A row of large, empty Champagne bottles stretches across the back wall of the canteen at ARM's headquarters in Cambridge.
Each of these bottles marks another milestone for the chip designer — another processor released or patent awarded, with the chip model scribbled on the label alongside the names of Champagne's great houses. The dozens of bottles are testament to the success of a British company in challenging the giants of Silicon Valley, all from an unremarkable industrial estate in the middle of England.
The chances are that you're never more than a couple of feet from an ARM chip. They not only sit in mobile phones, tablet PCs and MP3 players but also the likes of hard drives, digital cameras, home broadband hubs, anti-lock brakes, smart cards and embedded microcontrollers serving a range of industries.
ARM has, to a great extent, powered the mobile computing revolution, with its chips sitting inside more than 95 per cent of mobile phones — including the iPhone, iPad and nearly all Android devices. But it doesn't make the tablets or smartphones or even the processors that go inside them. ARM designs the cores of the processors that sit inside these devices – laying out the microscopic labyrinth of logic gates that shunt around the digital bits.
Designing chips and licensing those designs to third parties has been ARM's business since its formation in 1990. And now it's looking at new markets for its chips, preparing for opportunities created by cloud computing and the internet of things.
Today ARM has more than 900 deals that license companies, mainly semi-conductor manufacturers, to use its chip designs across many different markets.
Looking at the drained bottles and ARM's dominance of the smartphone market it would be easy to think the company's ascendency was a trail of uninterrupted success. ARM's overnight success, however, only came after more than a decade of hard slog, as Mike Muller, CTO and co-founder of ARM explains.
"People say 'ARM's been so successful' and you're thinking, 'for the first 10 years it didn't always feel like that'. Nobody was paying any attention and suddenly there's a whole lot of interest and it appears to have come from nowhere. Actually there's a whole lot of homework being done before that," he said.
In general ARM's business model means it takes years before a chip design delivers regular returns, with a processor design typically taking ARM two to three years to develop. A partner company will buy a licence that allows them to incorporate that design into their chips and from there it can then take another three to four years for the chip that incorporates ARM's design to ship, at which point ARM starts receiving royalties based on the sales.
The Cambridge-based company was born out of the PC maker Acorn. Acorn was a household name in the UK in the 1980s thanks to the success of BBC Micro, which was the staple computer of UK schools at the time and sold about one and a half million units.
ARM was formed in 1990, when Acorn decided to spin off its research and development division into a joint venture with Apple and the chip manufacturer VLSI. The company began life designing and licensing high performance 32-bit RISC (reduced instruction set) processors, which were used by Acorn in its Archimedes computer range — its follow-up to the BBC. To this day ARM still designs 32-bit RISC chips, with its first 64-bit chips announced last year.
One of the first devices to include an ARM-based chip was an Apple tablet designed to revolutionise computing on the move. It might sound familiar but it wasn't the Apple iPad — it was the Apple Newton, the handheld computer that reportedly cost Apple $100m to develop in the early 1990s only to be scrapped by Steve Jobs soon after his return to the company in 1997.
Throughout the early 1990s ARM focused on expanding its network of licensing partners worldwide, starting with UK semiconductor company GEC Plessey in 1992, then Sharp, Texas Instruments (TI) and others.
It was through TI that ARM fell into designing chips for mobile phones. TI had licensed ARM's chips for use in the automotive industry. But in the mid-1990s Nokia came knocking, and licensed a TI chipset based on the ARM 7 for use in its first 2G mobile phone, the 6110.
As mobile handset manufacturers seeking to maximise battery life were won over by how much performance ARM's Risc-based chips could squeeze out of a watt of power, more and more of them started incorporating ARM-based chips in their devices. By 1998 business was starting to take off for ARM, its partners had shipped 50 million of its products, it floated on the stock market and the company was able to move out the converted barn that had been its home for eight years.
ARM carved out a niche for itself; its Risc architecture chips weren't the most powerful or the cheapest but they struck a balance between energy consumption, processing power and cost. The balance is still what sells ARM-based chips today, says Muller.
"If you're just interested in performance, power or cost, there's probably another solution but if you're trying to find some balance, then ARM is probably the right choice," he says.
