Discover more from Best Anchor Stocks
All You Need To Know About Semiconductors
The most important industry in the world?
This article was first uploaded for Best Anchor Stocks subscribers. Best Anchor Stocks is an investment research service focused on high-quality, resilient businesses with significant growth runways ahead. The portfolio is currently made up of 13 companies, distributed as follows across industries:
Since inception (01/14/2022), the portfolio of Best Anchor Stocks has performed as follows compared to the QQQ and SPY:
If Best Anchor Stocks sounds like your kind of thing, be sure to check it out by clicking, there’s a two-week free trial.
The semiconductor industry is complex and there are few places where one can find an introductory article. This article aims to fill this gap.
Semiconductors are critical for technological development and, thus, our future.
The industry has evolved over the last two decades to become more efficient and collaborative, facilitating compliance with Moore’s law.
We go from the basics to the industry’s structure.
The goal of this article is to provide an introduction to the semiconductor industry, explaining the most important bits of it in a digestible way. Our goal is to fill a gap we found when we started our research on the industry. Back then, we found that the introductory material was scattered across the web, making it pretty difficult to get up to speed fast. After reading this article you should have a pretty good idea of the basics of the industry in case you want to investigate further.
We are invested in the industry through two companies, one of which is ASML (ASML). For this reason, you’ll find that some parts of the article make reference to the company’s position in the industry.
Note too that there are significant differences between analog and digital chips. While we feel that analog is a great place to invest due to its resiliency and lower risk of technological disruption, this article will focus on digital chips.
Without further ado, let's start with the basics.
What is a semiconductor chip?
A semiconductor chip is a collection of millions of transistors contained on a piece of silicon, which is the so-called semiconductor. Silicon conducts electricity on the middle ground between conductors (like metals) and nonconductors (also called insulators).
The transistors are wired together to perform certain tasks and act as switches that can be turned on or off to allow the flow of an electrical current.
High-end chips contain billions of transistors, although they are just a fingernail big. Being able to fit an increasing number of transistors in the same space is vital for our future, as technological evolution depends on this increased transistor density. The growth in transistor density follows one of the essential parts of the semiconductor industry: Moore's Law.
Moore's Law - Driving technological development
Dr. Gordon Moore co-founded Intel (INTC) in 1968 and led the company as its CEO from 1979 until 1987. However, his most famous "discovery" came in 1965 while working at Fairchild Semiconductor, where many of the first big names then worked. This discovery is what we now know as Moore's Law.
Moore's law is not a law of physics of any kind. Instead, it's an observation and prediction by the famous Intel founder about the rate at which transistor density would evolve. Moore predicted that the number of transistors included in chips would double roughly every two years while costs would go down by 50%. It seems crazy to think about it, but in the year 1970, chips contained less than 5,000 transistors. In 2020, we reached 50 billion transistors.
Gordon Moore made his prediction in 1965 and, judging by the graph below, the chip industry has managed to keep up with this frenetic pace up to today (at least in number of transistors):
As soon as we read about Moore's law, we started to ask ourselves if the chip industry would've been able to achieve what it has achieved without this law. Our gut leads us to believe that the industry would be behind where it is today without it, because the observation had more profound implications for the industry than what we might imagine. It pushed the whole industry to collaborate towards a common goal:
Moore’s law has tremendous implications - it motivates and challenges all of us. With a global consumer’s need and an orchestrated and cooperative effort from all industry manufacturers, suppliers, government organizations, consortia, and collaborations between universities and semiconductor industries, we are marching and keeping pace with Moore’s law.
Lin Zhou (Engineer at Intel) (Source)
This cooperation is evident everywhere we look. For example, equipment manufacturers such as ASML are successful because they actively collaborate with chip designers. Manufacturing an ultraviolet system that is not tailored to what designers require in 10 years would be of little use. At the same time, chip designers are in close touch with end-customers to cater their designs to their requirements because designing a chip that nobody wants would not be the best idea. This spirit of cooperation is vital in the semiconductor industry, and it's not that usual to see different players coming together in such a coordinated effort in other industries. Truth is that we are where we are today thanks to this cooperation, which might also be a reason why trying to localize the industry in certain countries might turn out to be a challenge.
Complying with Moore's law has profound implications for our future. As transistor density increases, chips become faster, more powerful, and consume less energy. When you hear or read that technology has a deflationary impact, that’s Moore’s Law at work. This deflationary impact increases tech penetration worldwide, which increases the demand for chips. It’s sort of a flywheel that makes the future quite difficult to estimate. According to ASML’s CEO, Peter Wennink:
Why didn't we see this massive and big demand for our products coming?
