Based on the sources provided, the following is a reproduction of the oral history of Sophie Wilson, recorded on January 31, 2012.
Oral History of Sophie Wilson 2012 Computer History Museum Fellow Interviewed by: Douglas Fairbairn Recorded: January 31, 2012, Cambridge, United Kingdom
Fairbairn: Okay, Sophie, can you hear me? Wilson: Yes. Fairbairn: Sophie? Are you there? Wilson: Well, I’m here. Fairbairn: Oh good. It’s working both ways now. My name is Doug Fairbairn and it’s January 31st, 2012 and I’m in discussion with Sophie Wilson, a primary architect of the ARM microprocessor which over the last 18, 20 years has become probably the most widely used microprocessor throughout the world. Delighted to be here, Sophie, welcome. Wilson: Hello. Fairbairn: Before we get into the general flow of the interview, could you spend a few minutes talking about what the ARM processor is; what’s unique and special about it and what impact it has had?
Wilson: Okay. So, I’d like to take you back to 1983. At the time, Acorn Computers was a reasonably successful producer of small microcomputers, a sort of British company like Apple. We had a machine, the BBC Microcomputer, that had been franchised by the British Broadcasting Corporation; it was, in some ways, quite like an Apple II. It had a 6502 microprocessor in it though we ran it faster, it had higher resolution graphics. It was mainly TTL logic but it had two custom integrated circuits inside it; a video processor and a serial processor.
The BBC Microcomputer was designed as a two processor system from the outset to get past an impasse in the company where some people wanted a cheap machine and others wanted a workstation. By having an IO processor and a second processor for heavy lifting, we could make all of that happen. We made lots of second processors based on existing microprocessors like the 80286, 6809, and NS32016. We could see what all these processors did and what they didn’t do. They didn’t make good use of the memory system, they weren’t fast, and they weren’t easy to use. They were too slow. Between the two things we felt we needed a better processor in order to compete with the flood of IBM PC compatibles.
We gave ourselves a project slogan: “MIPS for the masses”—Millions of Instructions Per Second for masses of people. Unlike RISC research at Berkeley or Stanford aimed at high-end workstations, we wanted the same thing at the low end. ARM kept Acorn alive for another 16 years. As a side effect of making it cheap and simple, we also ended up making it power efficient; that wasn’t intentional. We only had 25,000 transistors in the first one. We were worried about power dissipation for mass manufacture in cheap plastic cases without heat sinks. As the world has gone increasingly mobile, that aspect of ARM has mattered as well.
Fairbairn: That’s a great summary. Now I’d like to understand how you arrived at Cambridge. Tell me about your childhood and parents.
Wilson: I was born in Leeds in Yorkshire. My parents moved to a small village called Burn Bridge. My parents were both teachers, and if they had one ambition for their children it was “Don’t be a teacher.” My father was an English teacher, my mother a physics teacher. We all went off to university to do mathematics. My father did an immense amount of construction. He built the family car, boats, and half the furniture in the house. When my mother’s Physics Department needed instruments, we built them. We’d have a dining room table and everybody in the house would be sitting around building Heathkit multimeters. My father once had a 26-foot boat mast in the living room. Making things and making things work was a routine thing.
We had long summer holidays in the Lake District where we lived in tents or camper vans my dad built. There was no real limit to what my parents could do. We were a very “ready” family. We went to the library every week and pooled our tickets to get 12 books and read them all by the next weekend. There wasn't much electronics available, but when I was 13, we were given mechanical calculators—metal tablets with slidey things.
I was an unexceptional student in a streamed grammar school. I did modestly at O Levels. Secretly I hated chemistry but was fond of the math teacher. My parents expected me to follow math/physics/chemistry, but I revolted at the last moment and picked Maths, Further Maths, and Physics. I applied to Cambridge University and passed the entrance exams. Out of six pupils in the further maths group, only two of us got into Oxford or Cambridge.
I was extraordinarily naïve about why I was going to university. I did a year of maths, but it was clear Cambridge maths was substantially harder than the entrance exams. In my second year, I failed.
