TSMC May Win Bulk of A9 Orders Due to Better Production Yield

[gnasher729]

The difference between CPUs and other parts is that most things Apple buys from third parties are off-the-shelf products. As long as they meet a certain spec Apple can buy them and integrate them. However, the CPU is one of the few more complex things Apple actually fully designs in-house.

Now, I am not a chip manufacturing expert, that’s why I’m asking this question. What I hear from people who actually do this sort of stuff for a living is that you can’t just send the same tape-out to both. A tape-out needs to be created specifically with the foundry’s requirements in mind. Obviously you could create two different designs for two different manufacturers but the overhead and engineering headache for Apple would be substantial and possibly not worth it for only 20% of the production run.

I’d like MacRumors to actually do some digging and speak with people who commission big CPU production runs to see how feasible all this speculation actually is. If MacRumors can reveal that all this talk by business analysts is utter nonsense it could be a good story.

It's not quite as bad. The chip design is done by hand. The layout is done by some clever software. _Then_ you usually go in and make some changes to help the layout software. TSMC will have told Apple something like "if you stay below 195mm2 then we get x chips from a waver, otherwise up to 203mm2 we go down to only y chips", so depending on how close Apple would be to the 195mm2 they would work to get it under the limit (or not) by helping the layout software. So the layout would have to be done twice, but the design is the biggest part of the job.

Those complaining about contactless in the UK, where do you live? London is contactless central and one could, in theory, use Apple Pay with US issued cards in the UK so Apple Pay will be an easy roll out, it's just whether or not banks will be on board.

Their technology isn't identical, so their design rules are not identical. If you sent the TSMC tape (which isn't a tape anymore but is still called that) to Samsung, they couldn't build it, and vice versa. The design (which transistors and how are they connected) is the same, but the sizes and minimum distances between these transistors is different.

 

[AlecZ]

hey what do you know, samsung has poor product quality!

Their parts seem fine. I've got a Samsung Evo SSD in my MP. When it comes to consumer products, I don't like them. Samsung TVs have nice displays but horrible everything else, including boot up time and interface. I never knew someone could make it so hard to switch the input on a TV.

----------

So many rumors and there's no substantial backing to any of them. Does anyone really care which company is actually going to manufacture the A9 chip. No

People care enough to comment. On a national scale, almost nobody cares, but everyone on MR is in the minority.

 

[wizard]

All these analysts claiming the production run is split between two companies… I have yet to see the first person with actual chip production experience state that it’s even possible.

Anything is possible.

So far I only see experts claiming that it’s technically almost impossible to split advanced production runs because different companies have different requirements. You can’t send the same tape-out to both companies because tape-outs for TSMC wouldn’t be compatible with Samsung’s equipment and vice-versa.

Think about it this way, if both Samsung and TSMC are in the run here then both of them have to be fabricating A9 series chips right now for Apple to even know what the yield rates are like. Sure they can look at internal test runs but you still need to know what the yield rates will look like on your specific chip.

Even if Apple hasn't run test batches of A9's they still would have to have the A9's design in a form usable by both companies. Most likely a good part of the A series exists in one or more hardware description languages. This description would then be imported into the process tools associated with manufacture. Apple couldn't effectively pick and choose a vendor if they had to run an entire design cycle on a specific set of process tools.

Instead of printing speculation from “analysts” could MacRumors instead try to speak to a subject expert to get real insight? You know, someone who actually knows a thing or two about chip design and production?

That would be interesting! However I still believe a fundamental truth here is that Apple can't really shop around for manufactures if it is tightly tied to a set of tools. So given that I would expect them to have the same processors being pilot ran at both facilities.

In doing so the processors would deliver the same computational performance. More interesting would be the thermal performance. I suspect power usage would be as important to Apple as yields with the better vendor getting priority.

 

[firewood]

Obviously you could create two different designs for two different manufacturers but the overhead and engineering headache for Apple would be substantial and possibly not worth it for only 20% of the production run.

Not worth it? Well over 100M chips at near $40 a pop makes 20% a roughly $1B business. There are companies that spend as much on tape-outs for lesser business opportunities.

 

[mschmalenbach]

I don't think I follow... Why would these companies need different tape-outs and if that is the case, why couldn't Apple supply multiple tape-outs for the manufactures? Could you explain a little?

Different tape-outs will likely be needed for several reasons, even though Apple might be using 14nm (for example) processes from both TSMC and Samsung. "Why?", you might ask...

Samsung and TSMC are unlikely to have identical processes, even if they do use the same equipment in their fabs. Why different processes? Because they were developed differently - unless they licensed the process in its entirety from a 3rd party...

