Qualcomm Roadmap (2011-2012)

[wco81]

The only thing ridiculous about LTE is the number of power amplifiers required if you want to support multiple countries or even sometimes just multiple operators in the same country. If anything, that makes modules an even more attractive solution for tablet manufacturers.

What is the story on multi-network, multi-country requirements for LTE?

Different bands mainly?

So far, I've only heard of Verizon and AT&T ramping up. It seems a lot of operators are touting HSPA+ coverage now, with claims of 42 Mbps speeds in some areas.

Are LTE devices going to be interoperable with SIM unlocking? I've heard suggestions that Verizon will find ways to lock devices to their network and it sounded like something beyond SIM locking.

 

[metafor]

Actually integrated basebands are currently VERY common because:
1) Qualcomm is the market leader.

At the high end?

5) ST-Ericsson with the U8500. Ironically, some customers don't want the Nokia-designed 14.4Mbps baseband and would rather use the EMP-designed 21.6Mbps M5730 baseband next to it. In theory, that solution is called the U9500. While thz A9540 and A9600 are discrete, the U4500 is integrated and should be very high volume. They've also got integrated derivatives on their 28nm roadmap.
6) Mediatek with the MT6573 (7.2Mbps HSUPA baseband with a dedicated 650MHz ARM11).

What traction does the U-series have compared to the A-series? What do the respective design wins (if any) look like?

Actually Intel and NVIDIA aren't being too aggressive with baseband integration. The main reasoning behind their respective acquisitions was bundling and being able to provide direct support for the whole platform. I'd expect Intel to integrate a LTE baseband into their second-generation 22nm SoC and NVIDIA to include one in a lower-end derivative on 20nm... but not before.

It takes a while to come up with a design/validation process that can still pump out chip iterations for chips with an integrated baseband, so it makes sense that it wouldn't be instantaneous. Still, I question the bundling explanation for such an acquisition; there are far easier ways to pre-validate platforms other than buying the company.

You wouldn't claim Atheros was acquired just for easier bundling with MSM chips, for instance; even though it'll take a few more generations before full WiFi is integrated into the MSM and MDM series of chips and even longer before the APQ series gets that.

I completely fail to see how that is the case. If you only need 50Mbps LTE (e.g. AT&T) and don't need the 2600MHz band, Icera's ICE8060 does it in a 7x7 package with stacked DRAM and a die size of only 15.6mm². Their 28nm solution (which was recently still expected to be back from the fab in Q4) will support 150Mbps LTE with an even slightly smaller die size. What's prohibitive about that?

Stacked DRAM package compared to none at all? And we're talking Icera's solution here, which, depending on your need, may not fit. The MDM9600 is absolutely huge. The thing's 13x13. 28nm will alleviate this significantly but even then, we're talking a ~8x8 package at best.

With everyone racing to not only compete with Apple on features but also on price and profit margin, I expect at least high-end smartphones to prefer integrated basebands in the future for their LTE solutions.

 

[Arun]

At the high end?

AFAICT, the two highest volume high-end platforms in 2010 were the Qualcomm QSD8x50 on 65nm and the TI OMAP3 on 65nm & 45nm. I'm not sure which came first, but worst case Qualcomm couldn't have been far behind.

What traction does the U-series have compared to the A-series? What do the respective design wins (if any) look like?

I believe Nokia was mostly U8500 and non-Nokia was mostly A9500 (remember it's a Nokia-designed modem) but I could be horribly wrong, and who the hell knows what Nokia vs Non-Nokia volume will look like at this point anyway.

It takes a while to come up with a design/validation process that can still pump out chip iterations for chips with an integrated baseband, so it makes sense that it wouldn't be instantaneous. Still, I question the bundling explanation for such an acquisition; there are far easier ways to pre-validate platforms other than buying the company.

Oh absolutely. NVIDIA had already pre-validated platforms from Qualcomm, Infineon, ST-Ericsson, and Icera - and they'll continue supporting all of them at least in the short-term. In the mid-term, they still need to support CDMA via Qualcomm until Verizon is ready for single-mode LTE handsets (hopefully in 2014).

