Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

daywiz

macrumors regular
Original poster
Nov 12, 2012
128
0
I am not an expert at SoC information, but I would certainly would like to dwell more on this and more detailed explanation from any of the forum members would be welcome.

I would like to know if there is a substantial improvement if we move from a 32nm SoC to 28nm SoC. I know it won't be a drastic improvement, as compared to the 45nm (iPad 2,1) to the 32nm ( iPad 2,4) but what are the extra benefits, other than the decreased power consumption.

If there is substantial benefit, I would like to add some speculation to a future roadmap for iPad.

TSMC's 20nm quad-core "supposed" A7 is not expected to be in full production until late into the last quarter of 2013. At, the same time Apple wants to move production away from Samsung.

At that stage, it would be unwise for Apple to quickly jump into TSMC's 20nm. Wouldn't it be better for them to try TSMC by using A6 custom design with a 28nm design, and utilize that for the iPad 5. TSMC already has full production of the 28nm process.

Now let's suppose, Apple decides to launch the iPad 5 in June 2013 at the WWDC, with the product available at stores from late June ( 9 months after the 4th-gen iPad). Since the A7 won't be ready, even late into Oct 2013, they can use the 28nm A6, with a different moniker i.e " A6Y" for marketing purposes. It can help Apple substantinate their features such as improved battery life, thinner / lighter, if they can use it along with IGZO screen tech.

Apple can then push the A5X into a retina iPad mini in Oct 2013 ( and keeping the Mini one generation behind with the full-size iPad). My anticipation is that A7 will then make its entrance into the iPad 6, which may come out in April 2014 ( again 10 months later..and shifting back to the regular cycle) giving TSMC & Apple enough time to work with SoC & production capacity, while also tweaking as necessary.

What do you think?
 

KeepCalmPeople

macrumors 65816
Sep 5, 2012
1,457
659
Los Angeles, California
The A5X in the retina mini would put it two generations behind an A7X iPad within 6 months...? And the A5X chip needs a lot of battery to maintain 10 hours+ of use, if the iPad 3rd gen is any indication.
Am I missing something? Perhaps with a IGZO display it will be able to manage with a smaller battery I guess. It seems to me that performance isn't such an issue, but reducing power consumption of the chips is.
I for one hope they don't roll out new products quite so quickly. It will upset Apple devotees who prefer a yearly upgrade cycle.
 

lianlua

macrumors 6502
Jun 13, 2008
370
3
I would like to know if there is a substantial improvement if we move from a 32nm SoC to 28nm SoC. I know it won't be a drastic improvement, as compared to the 45nm (iPad 2,1) to the 32nm ( iPad 2,4) but what are the extra benefits, other than the decreased power consumption.
Decreased size means smaller transistors. Smaller transistors have less resistance, which means less power is needed and therefore less heat is produced. Smaller transistors and less physical distance to travel also imply less capacitance, which means they can switch faster (that's why you eventually need smaller sizes to achieve clock speed increases). You can also fit more onto each wafer because of the smaller physical size (or pack more transistors into the same size SoC), potentially reducing manufacturing costs.

But each process reduction piles on a number of roadblocks for manufacturing equipment design, leakage, and yield rates, among other problems. So it's not as simple as just printing everything smaller.
At that stage, it would be unwise for Apple to quickly jump into TSMC's 20nm. Wouldn't it be better for them to try TSMC by using A6 custom design with a 28nm design, and utilize that for the iPad 5.
If you're going to test a new fab and a new process size, it doesn't really matter which one you go with. The only reasons to go with 28nm over 20nm given the choice is if there isn't sufficient capacity or high enough yields at 20nm, or if the design itself simply doesn't scale down that far.

It's possible that 28nm shrinks down without much work but that 20nm requires some more extensive tweaking to the design to combat production issues. But assuming that's not the case and that there's enough capacity at a low enough defect rate, skipping straight to 20nm would be the way to go. There's no benefit (and huge financial and time/effort costs) in tooling up for an intermediate size.
 

daywiz

macrumors regular
Original poster
Nov 12, 2012
128
0
The A5X in the retina mini would put it two generations behind an A7X iPad within 6 months...? And the A5X chip needs a lot of battery to maintain 10 hours+ of use, if the iPad 3rd gen is any indication.
Am I missing something? Perhaps with a IGZO display it will be able to manage with a smaller battery I guess. It seems to me that performance isn't such an issue, but reducing power consumption of the chips is.
I for one hope they don't roll out new products quite so quickly. It will upset Apple devotees who prefer a yearly upgrade cycle.