Manufacturing a complex semiconductor involves a lot of steps, some of which require the use of pretty nasty chemicals to etch away other chemicals, thus forming various patterns and structures on the silicon wafer on to which metals like aluminum or copper are deposited (to form signal pathways) or to which dopants are added to form the N type and P type junctions that make up transistors.

The complex computer aided design systems used by Apple and other designers need to know about the characteristics of the target manufacturing process in order to make sure that dopant areas are big enough to enable the transistors to work as intended, that signal pathways aren't too close together that they might short out in manufacturing, or become too narrow and so perhaps affect system timings in a variety of subtle ways. Remember there are upwards of 1-2 BILLION transistors on each of these devices... they pretty much ALL need to work as expected...

So different manufacturing processes equates to different design rules - if you are going to manufacture on 2 different processes, even if they are very similar, you are going to need to simulate your system against each set of design rules, and do your chip layout according to the different process/engineering/layout rules of each - and layout and simulation represent the majority of design time... So you don't go for 2 suppliers with these complex parts for the hell of it - it's a critical decision to make... requires a lot of engineering resources...

But if you're going to be selling 40-80 million devices a quarter, and it's core to your business reputation and model, then you'd be almost negligent to put all your 'eggs in one basket', so to speak... or at least, all in one fab, even if it's only with one supplier...

Creating the mask sets needed for TSMC and Samsung (because they won't be the same) for the 10-14nm process node is likely to cost $10s MILLION per set, and my $10s MILLION we're talking closer to $80M than $30M! If not more...

In summary, it's very unlikely that 2 fab processes developed by 2 different companies, even if they worked together closely, will be sufficiently similar for only 1 tape out to be needed AND still get the maximum density and performance out of the manufactured chips...

... and the cost of each tape out and mask set for each supplier at these process nodes/geometries is VERY HIGH!! So you'd think twice, and then twice again, before going down the road of multiple supplier...

UNLESS it's so critical to your business that you can't afford the risk of poor yields / capacity & supply constraints from your supply chain.

For the procurement/supply chain side of this topic, investigate the financial risk vs supply chain risk 'grid' put forward by Kraljic (it's the basis of supply chain management studies these days, I believe). Google 'kraljic purchasing model'.

----------

Not worth it? Well over 100M chips at near $40 a pop makes 20% a roughly $1B business. There are companies that spend as much on tape-outs for lesser business opportunities.

Tape-outs & mask sets at the 10-14nm process nodes can cost 'in the region of $100M'...

So development by trial and error is out! Hence a LOT of simulation work... and that TAKES TIME and some serious computing power!

Also, just because we're talking Apple, TSMC & Samsung here - about the biggest buyer/suppliers of chips on the planet, doesn't mean that any of them will 'play nice and easy' when it comes to apportioning risk!

 

[mozumder]

Hope this isn't true.

TSMC has had a BAD history of delivering.

Wasn't A8 suppose to be bulk TSMC? They failed and Samsung had to take over.

Apple has wanted to stop using Samsung to make its parts but the truth is...they make the BETTER parts

Samsung > TSMC over chips.
Samsung > LG over displays.

TSMC > Samsung

The business models are different. TSMC is better if you have chip design expertise already. If you have no clue how to make chips, Samsung is better.

TSMC sells you produced wafers. It's your job to make sure the chip works and yields properly. If you don't design it properly, it won't yield well.

Meanwhile, Samsung sells you finished, tested chips.

 

[Iconoclysm]

Hope this isn't true.

TSMC has had a BAD history of delivering.

Wasn't A8 suppose to be bulk TSMC? They failed and Samsung had to take over.

Apple has wanted to stop using Samsung to make its parts but the truth is...they make the BETTER parts

Samsung > TSMC over chips.
Samsung > LG over displays.

Hope you understand that they both make the SAME part - it's all about who will produce a better yield. There's no truth to saying one is better without defining what you mean by better. Especially when we are talking about simple parts.

Samsung=TSMC on chips
Displays are a similar situation but that's not even up for discussion here.

 

[Rogifan]

So many rumors and there's no substantial backing to any of them. Does anyone really care which company is actually going to manufacture the A9 chip. No

Exactly. How would people even know if their A8 chip was fabbed by Samsung or not? Would a Samsung fabbed chip perform differently than one manufactured by TSMC?

 

[Bahroo]

Tbh hoping Samsung wins out here cause 14nm FINfet>>16 nm finfet

 

[sualpine]

Could care less about A9, I just want two damn gigs of RAM in my next iPhone.

