« March 2008 | Main | May 2008 »
April 29, 2008
It's engineering, Jim, but not as we know it
When synthetic biologists talk about what they are doing, they often point to the analogies between their work and what happens in engineering, particularly electronics engineering. You can point to some processes in living cells and describe them in the same terms as digital logic or oscillators - the kind of functions you find in a lot of electronic circuits.
The analogies don't stop there: the aim of synthetic biology is to develop a kit of parts from which you can build organic systems able to make fuels, drugs and chemical sensors. What are the parts? Professor Richard Kitney of Imperial College, London says: "We mean encoded biological functions: usually we mean modified bacterial DNA."
That modified DNA is injected into bacteria which has the machinery already in place to do the next bit, which is to make the parts work together to create simple circuits and, ultimately, create a system that does something. The annual iGEM competition, where undergraduate teams cook up modified bacteria to do unusual things, shows what can be done even at this stage.
You can have bacteria that smell differently based on whether their cultures are growing or have run out of room. The MIT team that did that put the genes for wintergreen - for a minty smell - and the chemical that gives bananas their characteristic scent in the bacterial chromosome and wired them to built-in sensors. The result was a rudimentary computer that looked at the state of the bacteria and reported what it found as smell.
From this, it sounds as though science is well along the road to being able to design biosystems. The relationship between parts of the bacterial genome seem so well understood that people have started to build tools to assemble the designer sequences: CAD for the genome.
It turns out that the technology to generate genetic sequences is outpacing bioengineers' ability to define them. Patrick Cai of the Virginia Bioinformatics Institute said at the BioSysBio conference last week: “We have more or less followed Moore’s Law for DNA sequencing.”
The technology has reached the point where sequencing company Blue Heron Biotechnology was able to produce a series of DNA chains that were combined to form a genome of 580,000 nucleic acids (bases) for the ‘artificial’ bacterium developed by the J Craig Venter Institute. However, this genome is based on a naturally occurring sequence.
The problem is that the ability to design novel sequences has failed to keep pace. Cai pointed to a competition run by Blue Heron, which would have provided the winner with $250,000 of sequencing in the form of a 40,000 base-long string of DNA. The company, apparently, did not receive a single entry.
The Virginia group has come up with Genocad: a tool that checks artificial gene sequences to see if they obey the 'grammar' of bacterial genetics. Jean Peccoud, associate professor at the institute wants to go further and build a computer language for defining DNA: "We would like to have something like the C programming language or Visual Basic," he says. You would define switches, oscillators, and AND and or OR gates using functions that look like this:
switch (ligand x, ligand y, reporter g)
A compiler would then take all those statements and compile them into a DNA sequence that would, when inserted into a cell, start to perform all the functions you defined. This language, which Peccoud calls XDL, is a long way from reality: "What I showed is only one part of what such a language would look like."
It will probably look less like C than the electronics hardware descriptions in use today, as they handle concurrency, which C certainly doesn't in its native form. Peccoud says: "We are looking at Verilog and VHDL, as used in electronics. We are looking with people who have an EE background to see if they are suitable."
However, this is where the analogies between electronics and IT begin to break down. If you look at a circuit diagram, you have lots of things like AND gates scattered around. In biology, as it stands today, you only get to use one. If you want to have two AND gates in the same biosystem, you need to find another gene that performs the same job but in a different way.
The one thing that electronics has over biology is the ability to define connections between parts. In bacterial cells, at least, you don't get that. Everything happens in what is effectively a little bag of soup in which proteins and molecules only get together by bumping into each other. And just because they bump into each other, it doesn't mean anything will happen: all the processes are statistical. It turns out that electronics is getting a lot more statistical, but it's a convenient abstraction to think as the processes as being deterministic.
Now, a biolanguage compiler could deal with a lot of that, simply picking compatible parts from a big list, such as the MIT Registry of chunks of DNA. However, as you add more genes into a system, you start to introduce odd little dependencies which means, in most cases, the thing doesn't work. There seem to be ways around this. One is to simplify the bacterial DNA chassis to the bare minimum. Another is to focus on doing more design downstream of the gene - it seems that nature makes a lot of use of what are called protein scaffolds. With these, you bring a bunch of functions together in one protein or design the proteins in such a way that only when certain proteins stick together does something happen.
