TSMC's 7nm process enhancement may keep Apple from using Samsung for 'A12' chip production...

AppleInsider · July 20, 2017 5:24PM

Samsung may not return to making Apple's A-series processors in 2018 as once thought, with a new report claiming chip producer TSMC's 7-nanometer FinFET fabrication process could help the firm retain its status as sole processor supplier for the upcoming 'A12' chip.

Taiwan Semiconductor Manufacturing Co. (TSMC) is reportedly battling with Samsung to secure orders for the "A12" processor, a 7-nanometer chip that is destined for use in a future iPhone generation. While reports earlier this week suggested Samsung had secured a deal to make the processors, industry observers advised to DigiTimes that this may not be the case, based on TSMC's technology.

The in-house developed InFO wafer-level packaging created by TSMC will make the firm's 7-nanometer FinFET technology more competitive than Samsung's equivalent, the observers told the report. The introduction of TSMC's second-generation technology, which is said to play a key role in chip production, will apparently make Samsung's chances of regaining application processor orders for a future iPhone "unlikely."

Tuesday's reports about Samsung's deal cites sources claiming one of the South Korean electronic giant's three co-CEOs visited the Apple headquarters in June, with the company allegedly using its close ties to OLED production as part of its persuasion effort. Samsung is also said to be close to completing testing of its new equipment for producing chips under a 7-nanometer process, in anticipation of receiving a final approval from Apple.

On Wednesday, a report surfaced claiming TSMC had started expanding its equipment suppliers, in order to implement a more efficient 7-nanometer process from 2018. TSMC is said to be producing processors for the "iPhone 8" expected this year, with the "A11" chip thought to use the company's 10-nanometer process.

While it is possible for Samsung to take on some of the orders for the "A12" chip, it is unlikely for Apple to completely switch away from TSMC, though it is plausible for Apple to split the order between the two firms while they embark on using the relatively new 7-nanometer process. Apple has used both TSMC and Samsung as processor suppliers, with the A9 used in the iPhone 6s produced by both companies, but since then, Apple has solely used TSMC for its A-series chips.

A switch to a smaller die size, such as the move from a 10-nanometer to a 7-nanometer process, typically introduces a number of benefits linked to squeezing faster processing technologies into a more confined space. Aside from a speed boost, this can also make the processors more power efficient than previous generations, in turn increasing battery life.

doozydozen · July 20, 2017 5:39PM

Sooo...what exactly makes the 7nm chip from TSMC better than Samsung's?

soli · July 20, 2017 5:51PM

repressthis said:

Sooo...what exactly makes the 7nm chip from TSMC better than Samsung's?

I'm not sure if "better" in regards to the chip is the right way to look at it. This process may simply make TSMC competitive next year, even if they aren't the most efficient option or even cheapest option compared to Samsung. My guess is Apple will likely use both vendors if that's an option because it can help reduce their costs and keep them being backed into a corner.

edited July 2017

9secondkox2 · July 20, 2017 5:56PM

Apple has used both suppliers before with better performance coming from the TSMC processors for some reason.

Perhaps Apple gets better pricing, better processors, and doesn't have to feed the mouth that bites it. Sounds like a win-win if it happens.

melgross · July 20, 2017 6:14PM

The preference is to use one chip vendor. It’s really expensive to use two, because the SoC design has to be modified for each process, and then taped out for both, maybe multiple times. Apple needs two design teams for that, and they have to be well coordinated. If one vendor is ahead of the other in certification then the other vendor holds up the process. It’s easier if one vendor is delayed than one moving forwards, and one not. Either way, there is enough tolerance in time that both will be on time.

what happened previously was that TSMC’s 16nm process was 20% more power efficient than Samsung’s 14nm process. But when both phones were tested, in system differences were about 5%, which is insignificant, showing that the system uses so much more power than just the SoC.

i imagine that Apple doesn’t want to use Samsung any more than necessary. And that’s not because of their rivalry, but because they rightfully don’t want to depend upon one vendor for so many major parts.

edited July 2017

stantheman · July 20, 2017 6:35PM

"TSMC's second-generation technology, which is said to play a key role in chip production, will apparently make Samsung's chances of regaining application processor orders for a future iPhone unlikely."