The basic reason NVIDIA acquired Icera is that they were damn cheap, the technology was damn good, and if they didn't buy them someone else would have because they were running out of cash. The bundling rationale is enough here because on their own Icera had difficulty convincing large handset OEMs to give them a chance but as part of NVIDIA the revenue opportunities are significantly larger. Of course, being able to integrate in the long-term is a very nice bonus, and NVIDIA probably realised this was their last chance to acquire baseband technology.

Stacked DRAM package compared to none at all? And we're talking Icera's solution here, which, depending on your need, may not fit.

Sure, it's still slightly more expensive for multiple reasons including the extra DRAM chip, but it's small stuff and the extra footprint at least is completely negligible. However it's true that ICE8060 only works if you don't need CDMA, TD-SCDMA, 100Mbps LTE, or the 2600MHz band. The 28nm baseband and next-gen RF will fix at least the last two, but they'll still need Qualcomm for CDMA (and it remains to be seen whether they bother with TD-SCDMA or only support TD-LTE for non-Chinese markets).

The MDM9600 is absolutely huge. The thing's 13x13. 28nm will alleviate this significantly but even then, we're talking a ~8x8 package at best.

Meh, it's not the end of the world: QSD8x50 was in a 15x15 package, and Infineon's slim baseband is still 10x10 or 12x12 if you use PoP DRAM! As for die size, I know the exact numbers for all MDM chips and I'm confident a shrink to 28nm (MDM9615) will get it very close to the die size of the MDM6600 on 45nm. Still much bigger than Icera but not too bad. Sadly the presentation in the original post makes clear it'll still be in a 10x10 package.

With everyone racing to not only compete with Apple on features but also on price and profit margin, I expect at least high-end smartphones to prefer integrated basebands in the future for their LTE solutions.

I don't buy it. The smartphone dynamics will be exactly the same as they are today for 3G. So you can be confident you'll see a mix of the two just like on 3G today. NVIDIA, ST-Ericsson, TI, Samsung are all clearly in the standalone camp for LTE in the next 3+ years. And there's little evidence Broadcom or Marvell will integrate LTE at least in the next 2 years. So that's the vast majority of the high-end market right there. Everyone is focusing on integrating 3G in the low-end, not 4G in the high-end! Qualcomm is the exception, not the rule.

 

[metafor]

AFAICT, the two highest volume high-end platforms in 2010 were the Qualcomm QSD8x50 on 65nm and the TI OMAP3 on 65nm & 45nm. I'm not sure which came first, but worst case Qualcomm couldn't have been far behind.

Well, 2010 was an anomaly as it was the introduction of Snapdragon devices. I don't believe Qualcomm had the same dominance at the high-end in 2009 (with the Droid) and may well be behind in 2011 (with Tegra 2). 2012 may be different with the introduction of Krait but that depends entirely on execution and marketing.

Qualcomm may be the leader by a wide margin in the overall handset space but I don't think they are in a good enough position to be considered dominant in the high-end; not yet at least.

I believe Nokia was mostly U8500 and non-Nokia was mostly A9500 (remember it's a Nokia-designed modem) but I could be horribly wrong, and who the hell knows what Nokia vs Non-Nokia volume will look like at this point anyway.

Well, my point is that obviously there isn't a clear preference for integrated modems at the high end. Else the A-series would be relegated to being a tablet-part only.

Oh absolutely. NVIDIA had already pre-validated platforms from Qualcomm, Infineon, ST-Ericsson, and Icera - and they'll continue supporting all of them at least in the short-term. In the mid-term, they still need to support CDMA via Qualcomm until Verizon is ready for single-mode LTE handsets (hopefully in 2014).

The basic reason NVIDIA acquired Icera is that they were damn cheap, the technology was damn good, and if they didn't buy them someone else would have because they were running out of cash. The bundling rationale is enough here because on their own Icera had difficulty convincing large handset OEMs to give them a chance but as part of NVIDIA the revenue opportunities are significantly larger. Of course, being able to integrate in the long-term is a very nice bonus, and NVIDIA probably realised this was their last chance to acquire baseband technology.

