TSMC Begins Trial Manufacturing of Apple A6 Processors

[pacmacman]

I own a Macbook Air, on my 2nd iphone, and will be on my 3rd iphone when it comes time to upgrade. I'm hardly an Apple hater. I'm just not an Apple cheerleader and I don't think Steve Jobs is god. Know the difference.

So in other words, your comments shouldn't be taken as a reflection on the devices you use or why you use them? Nor should anyone extrapolate that you find Steve Jobs to be your holy savior due to the fact that you prefer Apple products? Interesting. Tell me more?

 

[cmaier]

How is it not? Put your apple shaded glasses away for a minute and stop being such a senseless fanboy and actually think about what you're saying for a change.

LOL. I'm not generally considered an apple "fanboy" around here.

Apple's products depend on these processors and they have a steady and reliable supply right now through Samsung. Changing suppliers for a vital part of so many of your products is not something Apple is taking lightly, I assure you.

Ok. Who said anything about Apple "taking [anything] lightly?" I just said it's not a headache, and it's not. If you knew anything about designing processors, you would know it's far easier to use a fab that is heavily supported by EDA vendors (like TSMC or UMC) than it is to use one that is not (like Samsung).

Moreover, apple does not have to "contend" with anything. They don't manufacture any circuits, so what exactly are they having to "contend" with as to the manufacturing processes?

You obviously haven't the slightest idea what you are talking about. Apple designs the circuits. To do this they design macroblocks at the transistor level, and design logic circuits at the standard cell level. They place and route the circuits, and create gdsii files, which they fracture and tape-out to the fab. Apple pays for every wafer that is fabbed, and can only make money off of any die that are both functional and that work at the proper speed - if Apple screws up the design and the chips don't work, Apple still pays for them. If Apple only gets 10% functional yield, Apple still pays for 100% of the wafer.

Hence any difficulties due to (1) inadequacies of the EDA tools, library or transistor models; (2) inadequacies of the fab (such as process variability, yield, etc.) come directly out of Apple's bottom line - thus they have to "contend" with them.

 

[Rodimus Prime]

How is it not? Put your apple shaded glasses away for a minute and stop being such a senseless fanboy and actually think about what you're saying for a change.

cmaier is not what I would call an Apple fanboy. Now I would say he is an Apple fan and his opinion tend to error on the side of Apple but at the same time as far as I have seen he has never back pedaled and tried to use back ass fanboy logic to justify things. He tends to speak on how things work but he is not a fanboy.

 

[cmaier]

You are telling me that they could not find another client/clients that would fill Apple void.
I was implying that the fabs would not sit idle for very long if at all. So back to my orginal point.
.

Yeah, I don't think they will find clients soon. There are not a lot of customers eager to use their process, since most customers are far more familiar and comfortable with TSMC, UMC, GlobalFoundries, or even IBM. The only reason Apple used them is that Apple used their design in earlier generations, so Apple's engineers were comfortable with their process.

I mean, who do you think potential customers might be? Everyone already has their own fabs or already has a contract fab lined up - it's not easy switching from a real contract fab to jumping onto someone else's spare fab capacity - it's much easier to go the other way around.

 

[2IS]

Apple designs the circuits. To do this they design macroblocks at the transistor level, and design logic circuits at the standard cell level. They place and route the circuits, and create gdsii files, which they fracture and tape-out to the fab. Apple pays for every wafer that is fabbed, and can only make money off of any die that are both functional and that work at the proper speed - if Apple screws up the design and the chips don't work, Apple still pays for them. If Apple only gets 10% functional yield, Apple still pays for 100% of the wafer.

Hence any difficulties due to (1) inadequacies of the EDA tools, library or transistor models; (2) inadequacies of the fab (such as process variability, yield, etc.) come directly out of Apple's bottom line - thus they have to "contend" with them.

You're right, what I intended to say is that they don't make IC's.

And considering the process you just laid out brings up another interesting question... Some people think that Apple is looking to TSMC as an addition to instead of a replacement for Samsung. How feasible would that be if they're using different manufacturing processes?

 

[AppleScruff1]

So in other words, your comments shouldn't be taken as a reflection on the devices you use or why you use them? Nor should anyone extrapolate that you find Steve Jobs to be your holy savior due to the fact that you prefer Apple products? Interesting. Tell me more?

You just don't get it.

 

[cmaier]

You're right, what I intended to say is that they don't make IC's.

And considering the process you just laid out brings up another interesting question... Some people think that Apple is looking to TSMC as an addition to instead of a replacement for Samsung. How feasible would that be if they're using different manufacturing processes?

Not particularly feasible. I don't believe that for a second.

 

[MattInOz]

Not particularly feasible. I don't believe that for a second.

Question that comes to my mind...
If they need new memory modules to package with the new processor could they be considering moving to one of the newer non-volatile RAM types?

Thinking a tech like SPRAM which has already been used as a replacement to DRAM. Being Non-volatile it can have big gains in power use.

Hitachi, Tohoku Univ Announce Multi-level Cell SPRAM

If these guys are now taking the same tech towards the realm of Flash storage as well. It would seem like an idea target for one of Apple's "pre-order" deals.

 

[cmaier]

Question that comes to my mind...
If they need new memory modules to package with the new processor could they be considering moving to one of the newer non-volatile RAM types?

Thinking a tech like SPRAM which has already been used as a replacement to DRAM. Being Non-volatile it can have big gains in power use.

Hitachi, Tohoku Univ Announce Multi-level Cell SPRAM

If these guys are now taking the same tech towards the realm of Flash storage as well. It would seem like an idea target for one of Apple's "pre-order" deals.

If you're talking about inside the package, that's DRAM, not flash. SPRAM wouldn't be a suitable replacement for that.

SPRAM, itself, is nowhere near commercially viable yet, either.

 

[chrmjenkins]

Not particularly feasible. I don't believe that for a second.

Precisely. Why more than double your unknowns when you're trying to combat unknowns to increase yield?

 

[accessoriesguy]

With the performance of these small chips and tablets going in an upward trend.
---------->

And the size and power usage of Personal computers going in a downward trend.
<----------

I wonder when they will finally meet in the middle
----------> <----------

We will see the latest and greatest not only on the iphone and ipad, but also in desktop computers, one processor small enough, efficient enough yet powerful enough to rule them all and in the darkness bind them!

 

[cmaier]

Precisely. Why more than double your unknowns when you're trying to combat unknowns to increase yield?

It's double the design work - you need to meet two sets of design rules, meaning two cell libraries, two completely different sets of analog cells, etc. Alternatively you do one design that meets both sets of design rules, and with analog blocks tuned to work okay with both processes - this results in a design optimized for NO process, which runs like crap.

 

[2IS]

It's double the design work - you need to meet two sets of design rules, meaning two cell libraries, two completely different sets of analog cells, etc. Alternatively you do one design that meets both sets of design rules, and with analog blocks tuned to work okay with both processes - this results in a design optimized for NO process, which runs like crap.

Sounds like you're talking about Android.

 

[cmaier]

Sounds like you're talking about Android.

Same principle applies

 

[2IS]

Oh, I know. I'm just saying since Android runs on a variety of different hardware configurations, sounds like it makes the compromises you're talking about.