TSMC said to have locked up all 'A12' chip orders for Apple's 2018 iPhones

AppleInsider

Chip manufacturer TSMC has reportedly secured all of Apple's "A12" processor orders for this year's iPhones, once again scoring a coup over Samsung.

The chips will be made using TSMC's 7-nanometer process, DigiTimes claimed on Thursday. The A11 Bionic processor in the iPhone 8 and X is a 10-nanometer model -- shrinking die size makes it possible to pack more performance into the same space and/or improve power efficiency.

To keep ahead of Samsung, TSMC is also said to be preparing a 5-nanometer facility that should start construction later this year and embark on a trial run in the first half of 2019. A 3-nanometer factory valued at $20 billion should begin construction in 2020.

This summer Samsung was rumored to be competing with TSMC for "A12" orders. For years the company was in fact the exclusive manufacturer of A-series chips, making use of a factory in Austin, Texas. Apple decided to phase TSMC into production however, eventually supplanting Samsung completely.

TSMC may have won again this year due to its technological edge in 7-nanometer production. Samsung is expected to launch 7-nanometer lines later this year, with plans for developing 6- and 5-nanometer processes in 2019, and rolling out 4-nanometer chips in 2020.

Apple is believed to be preparing three iPhones for this fall: two OLED models, sized at 5.8 and 6.5 inches, and a 6.1-inch LCD unit intended to be more affordable.

iqatedo

I wonder what the lower limit is for use of grazing incidence X-ray optics and imaging?

iqatedo

Samsung might have 4 nm by 2020 but TSMC's will be smaller and more powerful... to paraphrase someone more powerful than me.

racerhomie3

What happens after 1nm?

Avieshek

racerhomie3 said:

What happens after 1nm?

Picometer.

blastdoor

Samsung? Pfft. They're done. 

The more interesting competitor to TSMC now is Intel, because Intel offers the potential to integrate the cellular modem into the SOC. 

But I wonder if Intel might have waited too long to get serious about going after Apple's business. These fab processes are difficult to assess by marketing name alone -- TSMC's 10nm is as good as Intel's 10nm, for example. But my impression is that the transistor density of TSMC's 10nm is better than Intel's 14 nm, and that TSMC's 7 nm MIGHT beat Intel's 10nm. If so, then Intel might have finally lost its fab process lead. And that would be a very big deal if true. 

But Samsung? Ha. They're done. 

fotoformat

Avieshek said:

racerhomie3 said:

What happens after 1nm?

Picometer.

Those guys at ifixit.com better have steady hands and tiny tweezers when they do that first strip-down!

cloudmobile

blastdoor said:

Samsung? Pfft. They're done. 

The more interesting competitor to TSMC now is Intel, because Intel offers the potential to integrate the cellular modem into the SOC. 

But I wonder if Intel might have waited too long to get serious about going after Apple's business. These fab processes are difficult to assess by marketing name alone -- TSMC's 10nm is as good as Intel's 10nm, for example. But my impression is that the transistor density of TSMC's 10nm is better than Intel's 14 nm, and that TSMC's 7 nm MIGHT beat Intel's 10nm. If so, then Intel might have finally lost its fab process lead. And that would be a very big deal if true. 

But Samsung? Ha. They're done. 

Intel would have to 1) start making ARM SOCs instead of x86 ones and 2) make SOCs that are competitive with the ones made by Samsung and TSMC. Right now, Intel doesn't even have the foundries to make ARM chips, nor the expertise to make them or operate them. Getting that would take years and billions of dollars. That is why Intel spent years and billions trying to get mobile device makers to use their mobile x86 designs instead, and when that failed they decided to exit the mobile business to focus on IoT. Now that Windows runs on ARM nearly as well as it runs on x86, Intel is probably freaking out. And the multiple security issues that have been identified with their chips will only give manufacturers more incentive to switch from x86 to ARM for Windows PCs that don't require professional/workstation type performance. Yes, Samsung is done, but they never really needed Apple's SOC business anyway. They've had 4 straight record quarters without it. Time to let the Samsung thing go. Apple's lawsuit against them was going on 9 years ago, and Samsung has made 20 times as much money supplying Apple with components than Apple would have ever gotten out of that lawsuit anyway. And there is also the not insubstantial fact that Apple has taken far more design ideas for the iPhone X (and 8/7/6S/6) from the Galaxy S and Galaxy Note lines than Samsung ever did from the 1st and 2nd generation iPhone for their iPhone knockoffs that didn't sell anyway.

ksec

cloudmobile said:

blastdoor said:

Samsung? Pfft. They're done. 

