Nokia's Present & Future

I think a much better way to think about this is to realise that as smartphones take over the world, the classical ULC/Basic/Feature/Smart market delimitations will no longer make sense. By 20nm we'll have good 3G "smartphones" with fast web browsing at the prices of EDGE basic phones in developing countries today.

If the SoC is such a small part of the bom for a smartphone, then I wonder why you are looking at process nodes for your estimate of price reduction?
 
It seems you partially misunderstand (on purpose?) or maybe I was unclear.
Of course there will be advantages to efficient integration - I wouldn't underestimate those $5 saved on 100 million devices. You know as well as I do that those $500 are not profit. Another $5 on their margin per device from every iPod/iPhone/iPad is significant. Additionally, there will be advantages in terms of power control = battery life = customer satisfaction. AND they keep control and give themselves options for differentiation in the future. Of course it makes sense for a player of Apples size and ambitions to invest in their own SoC know-how.
Wow... I did misunderstand a few things you said, but you managed to misunderstand even more of what I said ;)

When I say SoC, I mean application processor. Apple has no intention of integrating the baseband. They are following a LOWER integration roadmap than some of their competitors - but they're still saving money in the process because they don't have to pay gross margins to anyone else for the application processor part. And they get to co-develop their hardware and software which improves their time-to-market and software quality at launch. Their cost advantage has NOTHING to do with integration!

You also didn't understand what I said about that $5. Let's say Apple's per-chip cost for the application processor is $8 as I said. They're on 45nm today - by 20nm, they'll be able to fit about 4 times the transistors at the same cost point (simplification because I'm not taking either clocks or wafer costs into consideration, but not a bad one). So they can either increase the transistor count by 4x and keep differentiating on the SoC or save $5 and turn the hardware into a commodity. For a $500 product with up to $250 of gross profit, which do you think they're going to do? Only a complete idiot would save the $5 - especially when you have your own SoC design team! So if Apple keeps being aggressive on SoC specs, do you expect competitors not to bother and simply point out it doesn't make any difference? That would be fun to watch.

So - I gave the mechanisms I see for why hardware specs will diminish in market importance.
Where are your arguments for why they would increase?
I'm not saying they will increase in *importance*. My expectation is that 2011 will be the all-time peak in terms of hardware specs importance (based on the roadmap of various vendors) and it'll reduce slightly over time.

But higher importance would mean a higher share of the bill of materials going to it. I'm saying the share of the bill of materials will remain *constant* to slightly down; this implies, given Moore's Law, a significant boost of hardware specs over the next several years. To claim the contrary would either mean that something else would become relatively much more more important (and I honestly don't see what) or that the average selling prices are going to plummet very rapidly (also quite unlikely).

Btw, there are other reasons I didn't bring up for assuming that additional hardware capabilities won't get increasingly important in the marketplace. Most important that there are no applications that require much more than we already have, and we know that, because there are no such applications in widespread use even on PCs!
Ah, so no games use more than 2 TMUs @ 200MHz (i.e. SGX540 in the Google Nexus S) on the PC? Good to know! :D Obviously a key application for superphones versus multimedia webphones would be gaming, and other fairly expensive things like Augmented Reality (e.g. the camera image and GPS position are combined so you can see the street names in real time on your phone's screen).

If you give them an octa A15 processor with SGX565 parallelized to the gills, what do you expect them to do with it except bitch about the battery life?
I'm serious. These are the real questions. Not "what is possible".
Duh. My point is you can't do that with 50mm2 of silicon anyway and there are still plenty of very real improvements you can do on 50mm2 of 20nm silicon versus 50mm2 of 28nm silicon. And that's for phones - with tablets it's even more obvious.

If the SoC is such a small part of the bom for a smartphone, then I wonder why you are looking at process nodes for your estimate of price reduction?
Keep in mind I'm including the 3G baseband in the bill of materials (15 to 80mm2 on 40nm depending on architecture/performance/bitrate) and I'm also using that as a reference point for the timeframe in which you might expect that kind of thing. We'll start seeing 20nm superphones in 2015 iirc, so for low-cost integrated devices that should be 2016. Plenty of other subsystems will contribute to achieving that cost point by then.