Some insight into this has come from a group that focuses not on turning bacteria into factories but trying to work out why evolution has produced certain things. This is what is happening at the University of California at San Francisco.
"It turns out that signalling proteins are highly structurally modular," said Caleb Bashor, who is based at UCSF, at a meeting on synthetic biology late last year in Cambridge. Some protein enzymes in natural cells sit on a scaffold. Simply replacing and rearranging elements on one scaffold can ‘reprogram’ the protein. "It suggests that the complexity of networks is due to the interactions [between elements on a scaffold].
"Scaffolds appear to provide a powerful platform for exploring the plasticity of [cell] pathway signal processing: a tool for understanding circuit-design rules. Using them, we could rewire cells with new or modified behaviours," Bashor claims.
The problem that faces the synthetic biology community is working out just how concepts such as modularity fit into the discipline. It may be that engineers can only design these systems with a lot of computer support - with statistical tools working out what the potential interactions are between genes and other parts of the cell. It will be very different to what most people think of engineering. But the concentration on statistics might have other spinoff benefits - the inability to deal with random, infrequent events is so often what brings IT to its knees.
Posted by Chris at 11:28 AM | Comments (1) | TrackBack
April 28, 2008
Tech's forgotten takeover attempts
Marc Andreessen has a good dissection of the strategies that Yahoo could deploy to try to fend off a possible hostile takeover by Microsoft. However, most of it is of the form, "I wouldn't start from here if I were you". If Microsoft presses ahead, it seems likely that Microhoo is not far away. And, given the state of the markets, Yahoo's board may not be alone in looking over their collective shoulders.
Andreesen contends that Oracle-Peoplesoft was the first major hostile takeover in the tech sector: the argument being that tech companies depend far too much on soft assets, which are wont to leave even if the takeover goes through.
It should really be successful bids A number of attempts took place about ten years ago and a couple of years after the one that surprised everyone but which ended with an agreed deal: IBM's bid for Lotus Development.
The takeover spree of the late 1990s was fairly shortlived and, of close to ten big attempts, most of the bids were withdrawn. There was a lull and along came two more, away from the software business. At the time, financial analysts believed that a feeding frenzy was on the way and boards of directors queued up to put poison pills in place. Analog Devices only recently cancelled its - maybe it will start to reconsider its move. And yes, ADI's poison pill entered the company's byelaws in 1998. In the event, the feeding frenzy never took place.
However, the reasons as to why hostile takeovers are soon to be in fashion are similar: Andreessen is right to predict more of the same happening. And we already have interventionist investors trying to unseat boards - that is, doing takeovers the cheap way. Obrem Capital is continuing with its attempt to oust Ray Zinn and co from Micrel Semiconductor.
In the late 1990s, the big deal was AlliedSignal's takeover of connector maker AMP for close to $10bn (in 1998's money) - that's plenty major enough as Oracle picked up Peoplesoft for just over $10bn in 2005. It was a bitter campaign, as many of these things are, with the companies suing each other over poison pills and the like.
Why 1998? Basically, although the Internet revolution was picking up steam, the electronics companies were suffering from the hangover of the Asian crisis. Software companies had never had it so good. The people making the bits that went inside the computers they had to buy were facing dramatic slides in share price: making them juicy targets for anyone with a bit of spare cash. However, the slide was comparatively shortlived as electronics companies managed to get themselves back on the Internet bandwagon, at least for as long as it lasted.
In the end, in 1999, AMP was 'rescued', for the sum of $11bn, by Tyco International - yes, that Tyco. It all happened before Tyco boss, Dennis Koslowski, who liked to buy things - oh, he really liked buying things - was ejected from the company and ultimately tried for larceny.
When ADI cancelled its poison pill, I criticised the naming of them as "shareholder rights plans" as a misleading euphemism. They are surely there for the management. But there is an interesting paper that looks at the payoffs for companies that adopt them that concludes that poison pills can lead to bigger payoffs for shareholders.
