« October 2007 | Main | January 2008 »

November 29, 2007

Embarrassingly unoptimised

At a workshop on advanced architectures earlier this week, ARM's director of research Krisztián Flautner made an unprovable but seductive assertion. He said upfront that he had practically no evidence for it. But it's one of those statements that has the ring of truth about it. And at its heart is something that may prove to be the big problem that faces microprocessor makers as they try to work out how many cores they should put on one piece of silicon.

To make use of tens or hundreds of cores, you need applications that are "embarrasingly parallel". But Flautner is not so sure such things exist. His rule? That there is no such thing as an algorithm that is embarrassingly parallel, just an algorithm that hasn't been optimised enough yet. He pointed to situations in graphics where operations that used to be parcelled up and split across many processors in a dumb way have been replaced by approaches that apply more local intelligence but which are much tougher to parallelise.

I can think of one or two algorithms that remain embarrassingly parallel - they just aren't that useful. Anyone roving an electronics show in the late 1980s will remember people trying to push the Inmos transputer. If they weren't displaying scenes of shiny balls rendered by ray-tracing, they were processing the good-old Mandelbrot set.

People found lots of other ways to render scenes since then that don't parallelise so well. It's hard to think of a smarter way to generate Mandelbrot set images than the way they were processed then. But why would you want to?

Posted by Chris at 10:43 PM | Comments (1) | TrackBack

November 12, 2007

Intel goes heavy metal and turns it up to 11. But it's not the popular choice

Intel is spending a lot of time telling people, once again, how much of a breakthrough it has made in introducing metal gates - reversing a trend that Intel's founders began back in 1967 when they developed the first silicon-gate process at Fairchild. The Penryn processors are not just 45nm processors, they are "45nm hafnium-based high-k metal gate" chips.

The return to the metal-gate story is because the processors have entered production: reverse-engineering specialist Chipworks is etching away the top of its sample right now to see what lies underneath. So the 45nm generation is underway. And, based on the combination of metal gates and 45nm, Intel looks to be way out in front. But it's not necessarily the case. Foundry TSMC has started running customer wafers through its most advanced fab. We can expect the chips from those wafers to go on sale soon. Matsushita has already got manufacturing runs of its Uniphier chip.

Where Intel does seem to be going it alone is on metal gates. So, it should come as no surprise that the company is focusing attention on that part of the process. However, metal gates are not necessary for making 45nm devices, only those that are likely to burn a lot of power, such as PC processors. For those operating in the low-power segments, the metal gate is something of a luxury. And it's a luxury that a lot chipmakers are avoiding, at least for this generation. Only IBM has said it will use them and it has yet to show off any chips made on a 45nm process.

Although the combination of metal and a dielectric with a higher dielectric constant than silicon dioxide stops current flowing out of the transistor out through the supposedly insulating gate, that form of leakage is only one small part of the overall power loss. In a high-speed chip such as a PC processor, most of the leakage is down to the transistor passing electricity even when it is supposedly turned off. These days, transistors don't so much switch between on and off; they go from on to less-on. As a result, most of the changes needed to deal with leakage have to happen during the design phase. And, those changes cut gate leakage as much as the other, more bothersome form: subthreshold leakage.

So, other chipmakers do not share Intel's enthusiasm for metal gates. They don't want the extra cost of shipping in sophisticated and slow equipment that lovingly deposits the novel high-k dielectric one atomic layer at a time. Even Intel may not choose to use metal gates on the products it plans for the consumer-electronics market.

TSMC last year said it was developing a metal-gate process for 45nm but, as the process got nearer to production, found that a more conventional gate structure was all that it needed. The foundry is looking at the metal gate option seriously for the 32nm process and will, if a customer really wants it, think about putting a metal gate on the existing 45nm process. But they are really going to have to want it.

However, Intel's move won't shove up the cost of making Penryns that much, it seems. In comparison with previous desktop processors, the die is pretty small. The penny you see sitting on a wafer full of Penryns in the publicity shots gives a very good idea of how small the chip is. That means Intel can get a lot more onto a wafer, reducing its overall production cost. This, most likely, more than offsets the increase in cost caused by the use of the more complex metal-gate process. And it gives Intel a potentially strong lead in any price war that it might conduct with AMD to push the advantage it currently holds. There are other companies with 45nm devices, but AMD is not one of them.

Posted by Chris at 9:55 AM | Comments (0) | TrackBack

November 9, 2007

Searching for the young soul fansumers

I love the analysis at Bubblegeneration. But not in a good way. It's because the prophets of the microchunked hypersocial mediaconomy are laugh-out-loud wrong so often.

Take Umair Haque's "research note" trying to deflate the fansumer bubble. Yes, fansumer is an awful term. The approach, as described by Jeremiah Owyang and his acolytes, makes me want to scratch my palms until they bleed. But along comes Haque and you start to wonder whether Owyang is on to something. That really, really scares me.

Haque makes it clear:

"There are no fansumers. There are people who love products. But very rarely will they want to be pimped out and put to work on Facebook's (or anyone else's) digital streetcorner.

Are there really no fansumers?