More than just mobile
For decades, Intel has been the main company making the chips we use for computing, from home and office PCs to the servers in datacentres.
But the changing demands of computer users are chiming with ARM's energy-efficient chip architecture. The shift to mobile computing demands...
...smartphone and laptop chips that can run all day without draining batteries, while the growth of cloud services is pushing datacentre operators to seek new ways to reduce the electricity bills that swallow so much of their budgets.
As ARM's ambitions expand, the company finds itself encroaching on territory dominated by Intel, including the Windows PC and server market. This year saw Microsoft Windows and Office running on an ARM-based system for the first time, with the release of Microsoft Surface RT tablet.
Despite the server market being uncharted territory for ARM, Muller believes the time is right for the company to make its move.
The rise of people accessing software and services over the internet plays to the low-power strengths of ARM's Risc architecture, says Muller. Running these services doesn't demand chips with a surfeit of raw processing power, but rather a datacentre where processing power can be scaled rapidly to meet peaks in demand.
Of course Intel x86-based chips can scale, but if a task doesn't rely on individual processor performance, then the lower power consumption of the ARM architecture could start to look attractive to cloud service providers wanting to reduce energy bills. HP and ARM-backed server chip start-up Calxeda estimates it can produce a server at about 35 percent of the cost and 10 percent of the energy consumption of a conventional server.
"In a web-based service deployment it's about scalability," says Muller. "That opens up the prospect of 'Why don't you build those server farms out of more power-efficient components?'"
ARM doesn't expect to become a major player in the server market overnight, nor does it initially envisage itself challenging Intel in areas served by its high-performance Xeon processors. But by targeting web services ARM is going after business with some of the biggest names in tech – Facebook, Google and Amazon – not to mention the rapidly growing web services market. Facebook is already running large-scale tests using ARM and other chipsets.
"Initially we will be targeting those scale-out web workloads. The bit we're going after is the area where there's lots of volume and growth," said Muller.
In targeting the web services market, ARM is acknowledging the performance limits of its chips, which toe-to-toe can't match the processing power of the likes of Intel's Xeon processors. Sergis Mushell, principal research analyst with Gartner's technology and service provider research group, thinks ARM's server chips will be suited to what he calls the microserver market — servers used to perform tasks like web hosting and streaming video, which are I/O intensive but require low levels of processing. The microserver market accounts for about 10 to 15 percent of the nine to 10 million servers sold annually, he said.
However, incompatibility with legacy software will limit the market for ARM-based servers within large enterprise and government, Mushell says.
"You still have governments seeking software engineers for punch cards. Things linger on in corporations and governments for a long time. You do not rip and replace things just because they're more efficient."
"Things linger on in corporations and governments for a long time. You do not rip and replace things just because they're more efficient" — Sergis Mushell, Gartner
Mushell sees more of a market for ARM servers among large web companies like Google and Facebook but said there would still be the burden of managing x86 servers alongside ARM servers, and retaining the skills and parts necessary to run both.
The convenience of having a general-purpose, all be it more power hungry, Intel x86 server, which the company already has the skills and equipment to maintain shouldn't be underestimated, he says.
"X86 architecture, while it's not the most efficient architecture in the world, it's like the one wrench you can have in your toolbox that can do everything. Most people don't use a steak knife, a bread knife, a butter knife — they just want a good sharp knife in the kitchen," he said.
"It is not to say that ARM will not find application and usages. I think the opportunities are there for ARM to compete very aggressively with Intel. It is to say the world will not change over in a heartbeat."
Alongside servers, the datacentre market also provides an opportunity for ARM and its partners to sell chips for storage and networking. ARM has already made inroads into networking, persuading semiconductor manufacturers that sell into the networking market — the likes of Freescale and LSI — to switch to ARM-based chips from competitors.
"ARM will find a home within the datacentre no doubt. The question is where the server is going to be," says Mushell.
ARM's Muller sees its contesting of the server market as a case of 'slow and steady wins the race'.
"We started work on servers four to five years ago. People thought 'Why? What's the point? You're wasting your time'. Product is starting to appear now. Give us another year or two and it will start to take off."