AlliedSignal's move on AMP was mostly about money. The attempt by Mentor Graphics to take over Quickturn Design Systems was about putting an end to a bitter legal patent dispute - a fight that would lead to the boss of another company winding up with a long prison sentence for, among other things, trying to arrange the murder of a judge. Mentor was none too fussed if Quickturn people left, the aim was simply to stop the litigation.
Like AMP, Quickturn found a rescuing company: Cadence Design Systems. The patent war didn't end until the autumn of 2003, when Cadence and Mentor signed a deal, with Mentor paying $18m. It was more than ten times less than it wanted to pay for Quickturn, but also after a lot of legal fees had been paid.
Posted by Chris at 6:40 PM | Comments (0) | TrackBack
April 27, 2008
Talkin' bout a revolution
Last week, two events in London showed how far apart the views can be on what, to some, is the beginning of the future of chemical engineering and what to others is simply the beginning of the end.
At the IET's BioSysBio conference, which kicked off last Sunday, Professor Richard Kitney of Imperial College, London, argued that synthetic biology is the engine of a third industrial revolution. He pointed to the discoveries of the mid-19th Century and how they drove the rise of the synthetic chemical industry.
The problem that conventional synthetic chemistry has is that it is a brute-force process. It excels at producing simple molecules in high volume. But complex chemicals, particularly those needed for drugs, are expensive to manufacture. And it is no good for producing fuels because you have to put more energy in than you will ever get out.
Cells are chemical factories in miniature that are very good at producing complex chemical structures. Unfortunately, nature has not seen fit to evolve a petrol-producing bacterium. Synthetic biology opens up a future when gene reprogramming will make it possible to develop a bacterium that can turn sunlight and excess carbon dioxide into petrol or ethanol.
However, it does not take long for the ethical issues to surface. Opponents to this kind of technology worry the world will end up covered in toxic green goo pumped out by bugs gone bad.
At BioSysBio, scientists were concerned that synthetic biology could go the way of genetically modified (GM) crops, with stifling restrictions put on research and exploitation of the technology.
A few days later, it was easy to see why. Prof Kitney was also one of the speakers at the Royal Society of Chemistry's Engineering Life seminar. Biosafety was not meant to be top of the agenda, but it ended up that way. The first person I met, just before it kicked off, was a health and safety specialist from one of the London colleges.
During the seminar, Professor John McCarthy of the University of Manchester's Interdisciplinary Biocentre, wanted to take questions on the science before going on to the ethics and safety issues. However, the Q&A bypassed the science entirely and went straight for the safeguards.
As Baron Patrick Jenkin of Roding put it from the audience side: "We have moved immediately into this area. And that is why I am here...I don't think the public has begun to hoist onboard the huge potential risks and benefits."
Lord Jenkin said he planned to table a question in the House of Lords on biosafety regulation: "Is the current system of regulation sufficient to cover synthetic biology? If the public are going to be satisfied they need to be confident that the system of regulation is keeping an eye on it."
Although the Parliamentary Office of Science and Technology (POST) put out a report on synthetic biology at the beginning of the year, the debate at political level has barely started in the UK. It's not exactly racing ahead in other countries. Researchers from the J Craig Venter Institute, which is pushing ahead with research into synthetic genomes, co-authored a report on governance which largely concentrated on bioterrorism threats.
"It was an interesting assumption that this seemed to cover all the social and ethical issues," said Brian Wynne, associate director of the ESRC Centre for Economic and Social Aspects of Genomics at the Engineering Life seminar. He wryly quoted from one of the papers published by the JCVI last year, which described the transplantation of a genome from one strain of bacteria into the cells of another. One line said: "The societal and ethical implications have been explored". It referred to a paper from Science written in 1999 and the Options for Governance.
