Engineering Philosophy: Sophie Wilson

Key Takeaways

• She designed the instruction set in nearly every phone on Earth. Sophie Wilson specified the instruction set of the original ARM – the Acorn RISC Machine – at Acorn Computers, beginning in October 1983, with Steve Furber building the hardware. The architecture that grew out of that work has shipped in over 230 billion chips, more ARM cores than there are people many times over, and powers the overwhelming majority of the world’s smartphones.¹²
• Ruthless simplicity made low power almost an accident. The first ARM, ARM1, used fewer than 25,000 transistors – a fraction of its contemporaries – yet outran machines many times its size. Because so few transistors switched, it drew about a tenth of a watt, roughly a twentieth of what an Intel 386 needed. The minimal design was chosen for speed and for the constraints of a tiny team; the low power that later conquered mobile fell out of it almost by accident.³⁴
• A two-person project that beat the giants. Wilson and Furber designed a competitive 32-bit processor with a team you could count on one hand, after watching a single engineer at the 6502’s maker work on the next CPU and concluding you did not need an army to build one. Wilson modeled the entire instruction set in 808 lines of BBC BASIC before any silicon existed.⁴⁵
• She built Britain’s computing literacy first. Before ARM, Wilson co-designed the BBC Micro – prototyped in under a week – and wrote BBC BASIC, the language that taught a generation of British schoolchildren to program. (Her early Acorn and BBC work was originally credited under the name Roger Wilson; she transitioned in 1994.)¹⁶

The Principle

“Not knowing something is impossible has interesting effects on your work.” – Sophie Wilson⁶

Most processor design is addition. You start with a working machine and you bolt on features – new instructions, new modes, new special cases – because each one helps some program somewhere, and silicon keeps getting cheaper, so why not. The result is an instruction set that accretes complexity for decades, where every addition makes the decoder slower and the hardware larger and the next addition harder. Wilson’s instinct, sharpened by the Berkeley RISC research and by brutal practical limits, ran the opposite direction. You start with what the machine must do and you remove everything that is not load-bearing, until what remains is small enough to be fast, simple enough to be correct, and cheap enough that a tiny team can actually build it.⁴

The constraints were not theoretical. Acorn was a small British company, and the ARM team was effectively two people. They could not afford the transistor budgets or the verification armies that Intel and Motorola threw at the problem. So simplicity was not a stylistic preference – it was the only way the chip could exist at all. Wilson designed an instruction set where almost every operation completed in a single cycle, where the decoder stayed trivial, and where the whole thing fit in fewer than 25,000 transistors at a time when a comparable processor used five to ten times as many.³ Scarcity forced elegance.

And then the elegance paid a dividend nobody designed for. A chip with that few transistors switching draws almost no current – the original ARM ran on roughly a tenth of a watt while an Intel 386 of the era needed close to two.³ At the time, in a desktop coprocessor, low power was a curiosity. A decade later, when the question became how to put a real computer in a battery-powered phone, that curiosity turned out to be the single most valuable property a processor could have. Strip the design to its essence, and the efficiency you get for free can become the thing that wins.

Context

Sophie Wilson was born in June 1957 in Leeds, attended Harrogate Grammar School, and went up to Selwyn College, Cambridge, reading mathematics for the first two years before switching to computer science.¹ Before university she had already designed electronic systems for industry, and during a 1977 summer vacation she built a microprocessor-based controller – a cow feeder, of all things – around the MOS Technology 6502. That work brought her to Acorn Computers in Cambridge, the company at the center of the British microcomputer boom.¹

At Acorn she made her first mark on the BBC Micro. In 1981 the BBC, running a national computer-literacy project, needed a machine; Acorn won the contract, and Wilson was instrumental in the design – the prototype, by her account, was built in under a week.¹⁶ She then wrote BBC BASIC, the interpreter built into the machine’s ROM, complete with hand-written floating-point routines. Over a million BBC Micros sold, most into UK schools, and BBC BASIC became the first programming language a generation of British engineers ever touched.¹⁶ (Wilson’s Acorn and BBC work was originally published under the name Roger Wilson; she transitioned in 1994.)¹

