rlp
The semiconductor industry
Space invaders
Jan 7th 2012 | SANTA CLARA, CALIFORNIA, AND CAMBRIDGE | from the print edition
LAS VEGAS is a city of fast bucks, fast food and fast marriages. It could also
be the place where a long war was declared. On January 10th Paul Otellini, the
boss of Intel, will address the International Consumer Electronics Show (CES),
a vast gathering of gadget-makers, sellers and aficionados in Sin City. He will
introduce a phalanx of products showcasing the chips the world s largest
semiconductor company most wants to hype.
Up on the stage with Mr Otellini will be not just PCs of the sort that the
company has powered for decades, but also new slimline PCs known as ultrabooks
, which are being made by the likes of Toshiba and Hewlett-Packard (HP), and
even a couple of smartphones. They represent the front-line of an army of
Intel-powered kit going into battle against smartphones and tablets which use
processors based on designs from ARM, a British firm.
Intel and ARM, pretty much as different in size and approach as competitors can
be, have carved up most of the world of microprocessors the most lucrative bit
of the $313 billion global semiconductor market between them. Each has a well
defined patch in which it is pre-eminent. Intel bestrides the market for the
microprocessors at the heart of PCs and servers like a colossus; ARM s legions
hold sway in the wide open spaces of the mobile market, having expanded without
hindrance from their home turf in mobile phones to the booming world of
tablets. Neither side has shown the stomach for more than the occasional raid
into the other s domain.
The wares on show at CES provide the clearest indication yet that Intel is
escalating hostilities. It is trying to break into the rapidly growing
smartphone market (it already makes chips for some tablets). ARM, meanwhile,
has set its sights on the server business, where its low-energy chips should
appeal to customers worried about high electricity bills. And more of its
processors are likely to find their way into PCs in the coming years too.
The battle is not just about dividing up territories already occupied; it is
also about finding new lands to conquer. Both firms are keen to stake claims on
the largely uncolonised and still somewhat notional terrain known as the
internet of things : the myriad processors in industrial machinery, consumer
goods and infrastructure, ever more of which will communicate with each other
and with distant computers. Cisco, a giant American maker of networking gear,
estimates that by 2015 there may be almost 15 billion internet-connected
devices, up from 7.5 billion in 2010. Whereas the market for more phones and
other personal computing devices is limited by the number of persons the planet
has to offer, things, being more numerous than people, provide a lot more
long-term room for growth.
Intel, founded in Silicon Valley in 1968, is responsible for many of the
advances that have made today s semiconductors possible. It employs almost
100,000 people and mints money. In the first nine months of last year, the
company generated $40.1 billion in revenue and $13.2 billion in pre-tax profit.
Admittedly, in December it cut its revenue forecast for the fourth quarter of
2011 by 7%, to $13.4 billion-14 billion, because floods in Thailand had
disrupted the production of hard-disk drives and hence demand for its chips.
But that was a hiccup; Intel has no real rival in the market for chips that
power PCs and servers. Advanced Micro Devices (AMD), the distant second in
both, has been struggling. It replaced its chief executive in August after a
seven-month search and laid off 1,400 workers in November.
Empire and foundry-nation
Intel s business has been inextricably entwined with the steady fall in the
cost of processing-power known as Moore s Law, which is named after one of the
company s founders. Ever better chips mean ever increasing sales, which mean
ever more money with which to build ever better factories for ever better
chips.
The firm s top brass attributes much of the company s success in harnessing
this virtuous circle to the fact that it both designs and makes its chips, with
ten chip factories ( fabs in industry argot) operating and two more under
construction. Brian Krzanich, Intel s manufacturing head, says this means it
can bring chips to market faster and with fewer faults than rivals who use
external firms, known as foundries , to make their chips. This helps explain
why it dominates the market for high-performance processors in PCs and servers.
ARM s chips are, by contrast, designed to economise on energy rather than to
maximise processing power. And it makes none itself, instead selling licences
for its semiconductor designs or its architecture (a recipe from which
licensees make their own designs). Licensees pay a fee and a royalty of 1-2%
per chip.
In essence, ARM provides a development base on which others build. The costs
are shared, as are the resulting revenues and profits. ARM expects to recoup a
chip s development costs from the sale of the first ten licences. Royalties,
which flow later, make up just over half its revenues.