Well the point was, why acquire baseband technology at all? It's not nearly as profitable as app processors and requires a completely different skillset and organization than what nVidia does. Moreso, Icera wasn't exactly making money hand over fist despite their die size and power advantages. nVidia obviously has no trouble getting working reference designs using any other baseband company, nor would they have problems partnering with whomever bought Icera to provide modems.

I maintain that if the idea of single-chip integration in the long-term wasn't part of the decision, it would have never been made. Same for Intel and Infineon as well as Atheros and Qualcomm.

Sure, it's still slightly more expensive for multiple reasons including the extra DRAM chip, but it's small stuff and the extra footprint at least is completely negligible.

I don't know about that. 3G chips are cheap enough and integrated options thus far have somewhat lagged behind discrete components (at least on the HSPA side). LTE is a different story. The first to market in this case is the more expensive vendor (qcom) and with a bigger chip. Eventually, the economics may wind down to what 3G is like today.

However it's true that ICE8060 only works if you don't need CDMA, TD-SCDMA, 100Mbps LTE, or the 2600MHz band. The 28nm baseband and next-gen RF will fix at least the last two, but they'll still need Qualcomm for CDMA (and it remains to be seen whether they bother with TD-SCDMA or only support TD-LTE for non-Chinese markets).

I suppose it's true that certain models (particularly if you just target AT&T) will only need something along the lines of an Icera 50mbps LTE chip. However, this really all depends on what Apple does. If they put out an iPhone with at least universal LTE support for the U.S. bands (mind you, this won't be until iPhone6 at the very least) then others will follow suit.

Meh, it's not the end of the world: QSD8x50 was in a 15x15 package, and Infineon's slim baseband is still 10x10 or 12x12 if you use PoP DRAM! As for die size, I know the exact numbers for all MDM chips and I'm confident a shrink to 28nm (MDM9615) will get it very close to the die size of the MDM6600 on 45nm. Still much bigger than Icera but not too bad. Sadly the presentation in the original post makes clear it'll still be in a 10x10 package.

More to the point, integrated LTE will arrive to market on 28nm (and be in devices) well before the MDM chip on 28nm. We're really talking about a choice between 45nm LTE baseband vs 28nm Snapdragon with integrated LTE baseband. Or Icera's 28nm baseband I suppose, though as you pointed out, there are certain can-we-trust-this-vendor issues there.

 

[Arun]

Well, 2010 was an anomaly as it was the introduction of Snapdragon devices.

Agreed. Qualcomm was very strong with the MSM72xx in 2009 and is doing well with the MSM7x30/M8x55 in 2011 though... and it could actually be argued that the MSM7201A and MSM8x55 were high-end solutions in early 2009 and early 2011 respectively because OMAP3 and Tegra 2 weren't available yet in January of these years. And in 2008, the MSM72xx was clearly a high-end solution, and it dominated the Windows Mobile and early Android markets. So Qualcomm has a long history of being very competitive in the high-end at least.

Well, my point is that obviously there isn't a clear preference for integrated modems at the high end. Else the A-series would be relegated to being a tablet-part only.

Agreed, although I think it's even more correct to say that OEMs care more about having a good 21Mbps baseband than they care about saving a few dollars.

Well the point was, why acquire baseband technology at all? It's not nearly as profitable as app processors and requires a completely different skillset and organization than what nVidia does. Moreso, Icera wasn't exactly making money hand over fist despite their die size and power advantages. nVidia obviously has no trouble getting working reference designs using any other baseband company, nor would they have problems partnering with whomever bought Icera to provide modems.

I maintain that if the idea of single-chip integration in the long-term wasn't part of the decision, it would have never been made. Same for Intel and Infineon as well as Atheros and Qualcomm.

