Taiwan Semiconductor Manufacturing Company (TSMC) produces processor chips Apple and other big-name companies use. Photo / Getty
Taiwan Semiconductor Manufacturing Company (TSMC) produces processor chips Apple and other big-name companies use. Photo / Getty
OPINION
The chip war - in which the United States and its Western allies are trying to limit China’s access to the latest and most powerful integrated circuit technology - not only rages on, but is heating up thanks to the artificial intelligence boom.
To recap, after decades of outsourcingalmost all aspects of electronics manufacturing to low-cost China, the Western world has decided it wasn’t such a great idea after all.
Our economies can’t function without miniaturised chips running everything from transport to communications and weaponry: control chip development and production, and you control the world.
It has led to some bizarre situations, like US graphics card maker Nvidia being forced to build a slower version of its flagship Geforce RTX 4090 model, featuring a neutered chip for the Chinese market.
Apparently, the RTX 4090 card sold outside China, which costs the same, was around 10 per cent more powerful than export regulations allowed. Since graphics cards power AI applications as well as video games, the RTX 4090 couldn’t be sold as-is in China, and had to be detuned.
Western export controls notwithstanding, the Chinese aren’t giving up. Take Huawei, which lost its big telco network equipment markets in Western countries (including New Zealand) and then looked like it would have to exit the smartphone business as its chip supply dried up and Google couldn’t licence Android to the Chinese company.
For a company in a fast-moving and highly competitive technology sector, that’s like having both hands tied behind your back and getting shot in both legs. Huawei’s still hobbling along though, announcing new chips for mobile devices made with the finer processing technology required to bump up performance while lowering power consumption.
It’s not just mobile processors either, but chips for servers and personal computers. You’ve probably heard of American semiconductor companies Intel and AMD, but how about Zhaoxin and Loongson in China?
Intel workers installing an automated material-handling system track. Photo / Philip Cheung, the New York Times
The two latter Chinese companies design processors built around the Intel-AMD x86 hardware architecture, which is what server operating systems and Microsoft Windows are coded to run on. Loongsoon and Zhaoxin have come up with surprisingly advanced designs, and not only that, they’ve been turned into working processors that can go into computers.
Already, Zhaoxin has submitted software patches for the very popular free and open-source Linux operating system kernel, a piece of code that, among other things, controls computer hardware, physical or virtual - so, in theory, the processors could run a large amount of existing applications.
Nevertheless, despite China’s advances, like the state-owned Semiconductor Manufacturing International Corporation (SMIC) announcing it can make seven and even five-nanometre process nodes, it’s worth noting this represents technology from 2018 to 2020.
That matters more than you think. Finer processing nodes allow for even more minuscule transistors to be packed into processors, like the 3nm Apple M3 Max part, which has an insane 92 billion of them. In very simplified terms, more transistors in a processor design translates into being able to do more tasks, and faster.
This year, 2nm process node technology is supposed to be announced, and 1nm is on the horizon.
As a related aside, the “process node” metric used by the semiconductor industry is misleading. It no longer describes the actual physical size of important chip features, and doesn’t take into account design changes like stacked transistor layers for additional density, but it’s still used as a shorthand to mark incremental technology advances.
We are nearing what’s possible with current semiconductor technology: shrinking transistors and other processor features to the atomic scale is both extremely difficult and certainly not cheap as chips. In fact, it’s getting more expensive, with current semiconductor foundries requiring initial capital outlay in the US$20 billion range.
Newer “fabs” for smaller process nodes are budgeted to cost much more, and that’s before factoring in colossal energy and water use and the severe overall environmental impact of chip-making.
However, billions of dollars is chump change for world chip-making dominance, with the US and EU handing out subsidies to entice manufacturers to move production away from the Far East.
China will try to rise to that challenge, but even if it has the money to do so, it faces an uphill battle trying to catch up technologically while being squeezed by export controls. This looks set to be a fast-moving Big Nation game this year, one that could upset the world power balance.
