You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
Short version: Intel compared a 8GB Dram computer with a 256 Mega byte compute and ~8 Gbyte 3dXpoint(which is 10x cheaper than dram) sitting on the memory bus. They had almost the same performance. There's also potential for low-power , because non-volatility.
So now intel can come with this tech + a mobile soc , and build a substantially cheaper smartphone(windows , or even android) and come to Handset companies with an offer they couldn't refuse.
Another challenge for Intel is their cost structure. From R&D, design, manufacturing, marketing, to building expensive leading edge fabs, Intel needs to take care each of them. The problem is that mobile SoCs typically have lower unit price and lower margin comparing to Intel's traditional desktop and server processors.
Intel is fighting on too many battlefronts and I don't think they have the resource to keep doing that way.
>> So, the premise is: Superior Memory + Inferior GPU + Inferior Modem + Inferior Mobile CPU + Inferior Software = Intel Now Owns Mobile?
Let's talk about low-end and midrange phones because that's where the volume is going.
"Inferior Mobile CPU" - Maybe it's not the best, but it's probably good enough for those parts of the market. It's also part of some asus phones and users are generally happy with them.
"Inferior Modem" - Intel is in serious talks with Apple on LTE modems.
"Inferior Software" - Android sells very well.
As for iPhones ? it's easy for Apple to switch processors(including in apps) they've built is inside their dev tool. But it would be enough to just require Apple to fab chips at intel in return for 3DXpoihnt deeply hurt the fab ecosystem.
Is intel's cost structure that different from the fab ecosystem ? Intel generally has 80% margin, while the fab ecosystem has a 50% margin for the fab and a 50% margin for the fabless company - which together is 75% , pretty close to intel's 80%.
On the other, i agree with "too many battle fronts". definitely. They also face strong political risks with china and other countries wanting more control on computing.
The market reality is that Intel has to sell 3D Xpoint based memory products to whoever wants to buy. Especially this is a joint venture with Micron Technologies.
Apple is a very important customer of Intel today and buying a lot of rocessors from Intel. There is no reason for Intel to punish Apple by withholding 3D Xpoint product. Just because Apple doesn't want to buy more (such as mobile SoCs)?
>> The market reality is that Intel has to sell 3D Xpoint based memory products to whoever wants to buy. Especially this is a joint venture with Micron Technologies.
If i understand correctly, Micron is the minor partner in this project, With Intel investing most of the money, So i'm sure Intel has more control.Just a guess: It could just be a structure used by Intel to avoid anti-competitive charges and ease partners about second sources.
>> Apple is a very important customer of Intel
Apple has something like 8% of the declining pc market. So they are less(or much less) than 5% of Intel's business. That's big, but not sure if that important strategically.
Also, asking them to fab chips at Intel fabs under reasonable terms isn't that big of a punishment for Apple. Altough asking them to use Intel's core - that's a big deal so this may not happen.
>> Also, asking them to fab chips at Intel fabs under reasonable terms isn't that big of a punishment for Apple. Altough asking them to use Intel's core - that's a big deal so this may not happen.
Completely agree with the second part (asking Apple to use Intel's core). If Apple has decided to go vertical (A9 + iOS + iPhone), it would completely kill the strategy to buy Intel's core.
Disagree with the first part: if Apple use Samsung or TSMC foundry for Ax apps processor, it's because 1) advanced node availability at market cost, 2) power consumption behavior of these nodes.
Intel develops PC Processor and data center SoC on the most advanced nodes. These IC are: expensive and not known to be power conscious (when power consumption is as important than performance for AP).
Yes, it would be a kind of punition for Apple to use Intel fabs for their Ax AP!
I had a chance to sit with several Intel types at the recent IEEE EDPS event. They view their mobile SoC efforts as part of something bigger - the IoT, although they are quick to point out that isn't an application segment but a bunch of affiliated segments with common technology. They are putting significant energy into things like ISP engines for video, and really looking very hard at LTE Cat 1, Cat 0, Cat M, and NB-IOT with their modem efforts.
Three other comments from Intel:
-- In modems for phones/tablets, their market focus is China (including Apple). Period. At least until things change with US carriers.
-- Their IoT VP said that Android was a non-event in IoT circles, suggesting they are taking mobile SoCs a different direction (not surprisingly, Wind River is one of them).
-- 5G is where everything they do is headed.
They did not comment on 3DXpoint specifically, although there were some interesting glances during the HBM presentation. I work with Samsung so I'm fairly close to the V-NAND topic. Moving the game to phase-change instead of V-NAND depends on having fab capacity, and Samsung has lots of it. 3DXpoint, at least in the beginning, is an SSD play for Intel. It won't replace DRAM because of speed, even though it may come in lower price per bit.