-------

I think you got me wrong here. I meant that if indeed an iPad 5 is launching in July 2013, it would mean there wont b an A7 / A7X ready in time for it. A one year cycle at the current stage would mean, it gives competition a relatively long time to play catch up, if not move ahead. If they go ahead with June 2013 launch, they would have to utilize the same or further tweaked version of the A6X. It was at this juncture, I thought they might tweak the A6X to a 28nm SoC, downsized from 32nm ( and making the switch from Samsung to TSMC). If they are utilising IGZO for the Retina display, then the full-size and mini do not require such large batteries and since the iPad is relatively smaller, the A5X might be powerful enough. I don't believe apple will keep both the ipads at the same generation prcessors, as it will almost kill the bigger iPad.i feel it might be a good product line-up, if they start the iPad / Mini in June, iPhone / iPods in Sept and Other hardware in Oct. It keeps the year nearly full and builds up nicely to the holiday season
 

daywiz

macrumors regular
Original poster
Nov 12, 2012
128
0
If you're going to test a new fab and a new process size, it doesn't really matter which one you go with. The only reasons to go with 28nm over 20nm given the choice is if there isn't sufficient capacity or high enough yields at 20nm, or if the design itself simply doesn't scale down that far.

It's possible that 28nm shrinks down without much work but that 20nm requires some more extensive tweaking to the design to combat production issues. But assuming that's not the case and that there's enough capacity at a low enough defect rate, skipping straight to 20nm would be the way to go. There's no benefit (and huge financial and time/effort costs) in tooling up for an intermediate size.

---

Considering that TSMC is a new supplier to Apple, and that A7 isn't ready for full production till late in 2013, don't you think that would mean, they won't have any SoC upgrades for full-size iPad, while the Mini can get upgraded to A5X ( I don't think A6 / A6X will make the cut) as long as the bigger iPad hasn't moved up a generation. If Apple manages to get A7 ready in time, then thats a different story.

Also judging by the explanation you gave, I am thinking it might be alot easier for them to scale down from 32 to 28 ( as TSMC has been doing a full-fledged production of the 28nm..for a while now). Plus it gives, apple the ability to test their new foundry partner and buys them enough time to set-up a breathtaking quad-core set-up for the next iPad in 2014. A rushed up project may bring about results similar to the iPad 3rd generation.

adding to that, are you sure there isn't much benefit..or no benefit @ all? As for tooling costs, its bound to be there but won't it off-set the costs over time?
 

corvus32

macrumors 6502a
Sep 4, 2009
761
0
USA
I think the A5X is dead. There's no sense in bringing it back.

Apple could die shrink the A6 or A6X for the next Mini depending on if it goes retina or not.
 

Menel

Suspended
Aug 4, 2011
6,351
1,356
Decreased size means smaller transistors. Smaller transistors have less resistance, which means less power is needed and therefore less heat is produced. Smaller transistors and less physical distance to travel also imply less capacitance, which means they can switch faster (that's why you eventually need smaller sizes to achieve clock speed increases). You can also fit more onto each wafer because of the smaller physical size (or pack more transistors into the same size SoC), potentially reducing manufacturing costs.

But each process reduction piles on a number of roadblocks for manufacturing equipment design, leakage, and yield rates, among other problems. So it's not as simple as just printing everything smaller.

If you're going to test a new fab and a new process size, it doesn't really matter which one you go with. The only reasons to go with 28nm over 20nm given the choice is if there isn't sufficient capacity or high enough yields at 20nm, or if the design itself simply doesn't scale down that far.

It's possible that 28nm shrinks down without much work but that 20nm requires some more extensive tweaking to the design to combat production issues. But assuming that's not the case and that there's enough capacity at a low enough defect rate, skipping straight to 20nm would be the way to go. There's no benefit (and huge financial and time/effort costs) in tooling up for an intermediate size.
Well said. I just wanted to emphasize that these are huge IFs and engineering hurdles here that even most nerds don't realize.

And TSMC has other customers.

----------

I think the A5X is dead. There's no sense in bringing it back.

Apple could die shrink the A6 or A6X for the next Mini depending on if it goes retina or not.
Yes, A6X is already die shrunk from A5X. The engineering to bring A5X down a process level, and do all the qualifications, would make little sense.
 

ghellquist

macrumors regular
Aug 21, 2011
146
5
Stockholm Sweden
And now something completely different...

