We have all been watching and speculating on the mobile market for some time. It's abundantly clear that the market is infinitely dynamic, and that change is rapid and somewhat unpredictable. To make matters worse, all the players are involved in actions that have been undertaken to affect the market through propaganda, lawsuits, and other actions that should not really affect the ultimate outcome - but only delay natural outcomes.
Regardless of the unfounded allegations of the uninformed, and worse, the paid doubtmeisters, it's clear that Intel has a significant technological advantage over all the other potential suppliers of mobile SOCs. Intel (INTC) 3DFET's (Aka FinFET's) have a very significant technological advantage over the competition, consisting of TSMC (TSM), Samsung (OTN: OTC:SSNLF), and Global Foundries (privately held). Intel FinFET's are denser, faster, have better speed/power ratios than what the competition is proposing, and they are here now and have a 2 to 3 year lead over the competition in volume availability.
The competition has induced a number of willing or uninformed spokespersons to:
- Question the availability.
- Question the density.
- Question the power.
- Question the speed.
- Question the cost.
- Question Intel's truthfulness.
If the competition would only truthfully answer the questions they ask about Intel with respect to their own FinFET product (what product?) there might be less need to write this piece. For the truth is, the competition has no FinFET product. They won't have any until they can produce product in quantity, with acceptable yields, with known density, with known speeds, and known prices. So the sensible thing for them to do, at this point in time, is to stop talking until they have product. But we know that's not going to happen, so I'm writing this to put another nail into their coffin.
So, what does it take to own the top end of the mobile SOC market? How does Intel stack up?
- A reputation for quality. Unmatched.
- Competitive pricing. No competition, no argument.
- Leading edge features. Getting there.
- Straight talk. Better than TSMC and Global Foundries. At least equal to Samsung.
- On-time delivery. Yes.
- High yields. The best.
- High reliability. Designed in.
- Software integrity and continuity. Good rep on Windows, improving on Android.
- Security. Leading edge.
- SOC experience. See below.
A key objection has been raised by the Foundries and their stand-ins - and they have jumped all over this one - that Intel is a microprocessor company and doesn't know how to build SOCs. (A System On a Chip or SOC is an integrated circuit that integrates all components of a computer or other electronic system onto a single chip.) This is probably their last line of defense, because with all of the other advantages Intel has, it still needs to build good SOCs to keep the overall cost of building a customer's product under control. If they can't integrate a variety of analog functions onto their SOCs, including existing IPs from several different potential customers, then the ball game could be over before it starts.
Has Intel built SOCs before? According to the definition in the last paragraph, a computer constitutes a system, and with that in mind, the answer to the question is - yes! But philosophically, when one talks about SOCs in the context of mobile applications, they are talking more about the inclusion of analog IP and particularly, IP ported from other earlier processes. How does Intel propose to do this?
Intel's cadence of new process introduction is a two year interval between doubling the area density of the basic process, and in "odd" years, adding to the basic microprocessor oriented process, with minimal analog capabilities, additional features including variable lithographic pitch, high voltage capability, and analog transistors. This is pictorially represented in the slide below from Mark Bohr's presentation at 2012 Intel Developer Forum.
Analog is tough using FinFETs. An accomplished designer can just barely manage to build a passable circuit except when it comes to high voltage. In the abstract the following are the only two alternative approaches that seem reasonable to me:
1. Build a high voltage FinFET. (Don't laugh, there are published pictures of Intel high voltage FinFETs on the web) But - these don't solve three problems:
i. DC gain is poor.
ii. There is a need to put FETs in series and/or parallel for matching and for other reasons; as a result, circuits will grow in size.
iii. You can't easily port a customer's or Intel's analog IP designs to FinFET processes without lots of work.
Because of the difficulties in porting Analog IP to the latest Intel SOC processes, I have a strong tendency to believe rumors that Intel is working with outside companies to facilitate porting for its customers.
In the slide below again from Mark Bohr's presentation at the 2012 Intel Developer Forum, a variety of 22 nm SOC components are shown, including the cross section of a high voltage FinFET, in the lower left hand corner of the slide. Note the thickened gate insulation with respect to the logic transistor directly above it.
2. Build every chip as an SOC in an "odd" process (P1271, P1273, etc.). Mix native FinFETs with regions of higher voltage planar transistors, with a transition region between them. This results in:
i. High DC gain, to enable design of good opamps and other analog circuits.
ii. Planar circuits can be cascoded with FinFETs so no matching issues with planar.
iii. Anything that is planar IP can be ported to this process.
From a designer's perspective, the high voltage FinFETs are fine for digital I/O and some simple analog functionality, but for porting the broad range of existing IP to the latest and greatest Intel process, I would select alternative 2 as the most promising and universal method of constructing SOCs.
Alternative 2 is also tailor made for IOT (Internet Of Things) applications - a dense processor and the ability to include on chip any special IP (like a A to D converter, or a Bluetooth interface, etc.) that might be required by the particular Thing.
So as the market reacts positively to Intel's moves to gain share in mobile, the hammer is descending, and the last nail is about to be driven into the coffin. Intel is taking the top of the mobile market from the likes of TSMC, Samsung, and Global Foundries. The sooner they realize this, the more they can concentrate their energies on catching up - rather than trying to confuse the market about reality. Their continued attempts at misdirection will ultimately be recognized by the broad base of customers and investors - all to their long-term detriment.