The more interesting competitor to TSMC now is Intel, because Intel offers the potential to integrate the cellular modem into the SOC. 

But I wonder if Intel might have waited too long to get serious about going after Apple's business. These fab processes are difficult to assess by marketing name alone -- TSMC's 10nm is as good as Intel's 10nm, for example. But my impression is that the transistor density of TSMC's 10nm is better than Intel's 14 nm, and that TSMC's 7 nm MIGHT beat Intel's 10nm. If so, then Intel might have finally lost its fab process lead. And that would be a very big deal if true. 

But Samsung? Ha. They're done. 

Intel would have to 1) start making ARM SOCs instead of x86 ones and 2) make SOCs that are competitive with the ones made by Samsung and TSMC. Right now, Intel doesn't even have the foundries to make ARM chips, nor the expertise to make them or operate them. Getting that would take years and billions of dollars. That is why Intel spent years and billions trying to get mobile device makers to use their mobile x86 designs instead, and when that failed they decided to exit the mobile business to focus on IoT. Now that Windows runs on ARM nearly as well as it runs on x86, Intel is probably freaking out. And the multiple security issues that have been identified with their chips will only give manufacturers more incentive to switch from x86 to ARM for Windows PCs that don't require professional/workstation type performance. Yes, Samsung is done, but they never really needed Apple's SOC business anyway. They've had 4 straight record quarters without it. Time to let the Samsung thing go. Apple's lawsuit against them was going on 9 years ago, and Samsung has made 20 times as much money supplying Apple with components than Apple would have ever gotten out of that lawsuit anyway. And there is also the not insubstantial fact that Apple has taken far more design ideas for the iPhone X (and 8/7/6S/6) from the Galaxy S and Galaxy Note lines than Samsung ever did from the 1st and 2nd generation iPhone for their iPhone knockoffs that didn't sell anyway.

Oh dear No. Intel has been making ARM Chips, ( They are using ARM inside controllers ) and they do have the license to Fab ARM chips. Basically the highlighted word are false. Search Intel Custom Foundry.

tjwolf

fotoformat said:

Avieshek said:

racerhomie3 said:

What happens after 1nm?

Picometer.

Those guys at ifixit.com better have steady hands and tiny tweezers when they do that first strip-down!

Have them hire Trump with his tiny hands.

tjwolf

blastdoor said:

Samsung? Pfft. They're done. 

The more interesting competitor to TSMC now is Intel, because Intel offers the potential to integrate the cellular modem into the SOC. 

But I wonder if Intel might have waited too long to get serious about going after Apple's business. These fab processes are difficult to assess by marketing name alone -- TSMC's 10nm is as good as Intel's 10nm, for example. But my impression is that the transistor density of TSMC's 10nm is better than Intel's 14 nm, and that TSMC's 7 nm MIGHT beat Intel's 10nm. If so, then Intel might have finally lost its fab process lead. And that would be a very big deal if true. 

But Samsung? Ha. They're done. 

Would it not be possible (maybe not likely, but possible) for Intel to let Apple put their modem design onto the Ax SOC?  Sort of like what they did with Imagination's PowerVR processors - weren't they on the SOC also (before Apple decided to roll its own recently)?

tmay

ksec said:

cloudmobile said:

blastdoor said:

Samsung? Pfft. They're done. 

The more interesting competitor to TSMC now is Intel, because Intel offers the potential to integrate the cellular modem into the SOC. 

But I wonder if Intel might have waited too long to get serious about going after Apple's business. These fab processes are difficult to assess by marketing name alone -- TSMC's 10nm is as good as Intel's 10nm, for example. But my impression is that the transistor density of TSMC's 10nm is better than Intel's 14 nm, and that TSMC's 7 nm MIGHT beat Intel's 10nm. If so, then Intel might have finally lost its fab process lead. And that would be a very big deal if true. 

But Samsung? Ha. They're done. 