 

[usersince86]

I trust their processors are better than their logo.



Like them or not, Samsung has been a good production partner for Apple (and vice-versa, obviously).

 

[Illbay]

Wasn't A8 suppose to be bulk TSMC? They failed and Samsung had to take over.

Where'd you get that from? Last I heard TSMC was producing ALL the A8/A8x chips.

Are you sure you're not misunderstanding the nature of this business, and assuming that because Samsung makes a lot of Android phones, and Android phone users "hate" Apple, therefore Samsung must "hate" Apple?

The reality is, it's about what all these different folks LOVE: money. The pursuit of profit goes way above adolescent tech-weenie pissing contests.

 

[bobenhaus]

Not under sold but over priced for sure. Apple knows how to squeeze blood out of their customers

 

[firewood]

Tape-outs & mask sets at the 10-14nm process nodes can cost 'in the region of $100M'...

So development by trial and error is out! Hence a LOT of simulation work... and that TAKES TIME and some serious computing power!

Do you think Apple wastes the off-peak compute power available from all those massive iTunes and iCloud data centers?

That would fall right in line with "There are things only Apple can do".

 

[Morris]

Different tape-outs will likely be needed for several reasons, even though Apple might be using 14nm (for example) processes from both TSMC and Samsung. "Why?", you might ask...

[..]

Thanks, that's exactly the detailed background I was looking for. In essence, it's possible but very complex.

 

[macs4nw]

TSMC May Win Bulk of A9 Orders Due to Better Production Yield.

Article Link: TSMC May Win Bulk of A9 Orders Due to Better Production Yield

I find it surprising that Samsung, who despite their business antics, are still a manufacturer of quality parts, have reportedly been eclipsed on yield by TSMC.

Either TSMC have considerably stepped up their game, or this article is yet another one of those 'analyst interpretations'. Or perhaps wishful thinking by Taipei's MIC director Chris Hung.

 

[wiz329]

Different tape-outs will likely be needed for several reasons, even though Apple might be using 14nm (for example) processes from both TSMC and Samsung. "Why?", you might ask...

Samsung and TSMC are unlikely to have identical processes, even if they do use the same equipment in their fabs. Why different processes? Because they were developed differently - unless they licensed the process in its entirety from a 3rd party...

Manufacturing a complex semiconductor involves a lot of steps, some of which require the use of pretty nasty chemicals to etch away other chemicals, thus forming various patterns and structures on the silicon wafer on to which metals like aluminum or copper are deposited (to form signal pathways) or to which dopants are added to form the N type and P type junctions that make up transistors.

The complex computer aided design systems used by Apple and other designers need to know about the characteristics of the target manufacturing process in order to make sure that dopant areas are big enough to enable the transistors to work as intended, that signal pathways aren't too close together that they might short out in manufacturing, or become too narrow and so perhaps affect system timings in a variety of subtle ways. Remember there are upwards of 1-2 BILLION transistors on each of these devices... they pretty much ALL need to work as expected...

So different manufacturing processes equates to different design rules - if you are going to manufacture on 2 different processes, even if they are very similar, you are going to need to simulate your system against each set of design rules, and do your chip layout according to the different process/engineering/layout rules of each - and layout and simulation represent the majority of design time... So you don't go for 2 suppliers with these complex parts for the hell of it - it's a critical decision to make... requires a lot of engineering resources...

But if you're going to be selling 40-80 million devices a quarter, and it's core to your business reputation and model, then you'd be almost negligent to put all your 'eggs in one basket', so to speak... or at least, all in one fab, even if it's only with one supplier...

Creating the mask sets needed for TSMC and Samsung (because they won't be the same) for the 10-14nm process node is likely to cost $10s MILLION per set, and my $10s MILLION we're talking closer to $80M than $30M! If not more...

In summary, it's very unlikely that 2 fab processes developed by 2 different companies, even if they worked together closely, will be sufficiently similar for only 1 tape out to be needed AND still get the maximum density and performance out of the manufactured chips...

... and the cost of each tape out and mask set for each supplier at these process nodes/geometries is VERY HIGH!! So you'd think twice, and then twice again, before going down the road of multiple supplier...

UNLESS it's so critical to your business that you can't afford the risk of poor yields / capacity & supply constraints from your supply chain.

For the procurement/supply chain side of this topic, investigate the financial risk vs supply chain risk 'grid' put forward by Kraljic (it's the basis of supply chain management studies these days, I believe). Google 'kraljic purchasing model'.