As Jerry Seinfeld used to say, "That's a shame."

hexclock · July 20, 2017 6:57PM

It is really difficult to appreciate how small 7 nanometers is without some comparison. A strand of spider silk is about 7 micrometers wide, a full order of magnitude (1000x) thicker.
7 nm is also about 3 times smaller than the smallest virus.
We take it for granted, but it really is quite amazing that we can manufacture something that small - and smaller!

anton zuykov · July 20, 2017 7:06PM

repressthis said:

Sooo...what exactly makes the 7nm chip from TSMC better than Samsung's?

Competition affects pricing... That is a good thing.

But apart from that, tests actually demonstrated that TSMC SoC was a bit faster and a bit more energy efficient when that whole "ch_i_pgate" thing was happening.
So, if you go by that, you could say that TSMC chips are slightly faster...hence, better.

soli · July 20, 2017 7:55PM

hexclock said:

It is really difficult to appreciate how small 7 nanometers is without some comparison. A strand of spider silk is about 7 micrometers wide, a full order of magnitude (1000x) thicker.
7 nm is also about 3 times smaller than the smallest virus.
We take it for granted, but it really is quite amazing that we can manufacture something that small - and smaller!

And at such an enormous capacity.

melgross · July 20, 2017 8:28PM

Soli said:

hexclock said:

It is really difficult to appreciate how small 7 nanometers is without some comparison. A strand of spider silk is about 7 micrometers wide, a full order of magnitude (1000x) thicker.
7 nm is also about 3 times smaller than the smallest virus.
We take it for granted, but it really is quite amazing that we can manufacture something that small - and smaller!

And at such an enormous capacity.

Eh, what’s a billion transistors more or less? I think we’re getting blasé about it all these days. Next goal is 5 billion. I wonder when Apple will reach that?

soli · July 20, 2017 8:44PM

melgross said:

Soli said:

hexclock said:

It is really difficult to appreciate how small 7 nanometers is without some comparison. A strand of spider silk is about 7 micrometers wide, a full order of magnitude (1000x) thicker.
7 nm is also about 3 times smaller than the smallest virus.
We take it for granted, but it really is quite amazing that we can manufacture something that small - and smaller!

And at such an enormous capacity.

Eh, what’s a billion transistors more or less? I think we’re getting blasé about it all these days. Next goal is 5 billion. I wonder when Apple will reach that?

I seem to recall Schiller saying one chip was at 3.3 billion. And possibly around 500 million chips in the next year is a lot of transistors. I’m still amazed by all these numbers even while seeing it all happen gradually.

dachar · July 20, 2017 9:10PM

melgross said:

The preference is to use one chip vendor. It’s really expensive to use two, because the SoC design has to be modified for each process, and then taped out for both, maybe multiple times. Apple needs two design teams for that, and they have to be well coordinated. If one vendor is ahead of the other in certification then the other vendor holds up the process. It’s easier if one vendor is delayed than one moving forwards, and one not. Either way, there is enough tolerance in time that both will be on time.

what happened previously was that TSMC’s 16nm process was 20% more power efficient than Samsung’s 14nm process. But when both phones were tested, in system differences were about 5%, which is insignificant, showing that the system uses so much more power than just the SoC.

i imagine that Apple doesn’t want to use Samsung any more than necessary. And that’s not because of their rivalry, but because they rightfully don’t want to depend upon one vendor for so many major parts.

Apple knows is wise not to have all its eggs in one basket. I believe they take a strategic long term view by switching back and forth or better sharing suppliers. Plus, if there was some unfortunate incident that seriously disrupted manufacture of key components for many months or even more than a year they can look to another existing supplier to increase supply. If you only have one supplier then whatever the price advantage it could be very risky.

nht · July 20, 2017 9:18PM

If intel had the wafer volume Apple is better off going with Intel 10nm over any of the pseudo 7nm processes from GF, Samsung or TSMC.