Next to Intel s leviathan, ARM is a shrimp. It employs just 2,000 people. It
reported revenues in the first nine months of last year of 354m, or $568m, on
which it made a pre-tax profit of 107.3m (see table). But it lies at the heart
of a huge ecosystem of companies: a federation, perhaps, as opposed to Intel
s integrated empire. It has 270-odd licensees with 830 licences. Between them
they shifted perhaps 8 billion ARM-based semiconductors in 2011, half of them
in mobile phones and mobile computers, the other half embedded in consumer
items and elsewhere. According to IDC, a research firm, the market for
PC-powering chips that use Intel s x86 processor architecture, which Intel
dominates, was about 400m last year.
The ARM federation comprises not only chipmakers but also designers of
chipmaking tools, devicemakers and software companies. ARM uses collective
insights to design chips it thinks its partners will need; and they in turn
shape their products with ARM s processors in mind.
The dead past
There are intense rivalries within the federation, for example between NVIDIA
and Qualcomm, leading makers of graphics chips. And the dividing line between
federation and empire is not always clear. Design-tool firms such as Cadence
Design Systems and Synopsys work with both ARM and Intel. So do Microsoft, HP,
Apple and others. Intel itself is an ARM licensee. But that is for the most
part because those businesses straddle the divide, not because the divide is
not there.
ARM s impressive position in the mobile-device market was born out of what
seemed at the time like a string of failures. In the 1980s, when the trend in
chip design was to make the hardware on processors capable of ever more varied
and subtle types of calculation, Acorn, a then-marginal and now-defunct British
computer-maker, had a niche in designing chips good at carrying out only a few
types of calculation, but which did so very quickly. This
reduced-instruction-set computing (RISC) approach requires software that can
make up for the limitations of the chips, but uses less power than other
approaches.
In 1990 Apple, struggling itself, needed a chip for its Newton, a personal
digital assistant that was to restore the company s fortunes. It liked Acorn s
chip designs: the two formed Advanced RISC Machines as a joint venture with a
chipmaker. The Newton was a flop, and Acorn was wound up in 1999 the year that
ARM floated. Apple, which today has a cash pile of $81 billion, sold its 43%
stake because it needed the money.
When the mobile-phone market took off, ARM s parsimonious processors were ideal
for products in which battery life is at a premium and high-power chips and the
fans that cool them are therefore not an option. Today more than 95% of the
world s mobile phones contain an ARM-based chip. Tudor Brown, ARM s president
and one of its founders, adds that the shift to systems on a chip single bits
of silicon that package together not just one or more central-processing cores
but also graphics processors and other accoutrements also helped ARM. Its
stripped-down processing cores play well with others.
As mobile phones have become cleverer as well as commoner, ARM has gained
again. Cleverer phones use more and pricier processors. The average number of
ARM-based chips in a phone went up from 1.5 in 2006 to 2.5 in 2010. NVIDIA s
new Tegra 3 system on a chip for smartphones and tablets contains five ARM
cores as well as NVIDIA s own graphics-processing unit. And more, dearer cores
mean more royalties. A smartphone can in the best case bring the company eight
times as much in royalties as a basic phone, a tablet computer 11 times as
much. This shift is far from over (see chart).
There is no such growth in the market for new PCs, which increasingly depends
on the replacement of old computers by new ones with better, but not more
numerous, processors, and makes up ever less of the total market for
microprocessors. No wonder Intel is desperate for new territory, and willing to
fight for it.
It is bringing two powerful siege engines to the field. One is its low-power
Atom line of chips. The latest of these, code-named Medfield, is already in
production and will almost certainly feature in the phones Mr Otellini shows
off at CES. A study by Jefferies, an investment bank, says that Medfield is on
a par with several popular ARM chips when it comes to processing bang for the
energy buck. And Intel is working with Google to ensure that the search firm s
Android mobile operating system runs smoothly on Atom chips.
The company does not intend to stop there. The width of the circuitry on a
Medfield chip is a mere 32 nanometres (nm), or millionths of a millimetre.
Using a three-dimensional chip design Intel plans to shrink that even further
over the next couple of years, to 22nm and then 14nm, and sell chips that beat
the competition on both energy-efficiency and performance.