Between school and university, I had worked for ICI Fibres Research. They put me in a department outfitting production lines. I built a “wrap detector,” but I was told by the Union steward that I built it too quickly. Then I built a machine that counted droplets using an infrared beam. I wasn’t impressed with their RTL logic blocks, so I read the RCA CMOS book cover to cover. I rebuilt the drop counter with CMOS because it was noise immune and low power.
After my first year of university, I worked for a guy who had set up an electronics business. He wanted an electronic cow feeder. I bought a 6502—my first—for 76 quid. I designed the electronics to be waterproof and wrote the programs in machine code. I used Electrical Erasable ROMs (EEROMs) for storage, which was very advanced for the time. I had taught myself programming by using a PDP-8 at ICI, teaching it to play noughts and crosses.
When I failed maths in my second year at university, my Tutor suggested a one-year course in Computing Science. I resisted at first but did it. During that summer, I worked for Hermann Hauser. He wanted an "electronic pocket diary." I showed him my folder of designs for my home computer and cow feeder. He said, “Will you build one of those for me?” That became the Acorn System 1. I built the Verowire prototype and wrote the 512-byte monitor by hand.
Hermann had a consultancy called Cambridge Processor Unit (CPU). They won a contract to build an electronic fruit machine. Steve Furber and Chris Turner were working on it. I built circuitry to stop the machines from paying out when someone sparked a Piezo electric lighter near them. We passed the tests even when they struck electric arcs with a welding machine next to it. We changed the production machine to use my 6502 instead of Steve's 2650.
I spent most of my time at CPU Limited building stuff and a fraction of my time at university. I broke Martin Richard’s BCPL compiler with an elegant recursive program. Hermann attended my graduation and negotiated my job; I got paid 1200 pounds a year.
In the summer of '79, I went to work full time. I spent time writing an automatic assembler and then, over Christmas, wrote the first BASIC interpreter. For the Acorn System 1, I did everything: software, manual, PCB design, and shipping. We built a whole range of parts—memory cards, video displays, disk interfaces.
Then came the BBC Micro. Chris Curry wanted something for the home hobbyist. We were discussing a design called the Proton. The BBC wanted a computer for a TV series. Chris Curry told the BBC they should see our prototype, even though we hadn't built one yet. Hermann hoodwinked Steve and me by telling each of us that the other had agreed to build the prototype by Friday.
We started Monday. It was a 100-IC machine. We needed special Hitachi DRAMs for a high-performance memory system. We wire-wrapped the thing by Thursday night, but it didn't work. I went home at 2:00 AM to sleep so I could write the software. Hermann decided to take the in-circuit emulator out and put a native processor in, and it worked. I adapted the OS and bootstrapped BBC BASIC just hours before the BBC arrived. While Hermann delayed them on the stairs, I was poking registers to get a screen display. We sold 1.25 million of them.
By 1983, we were disappointed with processors like the 80286 and 68000. They were slower than the 6502 because they didn't use memory bandwidth effectively. We visited Western Design Center in Phoenix. We saw bungalows and college kids sticking Rubylith tape on things. We became convinced that if they could build a processor, we could too.
Andy Hopper put the first RISC papers from Berkeley and Stanford on my desk. I started designing an instruction set. Steve researched pipelines. We had a team of fewer than 10 people. We put a lot of effort into verification. Steve wrote a behavioral model in BBC BASIC. We sent the ARM to VLSI Technology; the first silicon came back on April 26, 1985. We plugged it in, and it worked the first time.
In 1990, ARM spun out of Acorn as a separate company with investment from Apple and VLSI Technology. I stayed at Acorn as a consultant to ARM. I wrote the "Acorn Replay" multimedia system and worked on set-top boxes and the StrongARM with DEC.
Eventually, Acorn was taken over because of the value of its ARM shares. I had been developing a 64-bit processor called ALARM (A Long ARM). We used that as the seed for a startup called Element 14. ALARM was renamed FirePath. We became a multi-core chip company for DSL modems and were bought by Broadcom.
I have two wildly successful microprocessor designs. I’ve always had the pleasure of seeing things I design used in volume. Now they’ve shipped over 32 billion ARM cores. It’s a bit scary to have 20 billion copies of a little bit of your brain running around the world. But I have the satisfaction of seeing my stuff be in things. Things are made to work.
End of Interview
No comments:
Post a Comment