Okay, put that way, I can't really disagree. I'd still frame it slightly differently though:
NVIDIA does not believe anything below the high-end market (depending on your definition that may include the upper mid-range) is worth pursuing in the next 2 years, and they do not believe an integrated baseband will be worth it there for even longer than that. But:
1) They do want to target the low-end market eventually and they know they need an integrated baseband for that. They don't want to be 'locked out' of most of the volume in a couple of years.
2) They fear being 'integrated out': when they started sampling Tegra 1, Icera/Infineon/Ericsson/Qualcomm all had (relatively) slim basebands on their roadmap. Now the first three have been acquired by companies targeting the application processor market and so (just like Qualcomm) it's risky to depend on them in the long-term (and it obviously doesn't help to have to redirect your customers to the sales teams of your direct competitors).

However, unlike the Infineon acquisition (which is definitely about integration in the long-term, sorry if I somehow implied otherwise, the problem for Intel is finding the right time to integrate and as I said that's probably the late 22nm generation), NVIDIA likely wouldn't have done it anyway if the price hadn't been so low and they hadn't been able to justify it financially just on a bundling basis in the short/mid-term.

I don't know about that. 3G chips are cheap enough and integrated options thus far have somewhat lagged behind discrete components (at least on the HSPA side). LTE is a different story. The first to market in this case is the more expensive vendor (qcom) and with a bigger chip. Eventually, the economics may wind down to what 3G is like today.

Strictly speaking Samsung beat Qualcomm to market by a lot... And that wasn't just a one-off as their solution is also used in the Samsung Droid Charge at Verizon next to a Via Telecom CDMA baseband. And GCT (with LG LTE IP) actually had a LTE data device based on the GDM7240 available at Verizon at the same time as Qualcomm (which they insist truly has integrated RF amazingly enough although I suspect it's still on 65nm).

Also I'm not sure why you say this is somehow different. Qualcomm has always been early to new standards (including WCDMA, HSDPA, HSUPA, and HSPA+) and they've always had bigger chips than some competitors (but also higher performance i.e. channel efficiency) which they partially compensated by a process advantage.

I suppose it's true that certain models (particularly if you just target AT&T) will only need something along the lines of an Icera 50mbps LTE chip. However, this really all depends on what Apple does. If they put out an iPhone with at least universal LTE support for the U.S. bands (mind you, this won't be until iPhone6 at the very least) then others will follow suit.

Actually, unlike their 21Mbps HSPA+ smartphone platform, Icera's 50Mbps LTE solution is mostly targeted at data devices and tablets. It's still based on a 130nm RF chip that requires an analogue baseband which increases their "active but low bandwidth" power consumption by quite a bit compared to their new 65nm 5MHz RF chip. Still before the acquisition they were trying to get smartphone design wins with it (since competitors' power consumption isn't great either) but not as aggressively. They had a 20MHz 65nm RF chip on the roadmap expected slightly before the 28nm baseband but I don't know if they'll bother releasing it before the 28nm baseband.

Anyway smartphone manufacturers have been making tons of variants of the same phone for a long time, and they'll need to do so anyway for Europe and Asia. What's one more variant for Verizon if it saves you three or four PAs (plus a tiny bit of money on the baseband since QCOM likely asks less for GSM-only despite being the same chip)?

More to the point, integrated LTE will arrive to market on 28nm (and be in devices) well before the MDM chip on 28nm. We're really talking about a choice between 45nm LTE baseband vs 28nm Snapdragon with integrated LTE baseband. Or Icera's 28nm baseband I suppose, though as you pointed out, there are certain can-we-trust-this-vendor issues there.

According to the presentation I linked, MSM8960 is 6 months ahead of MDM9615, and it is indeed one of the first LTE solution for smartphones that should genuinely result in good enough battery life.

However it's not the only viable solution for smartphones in that timeframe (besides Icera for 50Mbps LTE). I'm fairly confident the 40nm ST-Ericsson M7400 would be quite a bit better than the MDM9200.