The pivotal chapter began in October 1983, when Wilson started designing the instruction set for a processor of Acorn’s own: the Acorn RISC Machine, or ARM. Steve Furber led the hardware architecture – a three-stage pipeline and a barrel shifter – while Wilson defined what the machine could do and proved it would work by simulating the entire instruction set in 808 lines of BBC BASIC.⁵ The first silicon, ARM1, arrived from VLSI Technology on 26 April 1985 and worked first time – an almost unheard-of result for a brand-new processor.¹³ The most famous detail came moments later: the chip drew so little power that when it was first plugged into the development system, it sprang to life by pulling current through its I/O lines before its own power supply was properly connected. The team had built something so frugal it ran on what amounted to leakage.⁵ Wilson stayed close to ARM as a consultant after it spun out, and from 2001 became chief architect of the FirePath processor at Element 14 – later acquired by Broadcom – a signal-processing core used in ADSL broadband modems.¹ She was elected a Fellow of the Royal Society in 2013, made a CBE in 2019, and in 2022 shared the Charles Stark Draper Prize with David Patterson, John Hennessy, and Steve Furber.¹

The Work

The ARM instruction set: doing more with simpler, smaller hardware

Start with the instruction set, because that is where Wilson’s principle becomes silicon. A processor’s instruction set is its contract with software – the complete list of operations it knows how to perform. The temptation, followed by the dominant designs of the early 1980s, is to make that list rich: complex instructions that do a lot of work each, addressing modes for every occasion, microcode to interpret it all. The Reduced Instruction Set philosophy Wilson adopted made the opposite bet – that a small, regular set of simple instructions, each completing in a single clock cycle, would run faster overall and demand far less hardware to decode.⁴

Wilson pushed one idea past the textbook RISC consensus, and it is the cleverest thing in the architecture: conditional execution on almost every instruction. On a conventional processor, choosing between two actions means a comparison followed by a branch – a jump to one place or another in the code. Branches are expensive, because a pipelined processor has already begun fetching and decoding the instructions after the branch; when the branch is taken, that speculative work is thrown away and the pipeline stalls while it refills. Wilson’s ARM lets nearly every instruction carry its own four-bit condition code, so an instruction simply does nothing if its condition is false. Short stretches of conditional logic that would otherwise need branches become straight-line, branch-free code. The pipeline never stalls, and the hardware to support it is almost free – four bits per instruction and a little comparison logic.²⁴

That is the whole philosophy in one feature. Conditional execution removes branches – a bottleneck – not by adding a clever branch predictor with its own pile of transistors, but by making the existing instructions slightly more expressive at almost no cost. More capability from simpler, smaller hardware. Wilson validated the entire design this way before committing it to silicon, writing a simulator in 808 lines of BBC BASIC and checking that real programs compiled and ran efficiently on the imagined machine.⁴⁵ The discipline shows in the result: fewer than 25,000 transistors, and silicon that worked the first time it came back from the fab.³

BBC BASIC and the BBC Micro

Before any of the ARM work, Wilson built the machine and the language that made British computing literate. When the BBC launched its Computer Literacy Project in the early 1980s, it needed a reference microcomputer, and Acorn’s bid – a prototype Wilson helped pull together in under a week – became the BBC Micro, released in 1981.¹⁶ It was rugged, expandable, and tuned for teaching, and it sold over a million units, the bulk of them into British schools.⁶

The soul of the machine was the software, and that was Wilson’s. She wrote BBC BASIC, the interpreter burned into the machine’s ROM, including the floating-point arithmetic routines by hand. BBC BASIC was unusually capable for a home-computer language – structured, fast, with an inline assembler that let curious students drop straight from BASIC into 6502 machine code. For a generation of British engineers it was the first language they ever wrote, and the bridge from typing commands to understanding the processor underneath.¹⁶ That same instinct – model the machine in software, understand it from the instruction up – is exactly what Wilson would later use to design ARM: she prototyped a processor’s instruction set in the very language she had written for the previous one.⁵

How low-power-by-simplicity conquered mobile

Here is where the accident becomes history. ARM1 was conceived as a fast coprocessor for a desktop, and its tiny transistor count was chosen for speed and for the limits of a two-person team. The low power – about a tenth of a watt, against nearly two watts for an Intel 386 of the era – was a side effect of having so little silicon to switch.³ On a mains-powered desktop in 1985, that frugality was a footnote.⁵

Then the world changed shape. By the 1990s the interesting question was no longer “how fast on a desk” but “how much computing can you fit in a battery.” And on that question, the property that mattered most was performance per watt – and ARM, almost alone, had been frugal from birth. Acorn spun the design out into a separate company, ARM Ltd, in 1990, with a business model built on licensing the architecture rather than selling chips.¹ Licensees could drop a small, cool, efficient core into a phone, a music player, a router – anything where battery life or heat was the wall. The architecture Wilson had stripped to fewer than 25,000 transistors became the default processor of the mobile era, and over 230 billion ARM-based chips have now shipped.² The thing that won was not raw speed. It was the efficiency that ruthless simplicity had handed over for free, a decade before anyone knew they would need it.

