The Silicon Building Block of Technology

Last year, I said that TSMC “is probably the world’s most important company.” I’m not a politician, so I’ll stand by my words, but by my estimation, a close second would be the Dutch company ASML. It plays a critical role in the advancement of the most important technology in our world today: semiconductors.  

In my mind, the semiconductor is one of the most important scientific inventions in all of history. Semiconductors are the building blocks of technology. Without them, forget all other tech progress. So we should know a little bit about chips. Fortunately, I’m here to help.

The two key categories of chips are logic chips and memory chips. Logic chips work with binary code. The original logic chips are CPUs but added to that, we’ve now got GPUs and neural processing units (NPUs). We have memory chips that store information. There are two main types of memory chips: dynamic random access memory (DRAM) and NAND (“not and”)). Logic and memory chips are the most important, as computing cannot happen without them. 

But going one level deeper, we’ve got chips designed for very specific purposes. Take an iPhone: it’ll have the main processor (the CPU, GPU, and NPU). This is what we generally think about when we talk about chips. But the iPhone will also have chips specifically for the modem, Bluetooth, near-field communications (for Apple Pay), audio, NAND storage, RAM and sensors. Each of these chips is designed specifically for that purpose. Or a car. A car may have four wheels but it can have 1000s of chips.

As technology drives forward, we’re seeing more tech companies designing chips for a specific purpose. An example is the Full Self-Driving chip (FSD), a chip designed by Tesla specifically aimed at helping achieve level 4 and 5 autonomy. Google, Microsoft, Amazon, Tencent are developing specific chips for use in their data centres. Google developed the Tensor Processing Unit (TPU), a chip designed for neural networks.

We know the importance of chips, and they have become so mainstream that your average consumer knows whether their Mac runs on Intel or Apple silicon. Actually, that’s not quite true, they’ve re-entered the mainstream - I remember the days of the Intel Pentium chip adverts and that annoying Intel jingle. Still, give me semiconductor ads over betting ads every day of the week.

The semiconductor industry is one of the most globalised industries. It’s definitely one of the most technologically advanced industries. McKinsey's research suggests that when measuring research and development as a share of revenue, at 16%, semiconductors are the most R&D-intensive product. The semiconductor industry is a prime example of why we need massive technology companies: only they have the capital, expertise, global reach and supply chain management capability to design and manufacture technology like semiconductors. 

Semiconductors are a triumph of scientific research, precision manufacturing, and mass market production and industrialisation. There’s an incredible book by Chris Miller which explores all of this. To put the technology into perspective, the transistors in modern technology are smaller (at around 5 nanometers) than microscopic viruses. The bloody annoying Covid-19 virus is about 100 nanometers in diameter, which is huge!

It’s thanks to the engineers, scientists and the technology industry’s collective innovation that Moore’s Law has continued to hold for over 50 years. Moore’s Law continuing to hold is the most important thing in our lifetimes. For all the market power, market share, successful business models and insane revenues that Google, Amazon, Apple, Meta and Microsoft have generated, none of it would have been possible without cramming ever more transistors onto silicon. 

Nvidia is a company that has benefitted from the production of GPUs. GPUs were originally designed for graphics, mainly in gaming. The key feature of GPUs is that they can handle parallel processing. Now, this turns out to be incredibly useful for artificial intelligence, and many data centres use Nvidia GPUs. Nvidia ​​GeForce GPUs were originally released in 1999, so it’s been a two decade payoff for Nvidia.

But I started by talking about ASML and stating that it is critically important, so let’s get back to that. To understand why ASML is so important, let’s take a look at the semiconductor industry. 

The semiconductor supply chain is complex, so let’s think about it in four stages: research, design, manufacturing and application in devices. The semiconductor industry is centred around the West (particularly the US) and East and South East Asia. The US, Japan and Europe are upstream (research and design) in the supply chain, Taiwan and South Korea predominantly focus on design and manufacturing (Taiwan and South Korea account for about 40% of global chip fabrication capacity), whilst the assembly into final products, the most downstream parts, occurs in China. 

I mentioned Taiwan and semiconductors, so you’re thinking (points finger and nods) Taiwan Semiconductor Manufacturing Company. TSMC has a 50% market share in semiconductor manufacturing and fabricates more chips than anyone else on the planet. But TSMC doesn’t design any chips itself, they are designed and developed by fabless companies (named as such because they don’t have any fabrication plants). Upstream of TSMC are companies such as Qualcomm, Nvidia, and, of course, Arm. Arm is the dominant company designing chip architecture for CPUs. If you’ve got an Apple device, it probably uses arm architecture, which Apple licenses for its Apple silicon system-on-a-chip. If you have a modern Google Pixel phone it’ll have Google’s Tensor system-on-a-chip, which is based on ARM64 architecture. You can’t get away from Arm.

But none of these magnificent companies could operate without ASML. ASML has a monopoly as the sole supplier in the world of extreme ultraviolet lithography (EUV) machines which are needed to manufacture the world’s most advanced chips (chips smaller than 7 nanometers). EUV lithography is the process of using 13.5-nanometer wavelength light to project chip designs onto silicon wafers. The Economist ponders that ASML’s lithography machines may be “the most sophisticated equipment ever sold commercially.” I know I’m not debating it. It's a project that started in the early 1990s, but the first EUV-enabled commercial product wasn’t released until 2019 when Samsung’s Galaxy Note10 was released (no, I’m still not switching from my iPhone, sorry, Samsung).

EUV machines are doing a pretty good job of keeping Moore’s Law going, shrinking transistors, and fitting ever more transistors into an integrated circuit. The current cutting-edge EUV machines cost about $200 million and are solely bought by the leading chip manufacturers, such as TSMC, Intel and Samsung (unsurprisingly, ASML’s key customers). 

I reckon ASML may have the greatest competitive moat in business history. The scientific knowledge, the network of suppliers, the market power, the data it collects from currently deployed machines, and the simply ludicrous amount of capital you’ll need to try to dislodge them from market dominance. But without ASML doing ASML things, you don’t get a continuation of Moore’s Law. 

At some point, the competitive moat will be eliminated - it always is. Maybe it will be the development of a quantum transistor. Plus, there is a question about what happens when Moore’s Law stops, which it will at some point. Although I’m not going to add myself to the list of failed predictions about the imminent end of Moore’s Law. But you can’t shrink transistors forever and defy the laws of physics. As Elon Musk says, “Physics is the law, everything else is a recommendation.” But let me coin Mark’s Law - science and technology will provide a solution. We don’t know what that solution is or when it will arise, but science and technology always have, and always will, provide solutions. And I’m here for that.