 

[metafor]

Okay, put that way, I can't really disagree. I'd still frame it slightly differently though:
NVIDIA does not believe anything below the high-end market (depending on your definition that may include the upper mid-range) is worth pursuing in the next 2 years, and they do not believe an integrated baseband will be worth it there for even longer than that. But:
1) They do want to target the low-end market eventually and they know they need an integrated baseband for that. They don't want to be 'locked out' of most of the volume in a couple of years.
2) They fear being 'integrated out': when they started sampling Tegra 1, Icera/Infineon/Ericsson/Qualcomm all had (relatively) slim basebands on their roadmap. Now the first three have been acquired by companies targeting the application processor market and so (just like Qualcomm) it's risky to depend on them in the long-term (and it obviously doesn't help to have to redirect your customers to the sales teams of your direct competitors).

However, unlike the Infineon acquisition (which is definitely about integration in the long-term, sorry if I somehow implied otherwise, the problem for Intel is finding the right time to integrate and as I said that's probably the late 22nm generation), NVIDIA likely wouldn't have done it anyway if the price hadn't been so low and they hadn't been able to justify it financially just on a bundling basis in the short/mid-term.

You do have a point about the price being right. However, even they realize (though belatedly) that mid-end and low-end is where the money is. This is both true of their desktop/laptop parts as well as their mobile SoC's. I suspect this is the primary motivation for purchasing a baseband company. I agree that had the price not been low, they likely would've stuck it out another few years.

Strictly speaking Samsung beat Qualcomm to market by a lot... And that wasn't just a one-off as their solution is also used in the Samsung Droid Charge at Verizon next to a Via Telecom CDMA baseband. And GCT (with LG LTE IP) actually had a LTE data device based on the GDM7240 available at Verizon at the same time as Qualcomm (which they insist truly has integrated RF amazingly enough although I suspect it's still on 65nm).

As far as products to market, I believe the Pantec data cards came out before Samsung-based ones or perhaps at the same time and the Thunderbolt beat the Charge to market.

Also I'm not sure why you say this is somehow different. Qualcomm has always been early to new standards (including WCDMA, HSDPA, HSUPA, and HSPA+) and they've always had bigger chips than some competitors (but also higher performance i.e. channel efficiency) which they partially compensated by a process advantage.

Die area has never jumped this much. The move to LTE has resulted in an absolutely huge chip compared to previous improvements; at least for the MDM series.

Anyway smartphone manufacturers have been making tons of variants of the same phone for a long time, and they'll need to do so anyway for Europe and Asia. What's one more variant for Verizon if it saves you three or four PAs (plus a tiny bit of money on the baseband since QCOM likely asks less for GSM-only despite being the same chip)?

A universal LTE phone (at least one that works with all U.S. carrier bands) has its attractions and if it is indeed the direction Apple will move with their iPhone, I see it as a selling point that other manufacturers will likely try to follow. Whether or not it makes sense financially would depend on how much users flock to that model. Certainly it's been tried before but Apple has a tendency to set trends and make certain market models work where they did not work before.

According to the presentation I linked, MSM8960 is 6 months ahead of MDM9615, and it is indeed one of the first LTE solution for smartphones that should genuinely result in good enough battery life.

However it's not the only viable solution for smartphones in that timeframe (besides Icera for 50Mbps LTE). I'm fairly confident the 40nm ST-Ericsson M7400 would be quite a bit better than the MDM9200.

I'm not familiar with the features of the M7400 so I can't comment here. I'm mainly basing the emphasis on integration on the fact that a discrete 28nm LTE modem really won't be available as soon as an integrated solution.

 

[Arun]

You do have a point about the price being right. However, even they realize (though belatedly) that mid-end and low-end is where the money is.

Well then that is very recent change of heart because I can guarantee you that Phil Carmack and Mike Rayfield sincerly did not believe that back at Mobile World Congress 2011. Their position (especially but not only Rayfield's) was that the operator subsidies are large enough to make everything below the high-end irrelevant in North America and Western Europe. I think that's right (depending on your definition of high-end) but they seemed to be genuinely underestimating the potential of emerging countries and prepaid smartphones.