---

Really should get this discussion back to Nokia, I'm going way too OT here... 4M sales for N8 is fairly encouraging, we'll see if they can keep it up (which would have some implications as to whether word-of-mouth is good or bad).
One point about Nokia's hardware roadmap I am curious about: will the custom Broadcom 40nm SoC include 720p video, or will it just be VGA? I honestly have no idea!
 
If the SoC is such a small part of the bom for a smartphone, then I wonder why you are looking at process nodes for your estimate of price reduction?

Performance/Power ratio goes up -> high-end handsets get faster -> high-ends raise the bar of "what everyone wants" -> mid ranges get faster/closer to newest high-ends -> faster mid ranges set the bar for acceptable -> mass production of mid ranges lower the final price.



Kinda like how when HTC Legend came out (MSM7227), it was considered high-end-ish and now all the lowest-end Androids (save for Chinese handsets) have that same MSM7227.
I bet the MSM7227 is a lot cheaper now than it was a year ago..


I bet that in 8 months' time, no Android maker whatsoever will dare launching anything short of an 800MHz Cortex A8 / Cortex A5.
 
Last edited by a moderator:
I bet that in 8 months' time, no Android maker whatsoever will dare launching anything short of an 800MHz Cortex A8 / Cortex A5.
Mostly agreed for the USA/Western Europe, obviously completely disagreed for the rest of the world. There will be plenty of devices with 400-600MHz single-issue CPUs for a long time (sadly enough). Nokia's 600MHz ARM11 platform might look very high-end compared to some of them ;)
 
One point about Nokia's hardware roadmap I am curious about: will the custom Broadcom 40nm SoC include 720p video, or will it just be VGA? I honestly have no idea!

Custom Broadcom SoC? Wut? I've never heard of that, and google only took me back to your post.

Care to elaborate, please?

BTW, since the Symbian^3's lowest-end C6-01 has 720p recording, I don't see why they would remove that functionality in further handsets.
Even the lowest-end Symbian^1 5250 has decent VGA recording @ 30fps.
I don't see why they wouldn't raise that bar to 720p in Symbian^3, since all the high-ends will be getting 1080p recording in a few months' time.


Mostly agreed for the USA/Western Europe, obviously completely disagreed for the rest of the world. There will be plenty of devices with 400-600MHz single-issue CPUs for a long time (sadly enough). Nokia's 600MHz ARM11 platform might look very high-end compared to some of them

Hey, you took out my parenthesis :p
 
None of you guys took my Renesas comment on board I see further divestments
http://blogs.strategyanalytics.com/stp/?p=71
Oh yes, I obviously agree - although that's a 2H12 thing so slightly further out than the rest of the discussion here :) Certainly worth thinking about too though.
Custom Broadcom SoC? Wut? I've never heard of that, and google only took me back to your post.
See my first post in the thread:
Arun said:
- Nokia's mainstream Symbian platform will be based on a 40nm Broadcom integrated baseband (off-chip RF & PMU) - I don't know the overall specs, I suspect it will have decent 3D but only VGA video though. This directly replaces the 45nm ARM11 integrated platform from TI that is used in nearly all of their recent devices (including the N8 with the extra BCM2727).
For something more official: http://www.broadcom.com/press/release.php?id=s365393 (that has absolutely nothing to do with the BCM2727 obviously!)

I agreed it would be strange not to support 720p given that the C6-01 does, but maybe Nokia feels it's enough to have one set of solutions for 1080p and another for VGA. I would tend to believe 720p support is slightly more likely than not though.
 
Oh yes, I obviously agree - although that's a 2H12 thing so slightly further out than the rest of the discussion here :) Certainly worth thinking about too though.
See my first post in the thread:
For something more official: http://www.broadcom.com/press/release.php?id=s365393 (that has absolutely nothing to do with the BCM2727 obviously!)

Well if that press release isn't about BCM2727 (since it mentions Broadcom as being Nokia's "3G HSPA supplier"), I'd say it could be a SoC that simply joins a BCM2727 (32MB included) with the ARM11 and the baseband processor. The additional 256MB RAM would be in a separate chip?