What about open heart surgery? I mean, that is probably an area where it is easier to find reliably information than on what Apple is planning to do. And where we all are as much amateurs as on, ... , well, what Apple is planning to do

// Gunnar
Note: beeing a bit tired of reading of guessing. Take no notice, please continue!
 

darngooddesign

macrumors P6
Jul 4, 2007
17,913
9,416
Atlanta, GA
While it is impossibe to say for sure,the second gen iPad 2 had a similar reduction in its SOC and it gained around two extra hours of battery life with everything else being equal.
 

lianlua

macrumors 6502
Jun 13, 2008
370
3
Considering that TSMC is a new supplier to Apple, and that A7 isn't ready for full production till late in 2013, don't you think that would mean, they won't have any SoC upgrades for full-size iPad, while the Mini can get upgraded to A5X ( I don't think A6 / A6X will make the cut) as long as the bigger iPad hasn't moved up a generation. If Apple manages to get A7 ready in time, then thats a different story.
I'm sorry, I don't quite follow what you're saying here. TSMC isn't supplying any SOCs to Apple at the moment as far as I know, and it's unlikely that the 9.7" iPad is going to get a new processor before the second-generation mini is released.

There will never be an A5X mini. Any future upgrade is going to be to an A6 or better, regardless of the retina factor. This is far more important to the device than going to a 2048x1536 display, which makes very little sense for an entry-level tablet in the current market.
Also judging by the explanation you gave, I am thinking it might be alot easier for them to scale down from 32 to 28 ( as TSMC has been doing a full-fledged production of the 28nm..for a while now). Plus it gives, apple the ability to test their new foundry partner and buys them enough time to set-up a breathtaking quad-core set-up for the next iPad in 2014.
There's no point in "testing" at 28nm if your next move is to 20nm, nor does doing so buy any time. The process size decision is one based on price (i.e., tooling and yield) and available capacity.
adding to that, are you sure there isn't much benefit..or no benefit @ all?
None whatsoever, all things being equal. There's no point in paying for 28nm buildout if 20nm is available at the price and volume you need. "Testing" a new supplier just isn't done that way.
I think the A5X is dead. There's no sense in bringing it back.

Apple could die shrink the A6 or A6X for the next Mini depending on if it goes retina or not.
Absolutely agreed.
 

daywiz

macrumors regular
Original poster
Nov 12, 2012
128
0
I'm sorry, I don't quite follow what you're saying here. TSMC isn't supplying any SOCs to Apple at the moment as far as I know, and it's unlikely that the 9.7" iPad is going to get a new processor before the second-generation mini is released.

There will never be an A5X mini. Any future upgrade is going to be to an A6 or better, regardless of the retina factor. This is far more important to the device than going to a 2048x1536 display, which makes very little sense for an entry-level tablet in the current market.


There's no point in "testing" at 28nm if your next move is to 20nm, nor does doing so buy any time. The process size decision is one based on price (i.e., tooling and yield) and available capacity.

None whatsoever, all things being equal. There's no point in paying for 28nm buildout if 20nm is available at the price and volume you need. "Testing" a new supplier just isn't done that way.

Absolutely agreed.

Well, thanks for all the information. Now sincerely, hope the A7 / A7X variations are ready in time for the ipad 5.

a 20nm SoC combined to growing efficient Wi-fi / Bluetooth chips and a less power-hungry display ( i.e. MEMS display, IGZO) should easily pave way for a massive battery life improvement. TSMC claim is that the 20nm is 25% more efficient than its 28nm version ( so I am assuming around the same improvement over the Samsung 32nm - since there isn't too much difference between the two). MEMS / IGZO tech claims around another 40-60% improved battery life on display. Putting the new chip & screen together should easily crank up the battery life by at least 4-5 hours on normal use,which is mighty impressive. ( unless apple decides to rapidly decrease the battery size / capacity in a bid to get reduce weight / thickness and retain the same 10-11 hour life. I am not concerned about the thickness, jz d weight.
 

flameproof

macrumors 6502a
Jan 14, 2011
615
18
I would like to know if there is a substantial improvement if we move from a 32nm SoC to 28nm SoC.

No. It's just a different size, that's it.

If you use in your circuit 0201, 0604 or 0805 size resistors doesn't make any difference to the function of the product.
 

Menel

Suspended
Aug 4, 2011
6,351
1,356
No. It's just a different size, that's it.

If you use in your circuit 0201, 0604 or 0805 size resistors doesn't make any difference to the function of the product.
Not so simple.

Smaller die process has many potential improvements. Higher clocks for same power envelop. Or better battery life for same clocks. Etc.

See iPad 2b for example.
 

flameproof

macrumors 6502a
Jan 14, 2011
615
18
Not so simple.

Smaller die process has many potential improvements. Higher clocks for same power envelop. Or better battery life for same clocks. Etc.

See iPad 2b for example.

Maybe there are a few hidden benefits on the paper. Still no added functions to the customers.....

Same for multicore CPUs, how many cores doesn't really matter to the user. What I care is how the device works, whether it's achieved with 1 or 8 cores I don't care.
 

daywiz

macrumors regular
Original poster
Nov 12, 2012
128
0
There's no point in "testing" at 28nm if your next move is to 20nm, nor does doing so buy any time. The process size decision is one based on price (i.e., tooling and yield) and available capacity.