The wedge
The other thing Intel is counting on to help it succeed is new leadership. In
December it put Mike Bell and Hermann Eul in charge of a streamlined internal
unit focused on cracking the mobile-device market. Mr Bell, who joined Intel in
July 2010 after working at Palm and Apple, says the firm has hired more people
with a telecoms background and assembled a team to develop software to help
phonemakers get the most out of its chips. Intel has also acquired businesses
such as the wireless operation of Germany s Infineon Technologies to help with
systems-on-chips. Mr Bell is confident that combining all this with the company
s manufacturing might will make it a force to be reckoned with. We can move
this army en masse over to our mobile effort, he says.
But even if the chips prove effective, Intel will be hard put to build a phone
business out from its beachhead. Getting processors on a technical par with ARM
s, says Michael Rayfield of NVIDIA, is the easier of the two hurdles. The
software hurdle is staggering. Firms that have invested in ARM s
silicon-and-software combination will be reluctant to give Intel s chips a
chance until they are sure they can handle all kinds of software applications
as smoothly as ARM s. Intel will also struggle to match the extensive and deep
relationships its rivals have in the phone arena. The complex reciprocal
relationships that make up ARM s ecosystem, says Mr Brown, the company s
president, are probably our biggest barrier to entry .
The fragmented mobile-device market also requires lots of different
system-on-chip configurations, which Intel will find a challenge to match. And
makers of tablets and smartphones may be reluctant to commit themselves to an
architecture dominated by a single company that makes its own processors. With
ARM, when you are tired of Qualcomm you can go to NVIDIA or another company,
says Linley Gwennap, the boss of the Linley Group, a research firm. But in
Intel s case, there s nobody else on its team.
Birth of a notion
While Intel is mustering its forces to attack the mobile-device business, it
also faces an assault on its own redoubts. For years the firm has had an iron
grip on the PC arena thanks to Microsoft s decision to design successive
versions of its Windows operating system specifically to run on the x86
architecture. But last year Microsoft said that the next version of Windows,
which it wants to look and feel the same on mobile devices as on desktops, will
work with ARM chips too one of a number of cracks in the Wintel alliance .
This could encourage more firms in the ARM federation to try their luck in the
PC market, though Intel s extensive product lines and deep relationships with
PC makers make it very difficult to beat.
ARM itself spies a bigger opportunity in another of Intel s dominions: servers.
The server market is hitching a ride on the spread of smartphones, tablets and
other devices. The more data is sent to and from the cloud by them, the more
social sites they need endlessly to update, the more servers are required. And
the data farms in which these servers sit have a prodigious thirst for
electricity, a problem that ARM s chips were created to solve.
In November HP announced a project ambitiously named Moonshot to develop
servers using ARM-architecture chips made by Calxeda, a Texan company of which
ARM owns 25%. The chips are less powerful than their Intel equivalents. But
they are less thirsty and need less cooling, so whereas a standard rack (a
man-high cabinet with about a cubic metre of volume) in a data centre can only
house a few hundred Intel server chips, Calxeda thinks it can cram in almost
3,000. With 100 racks in a hall, you re talking megawatts with normal
servers, says David Chalmers of HP. Moonshot is designed to use a tenth of the
power of current server systems and cost 60% less.
Moonshot and other low-energy servers could appeal to, say, social media
companies and other web-based firms which do not need to carry out very
complicated processing which benefits from the architectures of more complex
chips or to do it very fast. But Reuben Miller of IDC thinks this segment is
likely to be no more than 10-12% of the overall server market by 2015. And ARM
s share of even that smallish slice may be modest to begin with. Just as
phonemakers are used to things working in an ARM-ish way, most server software
is written for Intel s chips, and reaps the benefits of its 64-bit
architecture, which makes accessing lots of memory, among other things, much
easier. ARM s architecture uses a 32-bit standard, and though the company
recently unveiled a 64-bit version, no chips making use of it are yet
available. Until they are, says Warren East, the firm s chief executive, we
can t even address probably 75% of the server market.
Meanwhile Intel isn t standing still: its investment in more energy-efficient
processors, such as those of the Atom line, can reap benefits in servers as it
does elsewhere. HP s Mr Chalmers, happy to work with both sides if it gets his
clients the servers they need, expects to announce servers based on Atom chips
and something similar from AMD this year.
Each side, then, seems to have defences against the incursions of the other.
But that does not mean the war will end in stalemate. Intel is more vulnerable
than it looks, for several reasons. One is that despite demand from emerging
markets such as China, the PC market is unlikely to grow anywhere near as fast
as it has done in the past. Intel s aggressive promotion of ultrabooks seems
like a somewhat desperate attempt to inject excitement into a category that has
lost momentum.
