ON THE one hand was a perfect storm of problems hampering supplies: a global pandemic, a trade war, a blaze at a manufacturing plant and atrocious weather including drought and snowstorms. On the other was the unprecedented demand for one of the world’s most sought-after products – a market worth $40 billion in January this year alone.
Even so, the news earlier this year that there was a global shortage of computer chips, pushing up the price of everything from laptops to fridges, took a lot of people by surprise. The crisis has seen nations and companies around the world scrambling to establish more chip-building capacity. It has also shone a light on an industry whose products have become ubiquitous and essential, leaving many to conclude that it is due a fundamental overhaul. For critics, the stranglehold on the market of just a few firms acts as a brake on innovation and increases the industry’s vulnerability to disruption.
Advertisement
Some big names are now backing an alternative model – taking the collaborative, open-source principles that have changed the way software is written and applying them to chips. For its proponents, it is only a matter of time before this becomes the new standard. But when the chips are down, does it have what it takes – and what does it mean for us?
Computer chips are everywhere – not just in our laptops, desktops and smartphones. They are in the myriad mysterious servers that run our webmail, online bank and other digital services we use daily. They are in many microwaves, televisions, washing machines and watches. The average new car today contains hundreds of them.
Zoom in and one 60-year-old invention still lies at the heart of all these chips: the metal-oxide-semiconductor field-effect transistor, or MOSFET. The brainchild of Bell Labs researchers Mohamed Atalla and Dawon Kahng in 1959, it acts as a tiny switch that, wired up in specific ways, can perform basic logical operations like addition, subtraction or multiplication.
The very first processor chips were custom-designed, with transistors wired up with specific computational tasks in mind. With computers becoming more complex and more commonplace, however, in 1964 IBM announced the IBM System/360, a mainframe computer that could run a range of different programs. Its processor came with a basic “instruction set” of operations – adding or comparing two numbers, say, or jumping to a certain spot in memory – from which more complex operations could be created.
“We have begun to reach fundamental limits on the size of transistors”
The process of programming software that used these chips rapidly came to be assisted by tools to analyse, debug and verify code that was specific to an instruction set and the chip “architecture” underlying it (see “How chips work“). That meant programmers no longer had to worry about how chips worked at the level of transistors. But it also meant handing a lot of power to a few established chip makers who provided well-understood, stable instruction sets and architectures.
That situation persists to this day. “The fact that [a chip architecture] has been there for that long means there’s a whole built-in infrastructure,” says Siva Sivaram at Western Digital, a company that creates hard discs storing 40 per cent of the world’s data. “Lots of software is written on it, and operating systems have been sitting on top of it. Everybody knows what it is.”
The main processors in your desktop or laptop computer, for instance, probably run on the x86 instruction set on chips designed by Intel or another US company, AMD, which initially reverse-engineered Intel’s chips to create compatible designs. Data servers might use the IBM z/Architecture, which traces its lineage back to the System/360. Mobile phones often use ARM architecture, owned by a UK-based company that evolved from Acorn Computers. ARM has .
All these companies keep strict tabs on their intellectual property. Intel, for example, has the sole say-so over the design and manufacture of its chips. “You want an Intel processor… they’re in control,” says Cecil Macgregor at data storage company Seagate.
Designing and making new chips is, after all, a complex and specialised business. That has become even more true as we have begun to reach fundamental limits on the tiny size of transistors – the current standard is just 5 nanometres across – and so how many can be crammed on a single chip. A standard central processing unit, or CPU, now consists of multiple “cores”, individual chips-on-a-chip that can work in parallel. Algorithms such as the artificial intelligence that underlies our phone’s virtual assistant, say, or those that do the high-speed 3D physics calculations that render computer game graphics realistic, increasingly require chips specially optimised to run the relevant instructions more quickly.
“As chips get more complex, security is becoming a hot potato”
The proprietary model gives stability and certainty to companies building hardware around a given chip, but it can also hamper their ability to innovate. “How much memory can you attach to a single core? Intel decides,” says Sivaram. “If you want to add more memory, they say ‘buy another core’. It’s like ‘do you want a new bathroom? Buy a new house’.” He says that his company buys tens of millions of chips a year, but still can’t get exactly what it needs.