FirePath, Broadcom, and a second act in signal processing

Wilson did not stop at ARM. After the architecture spun out and she stayed on as a consultant, she became chief architect of FirePath at Element 14 – a company that grew out of Acorn and was acquired by Broadcom in 2000. FirePath was a digital signal processor core for ADSL broadband, the technology that pushed high-speed internet over ordinary copper telephone lines.¹ DSP work is a different discipline from general-purpose CPU design: the workload is relentless streams of arithmetic, and the architecture must be shaped around throughput on that specific problem. That Wilson did consequential work on both a general-purpose instruction set and a specialized signal processor argues the underlying skill is not any one trick but the habit of asking what the workload truly requires and building no more than that.¹

The Method

Read across BBC BASIC, the ARM instruction set, and FirePath, and the same moves recur. Wilson’s method is less a slogan than a set of standing commitments.

Remove until only the load-bearing parts remain. The defining habit is subtraction. ARM was built by taking parts away – instructions, modes, microcode – until what was left was small enough to be fast and simple enough for a tiny team to get right the first time. The general lesson transfers far beyond silicon: the strongest design is usually the one with the least in it, and the discipline is to keep cutting until removing one more thing breaks it. It is minimum worthy product at the level of an instruction set – ship the smallest thing that genuinely does the job.⁴

Let scarcity force elegance instead of resenting it. Wilson did not have Intel’s transistor budget or verification army, and rather than treating that as a handicap she let it drive the architecture toward something cleaner than a well-funded team would likely have built. Constraints, taken seriously, are a design tool. The opposite instinct – Jim Keller’s “transistors are free” – wins when silicon is the abundant resource; Wilson’s wins when it is the scarce one, and knowing which regime you are in is the actual skill.³⁴

Prove it in software before you commit it to silicon. Wilson modeled the entire ARM instruction set in 808 lines of BBC BASIC and ran real programs against the simulation before a single transistor was laid down. The expensive, irreversible step came last; the cheap, fast iteration came first. It is the evidence gate applied to hardware – you do not believe the design works, you make it run and watch.⁴⁵

Add capability without adding cost. Conditional execution is the signature move: more expressive instructions that kill the branch bottleneck for almost no extra hardware. The lesson is to look for the change that buys you a lot of capability for a little complexity, rather than the one that buys a little capability for a lot – the same economy of means John Carmack used to wring impossible performance out of fixed hardware.²⁴

Build the thing that teaches. Before the processor that runs the world’s phones, Wilson built the language that taught a country’s children to program. BBC BASIC was deliberately approachable yet honest about the machine underneath – you could drop from BASIC into assembly. Making the powerful thing also the teachable thing is a rare and underrated discipline, shared by language designers like Bjarne Stroustrup who care that the tool can be learned as well as used.⁶

Influence Chain

Who Shaped Her

The Berkeley and Stanford RISC research. The Reduced Instruction Set idea – that a small, regular set of single-cycle instructions beats a large complex one – came out of the early-1980s academic RISC projects, and Wilson and Furber adopted it deliberately for ARM. They took the principle and pushed it past the textbook, adding conditional execution, but the foundational bet that simpler is faster was the research community’s. (Formative influence)

The MOS Technology 6502 and a single engineer. Wilson cut her teeth on the 6502 – the cheap, simple processor inside the early Acorn machines and the BBC Micro. When the Acorn team visited the chip’s maker and saw that essentially one person was designing the next version, the lesson stuck: building a CPU did not require an army. That observation made the two-person ARM project thinkable. (Direct influence)

Steve Furber. ARM was a true partnership. Wilson defined the instruction set – what the machine could do – while Furber architected the hardware that did it, the pipeline and the barrel shifter. Neither half is the chip; the collaboration is. They shared the Draper Prize for it nearly four decades later. (Direct influence)

Who She Shaped

Every modern smartphone. The instruction set Wilson designed, evolved across forty years, is the processor in the overwhelming majority of the world’s phones and a vast range of embedded devices – over 230 billion chips. Few engineers have shaped so much of what people physically hold every day. (Field-defining influence)

Apple Silicon and the performance-per-watt era. ARM’s licensing model and its efficiency made it the natural foundation for low-power, high-performance design, the lineage that runs through the chips in every iPhone and, eventually, the Apple Silicon Macs. The mobile computing era was built on the architecture’s frugality.