As far as products to market, I believe the Pantec data cards came out before Samsung-based ones or perhaps at the same time and the Thunderbolt beat the Charge to market.

I'm not talking about Verizon. I'm talking about TeliaSonera which had a Samsung-based data card on the world's first commercial LTE network in December 2009 (and yes, that's 2009, not 2010!) - link: http://www.samsung.com/us/aboutsamsung/news/newsIrRead.do?news_ctgry=irnewsrelease&news_seq=15946

Die area has never jumped this much. The move to LTE has resulted in an absolutely huge chip compared to previous improvements; at least for the MDM series.

That's wrong. The 65nm MSM6290 was slightly smaller than the 45nm MDM6600 while the 65nm MDM8200 was slightly bigger than the 45nm MDM9600. Of course, Qualcomm never tried to get the MDM8200 (or even the MDM8200A) into smartphones...

I'm not familiar with the features of the M7400 so I can't comment here. I'm mainly basing the emphasis on integration on the fact that a discrete 28nm LTE modem really won't be available as soon as an integrated solution.

Agreed, but that's only a temporary 6 months advantage for Qualcomm that will improve their market share in the overall LTE market. It won't change anything for anyone else except in terms of market share - and if ST-Ericsson delivers with the M7400 (sampling in 2Q11) then that should mitigate the impact somewhat.

 

[metafor]

Well then that is very recent change of heart because I can guarantee you that Phil Carmack and Mike Rayfield sincerly did not believe that back at Mobile World Congress 2011. Their position (especially but not only Rayfield's) was that the operator subsidies are large enough to make everything below the high-end irrelevant in North America and Western Europe. I think that's right (depending on your definition of high-end) but they seemed to be genuinely underestimating the potential of emerging countries and prepaid smartphones.

Judging by the (still) massive sales numbers of the MSM7227 and recently the 7x30, I'd say there's a big market even in NA and Europe for free-on-contract phones.

I'm not talking about Verizon. I'm talking about TeliaSonera which had a Samsung-based data card on the world's first commercial LTE network in December 2009 (and yes, that's 2009, not 2010!) - link: http://www.samsung.com/us/aboutsamsung/news/newsIrRead.do?news_ctgry=irnewsrelease&news_seq=15946

Ahh, that I wasn't aware of.

That's wrong. The 65nm MSM6290 was slightly smaller than the 45nm MDM6600 while the 65nm MDM8200 was slightly bigger than the 45nm MDM9600. Of course, Qualcomm never tried to get the MDM8200 (or even the MDM8200A) into smartphones...

MDM8200 at 65nm was about 60mm^2 for the die. MDM9600 was around 65mm^2. That's with a die shrink. MDM6600 was dual EVDO Rev A/B along with 14.4mbps HSPA+. So essentially, the area increase for LTE was equivalent to adding an EVDO modem...

Agreed, but that's only a temporary 6 months advantage for Qualcomm that will improve their market share in the overall LTE market. It won't change anything for anyone else except in terms of market share - and if ST-Ericsson delivers with the M7400 (sampling in 2Q11) then that should mitigate the impact somewhat.

What exactly is the die area and power usage of M7400? Is it dramatically better than MDM9200?

 

[Arun]

MDM8200 at 65nm was about 60mm^2 for the die. MDM9600 was around 65mm^2. That's with a die shrink. MDM6600 was dual EVDO Rev A/B along with 14.4mbps HSPA+. So essentially, the area increase for LTE was equivalent to adding an EVDO modem...

