- The bleeding edge of semiconductor manufacturers is one - Intel.
- The leading edge of semiconductor manufacturers is in disarray.
- It is early to buy Intel.
Intel (NASDAQ:INTC) is the largest semiconductor company in the world by dollar volume the company is also the leading semiconductor technology company by one to three nodes depending on the definition.
The state of affairs in the leading/bleeding edge of semiconductor technology is so full of rumors and self-serving "information" that even some of the "experts" appear confused.
I'm going to try to sort this out, as much for myself as for the readers.
Leading edge process: 28nm with or without High K Metal Gate.
Bleeding edge process: TriGate (Intel's term for finfet) certainly at 22nm going to 14nm and beyond.
For the purposes of this article we will leave out the memory manufacturers, Micron and Toshiba/SanDisk who also operate at the leading edge of semiconductor technology, but for much simpler devices.
At about $48 billion worth of mostly bleeding edge business, at what appears to be 78% gross margin, Intel has a commanding lead in the financial aspect of this semiconductor end game. Intel is the only company shipping finfet technology today. That technology is 22nm HKMG that is producing corporate gross margin of 65% while dragging along the $5-billion-losing mobile business. This is an indication of very high yields on the 22nm TriGate process. Intel has been shipping hundreds of millions of complex TriGate devices for over two years.
14nm TriGate devices are now in production at Intel for availability in consumer devices by the 2014 holiday season.
When the first 14nm based PC is in a consumer's hands, Intel will be three generations ahead of the nearest competition, two shrink nodes and the TriGate process itself.
There has been something of a delay in the introduction of 14nm at Intel, which is not surprising when you think of forming functional elements with dimensions of 14 billionths of a meter in size. This delay could be a technical problem, or, as is my opinion, there is no reason to release 14nm products when 22nm is virtually years ahead of the nearest competition and is yielding magnificently.
TSMC is the largest for-contract semiconductor foundry in the world. The company enabled the fabless business model and is involved in virtually every semiconductor product sold by companies who don't own their own fabrication facilities.
The founder of TSMC is Morris Chang, previously a high level manager at Texas Instruments (NASDAQ:TXN), Morris is arguably the most honest and truthful high level member of the semiconductor management universe.
TSMC is about $24 billion in annual sales with about $9 billion of that in the 28nm process which supports virtually all the FPGA, graphics chip and high end mobile business in the world, through fabless clients. $15 billion of the TSMC business is in processes of 45 nm and greater (not everyone needs the crazy technology).
Unfortunately for TSMC, the advent of complex, high performance, low power smartphone chips has brought Intel into direct competition with TSMC's clients and, therefore with TSMC themselves. Prior to this TSMC lived in a world that Intel didn't care about. Not anymore. If TSMC is to compete with Intel they must spend like Intel and that could be ruinous for TSMC. The cost to build a competitive bleeding edge fab could run to $10 billion requiring $20 billion worth of revenue to justify the expense. At $9 billion total leading edge (not bleeding edge) business, the problem becomes obvious.
Intel, on the other hand, has the $48 billion recurring x86 business to continually fund ever more advanced manufacturing technology. Personally, I can't see a way for TSMC to continue to chase Intel, staying two or more years behind, in the "node" race.
Two years ago Morris warned his customers that the inability of TSMC to keep up with Intel could produce consequences "too horrible to imagine" (for his clients). At that time he urged them to source other foundries for part of their requirements. I can't imagine who he might have had in mind since TSMC is by far the largest and best funded of the foundries. The link to this article has since expired, but recently Morris has made a statement recently that TSMC will lose market share on 14nm in 2015, but regain it in 2016 and beyond. That seems like a strange statement since TSMC will not have 14nm market share to lose in 2015
Even Dan Nenni, the self-proclaimed expert on all matters fabless, appears confused. He thinks Morris is talking about Intel as the foundry that will take the 14nm business. If that is the case, the business would never return to TSMC because Intel will never be short of capacity.
Dan's reasoning doesn't seem to hold water since somehow he must think Intel skipped 20nm on the way to 14nm. Apparently Dan has forgotten about the Intel 22nm TriGate. Dan, you know better than that, it's Samsung. Furthermore, for Dan to be right, Intel would have to agree to build parts for Qualcomm (NASDAQ:QCOM) and that ain't gonna happen.
