The Arm Story Part 2 : Archimedes to Advanced RISC Machines
How Acorn built the ‘fastest microcomputer in the world’, failed to sell it, and Apple helped twelve Acorn engineers jump ship to the ’ARM Barn’.
'ARM was part of a policy decision that a computer should be designed on silicon rather than cobbled together out of third-party components. This focus made Acorn one of a small, select group of genuine computer companies (Apple being the most obvious) that own their own technology from the ground’.
Hermann Hauser
Olivetti
At the end of Part 1 of the Arm story, it was 1986 and the Acorn RISC Machine (ARM) microprocessor had been developed at Acorn Computers in the UK by Steve Furber, Sophie Wilson and a small team of designers, all working under the guidance of Acorn co-founder Hermann Hauser.
The team’s objective had been to develop a successor to the 8-bit MOS Technology 6502 which could be used in Acorn’s home, educational and business computers. Following the Reduced Instruction Set Computer (or RISC) philosophy, they had designed, from scratch, a microprocessor that outperformed the latest 16-bit and 32-bit designs from firms such as Intel and Motorola.
Their focus on keeping down manufacturing costs meant that the ARM used less than 25,000 transistors and also had exceptionally low-power consumption.
But by 1986, Acorn’s business was struggling, forcing it to sell a controlling stake to Italian computer manufacturer Olivetti. The question now was what would Acorn and its new Italian owners do with the Acorn RISC Machine.
Olivetti was run by charismatic industrialist Carlo de Benedetti. De Benedetti had previously been chief executive of Italian carmaker Fiat, but his tenure in charge, had ended after a few months following industrial action, after he tried to lay off over 60,000 workers.
At Olivetti, he had set out on a strategy of cost-cutting and refocusing the firm away from office equipment and onto computer technology. In 1985, he had sold a 25% stake in the company to US telecoms giant AT&T. Aside from raising capital for Olivetti to expand, the rationale for the deal was stated as giving the firm access to AT&T’s technology, including microprocessors. By 1985 Olivetti had become the second-largest computer company in Europe behind IBM’s European operations. So with its financial strength enhanced by the AT&T investment, Olivetti was free to pursue deals, including its investment in Acorn.
By 1986 RISC was very much in fashion. The research efforts at University of California, Berkeley and at Stanford, that had provided the original inspiration for the Acorn team to develop the Acorn RISC Machine had each led to their own commercial developments. These would turn into SPARC (introduced by Sun Microsystems in 1987) and MIPS respectively. IBM, where much of the RISC concept had emerged, used its RISC microprocessor, the ROMP, in the IBM PC/RT which appeared in 1986.
And elsewhere, almost every other major firm seemed to be working on their RISC designs too. Hewlett Packard had PA-RISC introduced in February 1986. Even Intel had one in the shape of the Berkeley RISC inspired i960, which appeared in 1984.
Back at Acorn, the relationship with Olivetti wasn’t quite working out as the Acorn team had planned. They had believed that Olivetti would help sell some of Acorn’s unsold stock of computers, one of the reasons for Acorn’s financial difficulties. In fact very little of Acorn’s products were sold by Olivetti’s sales teams. Co-founder Chris Curry has recounted that a year after the takeover, Oilvetti’s de Benedetti challenged his management team in Cambridge as to why this hadn’t happened, only to be told: “Please Signore Benedetti, it is because you told us that we were not allowed to sell anything except IBM PC compatible products.”
Olivetti’s focus on IBM compatibles had come after a brief and unsuccessful flirtation with Zilog’s Z8000 microprocessor. Despite de Benedetti’s challenge, Olivetti never made significant use of ARM technology in their personal computer designs. The fate of ARM was to be intimately linked to that of Acorn for the foreseeable future. And the strategy for the use of ARM continued largely as before – as a replacement for the 6502 in Acorn’s own desktop computers.
Chris Curry left Acorn in 1985 soon after the Olivetti investment. He later fumed about how Olivetti had behaved after the takeover:
… we should have sued them because they breached the contract. But, of course, we didn’t behave like that, and afterwards Olivetti exercised its right to take 75% of the company and put in their own controlling management and gradually shut the place down …
By the end of 1985 Olivetti had injected another £4m into the company and owned almost 80% of it. Olivetti never owned Acorn outright, Acorn remained listed on the London Stock Exchange and external shareholders kept a small stake in the company. As a result, its ongoing problems and fragility would be painfully evident for all – including potential computer purchasers – to see.