Regulation may be the least of anyone's worries. Personally, I think the framework will not change all that much from what we have today, other than some additional safeguards that focus on the companies that produce DNA sequences based on recipes provided by researchers. That is not because of indolence but because most of the core technologies needed for synthetic biology were introduced in the early 1970s. The two changes since then have been the rise of DNA sequencing as an industry and an improved understanding of what genes do. It is worth remembering that, in the mid-1970s, scientists accepted tight rules over what could be done with the then novel technique of recombinant DNA which were then loosened slightly as it became apparent that some risks had been over-estimated.
The problem is the role of industry. If synthetic biology works - and there are few reasons to believe that it won't, even if the reality falls short of what the JCVI is promising - industrial exploitation is inevitable. But the antics of companies such as Monsanto did as much as anything to turn the public off even the idea of GM. A rush toward exploitation is likely to trigger a similar reaction - the antibodies are now in the system.
Posted by Chris at 10:23 PM | Comments (0) | TrackBack
April 25, 2008
"Words are there for themselves"
Today's Graph of Nonsense Award goes to the one reproduced by Erick Schonfeld at Techcrunch to try to explain the chocolately goodness that lies behind Radar Networks' plan to apply Semantic Web technologies to search. It's one of those classic startup-generated graphs that purport to show how existing, in their view clapped-out, technology is going to give way to the shiny new stuff.
The argument relayed to Schonfeld by Radar's Nova Spivack is that today's search engines, which are based on keyword searching, are running out of steam. Spivack's contention is that the volume of data is going to overwhelm keyword-based technologies soon and that what you need is to add meaning to the underlying text to help the poor old search engines out. And, in startup style, the argument runs that the established search engines cannot deal with a root-and-branch reworking of their algorithms.
For some bizarre reason, Spivack puts down tagging and natural language search as points on the way to semantic search, rather than having semantic search before natural language, which seems to be his argument. And then we get to "reasoning": presumably at the point where we hit the Singularity or something.
Before everyone gets excited about Google going the way of AltaVista, we should take a step back and have a look at what goes on with keyword searching and then listen to what one of the technique's creators had to say on the subject of the Semantic Web.
One striking aspect of search technology is the resilience of that old-hat keyword technology. The core of most search engines is an algorithm developed in the early 1970s, primarily by the late Karen Spärck Jones and Stephen Robertson, who is now at Microsoft Labs in Cambridge. At the time, people believed that computers would have to understand grammar. The breakthrough made by Spärck Jones and Robertson, among others, was that you didn't need to force the computer to process grammar. The world is just too complicated for a machine to handle more than simple phrases.
But statistical processing works amazingly well. The algorithm that resulted from this work is amazingly simple. You calculate the frequency of each word in a document but knock out all the words that are found in most documents. Documents only score highly for words that are found in a small subset of files. It's why names and specific terms work so well in locating documents when you are using Google.
Spivack contends: "Keyword search engines return haystacks, but what we really are looking for are the needles."
No, the TF-DF algorithm does indeed return needles, almost by definition, because you don't bother indexing most of the haystack. The failing of the technology is that you have to know what the needles are called. But there are ways around the problem that do not demand the introduction of an additional layer of data.
Spivack points to the problem of citation in driving the results provided by Google as a problem of keyword searching. It has very little to do with keyword search - that is Google PageRank at work, which is a technique that gets overlaid onto the other statistical mechanisms to provide the search engine with a bit more guidance as to which pages most people find useful. Search-engine researchers are sceptical as to how much even Google uses PageRank to rank pages these days. As spam and SEO techniques get more intrusive, I can see PageRank getting dumped entirely.
The important thing to bear in mind about today's search is not that it works using keywords: it's that it uses statistics, not any form of what you might call deterministic processing. The mechanisms have been augmented over the years into what the head of search specialist Autonomy, Dr Mike Lynch, calls "keyword plus".
Lynch contents that the Autonomy engine has gone beyond the TF-IDF algorithm of Spärck Jones and Robertson. But it still, very much, uses statistics. Autonomy uses Bayesian inference which dispenses with the concept of word frequency. Autonomy replaces it with an approach that uses the probability of finding certain words within a portion of the document. One of the problems with Bayesian inference is that the number of calculations you need to make is much higher than with keyword indexing if you are not careful. But there some tricks you can pull to reduce the workload.
