Intel Opening Foundry To Third Parties

angry tapir writes "Intel is exploring whether it can branch out as a foundry by opening its chip manufacturing facilities to more third-party customers. Intel has expanded its chip-to-order business by signing up additional customers to take advantage of its 22-nanometer process facilities." In particular, two FPGA design companies will be using Intel's fabrication plant, and "the unit has more than two customers but others are not disclosing their plans yet."

Comments

Just a suggestion...

[rickb928]

But AMD could use a good fab.

Just sayin'

Re:Just a suggestion...

[CxDoo]

Nah this is more like Intel going after AMD foundry too. kick them while they are on the ground and all that.

Re:Just a suggestion...

[UnknowingFool]

I think AMD is not Intel's only concern. The other chip foundries like TSMC, UMC, and GlobalFoundries probably worry Intel more. The increasing sales of ARM based mobile devices presents a challenge to Intel in the future. Other chip foundries are agnostic to the platform; they will make whatever chip you want. While this change will focus on FPGA, it could develop into other types including ARM. Just a thought.

Re:Just a suggestion...

[sexconker]

Nah this is more like Intel going after AMD foundry too. kick them while they are on the ground and all that.

AMD doesn't own their fab anymore. They are still big shareholders, however.
Global Foundries has been bleeding money for ages because they can't get their shit together. AMD should just dump their stock in the river and tell Intel to build them a 22 nm Piledriver chip.

Re:Just a suggestion...

[CxDoo]

I am aware AMD spun off Global Foundries, however they are still married to them.

OTOH, it makes no sense for Intel to go after AMD, as someone above mentioned. They own them in architecture and process.

I don't understand why would they offer their process to others when it is their biggest advantage.

Re:Just a suggestion...

[tlhIngan]

OTOH, it makes no sense for Intel to go after AMD, as someone above mentioned. They own them in architecture and process.

I don't understand why would they offer their process to others when it is their biggest advantage.

Intel offering fabs to AMD would benefit Intel greatly. Sure they're manufacturing their competitor's product, but that's the key - "competitor". There's no real competition in the x86 space - you have a few other x86 compatibles, but nothing matches the desktop line, except AMD.

And Intel needs AMD to survive because it keeps government off their backs. Just like Microsoft needs Apple around, and Google needs Apple as well (Apple offering competition has allowed Google to do stuff like acquire AdMob). If AMD dies and goes out of business, Intel will be a monopoly - nothing wrong, but you can bet there'll be increased scrutiny - did Intel contribute somehow to AMD's demise? Past actions with OEMs will come back to haunt Intel as well.

As for offering others access to high end processes - it's easy. Money.

A fab is horrendously expensive, and to get any form of ROI, it basically has to run continually. If you can't keep a fab working, it starts losing money. It appears that possibly Intel might not be able to keep their fabs under full utilization (probably due to the economy), so best to open it to competition and get back money building the fab than to let it sit idle and costing money while your competition is upgrading their fabs.

Apple would be a prime contender - and probably one of the main customers.

Re:Just a suggestion...

[symbolset]

I said this when we talked about this rumor before. It is as simple as it can get. Intel has a Fab business that is a part of their manufacturing operation. They turn sand into chips, and then to money. Fabs are horrifically expensive and new process fabs even more so. If people pay Intel to turn their designs into chips Intel gets that money to sink into future R&D and more wonderful fabs - and of course, profits. If Intel turns away that work, Intel's Fab competitors get that money to do those things with and are so better able to try to keep up with Intel on volume and process technologies.

It's just that easy. If Intel has excess capacity they should sell it at a profit - if not just for the profit then just to take some bread off the table.

nVidia not AMD

[gr8_phk]

Just imagine nVidia makes GPUs at Intel to get a full process node ahead of ATI/AMD. It seems unlikely they'd let AMD use their 22nm process to compete with themselves. OTOH, nVidia is now a competitor as well.

I was wondering if/when Intel would do this, since they are essentially a full process generation ahead of the whole world.

Re:nVidia not AMD

[Anonymous Coward]

It seems unlikely they'd let AMD use their 22nm process to compete with themselves.

It also seems unlikely they'd open themselves up to another huge anti-trust lawsuit.