BTW, the current ARM11+256MB RAM chip shipping in Symbian^3 smartphones is made by Samsung.
 
Well if that press release isn't about BCM2727 (since it mentions Broadcom as being Nokia's "3G HSPA supplier"), I'd say it could be a SoC that simply joins a BCM2727 with the ARM11 and the baseband processor.
No, you're wrong. That agreement is for an unreleased platform where Broadcom is responsible for the integrated Application Processor/Multimedia/Baseband chip (using Nokia Modem IP from before that division was sold to Renesas), RF chip, and power management chip. The BCM2727 is a completely separate (and earlier) deal. Broadcom might reuse the BCM2727's vector processor for multimedia - or they might not. That's an implementation detail that is not very important to the deal.

You also shouldn't assume it makes any sense to integrate 32MB of memory. It doesn't. DRAM processes are much more cost-efficient per megabyte than any eDRAM process will ever be (and that's before we even consider the ridiculously expensive possibility of using SRAM instead).

BTW, the current ARM11+256MB RAM chip shipping in Symbian^3 smartphones is made by Samsung.
No, the ARM11 SoC with integrated baseband is made by TI. The DRAM is made by Samsung. It is typical for Package-on-Package solutions to only have the DRAM markings visible because it's the top package and the processor is hidden at the bottom ;)
 
No, you're wrong. That agreement is for an unreleased platform where Broadcom is responsible for the integrated Application Processor/Multimedia/Baseband chip (using Nokia Modem IP from before that division was sold to Renesas), RF chip, and power management chip. The BCM2727 is a completely separate (and earlier) deal. Broadcom might reuse the BCM2727's vector processor for multimedia - or they might not. That's an implementation detail that is not very important to the deal.

You also shouldn't assume it makes any sense to integrate 32MB of memory. It doesn't. DRAM processes are much more cost-efficient per megabyte than any eDRAM process will ever be (and that's before we even consider the ridiculously expensive possibility of using SRAM instead).


Nokia is releasing tutorials on how to optimize apps and games for the BCM2727, how the developer shouldn't exceed the 20MB and how the driver runs directly in the GPU.
They're also commited on releasing firmware upgrades for Symbian^3 as an unified platform.

I'm just assuming that, if they want to have seamless updates and an unified platform for app development and respective performance optimizations, it should be in their best interest to keep the same hardware across all devices (as they've been doing until now) -> at least from a developer's point of view: ARM11 680MHz+BCM2727+32MB graphics memory+256MB application memory.

Kinda like the game consoles that just glue together their various chips across node iterations.



After all, aren't all Nokia's Symbian^1 devices using the same ARM11 434MHz CPU + 128MB RAM?
 
So you're assuming Nokia is going to stagnate on that platform for the next two years in both the low-end and the high-end? This is not a minor platform upgrade, it's a completely new platform that we won't see in shipping devices for quite some time. Are you going to complain when they release Symbian phones based on the ST-Ericsson U8500 in mid-2011 too? ;)
 
First off, there is some basic agreement.
Technology in smartphones is not going to gain importance in the marketplace going forward.
Nevertheless, it is going to evolve with lithographic processes and new devices.
To connect back to Nokia, they have been a mixed bag the last years - a lot of different, often leading, capabilities and functionality, but lacking in overall vision and execution. Implying that software and platform building is more important than Nokia expected or could handle.

Wow... I did misunderstand a few things you said, but you managed to misunderstand even more of what I said ;)

When I say SoC, I mean application processor. Apple has no intention of integrating the baseband. They are following a LOWER integration roadmap than some of their competitors - but they're still saving money in the process because they don't have to pay gross margins to anyone else for the application processor part. And they get to co-develop their hardware and software which improves their time-to-market and software quality at launch. Their cost advantage has NOTHING to do with integration!
Uhm. Who said anything about maximum integration? You outline perfectly valid savings/added value avenues above. Doing your own design avoids paying someone else for their work, and allows you to tailor the SoC to your specific needs - no need to include any functionality that your device isn't going to use. On the other hand you can enhance other aspects, and gain a leg up on the competition, particularly since hardware and software can be developed in parallel. At least, that's what the armchair expertise are saying.

