing here. There is an application case, let's say, for a feature phone. This application case does exist, even in a mature market like Europe. I don't know about China and the US, because those markets are very different.

Pierre Cambou:

But also on emerging markets like India. It's it's very popular because it has to be very rugged in some some places. I've seen places in Africa people having those feature phones. You need very rugged handsets so that it works on very little power, for example. So it's very interesting.

Keith Schaub:

Yeah, it's hard to imagine anyone without a phone these days. So I want to go back to what's going on with AI and edge technologies like smartphones. So, given this let's say new feature like you said, AI invading the smartphones, does Yole look at how that will affect the forecast of smartphone production in the future?

Pierre Cambou:

So, we're at 1.2 billion-ish smartphones, Do you foresee that AI is going to drive that or change that appreciably, or is it going to stay about the same? So I think the market has been maturing. What we've seen is the penetration of the number of devices per inhabitant on the planet, where it reached kind of a limit where almost everyone on the planet has a phone now, and basically during the COVID-19 episode, so 2021, 2022, there was kind of a bubble. People had some extra money, especially in the US, Europe, also China. They stayed at home, so instead of enjoying trips abroad or a vacation, they had to stay home and they had some extra money on mobile and consumer products.

Pierre Cambou:

So we saw the bump extra and the decrease in 2023. But overall, what we measure right now is that we are back on the previous trend of a slow growth in line with GDP growth. GDP growth is 3% a year, so this is our expectation. For sure, there is a requirement to attract customers with new features, such as embedded AI, that is one of those features, and so it will drive the growth of mobile in the next five years.

Keith Schaub:

Also, if you could talk a little bit about Rapidus, I think the audience would really appreciate that.

Pierre Cambou:

Yeah, so what we did is to cover the fund-raising capacity worldwide, trying to understand was it a push or a pull? We know there is a lot of government incentives with Chip Acts in Europe, in the US, but also in Japan. I think there was a lot of desire to reshore some of the capacity, such as the TSMC Sony Joint Venture and Kumamoto. So that was super interesting. And, the creation of Rapidus, and from our understanding I didn't know how strong was Japan. We saw the collapse of some of the big names like Toshiba and the creation and the restructuration of the Japanese ecosystem.

Pierre Cambou:

And really what we saw today is kind of a ranking of the different foundry capacities, Taiwan and Korea being first for sure, but Japan is just behind and this was kind of a surprise. America boasts about 10% capacity in the US and controls about 10% abroad. We know there are foundries from Micron in Taiwan and in Japan. So, this kind of capacity is US-owned but is locally, I mean geographically, located in Japan, Taiwan or China. So, Japan has a strong position today and is expected to have even stronger through this effort that we see today. And they were the quickest to have fabs online. So we saw, in two years just, TSMC being able to bring online their fab in Kumamoto. And so Rapidus, if I had a few bucks to bet on the Rapidus, maybe I would do that.

Keith Schaub:

So it's interesting. Normally when we bring a fab online or when anyone brings a fab online, it takes, I don't know three to five years before it's really up and running. And, like you said, it's a testament to Japan that they've been able to, and TSMC, that they've been able to just bring that online in two years. Anecdotally, Advantest is a Japanese-based company. Our headquarters is in Tokyo and I know a lot of employees that have recently joined Advantest, specifically for some of the reasons that you've outlined, that they're now, let's say, older in their careers and they want to revitalize the industry for their home countries. So I see that not just at the global scale, but even at this micro scale within Advantest, it's very interesting.

Keith Schaub:

Yeah, semiconductor supply chain within the United States and other countries have similar acts. Now China has invested well over 10 times that amount and I think they have plans to invest even more. And, of course, there are trade restrictions in place. There are geopolitical issues or challenges that we're dealing with. So can we talk a little bit about what's going on with the various acts and how that's affecting the semiconductor industry, even if it is affecting the industry?