Hauser became Olivetti’s Vice President of Research, overseeing a number of research and development centres in Europe and the US. No longer involved in the day-to-day management of Acorn, he retained his interest in the development of the ARM technology that he had been instrumental in creating.
‘Hail Archimedes. Fastest of all micros.’
And work on that technology continued. The original plan to create a successor to the 6502 based BBC Micro continued. If the new microprocessor was to be used in this role, then it needed some new support chips. So, having completed the ARM microprocessor, the ARM team turned their attention to three additional designs. These would perform memory management (with a chip called MEMC), control video and sound (VIDC) and other input and output (IOC).
The team also added some minor updates to the original ARM microprocessor design. Integer multiplication and combined multiply and add instructions were added to the instruction set. The new design would use VLSI Technology’s smaller 2um process, to create a CPU that would be known as the ARM 2.
Any new ARM machines needed an operating system. Acorn had set up a development centre, known as the Acorn Research Centre or ARC, in Palo Alto close to Xerox Parc, famous for its pioneering work on graphical user interfaces. The team there were working on a highly sophisticated OS for the new machine known as ARX.
But Acorn’s management became concerned at the slow progress of ARX and turned to Acornsoft, their Cambridge, UK based software team. According to Paul Fellows, who led the team, they were told 'the hardware is being made and we have got no software, you guys aren't doing very much at Acornsoft, can you make us a BBC-like operating system? You've got 5 months.’ The short timetable led to the software being called Arthur as an abbreviation for A RISC by Thursday.
ARX duly lost in a head to head demonstration against Arthur and was cancelled. Arthur, still incomplete and partly written in Basic, would be the OS that would ship in the Read-Only-Memories of the first ARM machines.
In June 1987, it was finally confirmed that an ARM-based successor to the BBC Micro would be launched. ‘Acorn user magazine’ announced that ‘RISC Machine is next BBC Micro’ and under the headline ‘The almost mythical Acorn RISC computer will be the next BBC Micro’.
The new machine would be called the Archimedes. There would be two distinct model lines, the A300 and the A400 series. The first machines were the A305, A310, A410 and A440.
The Archimedes A305 and A310 were still branded as being ‘BBC Microcomputer System’ despite protests from Acorn’s competitors, and they duly appeared on the BBC’s ‘Micro Live’ program (a successor to earlier programs made for the BBC’s computer literacy project).
The first reviews of the Archimedes were very positive. The UK’s Personal Computer World said that the Archimedes:
‘…felt like the fastest computer I have used, by a considerable margin. Benchmarks are inadequate to exude the feeling of power that it exudes in use; just about everything you do happens instantly …’
The same review also commented on Arthur:
“The overlapping, resizable windows are in colour, and resemble a hybrid of GEM, Windows, and the Macintosh.” (GEM being Digital Research’s windowing system, then shipping on the Atari ST).
For all Arthur’s limitations it did have some genuine innovation. It would be the first GUI to have a taskbar or dock, shipping the feature a year ahead of NeXT’s NextStep operating system.
An Acorn advertisement of the period trumpeted its major selling point – its performance – with the headline ‘HAIL ARCHIMEDES. FASTEST OF ALL MICROS.’ The accompanying table showed the Acorn machine dramatically outpacing the Atari ST, Amiga and a clutch of Intel-based machines (including one from Olivetti, which may not have gone down well with Acorn’s owners!) The fact that the tests used BASIC, and Sophie Wilson’s BBC BASIC was famous for its efficiency, probably flattered the results.
The entry level A305 was priced at £799 and was much cheaper than a comparable Mac or Intel 386-based PC, which would each cost around 3 times more.
A contemporary report identified the low cost of the ARM CPU as part of the reason for the low cost:
For a minimum order of 100 a 20 MHz Motorola 68020 costs around £300; 16MHz Intel 80386 costs £240. But a 10 MHz Acorn RISC Machine (ARM) CPU, which will comfortably outperform either, and without the need for expensive high speed static RAM, costs a mere £44.
With the Archimedes, Acorn had achieved an impressive degree of integration. The firm now had its own microprocessor design, and it was selling it in its own computers with a full complement of software, including a range of languages and an operating system with a Graphical User Interface. All this came from a firm with less than 500 employees.
