Birth of a world beater

THE EMERGENCE of ARM as one of the world's two dominant processor architectures can be seen as Britain's last laugh after its home-grown computer industry was swamped by the Wintel tide.

An estimated 20 billion ARM processors are installed worldwide, around four per person and up to four or five per device.

For Sophie Wilson (pictured at left), who wrote the original instruction set back in 1983 for what was conceived as a co-processor for Acorn's BBC Micro computer, they are still a "little bit awe-inspiring."

"From our perspective selling millions of BBC machines was a success. Selling 20 billion [of anything] we could never have imagined. It's still hard to imagine," she said at the Vintage Computer Fair at Bletchley Park.

The BBC Micro, sold as part of a Beeb project to promote computer literacy, was something of a compromise. Acorn co-founder Chris Curry wanted a mass-market machine, company techies wanted to create a more efficient design, and director Andy Hopper wanted a workstation class machine.

Wilson had the idea of splitting the design into an I/O subsystem and what she called a language processor that read and executed instructions. "The I/O system was what is now called the BBC computer and it was built from scratch to take a second processor. And the second processor could be anything."

Acorn offered models with a MOS 6502, a Natsemi  32016,  and a Zilog  Z80 in the second slot, and tried out versions with Motorola 68000 and Intel 186 and 286 chips. But the results were disappointing and Acorn engineers discovered that the performance of all the processors was dictated simply by the bus bandwidth, regardless of architecture and even clock speed.

Wilson recalled: "The different ways these people were building memory systems was leaving a lot of performance on the floor... The 8MHz 68000 was dreadfully wasteful because it used four clock ticks to run its bus, whereas the 6502 used only one." This meant the 68000's 16-bit memory bus was no better than the 8-bit bus of a 6502 with half its clock speed.

A visit Wilson and her colleague Steve Furber made to the Western Design Centre at Phoenix, Arizona, to check out a successor to the 6502 has passed into UK computing legend.

The two British engineers, expecting high-tech buildings bristling with expert brains, were astonished to find just two developers working in a bungalow with "a bunch of college kids" on holiday jobs. "We came away thinking if they can do it, so can we."

And so they did, with Furber concentrating on the hardware design. They knew from using the 6502 how to implement an efficient single-tick memory bus and put it into production. "So we thought why don't we build a processor to match, and stretch the memory system? So we built a memory system that was four times as good as the 6502's with a processor to match."

The instruction set, Wilson said, "came from that strange place inside my head where computer design comes from. Most engineers like to proceed from A to B to C in a series of logical steps. I'm the rare engineer who says the answer is obviously Z and we will get on with that while you guys work out how to do all the intermediate steps. It makes me a dangerous person to employ in IT but a useful one."

A major reason for ARM's success is the power efficiency of its designs, honed over the years with the help of it partners and the demands of the mobile handset industry. But Wilson confesses that they did not consider this aspect at all in their original design.

In fact when they plugged in their chip for the first time it appeared to draw no current at all. Everything ran perfectly and they cracked open some champagne to celebrate. Then Furber connected a multimeter and the power pin read zero. It turned out that a track on the circuit board had broken and the processor had been running on power drawn through two protection diodes.

"When we finally got round to measuring the power we found it was using much less than we had thought possible."

This was partly because of timing efficiencies but also because the chip had only 25,000 transistors - the RISC architecture used by ARM requires fewer than Intel's CISC chips. Even so that first chip was a power hog by today's standards.

"It was using about 100 milliamps at 5 volts. That's a huge amount of power. The [ARM] Cortex A8 only uses that amount at 120 times the performance. So it wasn't particularly well done."

The design team's major breakthrough, and one that came to define ARM's future, came in 1992. Miniaturisation had reached the stage that with ARM's low transistor count there was room to spare on a silicon die. So the I/O, video and memory controllers were integrated into the first-ever system-on-a-chip (SoC), which the ARM250 used in later Archimedes computers.

"We didn't know it was a first, and the world didn't know it was a first. It was a long way before anyone was talking about systems-on-a-chip," said Wilson.

By this time ARM had been spun off from Acorn in a joint venture with chipmaker VLSI and Apple, which planned to use an ARM chip in its Newton PDA. Acorn morphed eventually into Element 14, for whom Wilson designed ultrafast chips for DSL systems while remaining a consultant for ARM. She now works for Broadcom, which bought Element 14.

There is a huge irony in ARM's slow-burn success. Acorn's computer business failed because desktop computing needed a standard platform and coalesced around the unremarkable Wintel PC. The embedded and mobile industries coalesced around low-drain ARMs for much the same reason.

It could easily have been another architecture, and Wilson says the success could never have happened without ARM chief executive Robin Saxby. "His licensing model was just as important as ARM technology."

The model means ARM simply sells its core designs, from microcontrollers the size of pinheads to server workhorses, for other manufacturers to build SoCs round. Which is why Wilson, following rumours that Apple was thinking of buying ARM, believes the company should remain independent. "No manufacturer is going to want to rely on technology owned by its rivals."

Neither is she taken by some of the fashionable gadgetry that use ARMs. "I'm interested in computers that can do things, that can be programmed," she said, taking out her Samsung Windows Mobile phone and showing the screen. It is running the BBC Basic wrote back in 1982, still running on an ARM. µ