Pierre Cambou:

So the government help for me is a little bit blurry because it's unsure if this money really goes into real fabs and players that really do wafers. So what I'm more concerned with is the capacity that is coming online. So what I monitor is the real that is coming online. So what I monitored is the real position of players today. I think we overblow a little bit China's position. Where is China today? Basically, you have 15% of the world's capacity in China, I mentioned earlier. The US has 10%. So it is a significant player today, very similar to Japan in terms of capacity, not quality, because you will have mainly older nodes in China and much more advanced in Japan and the US. Then, when you look at what kind of fab you find in China, only one third will be owned by a Chinese player and serving the Chinese or serving Chinese players, because you will find some open foundry, also serving some demand abroad, and also you will find the foundries that are owned by foreigners. So in the end, china only owns 5%. For themselves, it's not that much and so far if they want to grow they have to spend all this capex. They want to become something like in between 20 to 30% of the world's capacity. So yes, I mean definitely, they will outspend the US, Europe and Japan, everyone else they will out spend to reach this kind of level, which is not unsimilar to the position of Taiwan and Korea today. Maybe they will pass Japan, maybe, but that's the order of magnitude they want for themselves. So it's a clear path they want to go.

Pierre Cambou:

I don't want to be alarmist, as many analysts have gone that way. They have their own weaknesses and strengths and we have to very calmly analyze what their strengths are. And one of their strengths is the clear path. Then, in Europe and the US, we see also this desire to maintain growth in the capacity, to maintain at least those 10% that I mentioned. So the US will not grow more than the industry, so it will maintain. So maybe some politicians will ask me why I say that, but you know the industry grows quite quickly. We need to remember the 6% to 7% growth in dollar terms. But in capacity, the order of magnitude is 4% to 5% in terms of wafer output that is added every year. So by 2029, 2030, this is something like a 30% increase in total capacity.

Keith Schaub:

Okay, so I want to ask a question on that. So, if I understood correctly, the wafer output capacity grows typically 4% to 5% per year. Did I understand that correctly? Okay, so we have, I don't know the exact number, but close to 30 wafer fab facilities that are under construction right now. Does that track what the capacity is going to be in the next two to three years and how will that affect with so much capacity coming online, or is it just enough?

Pierre Cambou:

So, this work we've done at Yole Group. There have been many announcers and we have to be super careful if those announcers will be fulfilled. But even if we just look at the announcers, I think we are on the path where we are just fulfilling Moore's Law. If we just pull the line, this 6% to 7% growth in revenue will come from this 4% to 5% of added capacity per year in the next five years. So all the announcers come in line, in fact, with this idea. But we are not told everything. Every fab in China will not communicate super clearly if they are online or not. We try to monitor the key players for sure. This is easy, but then you could find numerous smaller players that are more difficult to track. But overall, what we say is that, yes, China will carve a part of the market in the 20 to 30%. That's their goal, but maybe unlikely to go to 30%, but it's always possible.

Pierre Cambou:

Is there overcapacity today? Yeah, every new capacity that is added will displace some, so supply-demand will always equilibrate somehow, and so when we add more capacity ahead of demand, then some people lose. So in my opinion, is China adding too much capacity or overall all the players? I don't think so. So far, what's been announced is online. I mean it's not different from what we've seen in the past 20 years. We had some issues for lower nodes in the automotive industrial market, so there will be maybe some price lowering of prices, but overall what we say is that this capacity addition is similar to what we've seen. The craziness of Moore's law is that it's exponential, and so we will see exponential growth of capacity at the rate of 4% to 5% every year added for the next five years.

Keith Schaub:

Okay, that's good news for the semiconductor industry and good for anyone working in that industry, so I'm sure the audience will be glad to hear that message. I want to take that thought a little bit further. So most of the reports, most of the data that I see, typically goes to around 2030. And we have this one trillion goal where the semiconductor industry should reach around one trillion around 2030. What comes after that? Have you, or has Yole Group, looked at what might happen after 2030 and at least have some sort of say rails on upper and lower expectations?

