Semiconductor, possibly the most consequential technology of this decade
Disrupted's semiconductor industry primer
Hi friends👋,
This post is a business primer on the semiconductor industry, sprinkled with stories of its history and my thoughts on its future.
Usually, a primer is as concise as possible covering the essence of an industry, I will aim to do that, but as you know, the semiconductor is a complex industry in the whirlwind of the biggest geopolitical struggle of this decade, so covering even the fundamentals would take quite a length, hope you guys are ready for the read (~18 mins).
I have two videos on my YouTube channel for those who are not interested in reading, my coverage here will be more in-depth for advanced readers, you could also visit here for a comprehensive report on the topic from BCG and SIA. There are also many great links I’ve gathered in the process of writing this one, you can check them in the links graveyard below.
Lastly, I want to welcome the 30+ of you who joined us in the last 3 days. If you have not signed up yet, join +415 smart, curious, and critical people by subscribing here:
Here is why you should care about semiconductors.
I can say with 100% certainty that all of you reading this newsletter are viewing through a device built by semiconductors. And I can do one better. Chances are, close to 100% of you (!), have devices with one or more TSMC’s (Taiwan Semiconductor Manufacturing Company) semiconductors in them. (If you own a phone with Apple or Qualcomm chips, those are made by TSMC) This is how omnipresent semiconductors are in our lives.
Semiconductors are critical for power plants, telecommunication stations, satellite & dishes, ships, and fighter jets. Other than huge machines, they are also in smoke detectors, temperature sensors, washing machines, and even refrigerators. They are everywhere.
Not only are semiconductors part and parcel of our daily lives, but they are also mission-critical for guarding our privacy, making sure our financial systems run smoothly, and most importantly, keeping a nation secure.
Today, I will write about the semiconductor industry, going through its various components and exploring the question as to how we should understand the industry and why it is so critical for our future.
Let's dive in.
A brief history of Semiconductors
A semiconductor material has an electrical conductivity value falling between that of a conductor, such as metallic copper, and an insulator, such as glass. This unique property made it an important factor in the information age.
Thomas Johann Seebeck was the first to notice an effect due to semiconductors, in 1821.
The first semiconductor devices used galena, including German physicist Ferdinand Braun's crystal detector in 1874 and Bengali physicist Jagadish Chandra Bose's radio crystal detector in 1901.
In 1908, John Bardeen was born in Madison Wisconsin. After world war 2, John started his work in a research lab, this is where he met his colleague Walter Brattain.
On December 23, 1947, The first working transistor was a point-contact transistor invented by John Bardeen, Walter Houser Brattain, and William Shockley at Bell Labs. They called it the transistor effect.
A breakthrough in silicon semiconductor technology came with the work of Egyptian engineer Mohamed Atalla, who developed the process of surface passivation by thermal oxidation at Bell Labs in the late 1950s.
Another key moment of the semiconductor industry happened when Masatoshi Shima invented the first Intel Processor with only 2300 transistors.
After seeing the great computational potential of this technology, Intel doubled down on it. Under the observation of what’s called Moore's law, which predicts the number of transistors and the processing speed of an IC doubles every two years, the industry exploded.
Today, the state of the art Apple M1 chip has over 16 billion transistors in it, starting from a humble beginning of 2300 transistors in the first Intel chip, this is the power of exponential growth.
But of course, the history of semiconductors and the story of John Bardeen and Masatoshi Shima is a story of inventors and technologies, they are credited for the technology, but the money doesn't go to them. The true players behind these inventions are the companies.
If there is one moral I would like you to take from this story, it is that the semiconductor industry is invented in the US, but it is a collective effort by scientists and engineers around the world.
Though tech knows no borders, companies do. The semiconductor industry was shaped by unique historical circumstances, it was built by the best brains from all around the world who went to the US after the second world war for a better life. However, the companies that benefitted and own the semiconductor IPs were mostly US companies. This reality foreshadowed the chip struggle between the U.S. and China today, which I’ll talk about later.
A primer for the semiconductor industry
The semiconductor industry is complex, three core steps are involved in chip production, design, fabrication, and OSAT. On top of that, 5 specialized activities are behind the scene playing supporting roles, they are the companies behind the EDA tools, the IPs, the equipment providers, the wafer suppliers, and the chemical and raw material suppliers. Putting it together, the industry looks look like this.
Depending on the products, companies in the industry choose from two models to run their business, companies either do everything from design to manufacturing to testing, in which case it's called the integrated device manufacturers (IDM) model or it follows the fabless model which focuses on doing one thing really well and outsourcing others. Intel and Samsung are two big IDMs around the world, most other players like NVidia, Apple, and Qualcomm follow the fabless model.