ARM works slightly differently: it designs its cores, but licenses various basic “flavours” of them to end users to manufacture as they see fit. Seagate transitioned to this model just over 20 years ago. This gave more flexibility, says Macgregor, but did little for customisation: the company still had to pick from a handful of pre-determined options. “They’re also in it to make money, right? So they have to decide which flavour makes sense for the world,” he says. “They want to make sure a lot of people are going to buy it.” A very big company like Apple can get around this: it is currently moving from Intel to ARM chips, but has paid extra and used its enormous sway to create its own processor, called M1.
ARM’s proposed acquisition by Nvidia, a US company that designs graphics processing units (GPUs) and chips for smartphones, tablets and car navigation and entertainment systems as well as making games consoles itself, has also caused some unease. “If you were a competitor to Nvdia, you would be thinking hard because your core comes from them,” says Macgregor. When contacted by 鶹ý, an Nvidia spokesperson said all customers will be able to continue licensing ARM intellectual property, just as they can today. The UK government is seemingly unconvinced, with recent .
As chips get more complex, security and reliability are becoming hotter potatoes, too. Testing chips currently means either randomly probing one to find flaws, or trying to formally test every possible input and output. The first approach can easily miss problems, and the second quickly becomes unfeasible for all but the simplest designs. Intel recalled around a million Pentium P5 processors in 1995 after a flaw was discovered that could cause certain calculations to be done improperly. In 2011, a flaw in its Sandy Bridge chips cost an estimated $1 billion to rectify. Neither of these came close to two flaws identified in 2018, Meltdown and Spectre, that affected a whole generation of chips from a range of makers. Even Apple’s brand new M1 chip has been shown to , thought to be largely harmless, that allows two programs that should be strictly separated to share data.
Add to this mix the geopolitical tensions between the two biggest national players in the chip business, the US and China (see “International chip wars“), and you can begin to see why some manufacturers are itching for a different way of working. Many now think they see it.
Called RISC, for “reduced instruction set computer”, its origins lie in an academic project that started in the 1980s at the University of California, Berkeley. The original idea was that chips could be made leaner and simpler by having an instruction set with just a few instructions, rather than a vast library to cover every specific, esoteric task. A perennial roadblock to progress was the difficulty of getting to grips with companies’ secret, proprietary designs, so in 2010 RISC researchers set about creating their own chip and instruction set as an educational tool. They called it RISC-V, as it was the fifth RISC research project that co-creator David Patterson had worked on. As students who had learned their stuff on RISC-V chips moved on, the design started to crop up in real-world applications.
In 2015, the specification was spun out with a non-profit company, , overseeing its development. Its big plus is that it is open-source, applying a model that is increasingly a standard in software development. The Linux open-source operating system, notably, has become the basis of the Google-sponsored Android, the most popular smartphone operating system worldwide. It also powers .
Open season
The open-source approach allows whole communities of developers to constantly optimise the RISC-V standard, ensuring it works as it should do without any nasty security flaws or backdoors. “The best thing I can do is to be open-source,” says Sivaram. “Let a million people out there try to find a bug in my code.” The rationale for all design decisions is publicly documented, and any individuals or company can participate in development – and repurpose it for their own products for free. “You could make all sorts of flavours, you can add little tweaks to it,” says Macgregor.
The attraction for device manufacturers is clear: creating cheaper products with lower licensing overheads, using chips designed specifically for the task at hand, potentially meaning better performance and battery life. Some companies are already biting. Seagate is one, although it stresses that it hasn’t officially announced any product containing RISC-V chips. Macgregor won’t say what the company is using the standard for, only that customising chips offers advantages over competitors.
Others are less coy. Western Digital has now “fully subscribed” to RISC-V, says Sivaram. A RISC-V chip designed to have a particular functionality will be slightly smaller and give better performance for less power than an off-the-peg chip because it “doesn’t have legacy baggage to drag along with it”, he says. Back in 2016, Nvidia announced it would use RISC-V in its GPUs. When contacted for this article, it confirmed that it now does this, despite its plan to take over ARM.
Sivaram thinks these developments are only the beginning. “I’m not saying it is going to be instantaneous, but it’s only a matter of time before semiconductor cores are going to be open-sourced everywhere. When you buy your Ford or Toyota car, do you know who makes the engine?”
Patterson agrees. “People have this religious fervour about openness. They like this idea and want to make it happen,” he says. “I think RISC-V is inevitable and it’s going to be a lingua franca that’s going to get used from tiny computers to large computers.”