By now, though, Acorn and the Archimedes were facing massive competition. In the business market, IBM compatibles were dominant. In the home market there were Motorola 68000-based machines which, although less powerful than the Archimedes, had a much larger range of software, especially games.
The Atari ST had launched in June 1985 andthe Commodore Amiga in July of the same year. Both offered a graphical user interface and had machines that were priced much lower than the Archimedes. The Motorola 68000 was no match for the ARM 2, but by the time the Archimedes launched the two machines had built up a large range of games software. The range of software available for the Archimedes would never catch up.
So the Archimedes sales remained largely confined to the education market. Even there though, it faced competition from Intel-based machines including from Acorn’s owner Olivetti, who pushed Acorn to sell a rebadged Intel 8088 based machine they called, somewhat unimaginatively, the M19.
The Archimedes range continued to be developed by Acorn over the following years, but with limited commercial success. It did find a home in some niche markets where the Archimedes power and graphics capabilities were valuable. Successors of the Archimedes were used, for example, to generate the graphics for the first episodes of ‘Who Wants to be a Millionaire’. It was an aspirational machine for some. Having earlier programmed the 6502, Linus Torvalds who would go on to create Linux, wanted one:
What I wanted to upgrade to was Acorn Archimedes ... the thing that gave ARM its name, that was my dream machine for a while.
But Torvalds had to settle for the cheaper, 68008 powered, Sinclair QL. Acorn would only sold a few hundred thousand ARM-based machines, compared to over 1.5 million BBC Micros, 2 Million Atari STs and almost 5 Million Amigas.
But if Acorn and the Archimedes limped along, the Acorn RISC Machine that powered it continued to develop. In 1989, the ARM team finalised the ARM 3. This added a 4-Kilobyte cache and the clock speed leapt up to 33 MHz, whilst being manufactured on an improved version of VLSI’s 1.5 um technology. This gave the new CPU much improved performance, but not at the expense of significantly increasing the number of transistors. The ARM 3 was still low-cost and low-power when compared to the competition.
Apple Bites
By now, Hauser and Acorn’s Chairman Elserino had already decided that it was necessary to find a buyer for the ARM technology. It was clear that it would need increasing levels of investment to remain relevant, Acorn’s business wouldn’t support it and Olivetti had no appetite.
So they used Elserino’s extensive network of contacts to see if they could find a buyer. After talking to numerous firms, they nearly had a deal with German’s Siemens, but it ultimately fell through. SGS-Thomson, the forerunner of ST Microelectonics, was approached, but they had already invested in another British microprocessor company, INMOS, with its transputer technology. No one was interested.
In 1988 Hauser left Olivetti to start ‘The Active Book Company’ with £1m of his own money. The plan was to make a paperback sized computer powered with an ARM microprocessor with a screen that could be interacted with using a stylus.
Even though Hauser was no longer at Olivetti, he still had an interest in it future of the ARM Microprocessor. Its future looked bleak, but Hauser’s personal contacts would eventually pay off. Salvation, in the end, came from an unlikely source and an old rival of Acorn’s in the personal computer market developing a product that had quite a lot in common with Hauser’s new venture.
By 1989, Apple had already been developing the Newton handheld for two years. John Sculley was running Apple having ousted Steve Jobs in 1985. It was Sculley who would coin the term Personal Digital Assistant which would be used for the Newton and similar later products.
The first designs of the Newton used a central processor called the Hobbit, that was being developed by AT&T. The Hobbit was a derivative of CRISP, developed by Dave Ditzel and Rae McLellan at AT&T Bell Labs. Ditzel had co-authored with Patterson the paper ‘The case for a Reduced Instruction Set Computer’ one of the key early papers in the development or RISC and one of the papers that Steve Furber and Sophie Wilson had read before starting work on ARM 1. The Hobbit and CRISP weren’t pure RISC machines, and combined elements of the RISC concept with a stack-based architecture.
In 1989, the Newton project was struggling and Larry Tesler, who was leading Apple’s Advanced Technology Group took over the project. Tesler had previously shown Steve Jobs around Xerox Parc, a visit that had famously led to Jobs adopting the mouse and Graphical User Interface used in Apple’s Lisa and Macintosh. He soon became dissatisfied with what AT&T were delivering. According to Tesler:
The Hobbit was rife with bugs, ill-suited for our purposes, and overpriced. We balked after AT&T demanded not one but several million more dollars in development fees.