Intel would have to 1) start making ARM SOCs instead of x86 ones and 2) make SOCs that are competitive with the ones made by Samsung and TSMC. Right now, Intel doesn't even have the foundries to make ARM chips, nor the expertise to make them or operate them. Getting that would take years and billions of dollars. That is why Intel spent years and billions trying to get mobile device makers to use their mobile x86 designs instead, and when that failed they decided to exit the mobile business to focus on IoT. Now that Windows runs on ARM nearly as well as it runs on x86, Intel is probably freaking out. And the multiple security issues that have been identified with their chips will only give manufacturers more incentive to switch from x86 to ARM for Windows PCs that don't require professional/workstation type performance. Yes, Samsung is done, but they never really needed Apple's SOC business anyway. They've had 4 straight record quarters without it. Time to let the Samsung thing go. Apple's lawsuit against them was going on 9 years ago, and Samsung has made 20 times as much money supplying Apple with components than Apple would have ever gotten out of that lawsuit anyway. And there is also the not insubstantial fact that Apple has taken far more design ideas for the iPhone X (and 8/7/6S/6) from the Galaxy S and Galaxy Note lines than Samsung ever did from the 1st and 2nd generation iPhone for their iPhone knockoffs that didn't sell anyway.

Oh dear No. Intel has been making ARM Chips, ( They are using ARM inside controllers ) and they do have the license to Fab ARM chips. Basically the highlighted word are false. Search Intel Custom Foundry.

The truth is that Intel could certainly make ARM processors for Apple, but Apple would ultimately have control of pricing power via competition in fabs.

On the other hand, I consider it likely that both Apple and MS would be interested in ARM for the desktop, with or without x86/x64, and this is something that I could see as a viable market for Intel to pursue.

tjwolf

ksec said:

cloudmobile said:

blastdoor said:

Samsung? Pfft. They're done. 

The more interesting competitor to TSMC now is Intel, because Intel offers the potential to integrate the cellular modem into the SOC. 

But I wonder if Intel might have waited too long to get serious about going after Apple's business. These fab processes are difficult to assess by marketing name alone -- TSMC's 10nm is as good as Intel's 10nm, for example. But my impression is that the transistor density of TSMC's 10nm is better than Intel's 14 nm, and that TSMC's 7 nm MIGHT beat Intel's 10nm. If so, then Intel might have finally lost its fab process lead. And that would be a very big deal if true. 

But Samsung? Ha. They're done. 

Intel would have to 1) start making ARM SOCs instead of x86 ones and 2) make SOCs that are competitive with the ones made by Samsung and TSMC. Right now, Intel doesn't even have the foundries to make ARM chips, nor the expertise to make them or operate them. Getting that would take years and billions of dollars. That is why Intel spent years and billions trying to get mobile device makers to use their mobile x86 designs instead, and when that failed they decided to exit the mobile business to focus on IoT. Now that Windows runs on ARM nearly as well as it runs on x86, Intel is probably freaking out. And the multiple security issues that have been identified with their chips will only give manufacturers more incentive to switch from x86 to ARM for Windows PCs that don't require professional/workstation type performance. Yes, Samsung is done, but they never really needed Apple's SOC business anyway. They've had 4 straight record quarters without it. Time to let the Samsung thing go. Apple's lawsuit against them was going on 9 years ago, and Samsung has made 20 times as much money supplying Apple with components than Apple would have ever gotten out of that lawsuit anyway. And there is also the not insubstantial fact that Apple has taken far more design ideas for the iPhone X (and 8/7/6S/6) from the Galaxy S and Galaxy Note lines than Samsung ever did from the 1st and 2nd generation iPhone for their iPhone knockoffs that didn't sell anyway.

Oh dear No. Intel has been making ARM Chips, ( They are using ARM inside controllers ) and they do have the license to Fab ARM chips. Basically the highlighted word are false. Search Intel Custom Foundry.

I was going to comment/ask what is, from a foundry perspective, the difference between producing wafers that have ARM SOCs and Intel SOCs?  A transistor is a transistor is a transistor, no matter the instruction set (ARM vs. x86) or type of cache or...isn't it?

blastdoor

tjwolf said:

blastdoor said:

Samsung? Pfft. They're done. 

The more interesting competitor to TSMC now is Intel, because Intel offers the potential to integrate the cellular modem into the SOC. 

But I wonder if Intel might have waited too long to get serious about going after Apple's business. These fab processes are difficult to assess by marketing name alone -- TSMC's 10nm is as good as Intel's 10nm, for example. But my impression is that the transistor density of TSMC's 10nm is better than Intel's 14 nm, and that TSMC's 7 nm MIGHT beat Intel's 10nm. If so, then Intel might have finally lost its fab process lead. And that would be a very big deal if true. 

But Samsung? Ha. They're done. 

Would it not be possible (maybe not likely, but possible) for Intel to let Apple put their modem design onto the Ax SOC?  Sort of like what they did with Imagination's PowerVR processors - weren't they on the SOC also (before Apple decided to roll its own recently)?