“Because it’s an open-source licence, no one company, country or other entity controls it. Once it’s given out to the community, you can’t pull it back,” says Calista Redmond at RISC-V International foundation. The foundation announced in 2019 that it was moving its headquarters to Switzerland from the US to assuage fears from some of its members, unnamed but thought to be from China, that any one country could wrest control.
Certainly many Chinese companies are in the foundation’s list of partners. After US sanctions by the Trump administration, the Chinese retail giant Alibaba . India, too, has officially endorsed the , citing a desire to know what is inside chips used to build critical systems. The EU has chosen . Semico Research has estimated that there will be more than 60 billion RISC-V cores shipping in 2025. That compares with 6.7 billion chips using ARM technology that were built in the final quarter of 2020.
Redmond says “legacy” companies like AMD, Intel and ARM should be worried. “We’re accelerating the path that took them decades to build,” she says. “A lot of big folks are in the tent, and they’re deepening their investment.”
ARM’s chief architect, Richard Grisenthwaite, is less sure, despite its putative new owner’s involvement. Developing an instruction set is the relatively easy bit, he says; what is hard is creating the ecosystem of supporting tools to help develop chips on that platform. This has to be done from scratch for a new instruction set. “An instruction set lives or dies on the ecosystem, and to make that from scratch, rather than going for ARM, seems tricky,” says Grisenthwaite. “It took us 30 years to build momentum. Anyone who wants to go through that same experience that ARM has been through needs to put in the same sort of level of work.”
Another problem is keeping focus. The established players can control their instruction sets and approve changes, but an open-source design can be adapted and fork down different roads. Any fundamental divide that opens up could reduce its chances of success – or just create a new monopoly player. Android might be open-source, for example, but its main effect has been cementing Google’s commercial position.
“Observers say companies like AMD, Intel and ARM should be worried”
Whether manufacturers will ever be able to entirely wean themselves off legacy technology is less certain, too. For many, there is a need to retain backward-compatibility with older products and that may mean maintaining ARM or Intel chips in their devices. Modern devices already contain dozens of chips using different architectures and that is unlikely to change. And whoever designs chips, licenses them or orders them, the chip makers will still have to churn out the actual silicon. For the next few years at least – until new factories can be brought online – that will be the bottleneck.
In that sense, the RISC-V revolution, if it comes, is likely to go unnoticed by the end user. But if it allows hardware designers to make devices that are a little cheaper, a bit more specialised, a fraction more efficient, then we will all benefit. The chips will go on humming away, doing what we have become accustomed to them doing – silently enabling our digital lives, but perhaps just a little more openly.
INTERNATIONAL CHIP WARS
The position of chips at the centre of all computing hardware makes them an ideal place to insert “backdoors” or security flaws that might potentially allow unfriendly powers to spy on what individual computers are doing – and beyond, in a networked world. “For all we know, anything built out of China has 1-800-CALL-CHINA [written into it],” says Cecil Macgregor at data storage company Seagate.
That is a conspiracy theory, he says – but equally we just don’t know. With threats of a trade war between the US and China bubbling away, companies are already making choices about which chip architecture to use based on the nationality of the chip company. “If I were a superpower, why would I let some other superpower control me?” says Macgregor.
China is working on building up its own chip design capabilities, but for now its burgeoning technology sector still relies on licensing deals with US chip companies. US chip firm Xilinx has already had to suspend some sales to Chinese consumer electronics company Huawei .
It cuts both ways. In 1990, the US made over a third of the world’s chips, but that is now down to 12 per cent, with most chip manufacture outsourced to plants in Asia. The US is now getting the Taiwan Semiconductor Manufacturing Company to build a $12 billion chip factory on its own shores, with US president Joe Biden expressing his desire for boosting domestic manufacturing capacity.
HOW CHIPS WORK
A laptop or smartphone chip isn’t capable of directly understanding what all your various keyboard strokes, mouse clicks or finger swipes in an app mean. It is up to the high-level software and the operating system to translate those inputs down through layers of abstraction into a language the chip understands. One mouse movement may spark thousands of individual calculations to be run by several different chips inside your device.
A chip’s instruction set is the atomic level of all this. It is a framework of all the commands that a processor can carry out. One instruction in the set might be to bring a bit of data out of memory, another to store it again; one will add two values, another will subtract them. What a processor ultimately processes is a string of these instructions.
Today’s computers are built of layers of complexity. In the average laptop or smartphone, for example, millions or billions of instructions are executed by a variety of chips to run the operating system and software that you use.