They exist. And they are all around us. You work with these people. You might live with them (well, maybe you don't). You might even be one. But one thing is clear. They have no problem talking about their favourite brands. You might say they're almost obsessed with it. Of course, they almost certainly don't think of their favourite things in terms of being brands. And they certainly don't regard themselves as fansumers. But they are fans and they eagerly eat up the output of their chosen obsessions. That makes them consumers in my book.

Don't tell me companies don't want a piece of that. Only some of them can manage it - it's hard to envisage people 'connecting' with Mr Muscle Bathroom & Toilet Cleaner with Anti-Bacterial Action. But I have seen with my own eyes people with Nike tattoos. So, it is possible no matter how odd or even distasteful it seems to the rest of us who only go as far as watching the film, wearing the shoes or buying the CD.

Haque gives a stern lecture to the people talking about fansumers: "What we really need are better foundations, concepts which reflect economic reality. And building those takes much more critical thinking."

Damn straight.

Posted by Chris at 5:31 PM | Comments (6) | TrackBack

November 6, 2007

Another day, another mobile-phone consortium

Blimey, is there another mobile-phone software consortium? That's another one to add to the list. And, like the rest of them, the Open Handset Alliance is going to have to work miracles to avoid the curse of the phone-software consortium: lots of early buzz; very little in the delivery department.

Phone makers want two things: to create a must-have product that nobody else can make; and to spend next to no money making it. They haven't quite worked out that the two are mutually exclusive. But they are working very hard on the second thing on the list, as can be seen by yet more endorsements of yet another phone-software consortium. The only thing that's different about the latest effort is that we have a new player to consider: Google, which will offer access to remote applications over a highly unreliable connection. Genius. You kind of hope that they have something like Adobe's AIR up their sleeve or this thing is dead before even leaves the lab.

The rest of the cast is a list of the usual suspects, particularly serial joiner Motorola. That, perhaps, is the biggest warning sign over the Android 'platform'. Motorola has tried just about everything in phone software but it's yet to find a winner. The company seems unable to find anything in software to match its ability to design pretty, shiny cases.

Motorola has flip-flopped its way through the whole mobile-phone software business. It joined the consortium that funded Symbian but left, frustrated with the pace of development at the software company.

Motorola spent money on its own Java operating system for phones. Motorola ploughed money into Linux. It bought UK software specialist TTPCom. It got back into the Symbian game, but purely as a licensee rather than an investor. It recently took on half of UIQ – the Symbian-based user interface – from Sony Ericsson. It joined Limo. It's now joined the Open Handset Alliance. Oh yeah, and it makes some phones that run Windows Mobile. Did I leave anything out? And i haven't included some of the low-level specs consortia such as the Open Mobile Alliance and MIPI.

For reasons best known to themselves, phone makers continue to act as though you can assemble software for their products like rearranging Lego bricks. This is despite the fact that the hardware is sprouting additional microprocessors at a frightening rate, all of which introduce synchronisation and integration problems. These things are not pretty on the inside.

Even when they buy software from one source, they complain about the integration time. Nokia engineers are not exactly wild about the time it takes to port SymbianOS to a new handset. It doesn't get any easy trying to do a porting job on some kit of parts from a well-meaning consortium. It's hard to disagree with Fake Steve Jobs' characterisation of some lurching, slobbering, stitched-together monster tripping over its own tongue.

The problem for the Open Handset Alliance, like all the others before it, is that there is no company in there that will take a lead. Software companies such as Wind River are happy to provide a bit of low-level infrastructure but they've all seen the Symbian story and know how expensive it is to produce a moderately cohesive phone environment. Even if there was such a leader, the others would fight that move because they feared the consequences. Nobody really likes dealing with Microsoft because of that. This alliance actually has shades of the Microsoft at Work debacle around it: the hardware makers joined so they'd have a chance of killing it from the inside rather than end up enslaved, although Microsoft at Work fell apart of its own volition remarkably quickly.

It's either that or they all decided on an Underpants Gnome strategy:

1) Get into bed with Google
2) ?
3) Profit!

Posted by Chris at 8:45 AM | Comments (0) | TrackBack

November 5, 2007

Just what the world needs: another 32bit microcontroller

Microchip Technology, the company that stole the lead in 8bit microcontrollers from Freescale Semiconductor (back in the days when it was part of Motorola), is now going to have a go at the 32bit market. The company has licensed the M4K core from MIPS Technologies rather than give any of its own architectures another workover. Two years ago, the company turned its digital signal processors into general-purpose microcontrollers to try to break into the 16bit market. Now it has another architecture to push.

What does it mean? Basically, the decision provides designers inundated with choices of low- to mid-range 32bit micros with...more choice. Yay.

Now, this is a stunt that Microchip has pulled off once before. But only once. If you rewind time by about 12 years, you can look at the circumstances that gave an aggressive relative newcomer the advantage over a seemingly unassailable market leader. The problem is that, this time around, the circumstances are different. And Microchip has yet to demonstrate that it can turn success in the 8bit market into big sales in the 16bit sector, let alone the even more competitive 32bit space: a market space that a number of chipmakers have made the battleground for control of the future microcontroller business. And they all started earlier.

