POET Technologies Inc.

· Intel's fabs need more autonomy to better align with growth markets, rather than being primarily fettered to markets Intel is currently successful in.

To many people, Intel (NASDAQ:INTC) has been a fab company that also designs some processors, so strong is that part of the company. Even when listening to the investor meetings, Intel management constantly points to their fabrication technology as a primary differentiator. However, the times have changed, and Intel's situation with it. Are the fabs still the drivers for success?

The quick answer is no. Clearly, where Intel is successful, they would be even if the fabs were the same as their competitors. And where they are unsuccessful, the fabs obviously did not create success. This in no way implies fabrication technology is meaningless, just not generally salient.

There are several reasons for this. Interestingly, Intel management misuses the term Moore's Law, which died a while back. Initially, it meant the number of transistors in a given area would double every 18 months. When that failed, it was changed to 24 months. Now Intel just uses it to mean at some point transistors will get smaller, and hopes no one notices they changed the meaning. But, the reason it died also points to why fabrication technology advances matter less now.

For one, the advantages of a "shrink" do not even approach the benefits they used to. In some cases, some parameters even get worse with a shrink. This in itself would be enough, but in November of 2014, Intel took to misleading the community by misrepresenting the competitor's density based on published papers that were no longer valid or relevant. It's not clear what their motivation for doing this was, as it is easily exposed, but it does show a certain amount of concern on their part. This is understandable, since Apple's A8X has 48% more transistors for the area than Intel's Core M, despite being on an ostensibly less dense 20nm node, compared to Intel's 14nm node.

Let's take a step back and look at the three basic characteristics of a fab process - performance, power efficiency, and size/density.

Performance can be simply defined as how quickly the transistors can switch. This relates directly to the clock speed a processor can reach. Intel tends to place primary importance on this, although maximum clock speed has not improved since Sandy Bridge on 32nm, which is still considered by many the best overclocking part by Intel.

Power efficiency is not as simple as it might initially seems. It relates to how much power the device will use. But, it's important to keep in mind, power efficiency at one data point does not necessarily mean it is more efficient at all. For example, when Intel moved to 22nm, for most clock speeds the Ivy Bridge processor was more power efficient than Sandy Bridge, a 32nm part. But, as the clock speeds increased beyond 4 GHz, the power use on Ivy Bridge climbed quickly, whereas the 32nm node was more linear.

While power efficiency is important, it has to be balanced against gains in performance and density. Clearly for mobile parts power efficiency is very high priority, but less so for the desktop performance segment. As mentioned, although Intel has not been able to improve top clock speed to any real extent with 22nm, it was able to maintain it in the power envelopes they want to sell their chips in. Contrast this with AMD's (NASDAQ:AMD) move from 32nm to 28nm, where density was given priority over performance, with a consequent significant drop in maximum clock speed. The good old days where one gets a lot of both are clearly over, and companies have to make more difficult choices.

We don't currently know where the 14nm clock speeds will peak. We do know that the power characteristics at low clock speeds are quite good, and have allowed "big" processors in power envelopes that would have been essentially impossible in previous nodes.

The last criteria is density, which is where Intel continues to struggle, and place the least emphasis on. This is understandable from Intel's perspective, since it sells so many high-margin parts, and owns so many fabs. If the parts are bigger, it is something they can live with, rather than unduly compromise performance or power use. Even so, the cost of wafers for 22nm and 14nm have gone up 30% each generation, so there is still a desire to reduce size, even if it is generally a lower priority than performance or power use.

Of course, not just Intel makes these choices. As mentioned AMD, with GloFo, made the choice to focus more on density on the 28nm node, compared to the 32nm, with the result that the processors could not clock as high. Density can have significant positive performance implications, but only in certain scenarios like GPUs, where more transistors improve performance almost linearly, due to the parallel nature of the workloads they process. In the above example, you'll note the Kaveri based APUs have significantly more powerful GPUs, which is largely due to the improvement in density. With CPUs, this does not apply nearly as much, and the clock speed edge is more salient, especially for Intel.

I would be remiss for not pointing out that the names companies use for their processes is not directly related to anything in the actual transistor. For example, there's nothing 22nm in Intel's so-named process, or anything 20nm in TSMC's (NYSE:TSM) planar process, although they do try to convey some meaning about the general characteristics of the technology. So, although Samsung (OTC:SSNLF) and Intel are both using "14nm" processes now, they are not the same. They just chose the same name to frame the technology.

Intel's biggest problem is that the explosion of mobile parts has caught the company flat-footed and unable to respond in an effective way. Although able to move 46 million processors into tablets in 2014, this came primarily due to extraordinary pricing pressure, as the company lost $4.2 billion in that segment for the year. In Q4, Intel actually generated negative revenue in the MCG, so they gave a processor and paid companies to use it. Intel had virtually no sales in phones in 2014, although their SoFIA line should come out in 2015, and could gain some share in the low-end phone market. Perhaps revealing is ARM (NASDAQ:ARMH) has said they do not consider Intel's competitive threat to have any impact on their earnings in 2015.