Re:nVidia not AMD

[alvinrod]

If they were going to let nVidia use it for GPUs, they'd probably let AMD use it for GPUs as well. If they didn't AMD would probably die within a few years and leave Intel as the only major remaining x86 manufacturer and make them a de facto monopoly. Alternatively, Intel could let AMD make anything that they wanted to at Intel's fabs. Right now Intel has a much better architecture and is better in almost every way outside of integrated graphics, so they would still maintain superiority over AMD for several years. Hell, Intel might consider it just to keep AMD alive so that Intel doesn't come under increased scrutiny for future actions.

However, assuming that they're not interested in letting any of their competition use their fabs, it seems as though the number of companies that can will be quite limited. ARM SoC makers are obviously out because Intel has been trying to move into that market for a while now and wouldn't want to make things more difficult. AMD and nVidia might also be out, at least for most of their products. Cellular baseband manufacturers might be interested in using it, if only because it will reduce the power draw for those parts, but companies like Qualcomm are starting to build the baseband in to the SoC now so that seems less likely. It could just come down to FPGA makers and a few other niche markets that Intel doesn't compete in or has a very small presence and no interest in increasing it.

It will be interesting to see how this all plays out as it does present the opportunity for significant advantages to any companies that get their parts made there.

Re:nVidia not AMD

[Anthony Mouse]

It also seems unlikely they'd open themselves up to another huge anti-trust lawsuit.

By doing which, fabricating AMD's chips or refusing to?

Apple's rocky relationship

[jisom]

I wonder if Apple's rocky relationship with Samsung could be a motivator. Intel would love some of the iDevice revenue.

Re:Apple's rocky relationship

[Anonymous Coward]

^^^ This ^^^

Intel and Apple seemingly have a good relationship. If Apple kept Intel as their preferred CPU "manufacturer" it would be win-win for both.

Re:Apple's rocky relationship

[beelsebob]

Well, I'm under the impression that apple currently use TSMC for their CPUs, but yes, I could see them wanting a smaller, lower power processes.

Re:Apple's rocky relationship

[UnknowingFool]

Apple currently uses both Samsung and TSMC with Samsung manufacturing the bulk of the A5s at the moment as TSMC ramps up production. The lawsuit between Apple and Samsung might have been a factor but it just makes good sense to have more than one supplier of a critical component especially in different countries as both the Japanese earthquake and the Thailand floods show.

Re:Apple's rocky relationship

[RightSaidFred99]

Would be bad business for Intel in some ways. They are competing with those chips (ARM) and it seems silly to me that they would offer their _massive_ competitive advantage (top flight manufacturing with large capacity) to a competitor.

They could only manufacture using a previous generation process, I guess. Make their Atom chips on e.g. 14nm and only offer 22nm fabbing for third parties.

Atmel

[Anonymous Coward]

I'd love to see Atmel microcontrollers made on 22nm process. ATtiny85, 84, 861, etc running faster than 20MHz with twice the flash and RAM, anyone?

Re:Atmel

[Anonymous Coward]

lol

Re:Atmel

[Anonymous Coward]

Process and fab are not what keep the ATtiny low performing. It's by design. I guess it could drop the price like crazy by improving the yield per wafer, but you don't make Huffy's in a Toyota plant just because the Toyota plant can make a shitload more.

Re:Atmel

[K. S. Kyosuke]

I'd love to see Atmel microcontrollers made on 22nm process. ATtiny85, 84, 861, etc running faster than 20MHz with twice the flash and RAM, anyone?

Something tells me the idle power consumption would suck due to leakage. You don't need to go for 22nm to do 20+ MHz. You can get to a few hundred MHz just fine in the 500nm range.

Re:Atmel

[wmac1]

Perhaps you need 2GHz (because you are talking about 22nm). Besides ATMEL micro-controllers are very simple integrated circuits in comparison to x86 and 64 CPUs. Some of them are hardly even VLSI.

Re:Atmel

[AdamHaun]

Yeah, there are already lots of higher-performance microcontrollers out there if you're interested. If you really want speed and memory, go for something based on a 32-bit ARM. I see a 64 MHz Cortex-M3 with 1 MB of flash on DigiKey for $11. You can get 256k of flash for under $10.