The choice of business models depends heavily on the end products. In the 60s and 70s, there were no companies running on fabless models, the factor that necessitated the fabless model was the growing cost. Higher and higher amount of CAPAX and R&D is needed to build fabrication plants. This is when TSMC, along with UMC and Singapore’s CSM decided to specialize in making chips, so they can win over the market through higher production numbers from orders all over the world. TSMC’s latest 3nm foundry is eating up 28 billion USD in investment, it is too expensive to support two TSMCs in the market. This is why the industry must shift towards a fabless model for high-end chips production.
However, the focus on high-end chips is the narrative that is often seen in the newspapers, this reflected the world’s two superpowers’ growing anxiety towards TSMC’s power. This narrative is not representative of the industry. High-end chips production by TSMC is a critical factor of the semiconductor industry, but it is not the total equation because semiconductor isn’t just about the small M1 chip, it’s about a lot more.
Leading Edge & Trailing Edge
There are broadly three types of semiconductor products, logic chips, memories, and DAO products.
Their global capacity looks like this at different node sizes. Only two companies are competing in the 3 - 5 nm chips category, they are TSMC and Samsung, most others are in the larger node category on the trailing edge. ( leading edge is < 28 nm, trailing edge is > 28 nm)
Briefly, logic chips are used for computation such as determining your excel output and colors of a pixel; memory chips are specialized in information storage such as in your RAM or SSD; and lastly, DAO in the graph represents discrete, analog, and optoelectronics and sensors which exist in smoke detectors, fridges, remote controls and more. They specialized in converting real-world signals like temperature and light into digital signals.
Analog Devices is a 60 billion-dollar company in this category, as you may tell from the node sizes (55 nm and above), much of the DAO products are essentially commodities sold at potato prices, but they are everywhere, so the volume makes it a great business. (AD’s 2020 revenue is 5.8 billion and its net profit is 1.2 billion, +20% net margin is not bad at all)
I hope you can see the complexity here. Our discussion today will focus on the hotly contested process nodes, which are the 10nm and below, high-performance chips category, but semiconductor is about a lot more. In fact, in the higher node sizes category (40nm+), the fabless model becomes less necessary as CAPEX/revenue ratio is low, many companies in this category follow the IDM model, such as Analog Devices, Micron & Infineon.
If you want to know more about strategies of trailing edge memory and auto chips, I wrote this post a while ago on Singapore’s semiconductor strategy.
Now let’s zoom back to the production process of the leading-edge chips. As I explained earlier, the high Capex and R&D costs necessitated specialization.
Take Apple as an example, to make its state of the art M1 chip, the process goes like this. I borrowed the graph below from BCG’s semiconductor report to illustrate the process.
After Tim Cook decided that the M1 chip is a go, the first step is to design the chip, two independent elements are considered, IP, intellectual property licensing, and EDA, electronic design automation tools. These two are essential to make powerful processors that exist in our modern computers. These are the software architectures and the brain behind designing a modern chip. (Shown in the graph as steps 1 - 3)
The second step is fabrication, which involves buying the equipment, chemicals, and wafers for chip production, companies in this vertical are mostly from the US, EU, Japan, and Korea. (Shown in the graph as steps 5 - 8) These supplies will be shipped to a foundry for fabrication. Fabrication is the production process and the above-mentioned are the necessary components for production. (Shown in the graph as step 9)
The last step (Step 10) is OSAT, testing, and assembly. This is an area that is dominated by Taiwan and Mainland China companies (>50% of global capacity).
Over the years, the fabrication process has grown to become incredibly complex. In order to keep up with users’ demand for faster products, transistors on a chip are getting smaller. This is the reason why you now could buy an M1 Macbook Air that is as powerful as a Macbook pro from 2 years ago in graphic processing. To fit more transistors to improve processor speed, foundries have developed technologies over the years to make transistors tiny.
So, this is the basic framework of the semiconductor industry. You have 3 different business segments that go into the process, design, fabrication, and OSAT, and you also have 5 different types of suppliers that are fundamental to the production, that is IP, EDA, the chemicals, the equipment, and wafers.
Further, the semiconductor industry at the higher end is simultaneously Capex intensive and R&D intensive. What does it mean? It means the industry must consolidate to have a concentration of power in a few big players. Not only that, companies along the value chain must work with one another to ensure less redundancy.
The EDA tools provider for example essentially consolidated to only three players based in the US, Cadence Design Systems, Synopsys, and Mentor. For IPs, ARM is the king.
The foundry business is at the moment dominated by TSMC which takes over 50% of the overall vertical revenue, however, if we look at wafer capacity across process nodes, the U.S. is strong in the 10 - 22nm node size, Taiwanese fabs take over 90% of the <10nm node sizes.