And there was another key factor. Tesler didn’t trust AT&T to continue to back the Hobbit over the longer term. As a small division in a giant company, the Hobbit really wasn’t that important to AT&T.
This was not the first time that Apple had looked at ARM-based designs. A few years earlier, Tom Pittard and Paul Gavarini, members of Apple’s ‘Advanced Technology Group’, had noticed that Acorn was in a similar position to Apple in needing a successor to the 6502 and had taken an interest in the ARM project. As early as 1986, they had got so far as to build a prototype desktop computer using an ARM CPU as part of a project called Möbius. The Möbius machine had been fast enough to emulate both the Apple II and Macintosh II in software. But the power of the Möbius could potentially undermine the 68000 based Macintosh line that was now Apple’s leading design, and so Möbius was dropped.
So Apple had engineers who were already familiar with ARM. Given that ARM designs were available that were fully debugged and already low power, it made increasing sense for them to switch the Newton project to ARM.
There was just one problem. Acorn Computers was notionally a competitor to Apple. It was selling a low-priced and powerful desktop machine with a graphical user interface that competed with the Macintosh. The sales of the Archimedes may have been small, but Acorn was backed by Olivetti. It wouldn’t make sense to put the future of Newton in the hands of a competitor.
So it was agreed that Acorn’s ARM technology would be spun off into a separate company and that Apple and Acorn would both take large stakes alongside VLSI Technology, who were making ARM semiconductors for Acorn. The stakes owned by each firm were Acorn 46%, Apple 46% and VLSI 8%. Apple invested £1.5 million, which Tesler believed would be less than they would have had to pay AT&T in development fees, for their stake whilst Acorn injected the ARM technology.
Although Apple didn’t fully control the new company, its large shareholding gave it a massive say in the management and direction of the company. If the company struggled or Acorn lost interest in it, then Apple would always have the option of investing some more and taking control.
As the technology was no longer fully tied to Acorn, the name Acorn RISC Machine, was no longer appropriate. The abbreviation ARM had become popular though, and so ‘Acorn’ was simply replaced with ‘Advanced’. Advanced RISC Machines Limited came into being on 3 December 1990.
The new company would need some employees, so members of the Acorn team who had worked on the development of the Acorn RISC Machine would move over to the new company. In the end, just twelve engineers transferred. Although the new firm would be risky (as well as being RISCy) staying with the still struggling Acorn was not a safe bet either. One of the team who moved over, John Biggs, has said:
It was a bit like Hitchhiker's Guide to the Galaxy, when this great civilization decided to get rid of all its insurance salesmen, telephone sanitizers and other useless people, by ferrying them off to another planet. It convinced them that their home world was about to be hit by a terrible catastrophe -- a giant mutant star-goat -- and that they needed to escape on a space ship called the 'B Ark' to save themselves.
Perhaps surprisingly, neither Steve Furber who had designed the original ARM’s microarchitecture nor Sophie Wilson who had designed the instruction set, would transfer to the new company.
Furber wanted to follow an academic career, and in 1990 he moved to the University of Manchester, where he would become Professor of Computer Engineering.
Wilson wanted to remain at Acorn, where she had a key role in developing software that would run on ARM designs. She would do work for the new company, but on a contract basis, later saying:
So I stayed with Acorn, because I felt that joining a small chip company wouldn't let me do the things-- obviously Acorn contracted my services back as a consultant to ARM. ARM needed my services, particularly in the very early time until they could actually get it running themselves.
So twelve engineers would move to become the founders of the new company. The twelve were Jamie Urquhart, Mike Muller, Tudor Brown, Lee Smith, John Biggs, Harry Oldham, Dave Howard, Pete Harrod, Harry Meekings, Al Thomas, Andy Merritt and David Seal.
Saxby
The new company would also need a chief executive. Headhunters turned up one name that was already well known to the ARM team and to Hermann Hauser.
Robin Saxby had formerly worked for Motorola from 1973 to 1984 and had tried, unsuccessfully, to sell the Motorola 68000 microprocessor to Acorn.