The important thing is that every successful search engine in use today doesn't have to understand language by attempting to deconstruct sentences using grammar rules or trying to extract meaning: it just has to use statistics. Spärck Jones' final lecture, which was recorded about a month before her death from cancer at the age of 71, describes how successful statistical processing has been when dealing with language. In fact, even language researchers have found statistical models more successful than deterministic grammar models in recent years.
As Robertson put it to me when I interviewed him a week ago, there is plenty of scope for statistical processing in search and similar applications.
Statistical methods that evolve from the ones in use today probably have a lot brighter future than any kind of semantic search. The problem with the Semantic Web is that you have to add a new layer of information to the information you have already created through the medium of XML tags. I have nothing against annotating certain types of information. Microformats should make the meaning of data such as names and addresses more apparent to the machine.
But the idea of annotating massive chunks of text with helpful tags is, frankly unworkable. You have a bunch of words that, taken in context, mean something. With the Semantic Web, you are then adding another bunch of words to, hopefully, provide hints that disambiguate the text. And you are doing this at the same time that research is using statistics to allow the context around those words to provide clues to the meaning.
Or as Spärck Jones put it in her lecture: "Words are there for themselves. They are not being replaced by other codes."
One big problem with the Semantic Web is that, to make it useful to a computer, you have to produce a dictionary of meanings: an ontology. Building ontologies is hard. Very, very hard. People in fields such as medicine are still having trouble, and they've been doing it for a while.
Maybe tagging provides a halfway house. Some tags are useful but these turn out to be recommendations more than tags of meaning. Lynch explains: "Tagging has its uses but they are much more limited than people realise. It is fundamentally flawed as a retrieval technology because of specificity. There are not enough layers in the tagging."
The problem is that you need to add a lot of tags to provide the specificity you need, other than simply pointed to a key word itself and making it, well, a keyword. "People say they will add more tags," Lynch says. "But by the time you add them all, the probability that the person looking won’t go in the same order through the tag hierarchy is very high.
"The other thing is when you go to the 'right on' social ideas, such as folksonomy," Lynch adds. "People in different locales put on different tags. Then you get ‘metatoxins’: when they put the tag Britney Spears on a YouTube video it is actually them painting a wall."
"Where tagging is very useful is working out whether a document is helpful. Say, if you get people to tag stuff as tutorial documents or whether the document is useful, getting a human to make that comment is better than a computer."
If you talk to artificial intelligence researchers, they don't mind having the information provided by Semantic Web technologies or tags from keen Web 2.0 users. They will, as it were, take as much information about meaning as they can get. But, when it came to search and information retrieval Spärck Jones had no illusions about the effectiveness of the Semantic Web:
"The idea of a semantic web as a universal characterisation of knowledge strikes me as misconceived. There can be no ontology that will work for everything and everyone. There may be many specific ontoslogcal horses for particular courses. But the universal means for getting from one to another cannot but be a sort of lightweight ropeway.
"Something lke that does establish links and make it possible to move around. This is what natural language tools like statistical association between words provide. People will get a start moving from simple words and phrases...Natural language is general but it leads to particulars."
Ripping out the guts of today's search engines and replacing them with a very different engine is not going to work. The search engine of choice in ten years' time may not be Google: it could have gone the way of AltaVista. But it is unlikely to be replaced by something that does not use a statistical or probabilistic engine at its core.
Posted by Chris at 9:28 PM | Comments (3) | TrackBack
April 23, 2008
Apple gives the finger to the chipmakers
Late yesterday, Forbes reported that Apple has decided to buy boutique chipmaker PA Semi. So, the conference call later today where Apple announces its results for the second quarter of 2008 is going to be interesting. And there will be a bunch of silicon suppliers wondering what's going wrong for them.
Discarding the possibility that Apple has decided the move to Intel, and its rejection of PA's PowerPC processor in 2006, has been an awful mistake and it suddenly needs to press the architecture reset button, the move by Apple suggests that the company is not all that happy with the shape of today's integrated circuit (IC) business.