The chemicals are dominated by Japanese companies, the wafers are essentially supplied by 5 big names that control 90% of the market (Shin-Etsu, Sumco, GlobalWafers, Siltronic & SK Hynix).
With these materials ready, a typical silicon wafer goes through the following process to reach completion, below is a view from the lithography standpoint.
Steps of coating, adding photoresist, lithography, etching, doping was carried out once each layer of materials are applied. More detailed explanations are beyond the scope of this analysis. Essentially, as IC patterns are “printed” onto the wafer through this process, chips are made.
The semiconductor equipment market captures 64 billion in value, ASML’s Extreme Ultra-Violet (EUV) machine is essentially a monopoly in the cutting edge lithography vertical. A single EUV machine can cost +$150 million. Applied Materials, Lam Research, and KLA in the United States are other names that make machines for chip production.
Throughout this value chain, you probably noticed that I mentioned a few of these elements dominated by companies in one country, that is the reason behind the geopolitical tensions, we’ll talk about that in a bit.
Chip Shortage due to both demand surges and supply-side constraints
Supply constraints are short to mid-term.
Finally, the incredible interdependence throughout this process is the first reason behind the global chip shortage. No company in the world can control the entire supply chain and everyone depends on everyone else to make semiconductors. What that means is that it is incredibly hard to ramp up production, as Mckinsey Analysis noted. If Apple wants to create an iPhone with a new 3nm processor, it cannot do it alone, it has to work with TSMC in Taiwan for fabrication, and TSMC must also work with Dutch company ASML to purchase the machine and with Japanese companies for the chemicals and wafers. And this is not a simple game of coordination and conference call either, it’s research and breakthrough, and it means ASML engineers working side by side with TSMC engineers, literally. ASML has 2500 employees (about 10% of its total workforce) in Taiwan.
Apple's Chip design team may be able to design a chip with 3nm technology, TSMC and ASML still need to invest in R&D and building new factories to make that a reality. TSMC spent an estimated 28 billion dollars for the 3nm foundry in Taiwan in 2019, imagine the commitment and risk.
Anyways, this is the first reason for the global chip shortage, it’s about supply. The fact is that it is incredibly hard to ramp up supply because of the capital commitment and associated risks, not to mention the geopolitics.
Demand surge is here to stay.
Secondly, COVID has also brought incredible demand problems for the semiconductor industry. A Mckinsey report that spelled out the forecast vs. actual sales in the semiconductor industry in 2020,
It shows that consumer electronics, among various other product categories, has grown by +15% more than forecasted growth during COVID last year. The auto sector chip demand dropped during the pandemic by -16%, which caused the shortage today as the demand recovers. Demand for more semiconductor products was not anticipated, this meant the lack of production capacity going into 2020. Everyone is now working on full gears to increase capacity, but as I explained on the supply side, it ain’t easy.
Lastly, various strong industry factors also exacerbated the increase in semiconductor demand, 2 in particular, are robust. The demand for specialized chips in the AI, cloud, and the auto sector has grown strongly over the years and due to geopolitics, many Chinese vendors are also stockpiling chips to save up for a rainy day, I’m talking about a potential U.S. sanction of course. Since Chinese vendors are dominant in the area of smartphones, PC, AI & EVs, this created the incredible semiconductor demand we saw in 2020. The combination of the strong demand and the lack of growth in supply resulted in the global chip shortage today.
Chip Supremacy narrative’s problems
Power corrupts, absolute power corrupts absolutely. Chip Supremacy narrative is problematic as it negates the role smaller countries play in global supply chain. It also lacks honesty as neither China nor U.S. could easily “decouple”.
Since 2018, the US administration has slapped multiple sanctions on Chinese semiconductor companies banning them from accessing American technologies, here is why China is uneasy about this.
China imports more chips than it does crude oil, and the majority of those imports come from the U.S. Though China knows that U.S. companies will suffer more than Chinese companies if their trades are cut off, the U.S. ability to hurt China’s economy is not acceptable to the Chinese government. Remember, every % drop in China’s GDP means a proportional increase in unemployment (instability), and a proportional increase in social problems that translates to poverty and deaths.
One company that has taken the brunt of this geopolitical struggle is Huawei. Claiming national security concerns over CCP’s reach into the company, the US has cut off Huawei’s partnership with TSMC, the world's biggest advanced chip manufacturer that produces Huawei’s chips, and the results were devastating for Huawei. Its mobile devices business has fallen off a cliff for the foreseeable future.
While there are conflicting arguments on both sides saying it is unfair treatment on the Chinese side, or it is of national security concerns from the American side, it is beyond my scope of analysis for this video. What I’m interested in is the business aspect.