Hauser knew Saxby well enough that the two skied together in Austria on one occasion, leading to an accident where Saxby had fallen on top of Hauser’s then girlfriend. In spite of this, Hauser retained huge admiration for Saxby, later calling him “a consummate salesman, just a wonderful person”.
When Saxby left Motorola, he had headed up the European division of European Silicon Structures (ES2). ES2 used an advanced silicon compiler together with the Perkin Elmer AEBLE e-beam machine to create custom integrated circuits. ES2 had raised $100m in funding but ultimately failed due to the low productivity of the e-beam machines, and it eventually ran out of money. At the same time Saxby had headed up ES2’s US joint venture, US2 and commuted between Silicon Valley and Bracknell in the UK.
Before deciding to take the job, Saxby met the whole group of twelve Acorn engineers who would transfer across at a pub. It would be a key test of whether Saxby and the new team would work well together.
The engineers were late and were soon on the receiving end of a reprimand from Saxby for their lack of professionalism. Saxby and a colleague from ES2 grilled the engineers. At the end of the Saxby made his decision. ARM might just work and he would join the new company.
After deciding to make the move to the new company, Saxby gradually shifted his focus from ES2, starting to work a few days for ARM, whilst continuing to work for ES2, before finally becoming full-time CEO of ARM in January 1991.
‘We’re mere Davids against a lot of Goliaths but the little guys can still win if they’re sufficiently creative.’ Robin Saxby
He was quoted, in a piece headed ‘Roamin Robin Heads Up ARM’ in the Acorn staff magazine, as saying:
We’re mere Davids against a lot of Goliaths but the little guys can still win if they’re sufficiently creative.
The ARM Barn
The new company needed a home. Saxby found a repurposed barn eight miles outside of Cambridge in the village of Swaffam Bullbeck. The farmer who owned the barn told Saxby that no animal had ever died in the barn, which Saxby took as a good omen for the survival of his new company.
Somewhat ironically, given Arm’s future role in powering mobile phones, mobile communication was somewhat difficult when the ARM team moved in as the barn was in a ‘dead’ area. One of the team had to be altered on his ‘bleeper’ and then walk away from the barn to take a call.
The early days of the new company were ‘improvised’ in other ways too. With little money, Saxby looked to save wherever he could. He bought the furniture for the new company at a discount, after persuading the furniture company that they could use pictures of it in the barn for their brochure. The boardroom table was won on the toss of a coin, but the firm had no money for bookshelves, so the engineers had to stack their books on the floor.
The new company had two key customers that it was contracted to deliver new designs to. But that would be unlikely to be enough to sustain the business. Although the Newton was an interesting product, its future wasn’t guaranteed. Acorn’s financial and market position remained highly insecure.
They soon lost Hauser’s Active Book Company as a customer. It was taken over by AT&T and Hauser became president of AT&T’s EDO division, and the Active Book Company’s products switched to the Hobbit microprocessors that had been rejected by Apple. In 1994, AT&T closed the microprocessor division and the Hobbit came to an end. In the end Larry Tesler was proven right about the shallowness of AT&T’s commitment.
By now though, competition between high-performance microprocessors was fierce. Intel’s 80486 had been introduced in 1989 and was dominant in desktop business machines. There was a category of Motorola 68000 family powered machines with Graphical User Interfaces, with Apple at the high end and the Atari ST and Amiga at lower price points. A series of powerful RISC-based machines, using designs from MIPS and others, were leading the way in the workstation market. Each of these designs were backed by large organisations or well-funded newcomers.
Advanced RISC Machines, on the other hand, were replicating Hauser’s earlier model at Acorn, with little money and few resources. The spin out had bought them some time though, and a degree of control over their destiny.
So would the ‘Ark’ that the twelve Acorn engineers had jumped into survive, or would it sink?
We’ll find out what happened next in Part 3 of this story. In my view it’s the most interesting and yet, possibly, the least discussed part of the Arm story. Do subscribe to get it direct to your inbox.
If you’ve enjoyed this then you might enjoy the supplement to this post which is available to paid subscribers of “The Chip Letter”. This week we have links to over three hours of video, focusing on the Acorn Archimedes and lots of other materials on the early days of the Acorn RISC Machine.
There is even the first episode of ‘Who Wants To Be A Millionaire’, with graphics courtesy of an ARM-based Acorn computer.