----------



Tape-outs & mask sets at the 10-14nm process nodes can cost 'in the region of $100M'...

So development by trial and error is out! Hence a LOT of simulation work... and that TAKES TIME and some serious computing power!

Also, just because we're talking Apple, TSMC & Samsung here - about the biggest buyer/suppliers of chips on the planet, doesn't mean that any of them will 'play nice and easy' when it comes to apportioning risk!

+1 for most educational comment I've seen in a long time.

Have you worked in the semiconductor industry?

 

[groovyd]

I like how people on here talk about 'tape-outs' like they know more about it then they might have read off wikipedia maybe once if at all. ok, so lets assume this cumbersome process indeed costs 100m... does this even matter to them? We are talking about billions in sales here. To not spend an extra buck per unit to diversify your supplier seems a bit naive for a company that has been through the ropes a few dozen times already. A bit save the penny to lose the dime so to speak. Of course they have both companies fully ramping up no matter what the process involves. This isn't a dozen chicklets they are selling here, it is hundreds of millions of devices around the entire world... LOL

As a side note I have done microprocessor design at the transistor level and none of this push to manufacturing is really very difficult to do once the manufacturers themselves have a process that they already know to work which happens well before any of these negotiations even begin to occur. Sure yields will change as they refine their methods. For the most part porting a logic design into a particular layout is nowadays mostly automated, push go, sleep tight, wake up the next morning and you have your masks ready to go. Couple niggling timing issues, hand route a few bits here and there and spin.

 

[cambookpro]

Oh good, I've been missing the monthly 'TSMC will produce Ax chip for iPhone in 201x'. It seems every year this crops up.

Why do people get so worked up about who produces the chips?

 

[gavroche]

Samsung > TSMC over chips.
Samsung > LG over displays.

This may be true in the past. And this may be true in the present. But to suggest that all relative positions are set in stone going forward is not true. Who knows which company will make better products in the future. TSMC continually invests crap tons of money to improve. More importantly, there isn't really a need to choose. Apple can continue to receive components from multiple suppliers.

 

[redmac]

Hope this isn't true.

TSMC has had a BAD history of delivering.

Wasn't A8 suppose to be bulk TSMC? They failed and Samsung had to take over.

Apple has wanted to stop using Samsung to make its parts but the truth is...they make the BETTER parts

Samsung > TSMC over chips.
Samsung > LG over displays.

Show us one Samsung computer monitor that is better than LG's offerings. Just today they announced their 2015 AIO models with 27" 1080p displays. Samsung may be good at making over saturated displays in bulk, but in terms of panel quality LG is the king of real computer monitors.

Regarding the part about TSMC, I think you should read the article first.

 

[albusseverus]

What again?

This must be the most frequent story on macrumors. Someone starts a rumour that TSMC is going to supply the majority of chips…

Then they don't.

Somebody should chart this so we can see if there's any improvement, or this is just a TSMC PR press release (or equivalent) republished by lazy (let's be charitable and call them) journalists.​

And Apple is still using a competitor to make their chips, as they always were.

I'm glad there's an alternative supplier, but what we really need is a third supplier that can make millions a day. Maybe if the Sapphire thing works out, Apple could set Foxcon up to do that.

 

[AppleScruff1]

Show us one Samsung computer monitor that is better than LG's offerings. Just today they announced their 2015 AIO models with 27" 1080p displays. Samsung may be good at making over saturated displays in bulk, but in terms of panel quality LG is the king of real computer monitors.

Why not cite some sources showing LG monitors are the best in the business?

 

[hleewell]

Their parts seem fine. I've got a Samsung Evo SSD in my MP. When it comes to consumer products, I don't like them. Samsung TVs have nice displays but horrible everything else, including boot up time and interface. I never knew someone could make it so hard to switch the input on a TV.

I agree. Samsung makes very good Lcd panels and I own 2 of their 1080p LED displays and they work wonderfully with Mac Mini & MacBook. But their interface, and so-called Smart TV features are not their strongest offering.
Also, Samsung can work to improve the consistency of their edge lighting. You can see splotches of uneven lighting when viewing dark scenes of a BluRay movie.

----------

Why not cite some sources showing LG monitors are the best in the business?

Their high end displays are pretty damn good but their mid range & especially low end are atrocious sometimes.

 

[Michaelgtrusa]

Not sure why you portray the GTA sapphire contract as "unfavorable".

If GTA hadn't oversold its ability and delivered on the goods, the deal would have been anything but unfavorable.

GT oversold and Apple didn't investigate enough.