To understand why the semiconductor industry has such a delicate balance of power among players, we shall inspect the industry through a national security lens.
How does each country/region do in the value chain?
Dark blue color denotes U.S. value add in the specialized segment of the value chain. In the design segment, EDA tools and IP holders take their share of the revenue. You can think of EDA tools and IP holders as software-level raw material providers for Fabless chip designers, and they are mostly dominated by US companies.
In the Fabless market, the US again dominates, especially for logic chips. Big names like Qualcomm, Broadcom, NVidia, AMD are all from the US. But China now has a wild card that is Huawei, its highly advanced Hisilicon Kirin chips are competitive even compared to Apple’s M1, challenging Apple & Qualcomm chip design lead.
Moving on to the Fabrication segment. For any company to succeed in this area, it needs to purchase three critical resources, the equipment for the fabrication process, the wafers, and the chemicals required. Both the wafer production and the chemical suppliers are led by the Japanese, US, and EU companies. The equipment needed for fabrication is led by US companies like Applied materials and EU companies like ASML. ASML in particular is a monopoly in the market of advanced chip production in the lithography process. In the OSAT segment, Taiwan & mainland China takes over 50% of the market share, though OSAT is relatively easier to replace. Overall, as you may see, U.S. companies are the absolute champions across the value chain, capturing not only the most revenue but also the highest level of technology. This is understandable as US companies invented this industry, to begin with.
The most important leverage of China over the semiconductor supply chain is its chip demand because even though China isn't good at making chips (atm) following the three-step process, it manufactures most of the devices that use chips. 90% of the global handset and computer market is dominated by Chinese manufacturers like Xiaomi and Lenovo.
According to Statista, in the fiscal year 2020, Qualcomm earned $14,001 million from the Chinese market compared to a meager $1,129 million from the U.S. The writings are on the wall. On Top of that, the Chinese consumer market is also the second-largest in the world after the US (both are at around 6 trillion USD per year) and is projected to double that of the US in 15 years. This shift in consumption power from the US to China will have far-reaching consequences.
Firstly, U.S. companies will want to sell to the market which gives China significant leverage. Secondly, as the semiconductor industry is a highly capital intensive industry, losing the Chinese market could mean less capital to invest in the next-generation technology (such as chips for metaverse), TSMC for example needs 20 billion dollars in capital to build 3 nm foundries, without either the US or China market, funding future R&D would be impossible.
But IF China and the US must work together in Chips, why are the US and China fighting over this at all? Is that a waste of time? All of this boils down to trust. The US and China no longer trust each other as partners in their strategic supply chain. China feels necessary to challenge America’s dominance in the market, reflecting on the Huawei ban, and the U.S. also feels vulnerable about the real potential of China accomplishing its “made in China” 2025 goals.
They both want to make sure that they can live without the other, much like a couple going through a divorce. If you look at the value chain again from the geopolitical lens, if the US wanted to make chips, one way or another, the U.S. would be able to mobilize resources from the US, Japan, Korea, and Europe for the simple fact that there are US military bases on them. But for China, it is stuck. Chinese companies may be good at fabless (Huawei), OSAT (JCET) & fabrication in the trailing edge (> 28 nm) node process, it is still years behind leading-edge foundries in Taiwan, South Korea, and soon, the U.S. (by TSMC)
Taiwan with its great capability of chip fabrication becomes a critical flashpoint, as we know, China and Taiwan, in theory, are still in the middle of a civil war, this doesn’t make things easier. In the worst case, if winning a war depends on designing and producing advanced chips like how Germany needed oil to win WWII, neither China nor the US can allow the oil of the 21st century to be held in monopoly by the other, making Taiwan a critical piece in the puzzle.
Conclusion
Personally, for me, the superpower struggle narrative still lacks perspective. A few companies that have significant leverage over the entire industry are from neither America nor mainland China. A few chokepoints in the industry include the Taiwanese TSMC, the Japanese ARM, the Korean Samsung, and the Dutch ASML. American firms are super powerful in this industry, but so do Taiwanese, Korean, and Japanese companies. China may be the biggest customer of chipmakers, it is not easy to break into the high Capex and R&D-intensive Industry. All companies and countries should have an equal say in this, rather than simply following a black and white superpower struggle narrative.
If the most important commodity of the 20th century is oil, semiconductors might be the equivalent of that in the 21st century. As we head toward the third decade, much is changing and much is at stake, though the rhetoric of competition is high, the industry of semiconductor is unique and delicate, well the chip supremacy narrative focuses on the U.S. and China, my hope is, smaller players like Taiwan, Korea, and Japan get their fair share of voice, protect this delicate balance of the industry and promote better and faster products for all of us.
Your articles are so good!