The single-chip smartphone sideshow

There's been a very interesting discussion in response to my post about the Access Linux Platform, but one of the comments deserved a more lengthy response. Catherine White of Llamagraphics wrote:

"The cost of devices is a big factor for the carriers, who often subsidize the cost of the handsets. Having the OS and phone driver support on one chip, as it is for Symbian is an advantage in keeping the cost of the smartphones down so that the increase in revenues from increased ARPU from data traffic is actually realized on the bottom line."

The Register has been touting this issue since last December, and I've heard it a few times from Symbian as well.

The case for a single-chip smartphone makes intuitive sense. Most smartphones today require two microprocessors in them – one to run the smartphone operating system, and another to run the radio. This has some advantages. When there are two separate chips, it's pretty hard for a virus or other nasty software running on the smartphone OS to mess with the network. But having two chips adds cost, and when you're selling millions of phones even a small expense adds up. So a smartphone that needed only a single chip ought to have a major advantage.

To run on a single chip, the smartphone OS must have what's called "hard real-time" capabilities. That means the operating system manages its own activities rigorously, so it can get out of the way when the phone needs to do something time-critical, like transmitting your voice or answering a call. If the smartphone OS doesn't support hard real-time, it might accidentally cause a call to drop, or generate sputters in voice coverage.

Symbian OS was re-architected in version 8 to have hard real-time capabilities. Currently Palm OS doesn't, and The Register reports that Windows Mobile can't yet run on single processor phones either. If hard real-time is so useful, you might ask, why doesn't every mobile OS have it? One answer is battery life.

Most mobile operating systems save battery power by going to sleep all the time. Like a lazy dog on a warm summer day, a good mobile OS will nod off whenever it can. Even when the system looks attentive (the screen is lit and an application is running), the OS slows the processor and turns off features if it can. I remember a Palm OS engineer telling me proudly that the OS even caught a little shut-eye between blinks of the cursor. As long as the OS can wake up fast enough that you can't tell the difference, it's free to sleep almost all the time. This reduces the power used by the system, and extends battery life.

Adding hard real-time capabilities to a mobile OS is like bringing home your firstborn child. All of a sudden you're responsible for a bunch of tasks that might happen at any moment, and you have to stay attentive to them all the time. You never really sleep properly again.

So getting a mobile OS to work in hard real-time without trashing battery life is a neat trick, and it's great that Symbian did it. But I'm not sure how important of an accomplishment it really is. I've been watching that single-chip issue get more and more attention over the last year or so. A first it was very persuasive to me, but the more I looked into it, the less impressive it seemed. Now I think it's mostly a red herring. Here's why.


Do smartphone prices matter?

That's the first question to ask. Prices should matter – it stands to reason that if you make a smartphone cheaper, more people will buy it, and they'll use more data services. But that's true only if there are a lot of people out there who want smartphones and can't afford them. My sense is that most of the people who really want smartphones today can afford them. Smartphones are professional tools for people who need to do heavy-duty communication and data management. Professionals can generally afford a couple of hundred dollars extra for a phone. More to the point, people who honestly can't afford a smartphone are also going to have a lot of trouble paying for the advanced data services that go with them.

I've seen confirmation of this in the customer research I've been involved in. There's a big gulf between the people who want advanced phone features (a minority of all phone users) and those who don't. Price matters a lot to young people who want entertainment phones, because they don't have much money to begin with. But it's not the most important issue to mobile professionals who want databases or advanced communication. Those people will happily pay extra to get a better product. So the cost savings of one chip vs. two chips would come into play only if the two products were otherwise identical. If there are any other differentiators, a small difference in price takes a back seat.

This seems to be confirmed in the market. For example, PalmSource didn't see big explosions in demand when Palm OS licensees discounted their smartphones. But the most memorable thing to me was what happened when Orange in the UK discounted its first Windows Mobile smartphones. Demand went up – but only because less sophisticated users started buying the products. They didn't use much data, and more worrisome for Orange, those less sophisticated users needed oodles of tech support. We surveyed the users later and they reported making an average of seven support calls each. Tech support calls are enormously expensive for the carriers – a single call per user is probably enough to wipe out all of the cost savings from using one chip vs. two chips.

So saving a few bucks on the chip cost isn't decisive to most smartphone users, and probably isn't decisive to mobile operators. The only other party in a position to care is the phone manufacturers. Reducing the cost of a smartphone won't necessarily increase their sales, but if they make a few bucks more per smartphone they'll still be extremely happy. So, exactly how much money do you save by using a single-processor smartphone design?

Anyone have a hard number on that?

