In the 1950s I spent a significant chunk of my pocket money buying a transistor. It was a small metal cylinder (about 5mm in diameter and 7mm deep) with three wires protruding from its base. I needed it for a little radio I was building, and buying it was a big deal for a lad living in rural Ireland. My baffled parents couldn’t understand why this gizmo their son was holding between finger and thumb could be interesting; and, to be honest, you couldn’t blame them.
Now spool forward six decades. The A13 processor that powers the iPhone that I used to find a photograph of that first transistor has 8.5 billion of them etched on to a sliver of silicon no bigger than a fingernail – a “chip”. The next generation of these chips will have transistors almost as small as the diameter of a human chromosome.
This ability to cram more and more transistors into a finite space is what gave us Moore’s law – the observation that the number of transistors in an integrated circuit will double every two years or so – and with it the fact that computer power has been doubling every two years for as long as most of us can remember. The story of how this happened is a riveting tale of engineering and manufacturing brilliance and is brilliantly told by Chris Miller in his bestselling book Chip War, which should be required reading for all Tory ministers who fantasise about making “Global Britain” a tech superpower.
But with that long run of technological progress came complacency and hubris. We got to the point of thinking that if all that was needed to solve a pressing problem was more computing power, then we could consider it solved; not today, perhaps, but certainly tomorrow.
There are at least three things wrong with this. The first is that many of humankind’s most pressing problems cannot be solved by computing. This is news to Silicon Valley, but it happens to be true. The second is that Moore’s law will soon run up against the laws of physics. We’re getting to the point where the critical part of a transistor – the “gate” through which current flows – is approaching 2 nanometers (nm – a billionth of a metre) and a silicon atom is 0.2nm in diameter, which means that the gate is only about 10 silicon atoms wide, at which point all kinds of awkward interactions between atoms start to occur and manufacturing becomes really problematic.
The third problem is that there are two critical bottlenecks in the supply chain for the highest-end chips on which the tech industry is betting its future – including the Nvidia chips that are powering most machine-learning (or AI) systems such as ChatGPT.
Bottleneck A has come about because there is only one company in the world – ASML in the Netherlands – that can make the extreme ultraviolet lithography machines capable of “printing” 2nm transistors on to silicon.
These machines are, as MIT Technology Review has explained, quite something: “The size of a small bus and filled with 100,000 tiny, coordinated mechanisms, including a system that generates a specific wavelength of high-energy ultraviolet light by blasting molten drops of tin with a laser 50,000 times a second. It takes four 747s to ship one to a customer.” And the top-end machines cost $350m a pop.
Bottleneck B exists because only a handful of companies have the capacity to manufacture chips at the 2nm level, and the leading one of these by far – TSMC – is based in – guess where? – Taiwan.
Put these two bottlenecks together and you have what can only be described as panic in the leadership of western economies. ASML’s monopoly is perhaps not such a strategic concern: it is, after all, a European company. Its biggest problem is that manufacturing its machines is a delicate, complicated, time-consuming business. And according to the Financial Times it currently has a $40bn backlog of orders.
But the uniqueness of TSMC and its location on an island that the Chinese regime regards as part of the mainland is giving rise to strategic panic. Both the US and the EU are racing to try to ensure that they have 2nm chip-fabrication capability within their respective jurisdictions. The problem is that one cannot conjure up such capacity just by throwing money at it. TSMC itself has built a fabrication plant in Arizona. But in his speech marking the ceremonial opening of the facility last December, Morris Chang, the firm’s founder, said that it could not find enough qualified American workers to run it. It was sending every new American recruit to Taiwan for 18 months of training and was even importing engineers from Taiwan to make the Arizona plant operational. Hopefully it will all be up and running before Xi decides to “do a Putin” and we will no longer to be able to have chips with everything.
What I’ve been reading
Received knowledge
Terrific Substack post by Charles Arthur on research that challenges some myths about online filter bubbles.
Computing a cure
Fascinating London Review of Books blogpost by Liam Shaw on drug discovery by AI.
Fringe benefits
There’s an enjoyable dispatch by Helen Lewis in the Atlantic about a recent conference of rightwing cranks.
