The ARM Story: The Little Chip That Won

A Silicon Annex extra to Cobler's Electricity Fundamentals course: the course's story, continued from electrons into computation.

Reach into your pocket. The phone there has been awake since breakfast, has run maps and messages and a camera all day, and the battery is still going. It is barely warm to the touch. That combination, a real computer that does real work without a fan and without draining flat by lunchtime, is so ordinary now that nobody stops to ask how it happened. There is a question buried in it worth digging up: who designed the processor doing all that, and why is it so miserly with power?

The answer is a genuinely strange one, and this arm processor history has more in common with the war of the currents than with a tidy engineering story. It was not Intel, the company that owned the desktop. It was a team in Cambridge small enough to sit around one table, working for a company most people outside Britain have never heard of, who built their own chip because they could not buy the one they wanted on terms they could stomach. Frugality was forced on them. It turned out to be the most valuable thing about the design.

Why did a British computer company design its own chip?

Because it needed a faster processor for its next machine and could not get a suitable one it liked.

The company was Acorn Computers of Cambridge, and in 1981 it had built the BBC Micro, the machine that put computing into British classrooms for the BBC's Computer Literacy Project. It was a hit. But Acorn's next design needed more power than the 16-bit processors then on the market could give at a price and with a support relationship the company found acceptable. So a very small team decided to do the thing that sounds insane: design their own CPU from scratch.

Two people carried the core of it. Sophie Wilson designed the instruction set and Steve Furber designed the chip's logic. Wilson famously wrote the working reference model of the whole processor in BBC BASIC, in roughly 808 lines. This was not a corporate war room with hundreds of engineers. It was a handful of people betting the company's future on an idea most of the industry had filed under "academic".

What is RISC, and why did it make ARM1 so small?

RISC means keeping the instruction set simple, and simple silicon is small silicon.

Processors of the era, Intel's x86 among them, were CISC designs: complex instruction sets, with many elaborate, variable-length instructions, some of them doing a great deal of work in a single step. RISC, short for Reduced Instruction Set Computer, took the opposite bet. Use a small set of simple, uniform instructions that the hardware can run fast and, crucially, with far fewer transistors. You trade a slightly longer list of instructions to do a given job for silicon that is smaller, cooler and cheaper. If you want the ground-up view of how transistors become a working CPU in the first place, that is the subject of from transistor to CPU.

The result was startling. The first Acorn RISC Machine, ARM1, ran as working silicon on 26 April 1985, fabricated by VLSI Technology, and by legend it worked on the first attempt. It had roughly 25,000 transistors. Intel's 80386, released the same year, had about 275,000. ARM had built a working processor with something like a tenth of the parts. Low power was designed in from the start, partly for a very unglamorous reason: Acorn could not afford expensive ceramic packaging, so the chip had to run cool enough for cheap plastic with no heatsink.

The chip that appeared to run on no power at all

The most-told ARM story is the one about the chip that ran with its power disconnected, and it is worth telling honestly.

When the first ARM1 silicon was dropped into a development board, it worked, but the engineers noticed something impossible: the ammeter on the main power rail read essentially zero. The board had a fault that left the main power pins unconnected. The chip was running anyway, fed by tiny leakage currents creeping in through its signal and I/O pins. The design drew so little power that stray microamps were enough to keep it alive.

Flag this one as an anecdote rather than a lab measurement. It is recounted first-hand by the designers themselves, Sophie Wilson especially, which is why it is well attested rather than folklore. But it has been retold many times, the exact mechanism and numbers drift between tellings, and the detailed technical teardowns of ARM1 do not repeat it. Take it as a true story the team tells about the day they realised what they had, not as a spec sheet. What is not in doubt is the underlying fact it dramatises: ARM1 drew on the order of a tenth of a watt where the 386 drew a couple of watts. Nobody in 1985 was thinking about phones. This was a cost decision that quietly became the whole asset.

How a chip for a failed Apple gadget created ARM Ltd

ARM became a company because Apple wanted the chip but would not build a product on a rival's intellectual property.

In the late 1980s Apple was developing the Newton, an early handheld with handwriting recognition, and it needed a processor exactly like ARM: powerful enough to be useful, frugal enough to run off batteries. Apple did not want to depend on Acorn, a competitor in the computer market. The solution was to spin the processor out into a neutral company that both could own. In November 1990 Advanced RISC Machines Ltd was founded as a joint venture between Acorn, Apple and VLSI Technology. Apple put in around three million US dollars, Acorn contributed about a dozen engineers, VLSI supplied the tools, and Robin Saxby became the first chief executive.

The Newton itself flopped. That barely mattered. The company created to feed it was about to make a decision that would matter more than any single product.

The business twist that actually won: selling blueprints, not chips

ARM decided not to manufacture anything. It would design processors and license the designs to anyone who wanted to build them.

This is the move that won the war, and it echoes an older one from the course's own history. When James Watt could not sell steam engines fast enough, he sold horsepower, a unit that let customers reason about what they were buying. ARM did something similar with silicon. Intel's model was to design chips and build them in its own multi-billion-dollar fabrication plants, then sell the finished parts. ARM owned no fabs at all. It sold the blueprint: the processor core and the instruction-set architecture, licensed to chipmakers who paid an upfront fee and then a royalty on every chip they shipped.

That framing decided a technology war the way the transformer decided the last one. Because ARM did not have to build factories, it could let hundreds of companies build ARM chips in parallel, each tuned for a different job, without ARM spending a penny on manufacturing. Apple, Qualcomm, Samsung and MediaTek could all ship ARM-based processors at once. Intel had to build every chip it sold. ARM only had to be right about the design, and let everyone else fight over the fabs.

So how does the arm processor history end?

With the underdog winning, by a distance that is hard to overstate.

When smartphones arrived, they needed exactly what ARM had accidentally spent decades perfecting: real computing performance inside a power budget a battery could feed. Every major mobile processor, Apple's A-series, Qualcomm's Snapdragon, Samsung's Exynos, MediaTek's chips, is built on ARM's architecture. About 99 percent of the world's smartphones run on ARM-based processors, a round industry figure rather than a single audited statistic, but the picture it paints is correct: virtually all of them. In 2020 the story came full circle when Apple began moving its Macs to Apple Silicon, its own ARM-based chips, and by 2023 had retired Intel across the entire Mac line. Cumulatively, hundreds of billions of ARM-based chips have now shipped.

Here is the quiet punchline. The most common computer on Earth, the one warming your pocket right now, was designed by a team who could not afford to do it the normal way. They made the chip simpler because they had to, and simple turned out to be the thing the world needed most. The winning technology, again, was the one whose numbers worked, not the one with the biggest company behind it.

Go deeper on video

Reading explains; watching sometimes lands the picture. Full credit to the creators:

This is a Silicon Annex extra to Cobler's Electricity Fundamentals course. It pairs with from transistor to CPU, which builds the processor up from switches, and with the war of the currents, the other story in the course where a business model, not just the physics, decided which technology won.

The lesson travels well beyond silicon: the design that wins is the one whose numbers hold up under real constraints. That is the same discipline behind knowing where every kilowatt-hour in your facility actually goes. See how CobiNeural monitors your building's energy in real time.