The answer's not straightforward, because some mobile microprocessor companies are now making "dual core" chip packages that combine two processor cores and are priced very much like individual processors. This lets a non-realtime mobile OS get many of the cost advantages of being single processor without restructuring the whole OS, and with less fear that a rogue mobile app could interfere with the radio.

The most prominent of these dual-core processors is the Texas Instruments OMAP (Open Multimedia Application Platform) family. When I was at PalmSource, we calculated that a smartphone based on the OMAP chips could reach down into featurephone price points.

It's very hard to find OMAP price figures on the Internet. TI doesn't quote prices in public, but even if it did, the price you really pay depends on how many chips you buy and how badly TI wants you as a customer. Microprocessor Report estimated that typical OMAP prices are in the sub-$20 range .

Say you could cut that price in half by using a low-end single core chip. I sincerely doubt the savings would be that big, but let's use it as an example. Given typical manufacturer and channel markups, that might cut the end-user price of the handset by at most about $30. Nice, but not enough to dramatically change demand.

Qualcomm has also been working on dual-core chips for use in smartphones, specifically for 3G CDMA networks. If you're interested in speculating about future industry alliances, here are a couple of fun data points: First, Qualcomm is creating a version of Linux that will run on its smartphone chips. Second, Qualcomm and Access are already working together to enable iMode to work with the Qualcomm chips. Gee, I wonder if they've had any discussions about Linux.

Not all phone chip companies are making dual core parts, so an OS that can run single-core will have a wider variety of parts and probably somewhat better pricing as a result. But it's not nearly as big an advantage as you'd think.


So, to sum it all up, having a single-processor smartphone OS is very nice for a phone hardware company. It will give you a wider selection of processors, which may save you some money. That fact would probably help an OS company attract more licensees. But a single-processor OS doesn't have a decisive advantage in the market because the price savings aren't that enormous, and because price isn't the main barrier to smartphone adoption anyway.

I think the single-processor OS debate is mostly a red herring. It's a distraction from the genuine, urgent problem facing most smartphone companies – creating products that solve real problems, and that large numbers of users will actually want to buy.

12 comments:

Rafe said...

This is a really interesting post. I think its very important to challenge the assumption that single chip is the golden goose. It's not, it just part of it. I don't entirely agree with Michale on pricing. People always wants something cheaper. The support issue is a fair point, but it reflects something wrong with the usability / user experience and that has to be corrected seperately.

Freescale reckon their 3G single chip reference design for S60 is going to cut development costs by 50% which if true is quite impressive. I think the thing to appreciate is single chip it not just cost savings in component terms but also in development cost and time.

Another (related) point is that with single chip phones you can also develop a smaller manufacturing run and yet stll be cost effective (because of the development time saved).

However the most important thing about single chip is that it opens up the mid-tier device market to the mobile OS'. People may or may not be willing to pay a premium for smartphones but the same is not true of features phones. Feature phones are where the gigantic numbers are (the 100's as opposed to 10's of millions). That's why I think Symbian has an advantage with the single chip capabilities. Its not really that important in the high end (although arguably the P990 etc can cary the extras - wifi, better camera etc. without a cost increase because they are single chip), but it is very important in the mid/low end.

Simon Judge said...

"and because price isn't the main barrier to smartphone adoption, anyway"

I agree with most of what you say. However, I don't think the aim is "Smartphone Adoption". It's about Symbian OS adoption and making Symbian at least competitive with proprietary OSs.

Michael Mace said...

Ahh, two excellent comments. Thanks, guys.

Rafe wrote:

>>People may or may not be willing to pay a premium for smartphones but the same is not true of features phones. Feature phones are where the gigantic numbers are

Simon wrote:

>>It's about Symbian OS adoption and making Symbian at least competitive with proprietary OSs.

I think you're both right, and you're pointing out something that I should have said a lot more clearly in my post: Making a mobile OS single-processor matters, but it doesn't matter in the way that a lot of people have been positioning it.

If you're evaluating the mobile OS as a platform, something that enables the sales of lots of third party apps, then being single processor doesn't matter much because it doesn't move the needle on the sort of phones that drive a lot of app sales (for anything other than simple games). So when a reporter covering smartphones or a software developer like Llamagraphics asks about it, I think the answer is, "don't worry about it."

On the other hand, if you're talking about the competition for what will be the embedded OS for feature phones, then yes, being single processor is extremely important because even a savings of a few pennies makes a huge difference when you're selling tens of millions of phones.