A generation of British engineers. Through BBC BASIC and the BBC Micro, Wilson taught a country to program. The Computer Literacy Project’s machine, with her language inside it, was the on-ramp for an enormous share of the UK’s software and hardware talent.

The Throughline

Wilson is the mirror image of the series’ other great hardware mind. Jim Keller made his chips win by spending silicon freely – wider pipelines, more execution units, more transistors to kill every bottleneck – on the bet that silicon is the cheap, abundant resource. Wilson made her chip win by spending almost none: fewer than 25,000 transistors, because for a two-person team at a small British company, silicon and engineering effort were the scarce resources, not the free ones. Both are right, in their regime. And here is the twist the series keeps circling – Wilson’s frugality, chosen for speed and scarcity, produced the low power that made ARM the processor of the mobile age, which is a larger empire than raw performance ever built. Where John Carmack says master the fixed hardware you are given and Bjarne Stroustrup says do not pay for what you do not use, Wilson says: remove until almost nothing is left, and the efficiency you get for free may be the thing that wins. Keller spends to win; Wilson saved to win – and her savings shipped in 230 billion chips. (Series bridge)

What I Take From This

The lesson I keep from Wilson is that constraints are a design tool, not an apology. It is easy to read the ARM story as “look what they did despite having no resources” – but that misreads it. They did not succeed despite the constraints; they succeeded because of them. A team of two with a tiny transistor budget cannot build a baroque processor, so they were forced toward the simple, single-cycle design that turned out faster and far more efficient than the well-funded alternatives. When I find myself wishing for more – more time, more compute, more room – Wilson’s career is a reminder that the constraint might be pushing me toward the better design, if I stop resenting it and let it do that work.

The second lesson is about efficiency you cannot yet see the value of. ARM’s low power was, in 1985, almost worthless – a footnote on a desktop coprocessor. It became the most valuable property in computing a decade later, when the world wanted computers in pockets. Wilson did not foresee the smartphone; she just refused to spend what she did not need to spend, and that discipline compounded into an empire. The takeaway is not “predict the future.” It is the humbler, more durable thing: do not waste resources even when waste seems harmless, because the frugality that looks like a curiosity today can be the foundation everything stands on tomorrow. It is quality is the only variable read through a lens of restraint – the right design, done with the least, ages into something you could never have planned.

FAQ

What did Sophie Wilson design?

Sophie Wilson designed the instruction set of the original ARM processor – the Acorn RISC Machine – at Acorn Computers, beginning in October 1983, with Steve Furber building the hardware. That architecture now ships in over 230 billion chips and powers most of the world’s smartphones. Before ARM, she co-designed the BBC Micro and wrote BBC BASIC, the language built into it. Later she was chief architect of the FirePath DSP at Element 14, which Broadcom acquired.¹²⁵

How many transistors did the first ARM processor use?

The first ARM, ARM1, used fewer than 25,000 transistors – a small fraction of comparable processors of its time, which ran to hundreds of thousands. Because so little silicon was switching, the chip drew only about a tenth of a watt, roughly a twentieth of what an Intel 386 needed. That minimal transistor count, chosen for speed and for the limits of a tiny team, is the root of the low-power efficiency that later made ARM dominant in mobile devices.³⁴

Is it true the first ARM chip ran without its power supply connected?

Yes, and it is well documented. When the first ARM1 silicon arrived from VLSI Technology on 26 April 1985, it worked the first time it was tested – itself a rare achievement. Even more striking, the chip drew so little power that when it was plugged into the development system, it sprang to life by drawing current through its I/O interface before its own power supply could be properly connected. The design was so frugal it effectively ran on leakage, an early and accidental sign of the efficiency that would define ARM.³⁵

What is conditional execution in the ARM instruction set?

Conditional execution is one of Wilson’s signature design choices: almost every ARM instruction carries a four-bit condition code, so it executes only if its condition is true and otherwise does nothing. On a conventional processor, choosing between actions requires a branch, and a taken branch can flush the instruction pipeline and waste cycles. With conditional execution, short stretches of conditional logic become branch-free straight-line code – more capability from almost no extra hardware, which is the ruthless efficiency at the heart of ARM.²⁴

Sources