Pierre Cambou:

The question is why does the semiconductor industry exist? So what are the big underlying trends, the large cycles that it is powering?

Pierre Cambou:

So in the 80s, 90s, this was typically personal computers, and so it ended up with the internet and then mobile, and so mobile is what powered really the industry in the last 20 years, and now we have a switch towards something else. So even you know the AI, generative AI is, I think, a pivotal moment where it's not just information. We are going out of the information age because with generative AI, what can be done is robotics. We had to simplify so you can do typically the autonomous cars that we are dreaming of. Autonomous cars, maybe robots at home, but also industrial robots, for sure. So there is the potential. This new technology will require semiconductors, so huge investments today are in data centers to do this AI compute, and this is our hope that this will be the pulling force for semiconductors in the next 10 years, but I think it will not stop exactly in 2030. I think actually, we may need an extra five years to reach this trillion dollars.

Pierre Cambou:

I think this has the power to grow the industry to double. I mean, this is doubling the industry, more or less and for sure we will still have room for a next cycle, but we still have to discover what kind of cycle will come next.

Keith Schaub:

So we're looking at eventually maybe somewhere between five and ten years, according to Yole achieving this 1 trillion. But what are, or what could be, some of the limitations to that growth?

Pierre Cambou:

So, nobody is really seeing yet, you know, like a structural limit to Moore's law. So maybe a lot of people saying, we cannot scale anymore, or there are problems with scaling the technology, so we cannot add the number of transistors per square inch the same we could in the past. But eventually, I mean, Moore's law is still more or less alive, so we improve every year. But there are, in my opinion, from our research, we see, two issues coming on. First is the capital expenditure or cost. How much do I need to invest every year so that I keep improving the performance? Can I get the return on investments from this added performance?

Pierre Cambou:

This is probably one of the key issues that we have overcome in the last decades, and it's still to be seen. If we will, it might require even larger companies than the one that we see today, like Google, Amazon, Apple, super large companies able to invest in new generation of chips, and so that is one worry, and the other worry is the energy, the energy that the semiconductors consume, or electronics at large. Most of the compute will switch to data centers, and they will require even more and more energy to do this generative AI work, and this is the other issue and it could be a limiting factor. So today, the way the players are mitigating the energy demand or the impact on, for example, I mean their engagement to fulfill the Paris Agreement CO2 emissions. They buy green energy, but at some point, we will not find enough green energy to power the data center, so, that could be another limit to the growth of GAN-AI. It could be an issue at some point.

Keith Schaub:

So, do you track the energy consumption of the 10,000 data centers that are around the globe? I mean, where is that right now? Is it 2%, 4%, what is that?

Pierre Cambou:

So, we need to have a comprehensive picture of the carbon footprint of semiconductors, but I prefer to call it ICT, so all the information and communication industries, because semiconductor is just a small part. And then you have a system, electronic systems. So 50% of this power consumption comes from the internet. You know, power consumption comes from the internet and the other 50% comes from your edge, your computer at home or the office. So it's really the client, we call it, I mean some players call it client so your computer is 50% of the energy demand and all the back office is another 50%, and so this is typically what we call scope three. So scopes one and two are to create those computers and servers we spend energy, and the other when we use it. This is called scope 3. And again, there is a ratio between what's needed to create, to do those electronic equipment, and the usage, the energy that is required when we use it, and this is also in the order of 50-50. And therefore, when people say, I believe that the industry has that footprint, we need to ask, are you talking about scope one and two or scope three or all together? And in the end, the problem of the industry is the growth. Because if we grow so much and we don't reduce the energy consumption, especially for servers we know it's 50%, especially from usage this could create, especially from a public relations point of view, but probably even more if the energy is limited, then we cannot do what we want. So that's probably one direction of improving the technology to find ways to reduce, to continue to grow.