Apple is so large that Intel Custom Foundry is too small to absorb all of the business...maybe in a couple years.

melgross · July 20, 2017 11:32PM

Soli said:

melgross said:

Soli said:

hexclock said:

It is really difficult to appreciate how small 7 nanometers is without some comparison. A strand of spider silk is about 7 micrometers wide, a full order of magnitude (1000x) thicker.
7 nm is also about 3 times smaller than the smallest virus.
We take it for granted, but it really is quite amazing that we can manufacture something that small - and smaller!

And at such an enormous capacity.

Eh, what’s a billion transistors more or less? I think we’re getting blasé about it all these days. Next goal is 5 billion. I wonder when Apple will reach that?

I seem to recall Schiller saying one chip was at 3.3 billion. And possibly around 500 million chips in the next year is a lot of transistors. I’m still amazed by all these numbers even while seeing it all happen gradually.

Either it was the A9 or the A10, but I forget which. So the A10x has possibly 4 billion, what with the extra cores, and the doubling of the GPU cores.

melgross · July 20, 2017 11:36PM

dachar said:

melgross said:

The preference is to use one chip vendor. It’s really expensive to use two, because the SoC design has to be modified for each process, and then taped out for both, maybe multiple times. Apple needs two design teams for that, and they have to be well coordinated. If one vendor is ahead of the other in certification then the other vendor holds up the process. It’s easier if one vendor is delayed than one moving forwards, and one not. Either way, there is enough tolerance in time that both will be on time.

what happened previously was that TSMC’s 16nm process was 20% more power efficient than Samsung’s 14nm process. But when both phones were tested, in system differences were about 5%, which is insignificant, showing that the system uses so much more power than just the SoC.

i imagine that Apple doesn’t want to use Samsung any more than necessary. And that’s not because of their rivalry, but because they rightfully don’t want to depend upon one vendor for so many major parts.

Apple knows is wise not to have all its eggs in one basket. I believe they take a strategic long term view by switching back and forth or better sharing suppliers. Plus, if there was some unfortunate incident that seriously disrupted manufacture of key components for many months or even more than a year they can look to another existing supplier to increase supply. If you only have one supplier then whatever the price advantage it could be very risky.

Yes, it’s never wise to do that. When I had my firm, we attempted to have one supplier for a part, regularly, but compliance tested from a couple of other manufacturers In case we needed to move off the first.

the only time when having more than once current supplier is for chip manufacturing is I f you design your own. You’re going to run into problems with two suppliers. It was considered to be amazing, in the industry, that Apple didn’t run into problems with their SoC that year.

edited July 2017

analogjack · July 21, 2017 4:23AM

Not too long now and both companies are going to run out of nanometers to shave off, if they haven't already.

analogjack · July 21, 2017 4:24AM

repressthis said:

Sooo...what exactly makes the 7nm chip from TSMC better than Samsung's?

It hits Samsung where it hurts the most. It doesn't need to be better, it only needs to not be worse.

edited July 2017

lifeboat · July 21, 2017 10:35AM

melgross said:

Either it was the A9 or the A10, but I forget which. So the A10x has possibly 4 billion, what with the extra cores, and the doubling of the GPU cores.

TSMC is currently mass producing 21.1 billion transistor chips for Nvidia. (The V100 chip.) Each chip has 14,112 cores with 5376 FP32 cores, 5376 INT32 cores, 2688 FP64 cores, and 672 Tensor cores. To add a bit of extra power, they connect 4 memory chips to the top of each chip using 1,024 TSV connections giving you a total of about 156 billion transistors. Nvidia thinks this still isn't enough power so they are working on designs that will connect many chips together. (As AMD has done with their latest CPUs.)

melgross · July 21, 2017 2:52PM

lifeboat said:

melgross said:

Either it was the A9 or the A10, but I forget which. So the A10x has possibly 4 billion, what with the extra cores, and the doubling of the GPU cores.