Errr, I have no idea where those numbers come from - they're definitely not right. I wasn't going to release this publicly, but what the heck, here's my personal list of baseband die sizes:

Qualcomm MSM6290: >40mm² - 65nm
Qualcomm MDM8200: <110mm² - 65nm
Qualcomm MDM6600: ~40mm² - 45nm
Qualcomm MDM8200A: >60mm² - 45nm
Qualcomm MDM9600: >90mm² - 45nm
Infineon XMM6080: ~70mm² (*) - 90nm
ST-Ericsson M570: >50mm² - 65nm
Icera ICE8040: ~19mm² - 65nm
Icera ICE8060: <16mm² - 40nm
(*): XMM6080 is a two-chip solution: 40mm² for Shared MAC + EDGE PHY + Multimedia & 30mm² for HSDPA PHY (both 90nm)

I rounded the numbers for simplicity's sake but you get the idea. The MDM9600 is definitely the largest Qualcomm baseband that has ever gone into a mobile phone (excluding basebands with integrated application processors; I think the original 90nm MSM7200 was around 130mm²) but it's still smaller than the data-only MDM8200. And the 28nm MDM9615 should be slightly more than 45mm² (I'm not assuming a perfect shrink since the MDM9600 is on 45nm and not 40nm) which is well within historical norms.

What exactly is the die area and power usage of M7400? Is it dramatically better than MDM9200?

I know it's significantly below the MDM9600's ~90mm² but I don't know such specific information, sorry. Keep in mind it's based on NXP's EVP16 programmable architecture with accelerators (which I heard might themselves be attached to much smaller NXP DSPs with some Tensilica-like extendibility). It's not as elegant or efficient an architecture as Icera's but it should definitely be smaller than a traditional architecture. And there are already a few fairly traditional architectures with smaller die sizes than MDM9600's...

 

[metafor]

Errr, I have no idea where those numbers come from - they're definitely not right. I wasn't going to release this publicly, but what the heck, here's my personal list of baseband die sizes:

Qualcomm MSM6290: >40mm² - 65nm
Qualcomm MDM8200: <110mm² - 65nm
Qualcomm MDM6600: ~40mm² - 45nm
Qualcomm MDM8200A: >60mm² - 45nm
Qualcomm MDM9600: >90mm² - 45nm
Infineon XMM6080: ~70mm² (*) - 90nm
ST-Ericsson M570: >50mm² - 65nm
Icera ICE8040: ~19mm² - 65nm
Icera ICE8060: <16mm² - 40nm
(*): XMM6080 is a two-chip solution: 40mm² for Shared MAC + EDGE PHY + Multimedia & 30mm² for HSDPA PHY (both 90nm)

I rounded the numbers for simplicity's sake but you get the idea. The MDM9600 is definitely the largest Qualcomm baseband that has ever gone into a mobile phone (excluding basebands with integrated application processors; I think the original 90nm MSM7200 was around 130mm²) but it's still smaller than the data-only MDM8200. And the 28nm MDM9615 should be slightly more than 45mm² (I'm not assuming a perfect shrink since the MDM9600 is on 45nm and not 40nm) which is well within historical norms.

Interesting. I looked at the floorplan for the 9x00 modem but I suppose the standalone chip may have a larger die area. Still, the cost of a 90mm^2 die is very prohibitive. The 28nm shrink isn't for another 9 months at best.

I know it's significantly below the MDM9600's ~90mm² but I don't know such specific information, sorry. Keep in mind it's based on NXP's EVP16 programmable architecture with accelerators (which I heard might themselves be attached to much smaller NXP DSPs with some Tensilica-like extendibility). It's not as elegant or efficient an architecture as Icera's but it should definitely be smaller than a traditional architecture. And there are already a few fairly traditional architectures with smaller die sizes than MDM9600's...

Yes but how much smaller? And what are the trade-offs for that?

 

[Arun]

Interesting. I looked at the floorplan for the 9x00 modem but I suppose the standalone chip may have a larger die area. Still, the cost of a 90mm^2 die is very prohibitive. The 28nm shrink isn't for another 9 months at best.

Ahhh, the mdoem area rather than the full chip area, that makes sense. I'm very surprised that the MDM8200 modem was smaller (on 65nm) than the MDM9x00 modem (on 45nm) though since the chip was actually about 15mm² smaller. I suppose the non-modem digital blocks are larger than I'd expect (e.g. memory controller & USB MAC).

And another 9 months at best - that's for for production with end-product availability 3 months after that, right? Because according to Qualcomm's roadmap there should be engineering samples already.

