Chips with everything

Bastianpillai: What are the implications of the artificial intelligence (AI) boom for the semiconductor industry?

Miller: The immediate implication is higher demand for AI-related chips. The US semiconductor company Nvidia has been the most obvious winner, but many other companies have seen demand surge. Yet this first wave of spending on AI chips has been focused primarily on the data centre. In the longer term, many AI systems will be processed largely on the edge of networks, in devices like smartphones, cars and wearables. We’re just beginning to see a new set of chips for these devices that are optimised for AI.

Bastianpillai: Which innovation in the world of semiconductors excites you most at the moment?

Miller: The biggest immediate focus of innovation is the production of better chips for AI. Improved chips have been the primary driver of better AI systems over the past decade. Companies that manufacture graphics processing unit (GPU) chips, like Nvidia, and high-bandwidth memory chips, like SK hynix, can today produce chips that are multiple times more capable than a decade ago. This is why big AI systems are trained on multiple times as much data – and thus are far more effective than older generations of AI.

Bastianpillai: What are the most interesting use cases beyond AI?

Miller: I think AI will be the dominant theme for some time. In the longer term, I’m excited about the intersection between chips, biotech and medical devices. There are already an interesting set of start-ups building chips that can sense certain biological markers, for example.

Bastianpillai: How will the competition between the US and China affect the industry? Can it be a driver for innovation?

Miller: The downside of US-China competition is that supply chains are being duplicated, making them less efficient and adding to costs. The upside is that the US and other countries are pouring more money into research and development, which – if executed well – should accelerate innovation.

Bastianpillai: The industry is not able for its negative environmental impact. How can it best address this?

Miller: Making chips requires many toxic chemicals, but the industry has made substantial progress in mitigating their impact. The next major effort is to replace a category of chemicals called PFAS that take a particularly long time to degrade. This will take major innovations in chemistry.

Making chips

Semiconductor manufacturing process

Bastianpillai: Despite this, do semiconductors also have a part to play in the green transition?

Miller: Yes, they are critically important, though in a complex way. The short-term impact of AI will delay the green transition because new AI data centres will be powered with more fossil fuels. But semiconductors are the key drivers of electrification. New electric vehicles, for example, have over a thousand dollars’ worth of semiconductors inside, and many of them play a key role in managing the battery and power distribution.

Bastianpillai: Some 60 years ago, Moore’s Law observed that the number of transistors in an integrated circuit doubles every two years. That, in turn, means that chips can become ever smaller while doing the same (or even better) job. Most recently, we have moved from 7-nanometre (nm) chips to 5-nanometre chips. What does that mean – are 5nm chips actually 5nm?

Miller: The nanometre numbers are mostly marketing terms. It used to be possible to compare chips based on a single measurement. Now chips are 3D, so there are multiple dimensions on which one could measure them. Roughly speaking, between each generation (i.e. 7 nm to 5 nm) we get a tremendous increase in computing power.

Bastianpillai: Given the rising demand for semiconductors, are we going to see bigger capex spend to ramp up capacity?

Miller: We’ve already seen a major uptick in capital expenditure. I think chip manufacturers are cautious about expanding much more beyond their current commitments. They know the industry has historically been cyclical and want more visibility into what AI data centre demand will look like towards the middle of the 2020s. The key question is whether big tech firms will keep spending tens of billions annually on AI infrastructure, or whether they will pause and wait for more profitable applications to emerge.