TSMC is currently mass producing 21.1 billion transistor chips for Nvidia. (The V100 chip.) Each chip has 14,112 cores with 5376 FP32 cores, 5376 INT32 cores, 2688 FP64 cores, and 672 Tensor cores. To add a bit of extra power, they connect 4 memory chips to the top of each chip using 1,024 TSV connections giving you a total of about 156 billion transistors. Nvidia thinks this still isn't enough power so they are working on designs that will connect many chips together. (As AMD has done with their latest CPUs.)

While that’s true, and interesting, it’s completely different. GPUs are easy to manufacture though. They use transistors very efficiently, and they line up very well. In addition, that chip has no relevance in this discussion because we’re talking about a more complex chip than even that GPU, that’s also far smaller, and far less expensive. Those Nvidia chips costs thousands, while Apple’s SoC costs between $34 and $38, according to estimates by Micrprocessor Reports.

Inaddition, nvidia will sell hundreds, or low thousands of those chips, while apple will sell 200 million devices this year with their latest SoC.

edited July 2017

blastdoor · July 21, 2017 2:56PM

dachar said:

melgross said:

The preference is to use one chip vendor. It’s really expensive to use two, because the SoC design has to be modified for each process, and then taped out for both, maybe multiple times. Apple needs two design teams for that, and they have to be well coordinated. If one vendor is ahead of the other in certification then the other vendor holds up the process. It’s easier if one vendor is delayed than one moving forwards, and one not. Either way, there is enough tolerance in time that both will be on time.

what happened previously was that TSMC’s 16nm process was 20% more power efficient than Samsung’s 14nm process. But when both phones were tested, in system differences were about 5%, which is insignificant, showing that the system uses so much more power than just the SoC.

i imagine that Apple doesn’t want to use Samsung any more than necessary. And that’s not because of their rivalry, but because they rightfully don’t want to depend upon one vendor for so many major parts.

Apple knows is wise not to have all its eggs in one basket. I believe they take a strategic long term view by switching back and forth or better sharing suppliers. Plus, if there was some unfortunate incident that seriously disrupted manufacture of key components for many months or even more than a year they can look to another existing supplier to increase supply. If you only have one supplier then whatever the price advantage it could be very risky.

Also, the added fixed cost of the second supplier is *relatively* much smaller for a company operating at Apple's scale than other FAB customers. What may seem like a mountain to a company like AMD looks more like a molehill to Apple.

jdb8167 · July 21, 2017 3:45PM

Soli said:

melgross said:

Soli said:

hexclock said:

It is really difficult to appreciate how small 7 nanometers is without some comparison. A strand of spider silk is about 7 micrometers wide, a full order of magnitude (1000x) thicker.
7 nm is also about 3 times smaller than the smallest virus.
We take it for granted, but it really is quite amazing that we can manufacture something that small - and smaller!

And at such an enormous capacity.

Eh, what’s a billion transistors more or less? I think we’re getting blasé about it all these days. Next goal is 5 billion. I wonder when Apple will reach that?

I seem to recall Schiller saying one chip was at 3.3 billion. And possibly around 500 million chips in the next year is a lot of transistors. I’m still amazed by all these numbers even while seeing it all happen gradually.

Based on those numbers, just Apple using TSMC alone are manufacturing 1.65x10^18 transistors in a year. The Milky Way galaxy (ours) is estimated to have about 1x10^11 stars. The numbers are mind boggling. Apple is responsible for a transistor count 16.5 million times the number stars in the galaxy.

It is estimated that in 2014 total manufacturing of transistors was 250 x10^18. Expect at least double that now.

edited July 2017

TSMC's 7nm process enhancement may keep Apple from using Samsung for 'A12' chip production...

Comments