Re:Atmel

[robthebloke]

I was thinking along the same lines, although I think the z80 might be a more fitting choice.... It would be like the errant step-child coming home to roost ;)

Squeeze in the uint8 AVX instructions, add some internal ram, keep it within the original 80 pin package, and I'd finally be able to pimp my 48k speccy to a whole new level of awesome!!

Re:Atmel

[Technician]

The Intel Fin transistor design, the Tri Gate transistor cuts the leakage to a tiny value while keeping up the performance. Big transistors (500nm) have their power consumption when clocked as the device becomes a huge CCD device switching distributed capacitance from power to ground. Power per operation is high. Low K dielectrics, small dimensions, combined cut power in high speed operation. Low leakage cuts power draw in all operations. Going big dimensions to cut leakage, requires higher voltage to cut the transistors off, resulting in higher power use per clock cycle when switched.

Re:Atmel

[Anonymous Coward]

I'm talking about $2-3 microcontrollers and you're telling me to use $11 ones. Good job, Sherlock.

Re:Atmel

[Anonymous Coward]

The whole point of 8-bit micros is compatibility with standard 5V or 3.3V interfaces. 22nm will not handle anything above 1.5V even with special IO transistors. The old MCU tech is only on 180nm or older, because there's no need to go down.

Re:Atmel

[K. S. Kyosuke]

All right, you are correct, but...

Going big dimensions to cut leakage, requires higher voltage to cut the transistors off, resulting in higher power use per clock cycle when switched.

..that assumes that you have clock cycles in the first place. Why not make the circuit asynchronous? See ARM Amulet, c18, GreenArrays, etc. In microcontrollers, you should be able to afford it, they aren't that complex.

Re:Atmel

[symbolset]

What are you designing? Happy Meal toys?

Re:Atmel

[Anonymous Coward]

As somebody who works in Atmels fab, I know you would not like the price we would have to charge if we used Intel to make our wafers. Our stuff (and our competitors micros) are the price they are specifically because we do not use state of the art lithography.

Re:Atmel

[Anonymous Coward]

Yeah that makes perfect sense. Throw in a free $3 microcontroller with every $3 Happy Meal.

I'm guessing more intimidation

[Kjella]

Larrabee almost certainly lowered future expectations for discrete graphics chips, fearing that Chipzilla would enter their market. I'm guessing this is the same for the foundry business, scaring away potential investors in TSMC, UMC and GloFo. Huge, huge investment costs that take years to materialize and are extremely time-to-market sensitive, any uncertainty you can add to that is advantage Intel. I very much doubt that anything remotely competing with Intel will ever get their hands on their crown jewels.

Re:I'm guessing more intimidation

[Anonymous Coward]

Destabilizing funding of expensive process upgrades of other fabs might indeed be the most important goal in this for Intel. It could hurt those market segments where Intel products are not doing well by widening the process node gap...

Re:I'm guessing more intimidation

[afidel]

More like Intel using their fab capital as seed money for small startups that they might want to buy. If they're small and already on Intel process then it's an easy way for Intel to buy into a market that gets hot.

Tabula

[gweeks]

Tabula has some cool tech. Their current chips aren't dense enough for what we wanted to do with them, but at 22nm they might be.

GPUs!

[Anonymous Coward]

I guess it would be great if AMD and nVidia GPUs were made with Intel's superior processes.

Or if Intel made decent GPUs themselves, but it seems they are somehow challenged on that front.

I wonder

[squidflakes]

Any slashdotters familiar enough with FPGAs to know if they could be used to create digital camera sensors?

One thing that's been lacking from digital photography is an "affordable" large format digital camera that is good for anything other than still life. Currently, if you are going to do large format digital, you have to use a back that takes an optical scan of your ground glass, and that can take up to 30 seconds.

Re:I wonder

[UnknowingFool]

From my understanding FPGAs is a type of integrated circuit. A camera sensor like CCD or CMOS sensor are other types of ICs. Their manufacturing is different as well as their functionality. You cannot use a FPGA as a camera sensor any more than you can use your car transmission as your engine.

Re:I wonder

[Anonymous Coward]

No, FPGAs are Field Programmable Gate Arrays, and digital camera sensors are either active pixel sensor CMOS or Charge Coupled Devices (CCDs). They are two very difference kinds of technology. Intel's fabs are probably capable of lithographing digital camera sensors, but you really don't need such a high-tech process for that unless you're doing something like Foveon's deep-silicon imaging (and even then, you can probably get away with a few generation's old tech).