But that's the OS used as plumbing, where it drives neither end user demand nor application sales. For the five of us who enjoy debating Symbian vs. Obigo vs. Ajar vs. OpenWave it's interesting. But it's really not all that meaningful to users and developers.

Marty Fouts said...

Actually, you don't need hard realtime to do a smartphone. In the end, amazingly few things need hard realtime -- which is good, because few systems are capable of it.

Compared to, say, an industrial process control system, the realtime requirements of a smartphone are pretty minimal, and the scheduling windows have a lot of play in them.

There are other, more interesting problems with dual-core versus single processor, but they only appeal to tech-geeks, so I'll spare you the gory.

Dean Bubley said...

Michael

There's a lot of interesting things on this thread already. Just thought I'd add a couple of my own thoughts:

- for a handset manufacturer creating a broad range of products, a single-chip smartphone architecture may mean it can reduce its overall # of supported hardware platforms, with impacts on cost structure beyond pure "cost of goods sold".
- my understanding is that Microsoft and SavaJe are working heavily on single-core products. Also it's worth noting that lower-end Microsoft featurephones are on their way, courtesy of Intrinsyc and a couple of others working on adding functionality to a WinCE kernel.

For me though, the main issue is this central one of "What's a smartphone anyway, and who does the smartness benefit?"

The Palm OS community has always looked at the smartness being "for the end user" - installing assorted cool applications from myriad developers.

However, while I'd argue that's true for PDAs, it's generally not true for smartphones. I'd say that 40% of the benefit of smartness is for the manufacturer, 50% for the operator, and only 10% for the end user.

Some manufacturers like smartphone OS's because (once they're over the initial and non-trivial learning curve) they can churn out phones, tweaking apps, tuning for operators, making differential changes or easily adapting to different form factors. Nokia and HTC are the obvious beneficiaries here, while a lot of Symbian, MS and Palm OS licencees still haven't got over the initial hump.

(Some) operators like (some) smartphones because

(a) they can create custom "locked smartphone" handsets and UIs (3 and DoCoMo most obviously, and also Vodafone in the future);
(b) they can get easily-customised versions of a desirable high-end platform (eg all the operator-specific HTC variants);
(c) they get higher data usage from some applications
(d) they can deploy specific niche apps for particular services, such as a favourite email client, or a new "dynamic content" engine like Flash or Action Engine
(e) once in a while they may actually sell a 3rd-party app

I'm a classic example of a user who doesn't really care. I've got a Treo and an MDA Vario (HTC Wizard), and I've owned Symbian phones in the past. I've never bought a 3rd party app, and I've only used them when someone's given me a demo. I've recently downloaded a Skype client to the MDA, but otherwise that's it for me and "smartness", besides the fact it helps the WiFi and keyboard work more intuitively, and I might use Pocket Word once in a while.

So, would I rather have a cheaper, single-core WinCe or Linux? Frankly, it doesn't matter. Because I think the 90% (ie operators & manufacturers) WOULD want one, because it gets them closer to what they need - a flexible, low-cost, MULTI-TASKING, phone that they can customise easily & drop in new apps "at the factory".

(Multi-tasking featurephones are another pet topic of mine, which I won't harp on about here. Suffice to say I reckon they're going to be more important than "proper" smartphones).

Michael Mace said...

Marty wrote:

>>There are other, more interesting problems with dual-core versus single processor, but they only appeal to tech-geeks, so I'll spare you the gory.

Thanks. Actually, I think I could put up with a little gore, if you feel like posting a summary. One of the things I like about the comments here is that we're educating one-other, and this is a subject I'd love to get more educated on.


Dean wrote:

>>For me though, the main issue is this central one of "What's a smartphone anyway, and who does the smartness benefit?"

I agreed with all of your comments, and we are indeed getting to the heart of the matter.

When I say "smartphone," I'm thinking of a mobile device with an open application platform and the option to add a wide variety of third party apps. You're right, that's a legacy of thehistory of Palm, although for a long time Microsoft and Symbian also thought that way, and I think the S60 team still does.

But that's not what most of the operators have in mind, or perhaps I should say that open platform definition has been discredited because most open, app-enabling phones haven't sold very well.

On the other hand, many press people, analyses, and developers are still using the term "smartphone" in the former sense. So they track things like OS share and number of applications and sweat about who's winning. Thus the breathless discussion of single core OS's and how they're supposed to be a decisive advantage in the competition for what will be the dominant OS.

I think the reality is, if the OS is just plumbing and doesn't support a big third party development community, then the forces that would drive the market toward a single OS standard are largely absent. Embedded software like Ajar and Obigo and Symbian and CE can coexist more or less indefinitely.

