Chipageddon: the coming sequel

The global semiconductor shortage is wreaking havoc on industries from handheld electronics to car making. There is a web of complex economic, social and environmental reasons behind it – depending on where you read your news, it could be attributed to a combination of the pandemic, increased demand for supply, just-in-time manufacturing practices, Sino-US tensions and a drought in Taiwan.

Whatever weight you attribute to the various causes, the shortage – delightfully dubbed “chipageddon” – is unlikely to end soon. And serious as it is, it lays bare another problem in computing: reliance on a technology that – at some point – will become impossible to improve.

“Semiconductor” has a broader meaning in physics than in IT manufacture, but in both cases, the definition relies on the movement of electrons.

“There’s two kinds of materials in terms of electronic properties,” explains Michael Fuhrer, a professor of physics at Monash University and director of the ARC Centre of Excellence for Future Low-Energy Electronics Technologies (FLEET).

“There are metals, which have electrons that are free to move around at essentially negligible energy. It doesn’t take much energy to move the electrons out of metal and that’s why they carry current and they conduct electricity.” (Some non-metallic molecules – such as graphite – also fall into this category.)

“And then insulators, which have what’s called a band gap. So there’s a finite amount of energy to get the electrons to move. And – this is a non-technical explanation – semiconductors are insulators with a fairly small band gap. So they’re things that wouldn’t conduct, but if you give them a moderate amount of energy, they can conduct.”

It is this unusual property that makes semiconductors so useful in IT: they can conduct electrons sometimes, making them useful switches.

“Those switches are called transistors, and transistors are the basic units of making computer chips,” explains Fuhrer.

Transistors could theoretically be made from any semiconductor. Indeed, “the first transistor wasn’t made of silicon; it was made of germanium,” says Fuhrer. But there are a few advantages to silicon: it’s abundant, for a start, and the way it oxidises makes it clean and easy to work with.

In fact, there are some specific advantages to silicon that means that computer chips are almost always made with the same process.

“What happened is, sometime in the mid-to-late 70s, a process emerged for making computer chips out of silicon,” says Fuhrer. “It’s called CMOS: complementary metal oxide semiconductor.

“It wasn’t actually the best [process] at the time, but people could see that if we made chips this way, then next year we could make them a little bit smaller, and make the voltages a bit lower, and they’d still work better.

“And there was a path to making the transistors better and better every year. Whereas other technologies, you just had to redesign everything every year.”

The challenges posed by CMOS were engineering challenges, rather than challenges of basic physics. They were predictable, and the solutions could be developed as long as the right amount of time and money was invested. This is why, over the past 50 years, our computers have reliably become smaller and faster.

“For various reasons, many of them economic, we set off on this path of making computer chips,” says Fuhrer.

The problem with this is that manufacturers are now boxed into a very specific way of making a very basic computer component. The CMOS process has its own gravity.

“In order to push these technologies all the way to the end, the equipment that goes into those plants is just enormously expensive,” says Fuhrer. “And it’s not used for anything else. So nowadays, it costs $10 billion to build one factory that builds computer chips.”

Which is why the current shortage is unlikely to end in any hurry: building more factories is effectively out of the question, in the short term. The other problem with CMOS manufacture, according to Fuhrer, is that at some point, these chips will become as efficient as they can possibly be.

“When we get to the end of the road and we can’t make computer chips that are faster and more energy efficient, for a while, information technology is going to stagnate a bit.”

While this is unrelated to the current shortage, it’s a problem recognised by physicists and IT specialists.

“Now we’ve reached the point where we think we need kind of a revolutionary change, a completely different way of making transistors, maybe not with silicon, maybe not the field of transistors that work with charges and electronic current. It may be something totally different,” Fuhrer says.

And what might that thing be? Fuhrer says it’s impossible to tell, from this distance – it depends what innovations are made in physics: “All of that is really in the basic research realm right now.

“There’s not one clear winner. In some sense, it’s like there’s 12 different ideas for things that are all impossible, and people are just trying to figure out which one’s the least impossible.”

This is a problem that will take decades, not years, to solve.

Fuhrer’s own research at FLEET looks into some of these competing ideas. One project he’s investigating is focused on substances that don’t fit into the insulator/conductor categories.

“About 15 years ago, theoretical physicists figured out that there’s actually a third kind of material, that insulators actually come in two kinds,” Fuhrer says. “There are these things called topological insulators that are insulating the interior of the material, but they have these conducting states that are on their surfaces or on their edges if they’re two-dimensional.” These topological insulators can conduct well at room temperature.

Could they be the next revolution in computer chips? Come back in 10–20 years to find out.

“We hope to come to the rescue with something else,” Fuhrer says, “but it’s going to take quite a bit of time.”

Read more

• What is ransomware and how is it dealt with?
• First entanglement-based quantum network
• Cosmos Briefing: Supercomputing and Big Data