Thursday, March 6, 2008

Process: Intel’s competitive advantage.

If there is one reason why Intel cannot be matched by any other chip powerhouse, it’s because of their fabs and process prowess. They are at least a technology node ahead of the competitors. This difference will start becoming more significant as digital chips transition to smaller geometries.

Right now, it may look like Intel is mostly focused on PC /laptop microprocessor chips. That would be a dangerous fallacy. Intel has entered SoC space in a big big way. That’s how you cram multiple cores into a chip. That’s how you integrate Wimax. Bluetooth, GPS, RF might seem like separate markets. But with more integration and convergence of feature sets on a single die, Intel would be increasingly targeting those markets.

Their fab prowess would enable them to crank out high volume, high yield, small geometry products, and squeeze the competition on margins. Just look at what they are doing to AMD. This is exactly what they did to Cirrus Logic in the late 80s when they integrated the graphics chip into their core. Integration is the bane of all feature specific fables vendors.

This quote from Eric Savitz's blog is noteworthy.

Speaking at a meeting with analysts in Santa Clara this afternoon, Intel’s (INTC) Bob Baker, SVP of the company’s technology and manufacturing group, said the company has begun preparing for a switch to 32 nm process technology, even as it ratchets up production at 45 nm. And the company is looking ahead to 22 nm production, and smaller. Meanwhile, he says the switch to 45 nm is “ramping on track.”(This clearly was the geekiest of the afternoon sessions.)

Baker notes that the move to 45 nm process technology from 65 nm has allowed for higher yields, more good dies per wafer and lower cost chips. With the Xeon server processor, he notes, the chip can generate 38% more performance at the same power. With 32 nm, the trend should continue.

You would expect that smaller geometries would result in lower yield, or at least a transition phase of ‘yield learning’. Instead, their 45nm node is cranking out higher yields than the 65 nm already, which is further proof of their process prowess, and serves to reduces their cost structure.

Intel’s competitive advantage lies in transitioning to lower geometries. A lot of their process knowledge is in house, proprietary, and cannot be off-shored to an Asian foundry. Moreover, at lower geometries, leakage power becomes a dominant component of the chip. By having an in house fab, they are in a better position to tweak parameters to reduce or mitigate standby power consumption. I know Intel has always been notorious for churning out massively pipelined, high frequency, high power chips. But that’s because they decided to do that from a high level architectural perspective. Now that Intel has realized the future wave of low power applications, the power consumption on their new processor cores will increasingly be the number to beat.

The fabless model has worked great for the past two decades. As long as the process could be replicated easily, the companies could offshore the more cyclical portion of the business abroad, and focus instead on the more value added design flow. This worked in digital design, as tweaking process parameters offered little in extra functionality. By aggressively following Moore’s law, digital chips lowered costs and added functionality dramatically over the past two decades.

The caveat is the huge sunk cost at every technology node. Intel has a high fixed cost, and essentially reduces it’s variable cost through process optimization, yield enhancement and smaller geometries. Fabless vendors do not have to worry about the fixed sunk cost component. Assuming Intel churns out enough chips at a lower variable cost, their total cost would be lower than a fables vendor. At least that’s been their do-or-die strategy.

The Asian foundries would need to significantly augment their R&D in order to successfully transition to lower geometries. TSMC has been the best equipped to do so. Because of the prohibitively expensive start up costs, you are looking at increasing collaboration amongst fabs and fables companies in terms of costs and resource sharing.

A few years back, Silicon-on-Insulator substrate looked like the substrate of choice for better performance. Intel plodded ahead without going the SOI route, and instead introduced features like strained silicon. AMD, on the other hand, is now stuck with integrating their SOI chips with ATI’s regular substrate chips. Advantage: Intel.

Expect them to outperform and dominate the marketplace for many more years to come. Intel is a very strong buy in the current uncertain environment. I know right now I do not have a way to quantify this edge into a number. At the very least, Intel has the size and the economies of scale. Given that, it should trade at a premium valuation and multiple to its peers and the market in general.