For a $500 product with up to $250 of gross profit, which do you think they're going to do? Only a complete idiot would save the $5 - especially when you have your own SoC design team! So if Apple keeps being aggressive on SoC specs, do you expect competitors not to bother and simply point out it doesn't make any difference? That would be fun to watch.
I don't know what Apple will do. I doubt "saving money" is their main thrust. Then again, I don't think that "pushing the bleeding edge of technology" is either. Gaining independence and capability to innovate is probably the main idea. They do set the standard though - rumors say that Apple is procuring for over 50 million iPads next year, and God knows how many iPhones and iPods. We're looking at roughly (projected) 200 million iOS devices in total next year. Their next SoC is going to be very, very influential. If they do as they have done, they will use the same device across their product lines. (It may or may not make sense to do add processing capabilities specifically to the iPad.)

But higher importance would mean a higher share of the bill of materials going to it. I'm saying the share of the bill of materials will remain *constant* to slightly down; this implies, given Moore's Law, a significant boost of hardware specs over the next several years. To claim the contrary would either mean that something else would become relatively much more more important (and I honestly don't see what) or that the average selling prices are going to plummet very rapidly (also quite unlikely).
True.
I wouldn't be so sure about the ASPs though. If Android is hugely successful, I think they will drop quite a bit. And when they do, total volumes will explode far beyond what we see today. Component pricing will be under downward pressure. Hard to predict beyond that.

Ah, so no games use more than 2 TMUs @ 200MHz (i.e. SGX540 in the Google Nexus S) on the PC? Good to know! :D
Cheap argument - high powered PC gaming was implied not to be included in my "widespread use".
High power PC gaming is tiny compared to the PC market and tinier still compared to total cell phone market. It gains attention because it is unarguably the biggest group of PC users that can still be convinced that they need more processing power, and can thus be targeted for higher profit margin products. There's a reason all $100+ graphics cards feature a rendered girl with big boobs holding an even bigger "tool". It's not the same market Apple or Nokia targets.

Really should get this discussion back to Nokia, I'm going way too OT here... 4M sales for N8 is fairly encouraging, we'll see if they can keep it up (which would have some implications as to whether word-of-mouth is good or bad).
Yup.
But Nokia doesn't operate in a vacuum, and their degree of success will depend on where their competitors and the market goes. We have an excuse. :)
 
So you're assuming Nokia is going to stagnate on that platform for the next two years in both the low-end and the high-end? This is not a minor platform upgrade, it's a completely new platform that we won't see in shipping devices for quite some time. Are you going to complain when they release Symbian phones based on the ST-Ericsson U8500 in mid-2011 too? ;)



Well, Nokia did stagnate its platform for the last two years (not that I'm expecting them to repeat the same mistake this time).
I have no idea when this new broadcom SoC will come to handsets, but I'm expecting Nokia to keep this hardware platform at least until the end of 2011, for Symbian.

Furthermore, they already have their own market fragmentation planned with MeeGo for the higher end. Wouldn't a dual-core A9 + Mali400 make more sense for MeeGo?


I imagined that Symbian would stay for 640*360 , 2.8-3.5" devices, using a weaker but very efficient/cheap-ish hardware, and MeeGo would be to go all-out on 800*480 (eventually 720p?) 4-4.3" screens with A9s (eventually Atoms) and more potent GPUs.
Like having a "Wii-like" and a "Xbox360-like" platforms at the same time.
 
I imagined that Symbian would stay for 640*360 , 2.8-3.5" devices, using a weaker but very efficient/cheap-ish hardware, and MeeGo would be to go all-out on 800*480 (eventually 720p?) 4-4.3" screens with A9s (eventually Atoms) and more potent GPUs.

Why in the world would you expect them to migrate a platform established on Cortex-A9 to Atom? I don't believe Meego is intended to hide behind a VM like Android does. So they wouldn't want to change CPU platforms on it.

Word is that Meego phones in 2011 will be x86, so no plan to migrate from Cortex-A9 to Atom. But to me the whole thing seems like more of a corporate/political move more than a practical one. I don't see why Atoms would ever be a good choice for phones outside of a few very fringe benefits. It's also kind of hard to imagine Meego is going to be especially successful coming out pretty late in a pretty crowded market.
 