Because we simply did not connect all the dots. And it's still a challenge today to keep connecting all the dots. But it's the value of Moore's Law, which is basically reducing the cost per function that will drive our business, and we'll create these building blocks for growth and solving some of humanity's biggest challenges. And we are a strong believer in this. And Moore's Law is alive.
Source: ASML 2022 Investor Day
Understanding Moore's law is crucial to understanding why ASML is so important in the industry. The Dutch giant has a monopoly in EUV, the only lithography system that can be used below the 7 nanometer node. To give you an idea, that's much smaller than the flu virus, which is 80 to 100 nanometer (nanometers quotes in chips are more of a marketing term than the real size). Without its EUV technology, the chip industry would’ve fallen behind, negatively impacting technological progress. This does not mean that other companies are not essential, by the way. As discussed above, leading-edge chips are the result of a coordinated effort by many players.
Up to now, we have referred to chips as one category, but there are different types. Let’s go over these.
If you like these kind of posts and enjoy the content provided by this blog, be sure to subscribe. It’s free!
Note that we will not spam your inbox as we aim to upload quality posts which take time to write.
Types of chips
As already mentioned, chips are designed and manufactured to satisfy many distinct functions. Going through all of these functions would be of little added value for you, so we'll simply go over the two main chip categories: logic and memory.
Logic chips are responsible for processing information to conduct a task. They can be understood as the brain of any electronic device. The best-known logic chips are CPUs (Central Processing Units), although they have evolved to conduct more specific functions into GPUs (Graphical Processing Units) and NPUs (Neural Processing Units). The former is optimized for visual display and simultaneous computing, while the latter is designed to conduct deep and machine learnings tasks.
On the other hand, memory chips store information. We can differentiate memory chips further into two subtypes. First, we have the data that different applications of any device need to operate, which must be readily available while the device is turned on. The chips that conduct this task are called DRAM (Dynamic Random Access Memory). Once the device is turned off, this data is erased, so to say.
Secondly, there's information that the device's applications don't need to operate, which can be stored statically. However, this information can't be erased once the device is turned off. NAND flash chips are in charge of this type of memory. You can find these chips in USBs, memory cards…
Whereas logic chips are primarily manufactured by IDMs or Integrated Device Manufacturers, and Foundries, memory chips are typically manufactured in dedicated memory-chip factories. If you don't know these terms, don't worry, we'll go over them in this article.
Even though logic chips seem to be more complex by definition, we should not rush to conclusions. Complexity ranges from low to high in both chip types. There are leading-edge memory and logic chips, and the same can be said for trailing-edge chips.
Everything until now might sound a bit confusing, so we made the graph below to help you understand the main types of chips available. We hope it helps:
Important is that the manufacturing of these chips is all done with almost the same technology. Lithography systems are used for both Logic and Memory chip manufacturing.
In 2021, the logic market accounted for 52% of ASML's revenue, while the memory market accounted for 22% of total revenue.
Both chip categories are expected to enjoy significant tailwinds going forward, driven by increased technological adoption, the need for more advanced chips, and the need to store ever-increasing amounts of data.
Now that we know the basics of chips, it's time to see how the industry comes together to manufacture these microscopic high-tech structures.
What players dominate the semiconductor industry?
As already mentioned, the semiconductor industry is vital for the evolution of humankind. Moore's law drives it, but this law can only become a reality if the industry works together. The first thing that you should understand is that there's a lot of dependency between players in the chip industry; it's not a place where ego can be tolerated. However, the industry has changed quite a bit over the last decades. Let’s see how.
Transitioning to the fabless model
Not so many years ago, before the mid-1980s, chip companies were vertically integrated. This means that chip companies designed their own chips, manufactured them in their fabs, and sold the resulting products. Fab is short for “fabrication" and fabs is a shorter word for chip production facilities. Take into account that building a fab was somewhat affordable during these years. For example, building an advanced fab in 1994 cost approximately $2 billion:
However, these costs have skyrocketed since...
TSMC invested $9.3 billion in its Fab15 300 mm wafer manufacturing facility in Taiwan. The same company estimations suggest that their future fab might cost $20 billion.
These high upfront costs made it increasingly difficult for new companies to enter the industry.
Vertically-integrated companies are known as IDMs (Integrated Device Manufacturers), and while these companies still exist today (Intel, Samsung, Texas Instruments...), the industry has changed a fair bit.
The industry changed because it was the only way to continue marching along the predictions of Moore’s Law. Consider that chip manufacturing requires high upfront costs that can be considered fixed or sunk. These characteristics result in a need for high productivity to allow companies to reduce the cost per chip and be profitable. The only way to do this? Maintaining high capacity utilization throughout the entire year, something that was a challenge in a fragmented industry suffering from conflicts of interest. After all, outsourcing production to one of your competitors was not the best thing to do. They could steal your design or parts of it, so IDMs didn't interact. That also implied that excess production capacity could not be filled and thus increased the cost per chip during low capacity utilization periods.