Intel's inability to take advantage of the new markets is very dangerous for the company, because of the increased revenue for competing processor design and fabrication companies. As these companies continue to grow, and Intel fails to, there's a real threat of Intel's competitive situation deteriorating drastically. Already we see Apple (NASDAQ:AAPL) pushing fabrication companies to move to new nodes more quickly, and we also see how competent their new designs are. Qualcomm (NASDAQ:QCOM) has even announced they will be pursuing the server market, a market Intel currently dominates.

Intel management does not seem to see threats before they happen, or understand them clearly, perforce they take half-steps that are generally ineffective. Even now it seems they are not quite ready to take more drastic steps to redress the situation.

One of Intel's half-steps is to start allowing other companies to use their fabs. But, they arrived at this too late, and it is still not clear if they are committed to it. At this point, they are not competitive with Samsung or TSMC, and there's a question of whether they ever will be. Intel's processes are for their high performance parts, whereas both TSMC's 16FF, and Samsung's 14nm are both geared more for the mobile SoC market. Intel also does not have the development tools, or experience of companies like TSMC, nor the number of processes. Intel's fabs work on one process, then essentially ignore it and move to the next. TSMC generally has several processes for the same node (like 28nm), and also updates older nodes constantly with newer technology from newer nodes. To illustrate these points, in Q4 earnings, TSMC announced "in addition to the recently announced 55ULP ultra-low power technology, 40ULP, 28ULP technologies for ultra-low power application,... we are also working on 16ULP technology development. This 16ULP design kit will be available in June this year. It will be suitable for both high-performance and ultra-low power or ultra-low voltage, less than 0.6-volt application." Compare this to Intel's one node at 22nm and 14nm, and not updating older nodes with new technology, and the difficulty Intel faces in being a competitive open fab becomes all too clear. A fundamental change has to take place in the way the fabs do business, and so far we're not seeing it.

TSMC is not standing still either, and unlike the suddenly secretive Intel, has announced investing in "10-nanometer tools and facilities to be ready for customer product tape-out by end of this year." Will TSMC beat Intel to 10nm? Intel claims the delays they suffered at 14nm will not be experienced at 10nm, as they have improved their testing. But, that there is even the slightest doubt about this indicates just how things have fundamentally changed due the rapid growth in mobile revenue, and how Intel's position has become more precarious by failing to capitalize on it.

Interestingly, the only companies that would benefit from a higher performance process, like AMD, IBM, and Oracle would not be likely business partners. So, outside of relatively minor designs, like Altera, there's not much current opportunity for Intel. They need to take more than half-steps, after the fact, to be successful.

One such step would be to give the fabs more autonomy, whether it be by spinning it off completely, or keeping it as a wholly owned subsidiary largely independent from the rest of the company. This would cause the type of changes necessary at both the fabrication and design levels of the company.

The fabs would be much more inclined to court other companies, and would develop processes and tools more in line with that goal. This would also help their own designs, like phone processors, where cost is critical. It would also allow relationships with direct competitors, which is always healthy in the long run. Samsung and Apple prove this can and does work.

With only Intel as a customer, the fabs do not have the incentive to create an environment where SoCs can be made efficiently for other customers, or for their own mobile processors, like SoFIA. Since the process for a high-end processor and a low-cost mobile SoC are not the same, Intel is caught in a bit of a Catch-22. They will not generate significant fab business with their current, expensive process or have a cost structure that allows them to sell their own chips profitably in markets like low-end phones. Without committed customers, or strong sales of their own parts, the cost of developing a more dense node likely is difficult to justify.

As an autonomous unit, the fab group would see Intel's x86 line as a strong, but degenerating business, and would feel compelled to align more with thriving companies, in growth industries. Last quarter, over 3.5 billion ARM processors shipped, with 20% growth year over year in revenue for ARM. Intel can not even hope to approach those numbers, and with new ARM processor releases, and the move to FinFets, the delta between Intel and ARM-based processors can only grow. While its competitors are generating enormous profits from this growth, Intel is resigned to chasing a will o' the wisp with programs like contra-revenue. It does not address the primary problems Intel is facing, but instead obfuscates it, particularly with management making predictions that have no real evidentiary support, or logical basis.

But, this would be very new for Intel, as there has been little need in the past for a higher density node, and there was less opportunity to benefit from it. The market has changed, Intel has not in any material way. Thus, their continued dependence on a shrinking market, while their competitors show dramatic growth, and in some cases (Samsung) have higher CapEx for fabs than Intel. TSMC grew closer in 2014, and will pass Intel in 2015. Intel has to decide if it really wants to win in this market, and if it is willing make the dramatic changes it requires. Half-steps like offering a poor CPU design for the target market, or offering a process that is not well-tuned for what the mobile market is asking for, has little chance of long term success, even if it gives the illusion of some sort of progress.

Editor's Note: This article discusses one or more securities that do not trade on a major exchange. Please be aware of the risks associated with these stocks.

http://seekingalpha.com/article/2940506-intels-fabs-at-the-crossroads?auth_param=rdtll:1aemfnr:3c2db0e819c10f449e514008d7114579&uprof=46&dr=1