Summary

  • ARM1's first chips booted the Tube operating system and ran BBC BASIC immediately because Acorn had already exercised the instruction set through a behavioral model, instruction tests, faster simulators, interpreters and real software ports.
  • Sophie Wilson designed the instruction set and led validation work; Steve Furber wrote the first BBC BASIC behavioral model and developed the microarchitecture; named verification, compiler, physical-design and VLSI Technology contributors carried different parts of the result.
  • A successful first boot proved agreement across the tested architecture, implementation and system path. It did not prove exhaustive correctness, timing margin, power, manufacturing yield, qualification, volume readiness, ecosystem adoption or commercial success.

The board was waiting for an answer

The famous ARM1 story ends with a chip inserted into a second-processor board. On 26 April 1985, according to Sophie Wilson's Computer History Museum oral history, the Tube operating system booted, BBC BASIC ran and the team celebrated after a simple command produced the expected answer. The sequence is often compressed into a boast: the chip worked first time.

Compression removes the important part. The board could be ready because the machine had existed in several executable forms before it existed as silicon. Software had been written for an architecture that fabrication had not yet returned. The first chip did not meet a blank bench. It met a prepared chain of expectations.

That distinction turns a legend into an engineering method. First silicon can be a powerful receipt, but only when the claims behind it are visible: which behavior was modeled, which instructions were tested, which software ran, which implementation was fabricated, and which observations counted as agreement.

One correction clarifies the division of work

Wilson's own account is precise about attribution. She says Steve Furber wrote the first behavioral model of ARM in BBC BASIC. It was slow, advancing only a few cycles per second. Wilson designed the instruction set and led a larger group writing test programs to check instruction behavior, along with validation software and interpreters. The Royal Society likewise credits her with the ARM instruction set and jointly recognizes Wilson and Furber for ARM's design and analysis.

The correction matters because a model, an ISA and a chip are related artifacts, not synonyms. The instruction set describes the programmer-visible contract: operations, registers, addressing and effects. A behavioral model makes that contract executable. Furber's microarchitectural and block specifications described how the contract could be realized. Physical designers then turned those blocks into circuits and layout.

The result did not belong to a single “inventor” pressing a button. Wilson names Hugo Tyson, Jez Wills, Jon Thackray and David Seal alongside herself as writers of verification software, validation software and interpreters. Arm's later account of Seal describes ISA verification and validation as his principal Project A contribution. Wilson also recalls a compiler team led by Lee Smith and Harry Meekings. Each activity attacked a different uncertainty.

Slow truth, then faster questions

The first behavioral model's slowness was not necessarily a defect. Its job was to provide a reference for instruction meaning, not an experience of the future machine's speed. Test programs could ask whether shifts, arithmetic, conditions, loads, stores and control flow produced the expected architectural state. A slow model could expose a contradiction before it became a mask or a wafer.

The team then built faster, more selective instruments. Wilson recalled pure instruction simulators running hundreds of thousands of simulated ARM instructions per second on a 6502 second processor. That speed changed the kind of evidence available. Instead of exercising isolated instruction cases only, engineers could run larger programs and develop more of the software environment.

They ported BBC BASIC. They prepared the Tube second-processor operating system and other software. Interpreters and compiler work expanded the instruction combinations being exercised. This did not make the verification exhaustive, but it created a ladder. A microtest could isolate semantics; a simulator could explore more sequences; a language runtime could stress calling, arithmetic and memory conventions; an operating-system path could test boot, I/O and integration assumptions.

Real software was useful precisely because it was not written as a congratulatory test. It combined instructions according to the demands of a language and system. When interpreted ARM machine code performed strongly even before real ARM silicon existed, the team obtained both functional confidence and a signal that the architecture's performance assumptions were plausible. Neither signal was yet a measurement of the fabricated chip.

From specification to a physical object

Steve Furber's oral history adds another boundary. He describes short block specifications for units such as the ALU and register file, positioned between schematics and higher-level logic descriptions. Acorn's VLSI group took those descriptions, performed circuit and physical design and assembled the chip. Furber remembered three people working directly on ARM, with the group later growing as it worked on the wider four-chip system.

VLSI Technology supplied design tools and fabrication. Its own oral-history panel distinguishes Wilson and Furber as architects from the design group that created the chip and from VLSI Technology as foundry. This was a cross-organisational handoff, not a simple print operation. A behavior had to survive translation into blocks, circuits, layout, masks and manufactured samples.

Hermann Hauser's role was different again. He provided executive backing and an environment in which an exploratory processor idea could become a formal project. Later accounts celebrate Acorn's scarcity of money and people. Scarcity may encourage a small design, but it does not validate one. The validation came from artifacts, tests, review and implementation work performed by named engineers.

What the first boot actually received

When the chips arrived, the waiting board and prepared software made the test unusually legible. The Tube OS booting showed that the processor could execute enough of the expected instruction behavior and communicate through the second-processor environment to start the system. BBC BASIC running extended the path through a substantial, already-understood language implementation. The result connected ISA, model, simulators, ported software, microarchitecture, physical design, fabrication, board and interfaces.

That is why “worked first time” is meaningful. It was not merely that an oscillator toggled or a diagnostic pin changed. A stack of independent expectations converged in visible program execution.

Yet the receipt remained bounded. It did not show that every instruction corner case, interrupt ordering, electrical margin or long-duration state had been exercised. It did not measure power under representative workloads. It did not state wafer yield or prove that packaging and clocks would behave across temperature and process variation. It did not qualify the chip for volume production. It did not complete the support chipset, deliver a mature operating system or establish a commercial market.

Wilson's narrative itself keeps those stages separate. After first silicon, chips still had to be qualified; a complete Acorn machine and a more usable operating system followed. Arm's later corporate history adds still more layers: ARM2, the Archimedes, the 1990 formation of Advanced RISC Machines, licensing, partners and large markets. None of those outcomes was contained inside the first successful BASIC prompt.

A first-pass success is not a theory of RISC

The ARM1 was deliberately small—about 25,000 transistors in Wilson's recollection—and simplicity reduced the number of places an error could hide. A clean instruction set, disciplined conditional execution and a compact implementation helped. But “RISC” is a design family, not a guarantee that silicon will work on revision A. Other RISC projects could fail; ARM could have failed; later, larger ARM designs could reveal different risks.

Calling the result inevitable because the architecture was simple reverses cause and evidence. The team chose a tractable design and then spent heavily, in its own constrained way, on pre-silicon confidence. Models, simulators and software moved uncertainty earlier. Their success did not eliminate engineering; it showed engineering doing its work before fabrication made corrections expensive.

The credit map is part of the proof

Attribution is not decorative history. It reveals the interfaces that were tested. Wilson's ISA work and validation leadership linked architectural intention to executable programs. Furber's model and microarchitecture linked meaning to implementation. Tyson, Wills, Thackray, Seal and other engineers expanded the verification surface. Smith, Meekings and compiler contributors increased the software path. Acorn's VLSI designers made the physical object. VLSI Technology contributed tools, foundry work and a manufacturing relationship. Hauser provided organisational permission and resources.

The team was small, but it was not one person, and it was not one kind of work. The first-silicon receipt was strong because these roles met at a common observable boundary. A disagreement could be located: specification, model, test, interpreter, compiler, block, circuit, board or fabricated device.

This is the durable lesson of ARM1. A dramatic final observation is most valuable when it closes a chain of prior evidence rather than replacing it.

Sources