On the face of it, the M4K core selected by Microchip has everything a 32bit microcontroller user could want. Access to compilers and other tools? Check. Code compression? Check. Fancy-schmancy register banking or shadow register scheme that nobody ever really uses because operating systems don't support it and it takes too much time to set up manually? Check.

There is very little in the MIPS M4K that cannot be found elsewhere. However, that is true of most of the 32bit players - their cores pretty much do the same thing.

However, that was the case in the mid-1990s when Microchip began to catch up and overtake Motorola. What Microchip did different came down to two things: software and memory. And a boom-bust cycle that the semiconductor never really recovered from.

In 1995, Motorola was riding high, along with the other chipmakers. After a long recession, the chipmakers were selling everything they could make. And then some. Fab capacity was so short that some foundries were booking the same wafers twice, with disastrous consequences for whoever turned out to be the loser. Motorola found itself with the situation that it could not possibly provide 8bit microcontrollers to everyone who wanted one. Some customers were put "on allocation", which is industry code for "you're not getting any more for a while".

If you were in Motorola's position, what would you do? Who would you keep happy? The big guys, of course. And that's what they did. Smaller customers were turned away and so they looked for alternatives. As a company that was still recovering from a near-death experience in the late 1980s, Microchip had some spare capacity. But, more importantly, it had alighted on a sales strategy that not only won customers, it kept them.

It's hard to believe that people once argued over this point, but the ace that Microchip held was that it would sell programmable parts for about the same as those that had to be programmed at the fab. To justify this mask programming, you needed to order a lot of parts. And you had to get the order right as you could be in for a long wait if you found out sales were better than expected. Or you wound up with a lot of spare chips.

The advantage of mask program memory over one-time programmable is that it's cheaper to make. A mask memory is basically a matrix of wires: connections between them define ones and zeroes. You use a mask to define where those connections are. One-time programmable, on the other hand, needs a special kind of transistor. The result is that it takes more steps to make and works out bigger.

Incumbent microcontroller suppliers were loathe to provide parity between mask and one-time programmable devices even though the processes needed to manage those masks became increasingly burdensome. Microchip had no such qualms. So, once customers moved from their traditional supplier, they found they liked the new environment.

On top of that, Microchip made sure the tools they used to develop software were not too expensive. This encouraged very small companies and consultancies to use Microchip rather than anyone else, as the competition liked to charge as much as they could for the privilege of using their tools. Or they had to go to third-party suppliers who could not defray the software costs against silicon sales.

The strategy was so successful that a number of players have tried to emulate it since, at least in terms of software tools. It also means incumbent players are wise to the techniques used last time around. Cheap tools are not exactly rare in the 32bit space. In fact, you can get quite a lot for nothing, thanks to the Gnu and Eclipse efforts.

Memory is likely to play a part in the 32bit market. But Microchip does not have much in the way of cards to play here. Anyone who is serious about selling 32bit microcontrollers knows that they have to offer fairly serious quantities of flash and static memory to get anywhere. That is part of the reason why the processor core itself is not all that relevant. It comes down to how much you can use manufacturing decisions to push memory costs. The quantities of memory involved mean that these devices tend to look like memory chips with a processor core and some peripherals bolted onto the side.

This is where Microchip may end up being outpaced by companies already in the market. The Arizona-based chipmaker has chosen to go with TSMC's 0.18µm process – the company does not currently have any fabs able to deal with a sub-0.35µm process – whereas some competitors have decided to push density by moving to 0.13µm. Infineon Technologies in particular has a tactical advantage here as the company has had a working flash module on 0.13µm for several years. The foundries are catching up but, when it comes to pushing costs on what are becoming mature processes, the integrated device manufacturers tend to have more room to manoeuvre on price. Many of Microchip's competitors are sticking with 0.18µm for the moment, so Microchip is not that far behind in terms of process. But neither is it ahead.

And the company is coming into this market well behind Renesas, NEC and Freescale as well as ARM licensees such as Texas Instruments. And both microcontroller-oriented ARM cores – the ARM7 and the M3 – have picked up wide support among other chipmakers.

ARM has a distinct advantage if you take social factors into account. A lot of people who have migrated to 32bit from 8bit are picking ARM because it looks to be one of the dominant embedded-processor architectures of the early 21st century, similar to the way that Motorola dominated the 1980s and 1990s with its 68K family. Those social factors are hard to ignore - engineers are well aware that there are plenty of job ads for people with ARM experience. ARM's position in the mobile-phone and handheld market is paying off well here.

By opting for MIPS, Microchip does not have to compete head-to-head with so many players. But neither does it get the opportunity to pick up on customers who have embraced ARM because there is plenty of competition and plenty of engineering talent around for it.

It makes it look as though Microchip's position is more defensive than aggressive. If it cannot stop people moving to 32bit, at least get them to migrate to a 32bit micro supported by the same family of tools. But it's hard to be more optimistic about Microchip's chances than that, especially given the company's slow progress with non-8bit products in the past.

History does not look as though it is going to repeat itself in the 32bit microcontroller space.

Posted by Chris at 8:21 PM | Comments (0) | TrackBack