None whatsoever, all things being equal. There's no point in paying for 28nm buildout if 20nm is available at the price and volume you need. "Testing" a new supplier just isn't done that way.

Looks like our little discussion has had its effects. Rumour mills NOW picking up a potential move of the Apple A7 to be produced by TSMC on a 28nm process, while the 20nm is expected in mid-2014. Looks like Apple badly wants to leave Samsung and they may be able to squeeze out an improved performance ( albeit small but enough to keep people happy) while reducing battery consumption ( every tiny bit helps!). I think its a good investment, considering we have to wait approx 1.5 years for the 20nm.

And suddenly, I see Samsung massively endorsing their 28nm die-shrunk version of the 32nm, with slight improvements ( could it be possibly be to entice Apple to stay??)
 
Last edited:

lianlua

macrumors 6502
Jun 13, 2008
370
3
Putting the new chip & screen together should easily crank up the battery life by at least 4-5 hours on normal use,which is mighty impressive.
Not likely. More like a 2-3 hour overall improvement on the same basic specs. Efficiency claims tend to be misunderstood or based on synthetic information or preproduction testing, and you also have to contend with efficiency gains being offset by higher performance demands.

At this point, the main goal of having all-day battery life is generally obtained, apart from some particularly demanding use cases, and so I'd expect concentrated optimization in the power-hungry modes. Generally speaking, they're going to want smaller, lighter batteries rather than longer life for the next generations of tablets. It's all cyclical.
Looks like our little discussion has had its effects. Rumour mills NOW picking up a potential move of the Apple A7 to be produced by TSMC on a 28nm process, while the 20nm is expected in mid-2014.
Maybe. Still remains to be seen whether TSMC has the capacity to deliver the needed volume.
I think its a good investment, considering we have to wait approx 1.5 years for the 20nm.
Closer to two years, since you need ramp time, but yes, if they want to switch away from Samsung sooner, it's an essential investment.

I don't see it as terribly likely that Apple is actually trying to move away from Samsung so much as trying to continue and expand self-reliance and supplier options. They don't want to get stuck in a corner again.

The media pushes the emotional story, but the reality is that Apple will probably continue to use Samsung for a large number of components. Now that they are getting deeper into custom hardware and reaching stable volumes, they need to diversify suppliers and be more concerned about their IP. Samsung Electronics is not really a factor from the supply chain perspective--Samsung is a conglomerate of companies with varying degrees of autonomy. Samsung Electronics is just as much a customer of e.g., Samsung Display as Apple is. The components-supplying divisions aren't in a dispute with Apple.

The media just likes drama. The only concerns Apple has with using Samsung is sole-sourcing and the potential for spies from other Samsung units.
 

daywiz

macrumors regular
Original poster
Nov 12, 2012
128
0
According to my reading and research into the topic, I think the new A6X will indeed use a 28nm process, making it the new A7.

Here's the potential benefit of switching to 28nm compared to the 32nm, as its a half node approach.

1. About 13% higher speed
2. About 0.85 times the density
3. About 11% less power consumption

Now considering tweaked Die-shrunk A5 chip as an example (2 means square):

1. A5 on 45nm -> 10.09 mm by 12.15 mm - 123 mm2
2. A5 on 32nm -> 8.19 mm by 8.68 mm - 71.1 mm2 (41% smaller)
3. A5 on 28nm -> 6.00 mm by 6.00 mm - 36 mm2 (50% smaller than 32nm)

Let's c what can happen to A6/A6X, if we estimate around 40% decrease instead of the 50% shown in the A5.

1. A6 on 32nm process -> 96.71 mm2
2. A6 on 28nm process -> 58.03 mm2

Similarly A6X

1. A6X on 32nm process -> 124 mm2
2. A6X on 28nm process -> 75 mm2

That would make it nearly the same size as the chip on the modified ipad 2.

Along with TSMC, Apple can also utilize Samsung to make the 28nm SoC, reducing their dependency on Samsung alone.

Looks like there is substantial benefit switching to 28nm after all. Add this to an IGZO display, and utilising the PowerVR Series6 ( with nearly 10-20times performance ability) the ipad 5 could basically trash tegra 4 and other upcoming devices in terms of raw performance.
 

daywiz

macrumors regular
Original poster
Nov 12, 2012
128
0
On-die logic doesn't scale linearly with fabrication process.

maybe not, but at least we get a rough idea on the size, and in this case I have considered a much less percentage of decrease around 40% - when it should be 50%, but i did consider that the on-die logic wouldn't be linear, thus the difference.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.