Yes but how much smaller? And what are the trade-offs for that?

I don't know but in theory the trade-offs are similar to what you'd see from Icera, and that's obviously pretty good. In practice, it'll be quite a bit bigger given the much greater number of accelerators among other things. It is quite similar to CEVA-XC but probably slightly less area efficient overall. If I absolutely had to make a random guess based on practically no information, I'd bet on a die size of about 45mm² on 40nm (full chip) but who knows. I doubt anyone's going to bother checking for a pretty long time anyway and I probably won't have access to that information anyway.

 

[Lazy8s]

I could see Qualcomm securing a stake in IMG if they committed to their roadmap, but no buy out would likely be feasible.

http://news.techeye.net/chips/qualcomm-mulls-imagination-buy

 

[rpg.314]

Buyout is certainly out of the question, considering Apple and Intel's investments. A small stake is more likely though.

But Qcom licensing IMG would be a big deal. They have lots of volumes, especially in the low end. Also, what are they going to do with their in house gpu IP?

 

[metafor]

Buyout is certainly out of the question, considering Apple and Intel's investments. A small stake is more likely though.

But Qcom licensing IMG would be a big deal. They have lots of volumes, especially in the low end. Also, what are they going to do with their in house gpu IP?

http://www.youtube.com/watch?v=VuK-aLSZtNo

 

[Wishmaster]

http://vr-zone.com/articles/qualcom...give-you-the-details/13581.html#ixzz1YKRGXQv1
More or less all those informations were already known, except for highest-end quad core SoC.

APQ8094 - expected in 2013
Quad-core Krait CPU at 2.5GHz
Adreno 420 capable of 400 milion triangles per second(more than 5x faster than 220)
Dual core DDR4 at 1333MHz
1080p@60fps 3D video capture

One more new thing, slower(if 2GHz can be even considered slow) quad core SoC with 3G/4G modem is planned to support 4K x2K video capture at 24 fps, and 1080p at 120 fps.

Wonder how many of those things will really land in those chips.

 

[Lazy8s]

Qualcomm had gotten a steal on the acquisition of the mobile graphics team(s) from AMD, so the loss on investment would be little overall.

Considering mobile class processors will perform the majority of personal computing for people in the future and considering the significance of the GPU must rise above the other processors in a SoC, committing to anything less than world-class R&D in it is risky for a semi, especially one on the scale of Qualcomm.

 

[3dcgi]

Qualcomm had gotten a steal on the acquisition of the mobile graphics team(s) from AMD, so the loss on investment would be little overall.

Considering mobile class processors will perform the majority of personal computing for people in the future and considering the significance of the GPU must rise above the other processors in a SoC, committing to anything less than world-class R&D in it is risky for a semi, especially one on the scale of Qualcomm.

Are you implying you think Qualcomm is not committed to world class R&D with their current team or just stating that any competing company needs to be committed?

 

[Lazy8s]

The latter, but that provides the context to the matter of whether Qualcomm trying to compete with a dedicated specialist like IMG is more practical than partnering with them.

 

[darkblu]

The latter, but that provides the context to the matter of whether Qualcomm trying to compete with a dedicated specialist like IMG is more practical than partnering with them.

Isn't QCOM's Adreno division one of the 'dedicated specialists' in the field? Just curious what you think of them.

 

[Lazy8s]

It's a solid design. If all competitors were given the same budget of power and die, I think their architecture would come in second only to Mali in performance for lead PowerVR contender, ahead of Tegra, Vivante, and Pica (don't know enough to hazard a guess how the latter three would compare to each other, nor much at all about DMP's OpenGL ES 2.0 designs, nor whether VideoCore goes anywhere from here.)

While Qualcomm may start out committed to being competitive, their priority is their SoCs, and they'll ultimately make their graphics reach far enough for the needs of those SoCs while IMG must dedicate themselves to conquering the world with their graphics. An IP supplier like ARM would be like IMG in that way, investing what's needed to be better.