Why in the world would you expect them to migrate a platform established on Cortex-A9 to Atom? I don't believe Meego is intended to hide behind a VM like Android does. So they wouldn't want to change CPU platforms on it.



Wait! What?!

http://en.wikipedia.org/wiki/MeeGo

MeeGo is a joint-venture between Intel and Nokia.
The first release of the OS was made available for Atom netbooks and N900, and the development platform is the Moorestown-equipped Aava prototype.

There's no "migration" to be made here. MeeGo is definitely an OS to be used with x86 low-power CPUs from Intel in the long run.

Most probably, the only reason for MeeGo to support the ARM architecture right now comes from Intel's failure to mass-produce Moorestown CPUs that comply with a handset's power requirements in 2010 and the fact that MeeGo is an evolution of Maemo (OS for Nokia MIDs and N900).


The Atom architecture has always been intended to be Intel's entry-gate into hanhelds, in the long run. Do you think it's a coincidence that Intel launched a completely new low-power x86 architecture, 3 years after selling the XScale division to Marvell?


MeeGo phones with x86? It's not even a matter of "if", it's a matter of "when".
And after all tag-teamed kicks in the nuts Intel took during CES'11, I'd say their efforts to bring x86 to mobile phones will increase greatly.
 
I don't believe Meego is intended to hide behind a VM like Android does. So they wouldn't want to change CPU platforms on it.
MeeGo is all about universal binaries AFAIK. I think there's a chance Medfield has a win at Nokia for a MeeGo phone, and if not at least a win for a MeeGo tablet.

But ToTTenTranz is wrong; Nokia wants to use OMAP3/OMAP4/U8500 with MeeGo. That doesn't sound to me like a x86-only platform with one or two ARM phones as intermediary steps - although I'm sure that if Intel manages to execute with their handheld business, they'll get quite a few design wins there. Just certainly not a vast majority.
 
But ToTTenTranz is wrong; Nokia wants to use OMAP3/OMAP4/U8500 with MeeGo. That doesn't sound to me like a x86-only platform with one or two ARM phones as intermediary steps

This is crazy.
And why exactly would Intel start a company that develops OSs for their most threatening competitor architecture?
 
This is crazy.
And why exactly would Intel start a company that develops OSs for their most threatening competitor architecture?
You're thinking about it wrong. MeeGo is the merger of Moblin (Intel-only) and Maemo (Nokia-only). The only question from Intel's perspective is what they gain from merging with Maemo, and the answer is applications (anything that'd work on ARM would work on x86, not just vice-versa), an OS that can target phones rather than just netbooks/tablets, and a big foot in the door at Nokia. Given the relatively weak momentum behind Moblin, I think this makes a lot of sense for them despite your point.
 
Wait! What?!

http://en.wikipedia.org/wiki/MeeGo

MeeGo is a joint-venture between Intel and Nokia.
The first release of the OS was made available for Atom netbooks and N900, and the development platform is the Moorestown-equipped Aava prototype.

There's no "migration" to be made here. MeeGo is definitely an OS to be used with x86 low-power CPUs from Intel in the long run.

That is exactly opposite to the point I'm making. I'm saying why would NOKIA release Cortex-A9 MeeGo devices instead of releasing Atom ones from the start, if that's their ultimate goal? That should have been clear from my post, I think you read the first sentence then hit the reply button...

I didn't realize MeeGo is about universal binaries, that helps, but I wonder what that's going to make the compilation options like. It certainly isn't going to stop people from using assembly code, but I suppose it'll be discouraged. Hooray for making apps bigger on a mobile platform.

Most probably, the only reason for MeeGo to support the ARM architecture right now comes from Intel's failure to mass-produce Moorestown CPUs that comply with a handset's power requirements in 2010 and the fact that MeeGo is an evolution of Maemo (OS for Nokia MIDs and N900).

Those are assumptions you're making about Moorestown. I don't think Intel will be admitting to that any time soon.

Of course x86 phones are happening. But I still don't see the practical benefit.
 
Back
Top