Pierre Cambou:

How will the industry overcome this? That's the big challenge. How will the industry continue to grow and eventually limit energy consumption and eventually reduce its requirement in terms of oil? So maybe PV is an option, so renewable energy. So I think semiconductor is part of that trend. The transportation revolution to EVs also, semiconductor industry is really helping there. So the industry is offsetting a lot of their footprint through helping changing the way we produce energy or the way we transport ourselves. So I think in the future we will have lots of debate around what's the real footprint of semiconductors.

Keith Schaub:

For sure. It's unclear how that will eventually get solved. I mean, obviously we're improving the types of energy that we can create, but, like you said, there's scope one, two and three, and the entire industry and the world keeps growing, and so I would imagine that scope three is more of the bigger challenge than scope one or two, because you have the sustainability of the technology. Is that true? So how does that get divided between scope one, two and three?

Pierre Cambou:

So scope one and two is what happens in factories. So if you use electricity or you burn something, you know what it is.

Pierre Cambou:

And as I said, it's quite easy. Usually a computer is. You know a computer or server. You know what is the consumption and the lifetime of the server. So you know more or less, for a mobile phone is the same. I mean you know the number of cycles and so the energy needed during its lifetime.

Pierre Cambou:

What I see is the companies are more and more open. Intel, for example, was not so keen a few years back. Now I think they have a very nice view of their footprint. I saw Samsung also is improving gradually every year. TSMC also is trying to go the right way. So they disclose more and more one and two, for sure, but again with this rule of thumb that once you know one and two, basically scope three, can add another, you double it and you have the whole picture. And yeah, I mean, my hope is, that the industry will help change or will contribute to the way we generate energy, because if we go toward renewable energy, then scope three will go down significantly. And so that's probably some of the awareness I would try to convey the decision-makers of the industry. It's like the energy generation, for green energy generation, is super important for the industry and the more we can contribute to that, the better for the industry.

Keith Schaub:

So maybe just take a couple of minutes and talk about what you're seeing with 6G.

Pierre Cambou:

There is always this demand or we try to incrementally improve on previous states. So we've seen this with 6G. If we can increase the bandwidth typically, or try to from 5G, we want to increase again, increase the frequency. The standard is not yet fully disclosed, so there are several approaches on the table, similar to 5G in fact, and so, yeah, there are proposals. Also, I think what we see is space connectivity is, from my point of view, is probably part of that. You know how can we include space connectivity with our mobile? I think it's a very, very attractive proposition and this will be probably part of 6G.

Keith Schaub:

So, I do remember when 5G was still being worked on and it took an extra year or two before it kind of really became an important part of the mobile segment. So, I suspect that we'll see something similar with 6G. It does double the bandwidth, like you said. I think it's 320 megahertz, which that has impacts on test, but it's still early days and I think that you know we'll eventually see it. It's just a question of what that ramp will be. Do you expect that the ramp will be similar to what we saw with 5G or 4G, or will it be faster?

Pierre Cambou:

No, it's not going to be faster. I think as time goes, adoption rates tend to tame a little bit, especially if you don't have a strong use case. We were being asked, I remember, for 5G what's the use case? Why do we need that much bandwidth and we still need to fill the 5G promise.

Keith Schaub:

Yes, it's funny because inside of Advantest we try to steer away from the motto of build it and they will come. So some companies and some industries embrace that, where if we build it, then they will come. But, like you said, for 5G and for other Gs it was always what's the use case, why do I need it? But it always turned out that when it came online, the use cases were found and people found very interesting new ways to use this new capability and new features. So I think it's going to happen with 6G as well, and the same thing happens with AR and VR and people ask what's the use case? There's no real use case, but it just takes time for the market to sort of, let's say, get off the ground. So I'm pretty sure we're going to see that. All right, so we're nearing the end of our Advantest Talk Semi podcast. I'd like to thank you, Pierre, for coming on and sharing your valuable insights from Yole Group and looking forward to perhaps having you back on the show at a future date. So thanks again.

Pierre Cambou:

Thank you, Keith, and thank you for having me today. That was a great experience. Thank you very much.

Keith Schaub:

You're very welcome.