Breaking into the server market won't be easy for ARM. The incumbents have an advantage in that customers are already set up to use existing x86 architectures. But for the likes of web services companies, where datacentre costs swallow up large proportions of operating budgets, Muller believes the trials of transition will be offset by the lower running costs.
"If you can say 'I can make your datacentres 10, 20, 30 percent more power efficient, that really matters to their running costs. Therefore it's worth them putting in investment to go through whatever pain there is," he said.
How ARM designs a chip
The process of designing a chip begins with ARM consulting its customer base on what capabilities and features – the likes of media performance extensions or hardware-level security support — they want to see in future chips.
Using the list as a guide, ARM then defines and locks down the instructions that the chip will be capable of executing. The next step is defining the processor pipeline, the various stages of operation that a processor goes through when executing instructions, which is constrained by factors like the available power and the silicon die area.
Electronic design automation tools then take these higher-level processor designs and translate them into the required arrangement of logic gates within the processor core. Designers then go in and refine the arrangement of logic gates to perfect the design.
Third parties will then license that design, which in general will be a circuit description that sets out how data flows between registers on the chip, and incorporate it into designs for their own chips.
The Intel fightback
The perception might be that, as ARN encroaches into what has been x86 territory, the big boys are just sitting still while it attempts to eat their lunch. But it's by no means a one-way street.
Intel is mounting its own counteroffensive into ARM's home territory of mobile...
...with two major handset manufacturers, Lenovo and Motorola, announcing phones based on Intel processors. Intel, to back up its move into mobile and meet the current demand for energy-efficient chips, is focusing on developing its own low-power architecture. Next year it will launch its Haswell processor, a chip whose draw will be as low as 8W and which Intel claims will consume up to 41-percent less power than its predecessors.
There is a school of thought that, as Intel continues to focus on lowering power consumption of its chips, it will close the power gap between itself and ARM – removing ARM's ace card in mobile computing and its hopes for breaking into the server space.
And while Intel pushes down power consumption, ARM is moving in the opposite direction with some of its latest chips, such as the smartphone, notebook and server-focused A15, which consume more energy than other ARM models in the pursuit of better performance.
Nick Dillon, senior analyst in Ovum's devices and platforms team, believes Intel will face a struggle to make headway against ARM in the mobile market, given how many handsets are built using ARM-based chipsets.
"The ARM chipsets have it pretty well sewn up, for the mass market especially, and there's no real demand from the hardware manufacturers for any other alternative," he said.
"We're moving onto higher-power processors that play to Intel's strengths better, but at the same time those handset manufacturers are going to be very well embedded with ARM — a lot of the optimisations on the hardware and software side are around ARM. It's that classic disruptive technology that comes in underneath and has matured over time and now does the job at a lower price and greater efficiency."
ARM's Muller argues that ARM can retain a technical advantage at the low-power end of the market. But, more importantly, he believes that ARM's key advantage will come from the diversity of its ecosystem: ARM licenses its chip designs to hundreds of semi-conductor companies, which build that design into their own chips before selling them into a broad spectrum of markets. Markets – be it for servers or mobile computing – are better served, says Muller, by the innovation of hundreds or thousands of companies than by the smaller ecosystems of its competitors.
"The reason that ARM's been successful has more to do with the business model than the technology" — Mike Muller, ARM
"The reason that ARM's been successful has more to do with the business model than the technology. [By] licensing a number of different chip companies to do design we have diversity of competition within the ecosystem.
"I don't believe that any one company can design all the chips on the planet. You need to have different people, different specialisations and expertise to scale from little microcontrollers to supercomputers. The business model provides the diversity and that diversity is one of the reasons the ARM ecosystem has become so strong."
Muller claims that ARM's entry into the server market could bring similar diversity to the datacentre, with partners offering systems whose architectures are tuned for specific uses out of the box, be that supporting a search engine or serving SaaS.
Gartner's Mushell points out that while ARM's partners might be able to build many different server chipsets, each targeting a different type of server workload, the difficulty would lie in generating a market for each.