What is more, the well-fortified world of Wintel provided the PC market with
relatively juicy margins. In smartphones and tablets Intel will find itself in
a much more brutal competitive environment in which the advantages of its
integrated approach to design and manufacturing may well be outweighed by those
of agile competitors used to servicing a wide range of companies with lots of
different products.
The chips, like dust
Another big test for Intel will be the small but fast-growing market for
embedded chips the sensors and microcontrollers which will, as they become able
to talk to each other, make up the internet of things . Renesas Electronics, a
Japanese company, holds the largest fief in this fragmented terrain. ARM also
has a worthwhile chunk of it. But it is a lawless and fragmented territory,
largely served by in-house designs and software that both Intel and ARM see as
ripe for replacement.
In 2009 Intel splashed out $884m on Wind River, a firm that specialises in
software for things that you might not expect to need any, in order to give its
efforts in the embedded-chip market a fillip. It has since been able to ink
deals with car companies, makers of digital signage and other firms that put
chips into their various wares. The company says that annual revenues from
embedded-chip sales are now running at $1.5 billion, and it expects these to
grow by 25% in each of the next three years.
Yet ARM s flexible business model, allowing for lots of different chips for
different applications, and its happiness in lower-margin businesses, may well
give its federation an edge in this business too. Its long experience of
producing low-energy chips should be another advantage. Tiny embedded
processors will not use huge amounts of processing power, but power
consumption will become more and more critical, says Ganesh Ramamoorthy of
Gartner, a research and consulting firm. ARM already makes a quarter of its
revenue from embedded chips. And for the newer embedded processors in what the
company calls the Cortex-M family, nine-tenths of the licences so far sold have
yet to lead to products, and thus royalties. Having your own fabs can be handy.
But when it comes to invading virgin territory quickly, having lots of allies
to help you is absolutely fabulous.
This article fails to mention a major reason in Intel's dominance over the PC
market
The article misstates Intel's competition. Intel is the world's best baker.
Their competition used to be (IBM's) Global Foundries (before AMD moved to)
TSMC and their cousins in Taiwan, and now (the up and coming giant in the room,
larger than Intel) Samsung (and its Chaebol) in Korea. A given instruction set
architecture and RISC v. CISC is largely irrelevant save for the developer
community it supports.
System-on-Chip implementations (as required for mobility and sipping power)
make the general purpose processor largely secondary to everything-else. This
change in orientation alone is a major challenge to Intel. But Intel has a two
generation lead in process technology over all their competitors' so the
opportunity is Intel s to seize. But, like Microsoft, they need to manage this
opportunity without destroying their (past) core business in the transition
i.e. eat their own children at a profit maximizing rate while maintaining their
fabrication technology lead - which requires 10x the investment of a technology
follower who can ride on their coattails a generation or two back while doing a
more focused, likely better job on everything-else.
At semiwiki (www.semiwiki.com), a social media area dedicated to the semi
conductor (SC) industry, we (the bloggers) have written numerous blogs about
this battle, or even war, between ARM and Intel.
The first blog was:
http://www.semiwiki.com/forum/content/
419-arm-vs-intel%85performance-power-os-support-ip-library.html
As said, numerous blogs have followed, all of these being written by SC
industry veterans, reading it may help the reader to find technical
informations not available in this excellent article from The Economist. Which
make sense as the article if for a large audience, not necessarily up to date
with the latest, deep technical trends of SC industry. Feel free to visit
semiwiki!
As someone fairly knowledgeable in the exact technical matters, ARM has always
had the design with an advantage in power usage for computing performance.
Recently this has been ramped up enough to take hold of the new market of
smartphones and tablets, who's primary need has been power management.
However, Intel has always had the computing performance for cost layout
advantage. Frankly, it still does by a good margin. Which gives Intel an
advantage in any device that doesn't need to worry as much about power usage.
Intel is also pushing faster and further into giving a good ratio of computing
performance to power usage than ARM is pushing into Intel's territory.
In the long run new battery technology, an incredibly fast paced and highly
invested section of R&D (more for the sake of the electric car replacing gas
than anything) will push mobile devices away from needing as much power
efficiency as today's devices need. People care if their phone lasts one day or
two before charging, they'll care a lot less whether it will last ten day or
only six.
Thus, in the long run Intel has the advantage. If they can keep ahead of the
game in terms of computing power to cost they'll win out in the end, and ARM
may well be relegated to history within a decade or two.