A world without 14 nanometres. Or how things could fall apart.
- Published
- in Business Models
I recently wrote about how our current generation of technical achievements is critically dependent on just one company in Taiwan. TSMC is the only chip company in the world that is capable of making the most advanced chips that drive our latest smartphones and data centre servers. Without them, we might need to wait another five years for AI or 6G to progress, and Apple would probably be unable to make anything more recent than an iPhone 12.
It’s an example of how so much of what we rely on is bottlenecked by a single company. In this case, it’s also a single country, as it’s proving difficult to replicate TSMC’s engineering experience outside Taiwan, despite hundreds of billions of dollars being thrown at the problem. These bottlenecks aren’t uncommon. They may be due to a company that make a component, a company that makes the tools to manufacture that component, or even the raw materials. If any part of this chain is disrupted, things will start to fall apart.
We saw some elements of this during Covid, but those were mostly supply chain issues, where things were in the wrong place. Manufacture didn’t stop; shipping became difficult and some industries failed to plan for the resulting extended lead-times. But what if, at some point in the future, manufacturing did stop?
What made me think about this is a seminar that the British Computer society is organising on Digital Sovereignty. It’s looking at the UK’s technical vulnerabilities and how the IT industry and infrastructure would cope if key technologies were to become unavailable for any reason. It’s a good question.
Like most nations, the UK has always had a concern about maintaining its defence capabilities. The thinking around that has historically (and actually, very recently) been about heavy engineering, with much hand-wringing whenever a steel foundry is shut down. Billions of pounds of public money go into maintaining the ability to build ships. But far less thought seems to be directed at the vulnerability to more modern tech.
Britain has always performed an interesting balancing act. Since around 1750 we’ve not been self-sufficient in food. Back then we could feed a population of around 7 million. However, the growth of a merchant navy led to an increasing reliance on imports. That policy showed its weakness at the end of the second world war, when the nation was close to malnutrition. Even at the end of the war, food rationing continued for a further nine years. As manufacturing has moved overseas, a similar vulnerability has developed across all areas of technology. Most people forget how dependent we are on tech for every moment of our daily lives, largely because it’s invisible.
It’s interesting to look at the steady development in semiconductor technology, which is embedded in so much of what drives our economy. While many products, like toasters, light bulbs and thermostats use chips made with production processes that date back thirty or forty years, most of what controls our lives – phones, computers, data servers and the networks which connect them, use far more complex semiconductor processes. The industry doesn’t deliberately design obsolescence into these products, but as we crunch more data and stream more video we need devices to work faster and at lower power. That means cramming more transistors into each square millimetre of silicon.
The measure of complexity of these chips is generally classified by the process node – the width of the smallest lines which can be etched onto a silicon wafer. Twenty years ago, that process node was 65 nanometers (nm) – about one thousandth of the width of a human hair. That allowed us to build chips with around one million transistors per square millimetre. At that level of complexity, we were about to make the first chips that supported cloud computing, starting the steady growth of data centres.
Every few years since then, chip manufacturers have managed to make something smaller, squeezing more transistors into the same square millimetre of silicon. Today, with the latest 2nm process, it’s possible to put around 300 million transistors into each square millimetre. That’s good, because you need complexity if you want to do 4G, or 5G, or AI. The introduction of these technologies is closely linked to the ability to make chips that are fast enough to support the complexity of every more difficult processing tasks. That means ever greater densities of transistors are needed, as you can’t make the chips any bigger.
The downside is that this provides a gate. Unless you can make 2nm chips, it’s going to be difficult to make 6G phones. If you can’t produce 10nm chips, there would not be any more 5G phones. Without 14nm chips, your PC would not be able to run Windows 11.
The problem the world has is that not many countries can make chips this complex. At 2nm, you need to go to Taiwan. At 14nm a few more countries have facilities, but not many. Intel has a plant in Ireland and two in the US, but outside those, you’re back in the Far East. If anything happened to those plants, we wouldn’t be able to make PCs to run Windows 11. Despite the fact that the technology to make these chips appeared in 2014, there are still pitifully few fabs that have that capability. Not least, because it’s really difficult.
We rarely make these connections. In the figure above, I’ve tried to map the major applications to the process nodes to illustrate this vulnerability. It is a broad brush representation. However, the smaller the process node, i.e. the higher up it is in the table, the fewer the number of countries that have wafer fabs to make it. At the bleeding edge, that capability is very limited. Should anything go wrong, it would take a long time to fix it.
If you watch dystopian films like Fallout, you always see some of the factions with amazing futuristic tech, but you need to ask how they’d make it. It doesn’t need many parts of the manufacturing chain to disappear for the knowledge to be lost. Sondheim has a very pertinent song in his musical Assassins, which is “It takes a lot of men to make a gun”. Once you start taking things to pieces you realise how complex they are; how many layers of expertise goes into them, and how difficult it would be to revive that knowledge and evolve it once it’s lost. There is an amazing article written in 1958 titled “I, Pencil”, which everyone should read. It is told by a pencil, marvelling at the complexity of its creation and deducing that its existence could only be due to the existence of a God. That’s just for an everyday pencil. Could you make one from scratch? How many people could contemplate a 2nm chip fab?
Which brings us back to the complexity of what we rely on in our everyday life. Events around the world have shown us how fragile things are. It also illustrates that many of our national concerns are probably outdated and ill-placed. Ukraine has learnt that it needs to be self sufficient in servo motors and fibre optic cables, as that’s what keeps drones pouring out of its factories. Nobody predicted that. That conflict has also raised the stakes in cybersecurity, as have the recent high level hacks of retail companies. When you can’t get your mascarpone delivered, civilisation as we know it is surely at risk.
We’re also seeing evidence of the fragility of our infrastructure. Much of that is down to its increasing complexity. The day-long electrical power outage in Spain and Portugal last year shows that it’s not just malicious actors, it’s the pure complexity as infrastructure becomes ever more advanced. Distributed generation should help reliability, particularly with local generation, but managing that within national infrastructure is vastly different from the traditional grid. The same applies to almost all of our interconnections.
Which is why the BCS debate is timely. We need better preparedness for any future conflict, but equally, we need to understand what to do in the event of unexpected emergencies. Particularly within our infrastructure, there are many critical items which have replacement lead-times of months or years that will be very difficult to “muddle through”. We’re also losing much of the expertise which designed, installed and maintained these systems, as a generation of engineers involved in their design retire. It is timely to talk about these issues. As the Boy Scouts used to say, “be prepared”.