Silicon Valley Watcher - Former FT journalist Tom Foremski reporting from the intersection of technology and media

Intel's Battle With ARM Is About Making Its Future Fabs Viable

Posted by Tom Foremski - September 24, 2012

Many Intel watchers are concerned that the world's largest semiconductor company hasn't been able to break into the smart phone and tablet markets, which will lead to serious consequences.

Intel barely registers a blip as a supplier to smartphone and tablet makers yet it dominates all other microprocessor markets. British chip design company ARM has a near monopoly because they use less battery power and can be easily customized for each product.

How is it that a small 27 year old company like ARM, which reported just $213 million in revenues in its most recent quarter, is considered such a huge threat to Intel, which reported $13.5 billion, nearly 64 times as much?

Even if Intel won over ARM's entire business tomorrow it would boost its quarterly profits only by about $100 million - almost a rounding error on its most recent quarter's $3.8 billion in operating income.

Yet Intel's critics are claiming ARM is a chip giant killer.

Clearly that's not the case, Intel's business would have to shrink at catastrophic rates for many years for that scenario to come true.

However, there is an Achilles' heel weakness in Intel's future roadmap but it's to do more with the brutal economics of chip making.

Intel's clean-room secret...

The secret to understanding Intel is that it's a semiconductor company first; and a microprocessor company second.

It is absolutely brilliant at making chips. Its chip manufacturing technology is stellar - often several years ahead of rivals. Its chip fabs are the most advanced in the world and cost more than $5 billion each to build.

Intel just happens to make microprocessors because they happen to be the most profitable chips to make. If you are great at making chips then you should concentrate on making the most profitable type of chips.

Few people know that Intel did a pivot in its early history. It used to make a living making memory chips until they became commoditized by Japanese rivals in the 1980s and margins plunged. It dumped DRAMs and focused on the far more profitable microprocessor markets which were beginning to spike in demand. And because of its experience in making lots of memory chips, it knew how to make microprocessors in volume.

High yields...

Intel's great skill is that it can make far more useable chips per wafer than any of its competitors. Rivals might have to trash 20% to 40% of the chips on each wafer. But Intel's production yields are in the high 90% range.

Since manufacturing costs per wafer are about the same at each company, the fact that Intel can harvest 30% to 40% more chips per wafer is a tremendous competitive advantage. This is why Intel is so profitable.

But running a chip fab is not easy, you have to run it at near full capacity or you start losing a ton of money.

Chips take about 3 months to make and they are put through more than 300 separate processes. You can't just stop and start production because the machinery is so finely tuned, and so sensitive to outside factors, that it has to be kept in a stable state, running all the time.

Atom to elbow ARM out...

Intel has every reason to be concerned about ARM as a strategic market that might spill into desktop and even into server markets, where it already has a small toehold. Intel's competitor to ARM is its Atom family. While Atom has come a long way and the architecture isn't as power hungry as its desktop and notebook processors, they aren't a match for ARM.

This is a problem because Intel wants to make a lot of Atom chips, not just because it wants ARM's market for strategic reasons, but also because it needs to be making a lot of chips to keep its future chip fabs running at optimal performance.

Can't stop the fabs...

Today's silicon wafers are about the size of a dinner plate at 300 mm in diameter. Intel plans to shift production to 450 mm silicon wafers -- the size of a small coffee table. That's about 40% more chips per wafer at a fraction of additional production costs.

With the massive silicon wafers, and the move to smaller geometries below 22 nm, which squeezes more chips into the same area, Intel's new fabs will be able to make mind blowing numbers of chips.

But before Intel spends $7 billion plus on 450 mm fabs it has to ensure that it will have mind blowing demand for its chips. It needs high demand because it has operate its fabs at near full volume and 95% plus yield rate.

If Intel were making hundreds of millions of low cost Atom chips, replacing all those ARM chips, it would have the volumes it needs to keep its fabs running smoothly. That means it has the means to manufacture all of its highly profitable server and desktop microprocessors, at the lowest possible production costs.

There's another very important element in this story: a high volume chip like Atom is essential in helping to quickly ramp up production of all of its chips.

High volume products qualify chip production machinery and fine-tune hundreds of complex chip processes. They are essential to Intel being able to maintain high levels of usable chips per wafer.

This is why ARM markets are important to Intel and why it must defeat it with Atom, or something similar - it needs the numbers. Not because it needs the microscopic revenues that feed ARM.

Intel needs those markets for the sheer volume of chips that device makers such as Apple and others buy, so that its hugely expensive future 450 mm fabs will be viable producing its high profit chips.

Story link | Subscribe free | Categories: A Top Story, INTC


The Holmes Report names Tom Foremski one of the top 25 Innovators of 2013.