But you may not be entirely out of luck. Depending on how big of format you're talking about, lots of companies will sell you CCDs and/or CMOS sensors at very large sizes.

Re:I wonder

[Anonymous Coward]

Right... right... but they ARE etched semiconductors integrated onto a silicon substrate, no?

About the same size and shape... essentially the same style of packaging...

Re:I wonder

[UnknowingFool]

Only in the similarity that a x86 chip is manufactured generally in the same process as a DSP. But specifically you cannot take the chips out of your stereo and build a computer to run Windows. Specifically making a CCD or CMOS sensor is very different than making a FPGA. Specifically they have different functionality and are not interchangeable.

Re:I wonder

[robthebloke]

No. I'd imagine that some digital cameras use FPGA's though, but only as a way to get high speed image processing capabilities (performed in parallel)

Re:I wonder

[Anonymous Coward]

FPGAs are neither extraordinarily fast, energy-effient nor cost-effective, but certainly those chips compete well on performance with completely general-purpose microcontrollers for a specific task and on price against creating own IC design from scratch for relatively small product runs. With these factors considered, most likely places to see FPGAs on cameras would be scientific/industrial cameras and very high-end cameras, and I wouldn't even be too sure of that high-end variety.

Re:I wonder

[squidflakes]

Ahh, good to know. Thank you!

You can't make CCDs using FPGAs

[Brannon]

they are different electrical devices, FPGAs only allow you to implement different logical functions.

Re:You can't make CCDs using FPGAs

[dbc]

Soooo much misunderstanding here, all around.

CCDs are one kind of sensor. I'm not very familiar with the physics. In any case, no FPGA that I know of contains CCD cells.

CMOS sensors are another way to sense light. The physics behind a CMOS sensor is that if you have a CMOS RAM cell, and expose it to light, the light depletes (IIRC, maybe it enhances??) the charge. Anyway, A CMOS image sensor is a big RAM that has been optimized for photonic effects and laid out with a geometry that makes image retrieval practical. You can take a CMOS ram chip and put a transparent lid on it and have a CMOS image sensor. It will suck, because it has been optimized for data storage and the pixels will likely have strange spacing and obnoxious addressing.

FPGA's usually have embedded RAM. They also store the logic configuration and routing in RAM. You could, in principal, put lids over the RAM blocks and block the light hitting the routing and function control cells. It would be very hard, and would yield a really bad image sensor with a useless imaging pattern.

So to answer the OP, take your choice of:
A) Yes, of course. It is simple physics.
B) No way, it's impossible. The results would suck, even if you could pull off the packaging challenges.

Hope that clears things up a little.

DEC

[Anonymous Coward]

Ask the folks over at Digital Equipment Corp how well that worked for them. -oh wait.
http://www.wired.com/techbiz/media/news/1997/10/8024

spinning off their foundries?

[schlachter]

I wonder if this will ultimately lead to them spinning off their foundries...in the past, I thought they have viewed their foundries as a competitive advantage against other chip makers. I wonder if that's still the case.

Re:spinning off their foundries?

[chrysrobyn]

I thought they have viewed their foundries as a competitive advantage against other chip makers.

They may still. If they offer their foundries' services at a premium, that would be one way to profit off of that advantage. The 22/20nm node is still years away for other foundries, so the combined "3d transistor" and more chips per wafer that 22/20nm bring should provide some serious cash. Not that anybody cares if their chip is made at 22nm, but if it means they can yield 25% more chips from a wafer, that's the kind of premium cost Intel can demand for that individual advantage alone.

Their bargain chips have always been the largest piece of Intel's volume, and those bargain I3s and Atoms are now getting pretty small -- which may have opened up some manufacturing capacity.

Re:spinning off their foundries?

[RightSaidFred99]

Intel is a manufacturing company. Spinning off their foundries would be like McDonald's spinning off their hamburger business.

Re:spinning off their foundries?

[symbolset]

Strangely enough, only 15% of those hamburger businesses are owned by McDonald's. So it's more likely than you might think.

IP Snooping, anyone?

[segin]

Maybe Intel is hoping nVidia or AMD will use their foundry to fab their GPUs. And then Intel magically comes out with a GPU that doesn't suck 6 months later.