I think the issue for a developer like the nice folks at Llamagraphics isn't "which OS is winning," it's "will there be any substantial market for my mobile software at all?"

That merits a whole different post, one I'm hoping to write as soon as I get a little free time.

Marty Fouts said...

Well, since you asked:

You've got the single-part versus multiple-part tradeoff down pat: more parts (almost always) equals more real estate plus more glue, so means more expensive equipment.

So we decide to put the pieces on a single die and tried to decide whether to have two smaller processors or one beefier one -- assuming we deliberately leave the die size the same in either case.

(Aside: bigger dies have poorer yield and are therefor more expensive. Small is good, unless you're Intel...)

The advantage of two smaller processors on a single die ('dual core') is that you get to keep your existing programming model. This is a GoodThing because your investment in sofware may well outweight your savings in BoM.

The disadvantage is that one of that the workload doesn't split nicely, so one of the processors is always going to be busier than the other. So, you lose performance. The same die size / chip cost gives you a simpler system but at the cost of poorer overall performance.

Or you can decide to go with the beefier single processor and let the software guys sort it out. The good news is that you can handle a bigger load in the system, so you can do more cool things, or more things at once. the bad news is that the software is harder to write, and you don't get to use your existing investment.

(I simplify, but I hope that gives you a taste of the problem.)

on another front, there's the whole problem of the FCC and software defined radio (SDR) which they call 'cognitive radio'. as a phone software maker, you do not want the FCC to think you've got an SDR, because the regulatory headaches aren't worth it. It's easier to convince them of that if you've got dual cpu or dual core than if you've got single chip.

Michael Mace said...

Hi, Marty.

Good stuff! Thanks for posting it, and I didn't think it was gory at all.

A quick web search on the term "cognitive radio" turned up a lot of interesting stuff. The terms is generally used to refer to radios that can dynamically pick an available chunk of spectrum and use it on the fly -- kind of like the way airliners of the future will pick their own routes rather than flying in pre-set corridors.

I can see why the FCC would get twitchy about that concept, even though to me a single-processor mobile phone is a very different beast. I assume they're just afraid that a single-core phone could somehow be hacked to act as a cognitive radio. That sort of mirrors the operators' unfocused fear that open smartphones create a vulnerability for attacks on their networks.

There are some fascinating posts speculating on what the world will be like when cognitive radio technology becomes widespread.

Marty Fouts said...

yeah, 'cognitive' is the tip of the iceberg. it bothers the fcc from a regulatory perspective because their 'type accepted' model relies on radios being simple enough for the fcc to be able to easily test to spec. but 'cognitive' as part of software-defined means that the radios are more flexible, making them more difficult to test -- and easier to abuse.

This is already going on in wi-fi land, where the power levels of the radio are software controlled. The FCC is relying on an honor system to keep people from using more power than they're supposed to on US wi-fi, but we all know someone who has tweaked a WAP to transmit at a higher power.

Spectrum is a limited resource, and the FCC has been very conservative in the past about changing allocations. This has become a bottleneck to rf innovation and 'cognitive' (to change metaphors) is the tip of the camel's nose in the tent -- except that wi-fi waps are are camel's paw and it's already in.

the fcc likes the way sofware radio in phones work because they're tightly controlled -- even handset vendors don't often have the source code for the radio software in a broadcom or enfora radio. So, unlike waps, where it's easy to find a linux distribution and 'tweak' the radio out of spec, so far phones are more tightly controlled.

obviously, another step after dual core is to put more radio functionality in the processor and less in the radio, which will appeal to a lot of innovators, but drive the FCC crazy. (Check out, for instance, GNU Radio)

Catherine E. White said...

Michael Mace recently wrote:

I think the issue for a developer like the nice folks at Llamagraphics isn't "which OS is winning," it's "will there be any substantial market for my mobile software at all?"

That merits a whole different post, one I'm hoping to write as soon as I get a little free time.


Thanks so much to you and everyone else who is participating in the discussion for your insights on the single chip topic!

And yes, my gut feeling is that there's some kind of sweet spot at the intersection of the technical APIs and the business channels that makes an OS attractive for us as developers trying to serve existing customers with appetites for more sophisticated mobile software applications. That intersection may or may not be in the same place as the overall "winning" market share of the OS as reported by most industry analysts.

Anonymous said...

the microprocessor report link is broken, could you direct me where to go?

Michael Mace said...

Thanks for the heads-up, anonymous.

I haven't been able to find another link to that Microprocessor Report document, unfortunately (it was very hard to find in the first place).

Sorry. I will keep looking.