Yes, that would be possible. But as you say, not likely. 

Intel made a massive error in rejecting Apple for the iPhone. That decision has pushed literally tens of billions of dollars into the coffers of TSMC (and Samsung), allowing TSMC to catch up to Intel in process tech. Right now, today, TSMC is manufacturing more advanced chips than Intel (I mean the manufacturing process). That's because TSMC 10nm is better than Intel 14nm. When Intel comes out with their 10 nm, they will surpass TSMC again, but that may be very short lived if TSMC 7nm beats Intel 10nm (it seems possible that it will, but we won't know until we see shipping products). 

For decades, Intel's profit margins have depended on two things (1) x86 near-monopoly and (2) manufacturing process leadership. Now both are in serious jeopardy. 

If Intel is smart, they will do WHATEVER IT TAKES to fab fully integrated SOCs for Apple. 

macky the macky

tjwolf said:

I was going to comment/ask what is, from a foundry perspective, the difference between producing wafers that have ARM SOCs and Intel SOCs?  A transistor is a transistor is a transistor, no matter the instruction set (ARM vs. x86) or type of cache or...isn't it?

Intel could charge hundreds of dollars for a x86 chip where they’d virtually have to give away an ARM chip, and they (incorrectly) felt that they didn’t need the low profit ARM buseness. With the ARM chips becoming quickly much more powerful than expected, the high-profit x86 business may evaporate much more quickly than expected...

tjwolf

blastdoor said:

tjwolf said:

blastdoor said:

Samsung? Pfft. They're done. 

The more interesting competitor to TSMC now is Intel, because Intel offers the potential to integrate the cellular modem into the SOC. 

But I wonder if Intel might have waited too long to get serious about going after Apple's business. These fab processes are difficult to assess by marketing name alone -- TSMC's 10nm is as good as Intel's 10nm, for example. But my impression is that the transistor density of TSMC's 10nm is better than Intel's 14 nm, and that TSMC's 7 nm MIGHT beat Intel's 10nm. If so, then Intel might have finally lost its fab process lead. And that would be a very big deal if true. 

But Samsung? Ha. They're done. 

Would it not be possible (maybe not likely, but possible) for Intel to let Apple put their modem design onto the Ax SOC?  Sort of like what they did with Imagination's PowerVR processors - weren't they on the SOC also (before Apple decided to roll its own recently)?

Yes, that would be possible. But as you say, not likely. 

Intel made a massive error in rejecting Apple for the iPhone. That decision has pushed literally tens of billions of dollars into the coffers of TSMC (and Samsung), allowing TSMC to catch up to Intel in process tech. Right now, today, TSMC is manufacturing more advanced chips than Intel (I mean the manufacturing process). That's because TSMC 10nm is better than Intel 14nm. When Intel comes out with their 10 nm, they will surpass TSMC again, but that may be very short lived if TSMC 7nm beats Intel 10nm (it seems possible that it will, but we won't know until we see shipping products). 

For decades, Intel's profit margins have depended on two things (1) x86 near-monopoly and (2) manufacturing process leadership. Now both are in serious jeopardy. 

If Intel is smart, they will do WHATEVER IT TAKES to fab fully integrated SOCs for Apple. 

It seems to me that this (getting their modem into Apple's SOC) would be a pretty desirable goal: while it's obviously not as profitable as selling an x86 chip, volume could potentially offset per-unit profits.  More importantly, though, is that it would pretty much guarantee that Intel would become the sole modem supplier for Apple.  Currently the modem is a discrete part and Apple uses availability of multiple vendors (mainly Qualcomm and Intel) to negotiate lower prices.  For Apple, in the past, that bargaining power was more important than any technological benefits of placing the modem on their SOC.  With relations with Qualcomm souring badly, Apple might be amenable to a change in this regard (the only complicating factor is that Intel's modems still don't do CDMA, but with everyone moving to LTE and beyond, that might not be a consideration in a year or two?)

tshapi

Isn’t intel working with Apple on a 5g modem chip? And isn’t Apple working also on a power management chip?  It’s only a matter of time before all the important chips in the iPhone end up in the Soc

harry wild

A12 @ 7 nm 5.8" iPhone is my in December 2018!

fastasleep

racerhomie3 said:

What happens after 1nm?