"ARM server folks — Calxeda, Applied Micro, AMD — will make many different chips in the microserver market. The big question is: which one of the them will hit the mother lode?" he said.
"With a market of nine to 10 million servers you've got to turn over a significant amount of that to make it a meaningful market for yourself. You've got to target a sweet spot, a big customer, with a big application, with a big need for reduction. All those stars need to line up in your favour at the same time."
What could count in Intel's favour is its control over the chip manufacturing process. Over the course of decades making chips in its foundries Intel has managed to achieve the highest yields for its chips in the business. It also has direct control over improvements at its chip fabs, while ARM has to rely on innovations by its chip-making partners like TSMC, Samsung and Global Foundries.
Muller said that Intel does have the lead in manufacturing technologies "on some levels" but that when it came to manufacturing low-power, low-cost chipsets ARM's partners like TSMC can ramp to volume more effectively, because "that's what they do for a living".
However any advantage that Muller perceives for ARM and its partners over Intel in manufacturing low-power chips is not guaranteed to remain in place as Intel shifts its manufacturing process to produce more low-power chips like Haswell.
What about ARM's core market? Where next for ARM in the smartphone and tablet space? With the latest smartphones able to stream high definition video and push around 3D graphics with relative ease – will demand for mobile performance tail off?
Not so argues Muller, who believes that users will demand more processing muscle from their handsets as we move from prodding and poking mobile computers to talking to them and even more exotic forms of interaction.
"The PC market seemed to reach some sort of plateau in performance. People went 'My PC is good enough'. I don't think we're anywhere near there yet in mobile phones," he says.
"There are a whole loads of use cases, different ways of doing UI when freed from that keyboard and mouse paradigm.
"Voice recognition is an obvious use case on phones. Could you make that a lot better? Yes, but that requires more compute. The augmented virtual reality stuff that people do with phones? That could also get a whole lot better."
ARM is continuing to push the performance of its chips destined for the mobile market. Earlier this year ARM announced its 64-bit Cortex A50 chip series that it said will deliver up to three times the performance of today's top of the range smartphones.
Looking beyond mobile, Muller says that a big market opportunity for ARM is in the machine to machine (M2M) comms market once the internet of things of things takes off. The internet of things is a vision of the future where low-power, networked computer chips are embedded in everyday items, from lightbulbs to car keys to food packaging to door locks. Giving the objects around us the ability to communicate the likes of where they are and what they are doing, and also react to messages, opens up the possibility of locating a lost pair of glasses with an internet search or a shopping basket that suggests healthier alternatives to your purchases.
Muller sees a massive future role for ARM in providing embedded computer chips for the sensors and the network infrastructure needed to support the internet of things. An example is ARM's Cortex-M0+, whose small size and very low power-consumption is suited to use in the sort of tiny, sub-$1 embedded sensors that will be needed.
"The price point for the silicon is at a point where you could put the required intelligence in for marginal cost and the comms networks and infrastructures are starting to become pervasive," he says.
"It's starting to happen now," Muller adds, predicting a five-to-10-year build-up to it becoming a significant market for the IT industry.
"The PC market seemed to reach some sort of plateau in performance. People went 'My PC is good enough'. I don't think we're anywhere near there yet in mobile phones" — Mike Muller, ARM
"It's a really exciting market and people don't quite know what it is. People don't know what they are going to do with it, what are the services it will enable.
"It's like the early days of the internet, people were talking about stuff that sounded stupid and most of it was, and then other stuff appeared that you never thought of, that turned out to be what everybody wanted."
Whatever the future holds for ARM, Muller is braced for change. He was one of the 12 engineers who has been with ARM from the start, from when it was a barn on the outskirts of Cambridge with only one licensing partner to its name.
ARM's headquarters might be located in the same city but that's about all that remains the same. Today the company employs about 2,200 people worldwide, shipped 7.9 billion chips by its customers and last year reported £230m profit before tax.
"I sometimes still naively think I work for a small company. You used to know everybody, what they did and who their wives were. Now we occasionally have big company meetings and I think 'I don't know who half these people are. What do they do?'.
"It's become a completely different business. That's why it's stayed interesting because the problems today are completely different from the problems 20 years ago," Muller says.