Why Is Intel Fabbing Achronix FPGAs?

Why on earth has Intel decided to fab Achronix FPGAs?

Achronix says it will use Intel for the 22nm process..

Intel is not thought to have previously acted as a foundry.

Indeed when, many years ago, Actel approached Intel to fab its FPGAs Andy Grove gave a very firm No to the request.

So why the change in attitude?

Well Intel could be wanting to get back into the FPGA business – it had a stab at it in the 1980s and made a hash of it. Intel could buy Achronix any time it wanted and this could be a gentle way in to see if there’s any chance of a second chance.

Intel is certainly thrashing around looking for new markets and Achronix’s high-performance FPGAs would suit Intel’s temperamental inclination for speed. Is the $4 billion FPGA market worth Intel’s attention? Well, maybe.

On the other hand, FPGAs with their regular structures are good devices with which to de-bug processes. Intel may think the Achronix FPGA would be a useful de-bug vehicle for its 22nm process.

The only other reason I can think of – that Intel wants to get into the foundry business to defray fab costs. After all Samsung has a substantial foundry business which it is trying to grow.

So foundry could be seen as another new business area for Intel. And a foundry business would have the plus of giving Intel wider experience of processes designed for other things than processors.

But to operate in the foundry business you have to co-operate closely and amicably with your customers.

TOMORROW MORNING: The Ten Best Performing Semi Shares This Year.


Comments

16 comments

  1. Sorry to comment so late … but I have the solution !!!
    Intel is fabing Achronix just because Intel intend to had a small FPGA array inside its chips, in order to allow the addition of specific instructions to the core, these instructions being defined by the software.
    The low hanging fruit has been to add FPGA as coprocessor to PC processor, and this has been done for something like ten years. But this is what is called losely coupled coprocessing.
    The communication overhead is quite significant, and thus such coprocessing is not always relevant. Moreover this solution is by definition not integrated (cost++, power++)
    To performs what is called closely coupled coprocessing, the FPGA matrix has to be integrated along the CPU core itself. Ideally in such an intricated way that even adding operators to the pipeline could be possible.
    The achronix deal is just the first step of this strategy: to see how well (or how bad) the FPGA port to Intel process is done. Then second step, to integrate FPGA technology on die.
    I have just reviewed tens of articles (+comments), the foundry business for Intel is a dead duck. To summarize:
    – customer which want high perf process are all direct Intel competitors
    – customer which do not want leading edge process will find cheaper (and safer) elsewhere.
    – Intel has an extraordinary reputation (well deserved) of being not a reliable partner. It has entered various markets, and then abrutely exited them, without any respect for their naive customers.
    Beside its core PC business, to trust Intel on anything else is just toal madness (for exemple, FPGA, graphic chips, phone soc, networks, and so on).
    To use achronix as a process test vehicule is just laudable. Why Intel would use an external design to do so !! With all the trouble associated to ? (Intel is a Paranoid company, why bother to communicate with a small start-up ???)
    See how well the ATOM+TSMC deal worked (you can find my comment on this on your blog, Among other thingh I was saying that nobody will use the ATOM core at TSMC, a prediction 100 % accurate), and for the obvious reasons I explained at the time. this reasons still stand to says that nobody significant will ever use Inrtel as a foundry.
    To merge cpu core + fpga is already done in some company (STm and Atmel for exemple). It is a logical step for Intel, and sp far the most sensed explanation to the deal (moreover it is a way to recover from Larrabe debacle, from a technical standpoint, and to counter the Fusion things from AMD.
    Achronix is the logical choice in this explanation, since their Matrix are just cloked 3 times highier than the one of Xillinx or Altera. Thus Achornix matrix are the only one suitable for an integration with an Intel core, in terme of targetted frequency.
    (moreover Achronix is a small fish to swallow).
    Fabrice

  2. If Achronix is going to do Military/Space/HiRel business they may have a need to use a Trusted Fab which Intel may be interested in doing business as a Trusted Fab.
    IBM is a Trusted Fab as I remenber.
    Edward

  3. Also, intel produced chip for phillips semi/lab back in alb when i was working at F11X in rio rancho. They had some kind of fire in there fab and asked intel to take over there wafer starts until the damage was repaired.