Measurements aren't limited to integers.

colinng

tjwolf said:

More importantly, though, is that it would pretty much guarantee that Intel would become the sole modem supplier for Apple.  Currently the modem is a discrete part and Apple uses availability of multiple vendors (mainly Qualcomm and Intel) to negotiate lower prices.  For Apple, in the past, that bargaining power was more important than any technological benefits of placing the modem on their SOC.  With relations with Qualcomm souring badly, Apple might be amenable to a change in this regard (the only complicating factor is that Intel's modems still don't do CDMA, but with everyone moving to LTE and beyond, that might not be a consideration in a year or two?)

I think Apple is working on their own modem and GPU initiatives. I recently looked at modem firmware, and it's a disgusting mess with at least 3 different languages and frameworks all cobbled together, and multiple APIs (in different styles) to access the functionality. It's surprising anything works. FYI the modem in the original iPhone had so much processing power excess that the iPhone outsourced its sound generation functions to it. 

If Apple makes a cellular modem, it will probably blow away its competition for power and space efficiency, and it will probably be evolved and maintained more easily. The Qualcomm lawsuits serve as a "test" of whether they can invalidate some of the key patents, or whether they can finally get FRAND terms. If they can legally build their own without having Qualcomm et al suck them dry for work they didn't do, I think they'd just roll their own.

37 million iPhones sold per quarter, the world deserves a better cellular modem. I think Apple has proven their radio prowess with W1 and W2 - the most functional, reliable, power efficient, and possibly smallest - Bluetooth radio chips on the planet. And it's like their first in-house Bluetooth silicon. Imagine what they'd do with a cellular chip! 

As for whether the modem is integrated on-chip - I don't think Apple cares that much. The lines are blurring. The Apple Watch 3 has an LTE modem that is on-package. Is that really all that worse than on-chip? Modern mobile processors seem to include multiple components from different manufacturers, colocated on the same package. Even the humble Raspberry Pi stacks RAM from another party on top of the Broadcom ARM processor. 

ksec

tjwolf said:

ksec said:

cloudmobile said:

blastdoor said:

Samsung? Pfft. They're done. 

The more interesting competitor to TSMC now is Intel, because Intel offers the potential to integrate the cellular modem into the SOC. 

But I wonder if Intel might have waited too long to get serious about going after Apple's business. These fab processes are difficult to assess by marketing name alone -- TSMC's 10nm is as good as Intel's 10nm, for example. But my impression is that the transistor density of TSMC's 10nm is better than Intel's 14 nm, and that TSMC's 7 nm MIGHT beat Intel's 10nm. If so, then Intel might have finally lost its fab process lead. And that would be a very big deal if true. 

But Samsung? Ha. They're done. 

Intel would have to 1) start making ARM SOCs instead of x86 ones and 2) make SOCs that are competitive with the ones made by Samsung and TSMC. Right now, Intel doesn't even have the foundries to make ARM chips, nor the expertise to make them or operate them. Getting that would take years and billions of dollars. That is why Intel spent years and billions trying to get mobile device makers to use their mobile x86 designs instead, and when that failed they decided to exit the mobile business to focus on IoT. Now that Windows runs on ARM nearly as well as it runs on x86, Intel is probably freaking out. And the multiple security issues that have been identified with their chips will only give manufacturers more incentive to switch from x86 to ARM for Windows PCs that don't require professional/workstation type performance. Yes, Samsung is done, but they never really needed Apple's SOC business anyway. They've had 4 straight record quarters without it. Time to let the Samsung thing go. Apple's lawsuit against them was going on 9 years ago, and Samsung has made 20 times as much money supplying Apple with components than Apple would have ever gotten out of that lawsuit anyway. And there is also the not insubstantial fact that Apple has taken far more design ideas for the iPhone X (and 8/7/6S/6) from the Galaxy S and Galaxy Note lines than Samsung ever did from the 1st and 2nd generation iPhone for their iPhone knockoffs that didn't sell anyway.

Oh dear No. Intel has been making ARM Chips, ( They are using ARM inside controllers ) and they do have the license to Fab ARM chips. Basically the highlighted word are false. Search Intel Custom Foundry.

I was going to comment/ask what is, from a foundry perspective, the difference between producing wafers that have ARM SOCs and Intel SOCs?  A transistor is a transistor is a transistor, no matter the instruction set (ARM vs. x86) or type of cache or...isn't it?

License. Especially for Intel. You cant just hand out all the ARM IP to Intel and expect them to Fab it for you, since they are not a pure play company and ARM needs to guarantee its right and IP are preserved and protected.