The search for a more efficient manufacturing process joint with the high upfront costs required by fabs gave rise to fabless companies. These companies receive the adjective "fabless" because they don't own a fab. Instead, they conduct specific parts of the chip value chain that don't require manufacturing. Of course, the emergence of these companies was coupled with the emergence of pure foundries, companies that focused on chip manufacturing without the design. This was not only more cost efficient but it also ended the inherent conflicts of interest, as design and production were split up and companies didn't have to be afraid anymore that their designs would be stolen.
The industry did not completely change; it was simply split into two parts:
In the design part of the industry, we can find several players today. The better-known design companies are companies like Nvidia (NVDA), AMD (AMD), Qualcomm (QCOM)… However, there is a relatively new trend in which consumer-facing companies are starting to design their own chips. For example, Apple (AAPL) has been designing its chips for a long time now (since 2008). Amazon (AMZN), Tesla (TSLA) and Meta (META) also do this today. There are also rumors of companies such as Ford (F) jumping on this train soon. In a world where technology is seen as a differentiating factor, designing your own purpose-built chips can mean more bang for your buck.
However, the "legacy" design companies are in a really strong position nonetheless, as they own most of the IP (Intellectual Property) that newcomers need in order to design most of their chips. Without this IP, newcomers would take decades to design high-end chips as they would also have to design "simple" patterns that were already solved many years ago.
A niche of the industry is expected to benefit significantly from this new trend: the design software companies, commonly known as EDA (Electronic Design Automation). Complex chips can only be designed with the design software companies like Cadence (CDNS) or Synopsys (SNPS) provide:
Newcomers to chip design mean new seats sold by EDA companies, so they will most likely benefit. Design is an essential part of the value chain, and design companies spend billions of dollars in R&D to design high-end chips, a process that takes several years. The fact that the process takes several years also makes EDA companies quite resilient, as customers are unwilling to stop the design process even when times are tough.
If we turn to manufacturing, we have two differentiated players: foundries and equipment manufacturers. Foundries (such as TSMC (TSM), GlobalFoundries (GFS) and UMC (UMC)) manufacture the chips. Capital expenditure is extremely high in this segment of the industry (averaging around 20-30% of revenue), which has seen tremendous consolidation over the last few years for this reason. TSMC, Samsung, and GlobalFoundries make up almost 75% of the industry:
Equipment manufacturers are also get a supply of critical pieces from other equipment manufacturers. For example, an EUV system made by ASML has around 100,000 parts, so there’s a complex supply chain behind the semiconductor industry’s supply chain. That shows you how interwoven and complex the supply side of semiconductors are.
We will not go much more into detail in the industry as a whole. We have built the following graph to summarize what we have seen in this section.
Note there are companies missing here. The goal was only to include some logos for illustrative purposes.
If you like these kind of posts and enjoy the content provided by this blog, be sure to subscribe. It’s free!
Note that we will not spam your inbox as we aim to upload quality posts which take time to write.
How chips are made
Now that we understand the different pieces of the industry and how they come together, we can look a bit more in detail at how chips are made, from the design phase to the assembly on the device:
The process of manufacturing a semiconductor chip is a collaborative one. Even though the tasks inside the industry are distributed across independent players, none of them are really independent of one another.
In the design phase, the four most important players are:
Chip designers, which can be pure-play such as Nvidia or AMD, or consumer-facing companies that decide to design their chips such as Apple or Tesla.
End-customers who outsource the design of these chips to pure-play companies.
Equipment manufacturers such as ASML or Lam Research that build the equipment necessary to manufacture these designs.
EDA companies that provide the critical software to design chips.
End-customers are responsible for letting the designer know the specifications they will require from the new chip (i.e., what functions they expect it to complete). Once the designer knows these requirements, it has to do two things. First, it has to start designing the chip with the software provided by EDAs. Secondly, it has to be in close contact with equipment manufacturers to let them know the type of chip coming to the market in the future. This last step is vital because all the work is useless unless the lithography systems are able to build these chips.
Chip designers need two "tools" to design a chip: architecture and software.
The architecture can be understood as the coding language. There are two main types of architecture, RISC (Reduced Instruction Set Computing) and CISC (Complex Instruction Set Computing). Under the RISC architecture, instructions are simpler but are executed faster, whereas, under CISC, instructions are more complex and take more time to execute. As a result, RISC chips are more energy efficient as the tasks they complete are simpler, which is one of the main reasons they are the most widely used chips in smartphones.