The awful bottom line for TSMC is that they will go into 2015 with no 14nm business and maybe no 14nm process, or both. I can't remember a semiconductor company that missed an important node and make it up later, especially when there needs to be $20 billion worth of sales in place to justify the expense of 14nm.
I think Morris is gently throwing in the towel on 14nm. Competing with Intel is a no win situation, trust me, I know.
Samsung certainly has the financial capacity to match the Intel spending on advanced technology nodes if they choose. Whether they have the technical know-how is something of a question. Intel management has often said that there will likely only be two manufacturers standing at the end of the party, presumably they meant Intel and Samsung.
Samsung has a great deal of internal requirements for leading/bleeding edge technology - they are also running a business and if in-house sourcing cost much more than outsourcing, outsourcing (Intel) will carry the day.
Samsung has been the Application Processor supplier to Apple (NASDAQ:AAPL) since day one, but have apparently lost the business to TSMC with a 20nm planar part for the iPhone 6. Apparently TSMC has been building the Apple parts for months in order to support the iPhone 6 introduction, thus squeezing traditional customers like Qualcomm, Nvidia (NASDAQ:NVDA) and AMD (NASDAQ:AMD) out of access to the 20nm process.
Samsung has also teamed with Global Foundries on future 14nm efforts. Things like that just make me go hmmmm. The blind leading the blind comes to mind.
Samsung is publicly skipping 20 nm in favor of 14 nm finfet. Node hopping (28nm to 14nm) and bringing on finfet in one move is an extremely difficult thing to pull off. The jury is out on whether they can pull this off. I would bet not.
Global Foundries is the second largest semiconductor foundry in the world. The company's roots are as the AMD fab. AMD is another failed case of competing with Intel. AMD finally had to sell their fab to Middle East interests in order to stay in business.
The history of Global Foundries is best described as over-promising and under-delivering on bleeding edge technology. They do a competent job on 28nm after a long delayed start.
At one time IBM made more semiconductors than the rest of the industry combined. Their laboratory semiconductor technology was the best in the world.
IBM was the leader of the "Common Platform," an effort to unify the industry under a standard process using a "gate first" approach. Gate first was the easiest and cheapest process at 120nm, but got increasingly more difficult with each node shrink and became impossible at about 32nm. Samsung, Global Foundries and others followed IBM down the gate first path. Apparently IBM didn't believe anyone could shrink to 32nm.
Intel went "gate last," apparently because they could see a path to 32nm and beyond. TSMC after much internal discussion went gate last simply because Intel did.
IBM has recently put their semiconductor operation up for sale. In the ultimate humiliation Global Foundries turned down an opportunity to buy the IBM chip business, saying that the IBM fabs have no value.
To call the current state of the leading/bleeding edge semiconductor business a mess is a gross understatement.
This segment of the business, at about $110 billion, is about one third of total worldwide semiconductor sales. Most of the leading edge business needs to move to bleeding edge technology.
Bleeding edge technology must include finfet.
Intel is alone in the bleeding edge finfet technology with a 2-3 year lead. No one else is delivering a finfet device, no one.
Apple is quickly moving to TSMC for 20nm planar for iPhone 6 and then moving back to Samsung for a "maybe" 14nm finfet process for iPhone 7, and then moving back to TSMC for lots of 14nm for iPhoneX. Does anyone believe this?
The spending to follow Intel is a bone-crushing deal for anyone without that wonderful, high margin, $48 billion x86 business to fund fabs.
I've suggested in other articles that Intel can snuff all competitors in the smartphone AP business by putting DRAM on the processor chip. Not everyone is buying this. You need to read this link to see how rigid and stuck in the mud some of the industry pundits are. At some point before the end of Moore's Law, DRAM-on-processor-chip becomes free to do. I think 14nm is at or near that point. A smartphone chip application processor with integrated LTE modem and integrated DRAM will reduce Qualcomm to the royalty collecting company that they used to be.
Right now I think Intel is the last man standing in the marathon to the end of Moore's Law. There are some twitches from the dying corpses, but I think it is really over.
I'll leave it to others to project the financials on a $100 billion Intel at 75% gross margin, right now I'm busy buying all the 2016 LEAPs that I can reasonably afford.