Intel has a lot of money and a lot of patience, this is going to be a long game.
>> 3DXpoint, at least in the beginning, is an SSD play for Intel. It won't replace DRAM because of speed, even though it may come in lower price per bit
I think Intel at first will aim at high-end SSD market - it just makes sense to iron the kinks that way and get premium.
As for volume: Samsung does on the order of ~24TB storage/Wafer of 48L V-nand (leaving aside yield issues) - [with 2600Mb/mm2 * 74000mm2 per 300mm wafer]
Assuming that what that Intel guy says( at the video in the link in the top) is correct and they can replace most of the ram with 3DXpoint without degrading performance, replacing the RAM for 1/2-1/3 of the Android phones would take say 0.5Billion * 2GB = 1 Billion GB = 1 Million TB. That's 44,000 wafers of 48L V-Nand. Assuming Intel gets similar density with 3DXpoint, 44,000 wafers/year isn't a big number, it's closer to a monthly output from a fab , no?
>> Their IoT VP said that Android was a non-event in IoT circles,
When looking realistically, it's hard to see how the IOT even comes close to the silicon volume of mobile, unless most nodes will use something like computer vision or AI. But that's very far from any useful vision of the IOT i've read about.
It wouldn't surprise me to see something like an enhanced 3DXpoint microSD soon, as well as high end SSDs.
I should clarify that Android "non-event" comment. What the Intel guy said was most of the embedded Android applications he's being pitched look just like a tablet, nothing differentiated about the SoC itself. I strongly believe (as does Intel) IoT devices will ultimately need differentiated SoCs and operating systems. Of course, a mobile device is the on-ramp to the IoT - a gateway. Intel, like everyone else, is grappling with defining what an IoT chip is.
5G is indeed the shiny new object, the magician's voluptuous assistant. We saw this in 3G, and 4G, among those who failed to master 2G and 3G. I'm always skeptical of those who promise to dominate future G's, when they have had no success at leading the technology curve in previous G's. 5G won't be easier to define and implement than those G's before it. To the contrary.
5G is indeed the shiny new object, the magician's voluptuous assistant. We saw this in 3G, and 4G, among those who failed to master 2G and 3G. I'm always skeptical of those who promise to dominate future G's, when they have had no success at leading the technology curve in previous G's. 5G won't be easier to define and implement than those G's before it. To the contrary.
My assumption is by the time the 5G finally arrives, there will be more than one company to provide the modem related products. It's less likely for them to command a premium high price. And if their 5G modem products are targeting China market, then we can almost drop the "make big money quick" dream immediatelly.
>> I strongly believe (as does Intel) IoT devices will ultimately need differentiated SoCs and operating systems
Why ? IOT devices are basically general embedded devices, and those, most often aren't highly differentiated by hardware, and only rarely differentiated by software. Why should the IOT story be any different ?
>> I strongly believe (as does Intel) IoT devices will ultimately need differentiated SoCs and operating systems
Why ? IOT devices are basically general embedded devices, and those, most often aren't highly differentiated by hardware, and only rarely differentiated by software. Why should the IOT story be any different ?
I strongly believe that Intel has embedded devices for sure, but not the type of devices to be used in IoT edge devices (low power, low cost) and I have written about this in this blog:
But IoT will require strong cloud support, and Intel has a big stake in data center SoC. This side of IoT will benefit to Intel...
Mostly in software, not in chips. If Intel copies something like the thingworx platform(great easy to use platform, the most productive there is, but expensive), offers it for cheap/free(with conditions that appeal to customers with regards to lock-in) as part of a bundle of cloud + AFFORDABLE gateway with a software-defined-radio + a large ecosystem of IOT node providers that only use their mcu's(and with condition their mcu's offer good enough battery life) - they would probably rule the IOT. But:
1. I don't see them aiming at that software at all, and they aren't that good at building software or developer ecosystems.
2. I'm not sure their mcu's are good enough, even under this favorable scenario. But maybe with time.
I do think they will need to keep on fighting many battefronts until they get enough share in one of the markets to complement their server business. The cost of a fab for each node goes up meaning more revenue is needed to pay back the cost. Likely something complementary is needed to their server/back-office business to generate enough revenue in the future.
Latest news is that Intel has killed off Sofia: Forbes Welcome
As a result of these reviews, I learned today that Intel’s senior management has determined that they will be ending their SoFIA projects (specifically 3Gx, LTE, LTE2) as well as their Broxton SoC for smartphones and tablets.