We don't want to outline more details of the architecture because we don’t think it will add much value and we don’t consider ourselves knowledgeable enough on the topic. Instead, we'll leave this table with some differences between both, in case you want to dig deeper:
Many years ago, there were no high-tech tools to design and test a chip, so chip designers had to design chips by hand or use their own custom-built software. Then, once designed, they needed to manufacture them to test them as there was no way of simulating their performance. This was a very costly process both financially and time-wise. At the end of the 1980s, EDAs emerged to fix this problem by providing software that allows chip designers to design and test this design before mass production.
With these two elements (architecture and design software), the design companies can design chips, which go into the manufacturing phase when the design is finished and the photomask with the design has been manufactured. The photomask is like a mold that foundries need to print the required pattern:
Once the design is ready, the chip can go into the next step: high volume manufacturing. Foundries and IDMs (in their own fabs) are responsible for mass production. The places where this production takes place are commonly known as fabs.
Fabs are massive industrial complexes full of equipment, with the bulk being lithography systems, both EUV and DUV. Here is where ASML comes into play, as it's the leading supplier of DUV systems and the sole supplier of EUV systems. Fabs are also known as clean rooms because manufacturers have to keep them as clean as possible. Take into account that the smallest dust particle falling on a wafer can ruin the whole chip.
There are many steps involved in manufacturing a semiconductor chip. Lithography is a critical step, but it’s not the only one.
To “print” the photomask’s design, chipmakers need several things: a light source and a surface to print it in. Let’s start with the surface first.
Chips are “printed” on wafers. A wafer is a thin slice of semiconductor material, usually silicon:
The “printing” process increases in complexity together with the complexity of the chip’s design, because the more transistors you want to fit in the same space, the more resolution you need. This is precisely why R&D in the industry is oriented towards achieving higher resolution. High resolution translates into smaller, faster, cheaper, and lower-consumption chips, which everyone is after. The fact that your phone now has the computing power of a NASA computer 30 years ago is thanks to an "ever-improving" resolution.
The wafer is then coated with a light-sensitive material that changes when exposed to light called photoresist. Once the wafer has been covered, it enters the lithography machine. In the case of EUV, the machine generates extreme ultraviolet light, which is reflected on several mirrors to increase its resolution, passed through the photomask, then reflected on more mirrors and finally projected on the wafer. As soon as the light hits the photoresist, the circuit is printed on the wafer.
This process is repeated until the wafer is completely covered with many identical patterns.
Although this process already seems complex by itself, there are plenty more steps. Once the wafer comes out of the lithography machine, it’s baked and developed to make these changes permanent. After the baking process, there’s an etching process through which material left in the open spaces is removed, leaving a 3D pattern. Does it look complex enough? Well, this process has to be repeated once per layer, with advanced chips having up to 100 layers.
If we were to put it graphically, it would look something like this (assuming that the design has already been finished):
We could go more into detail on this process, but we think this is enough to portray the main message: it’s a very complex process. However, this article explains it quite well and in detail for those interested in knowing more about the process.
Manufacturing semiconductor chips is very complex, and there can be errors during the process. This is precisely why chips must be tested to see if everything is running as expected. KLA is the main player here, as it provides metrology and inspection systems to its customers. This equipment helps measure the chip's quality and possible defects. Note that ASML is also a relevant player in this niche, which falls in the company’s strategy of providing holistic lithography:
The testing phase is not only important to analyze if the chips can go into the next step of the process, but it's also critical to find possible manufacturing errors early on so that the lithography systems can be adjusted for these errors and the following chips don't come out with the same defects.
It’s an iterative process through which chip manufacturers aim to increase the yield. The yield is the number of working chips on a wafer. For example, if one wafer has 100 chips and only 40 work as expected, the yield is 40%, and the objective should be to increase it. Low yields are very costly for the fab because most costs are fixed or sunk, and thus lower productivity results in a higher cost per chip.
Once the chip has been tested, it must be packaged. Packaging is simply the process of encapsulating the chip in a supporting case to prevent damage and corrosion. It's basically like putting it inside a box. In the following illustration, you can see how the integrated circuit (which is the output of the chip manufacturing process) is packaged into a case:
In real life, it looks like this:
Once the chip has been packaged, we can go into the last step: assembly.
The assembly process is relatively simple and is usually conducted at the customer's site. It's simply putting the chip into the electronic device. For example, Apple gets the packaged chip from TSMC and puts it inside an iPhone. This is what an already assembled chip would look like:
As you can see, the process that goes from the design to the assembly into an electronic device of a semiconductor chip is very complex, which is why it can take years to be completed.
If you enjoyed this post be sure to share it with people who might be interested and to subscribe (both things are free!)
We hope this article helped you understand the semiconductor industry a bit better and helped you learn something new. It's not the most detailed industry analysis, as we would need a book for that. However, we think it should help you get a good knowledge base to dig deeper into the stocks you are interested in. This is a truly a fascinating industry.
In the meantime, keep growing!