  4. Another couple of points:
    – Some reports say Intel’s 22nm is going to be SOI
    – Achronix has got lots of support in the US space and defence industries
    – SOI is great for these apps…
    – Variability is going to be massive at 22nm – a good fit with asynchronous that will go as fast as the individual circuits allow without causing errors (accounts for the support above)
    This also fits with US manufactured FPGA comments already made.
    Intel’s move is also, I think, a vote for Xilinx’s “programmable imperative” so will please the rest of the FPGA industry.
    Cheers,
    CG

  5. Thanks Tom that all sounds eminently plausible.

  6. When you have 22nm process and are heading for 450mm wafers one fab can make an awful lot of chips. If Intel wants to keep scaling its fabs sooner or later it will either have to manufacture for other people or dramatically broaden its product range.
    Achronix has FPGA and is asynchronous. Both FPGA and going asynchronous could be interesting to learn about for a company whos own technology is running out of steam in terms of maximum parallelism and maximum clock frequency.
    Government people may want to have a source of FPGAs (and possibly other chips)designed and manufactured entirely in the US. There could be a quid pro quo when Intel asks for help to build extremely expensive fabs in the US.
    Finally Achronix is small enough to use as a learning experience without worrying.

  7. Robert, on the cherchez le Intel-homme theme: Achronix CTO and founder, Dr. Virantha N. Ekanayake, is an ex-Intel-ite. Maybe you’ve cracked this little mystery.

  8. Thanks for an intriguing insight, Robert.

  9. From my experience when companies depart from their normal behavior, there is typically a personal relationship behind the business “strategy” change. I’d be looking at the R&D resumes to see who’s ex-Intel.
    This sort of issue is two sided, first someone, in a very powerful position at Achronix, has to believe, in his heart of hearts, that access to Intel’s process is the only solution. Secondly Intel has to believe, that there is no downside.
    What usually kills these deals is that there is really no economic upside for Intel, so the deal only gets done as a personal favor.

  10. Yes, greg, it’s always been Xilinx’s contention that the market for such fast FPGAs is tiny While both Altera and Xilinx have parts to compete in the 400G and LTE markets at which Achronix says its 22nm family is targeted.

  11. I think it provides a window into their stategy. Im sure the intel process is fast but also isn’t as cheap as some alternatives they could have picked. I wonder if they paid a premium for this. So achronix thinks they can get enough business from a screaming FPGA to sustain their existence? I’m sure there are people out there that want it but I doubt theres enough. Its a differentiator but I’d rather put my money on them wanting to appear as fat and juicy as possible.

  12. My guess would be its a far quicker task to lay out an FPGA than a processor. In the paat Intel have used SRAMs as technology test beds but this has never really been satisfactory.
    And as Elmer said, Intel have long fabbed chips for other people. The Itanium is effectively a HP custom chip – no one else wants it πŸ™‚
    Back in the early 90s they also did some custom 386SX cores with other blocks on for communications use.

  13. I certainly will Elmer. The Achronix people are coming in on Thursday to talk to a colleague and that’s the first question he’ll ask them.

  14. Another note.
    Intel had an ASIC operation in Folsom in the 80s after picking up IBM’s ASIC IP in exchange for manufacturing rights to the i386. They produced products for multiple customers including IBM’s uChannel chipsets before quitting the business.
    So why FPGAs now and why Achronix?
    When you figure that out please let us know. πŸ™‚

  15. Thanks Elmer, that’s interesting. As you say no one (including me) seems to have got this right.

  16. I wish someone would get this right. Intel has fabbed material for HP and Qualcomm in the past, to name just 2. Otherwise I’d agree with you that intel’s motives here are unclear.

Leave a Reply

Your email address will not be published. Required fields are marked *

*