One possibility is that Apple has decided it needs more in-house chip designers and buying PA was a quick way to staff up. That's not unusual in this business: it's a surprisingly common way of getting hold of people who can design the analogue circuits that most electronics engineers fear to touch. Even after you've bought in a bunch of processors and memory, there are other places a computer maker can use experienced IC designers to get an edge on its competitors. You don't see that much in the PC business but it's a lot more common in places like the phone market.
This particular team has a string of famous processors behind it, as it's the team led by former Digital Equipment Corporation architect Dan Dobberpuhl - responsible for the Alpha, famed for being the world's fastest processor for a while, and the StrongARM, which Intel ended up using to push its way into the PDA business before passing the design on to Marvell Semiconductor. With that kinds of background, it's hard to see this team being happy to work on glue logic. Apple might be able to lock the senior people in for a while but, if the project isn't a processor or something similarly complex, you would expect most of them to drift away quite quickly. So, it's fair to assume that Apple is serious about having its own processor design team, if not the PA processor itself.
PA's PWRficient design uses a number of tricks to get the power of a 2GHz processor down to manageable levels - below 20W rather than the 80W of a typical laptop processor. But what PA has today does not fit too well with what Apple is doing these days. It's OK for telecom basestations but even the best-case power consumption of 5W is too rich for a palmtop devices: you really need to be under 3W to stop the thing turning into an expensive hand heater. Forbes, quite reasonably, speculated that Apple wants the team to work on a processor for a future iPhone or mobile internet device (MID).
Apple's decision is not just a poke in the eye for Intel, which would have liked to sell its Atom into the world of MIDs. There are a ton of companies out there with processors based on ARM all desperate for a slice of the phone and MID markets. And it looks as though Apple looked at all of them and found them wanting.
Today, the iPhone uses an ARM-based processor from Samsung Electronics. The company has used a variety of similar processors in its iPod line-up over the years. PA made a big issue of its use of clock gating and power supply control - but similar techniques are in use on ARM-based processors. For example, ARM has its own software-controlled energy manager for the ARM11 that is broadly similar to PA's dynamic power supply control. You can buy tools from companies such as Calypto to insert gated clocks into circuits automatically, which lets designers do the much finer-grained clock control that PA claimed to have.
So, why take on the risk of doing your own processor when you have 20 people beating down your door to offer you one with broadly similar features? Because you think, with your own team, you can do a better job. The StrongARM was an example of what could be done with the ARM architecture if your focus was to get a much faster processor without driving up power consumption too much - and that was not ARM's focus at the time.
It would make sense for Apple to continue with ARM in this class of device simply because the software has now been ported to the architecture rather than trying to do an even leaner PowerPC. So, one option may be to get Dobberpuhl and co to do StrongARM 2, or to get them to have another go at the PowerPC architecture they already have and bear the cost of porting the iPhone software to that processor. Having gone through several architectural changes, Apple is getting pretty good at it. I'd opt for the ARM approach as that gives Apple the most room to manoeuvre if its plan does not come off.
Whichever way Apple goes, it's not good news for the mainstream chipmakers as this acquisition largely tells them that, whatever they are doing, Apple reckons it's just not good enough. It's not as if Apple will have a problem getting the resulting products made: there are a number of foundries who would only be too happy to take Apple's business.
Posted by Chris at 9:47 AM | Comments (2) | TrackBack
April 21, 2008
TSMC's 5nm difference
There was a telling moment in the conference call hosted by Altera ostensibly to talk about its 1Q08 results but also drop in a few hints about an upcoming family of programmable logic chips.
Historically, Altera and market leader Xilinx have taken lumps out of each other as they vied to be first onto each manufacturing process. But something changed at the 65nm process node. Xilinx was quick to get its high-end parts out on the 65nm technology, but nowhere near as quick as the company's claims over having first silicon on that process. Altera was behind but took the opposite tack: putting its cheaper Cyclones onto TSMC's 65nm process first. Then it all went quiet. The launch of the 45nm process went off with both Altera and Xilinx being uncharacteristically quiet. It began to look as though a kind of chip that had become the foundries' banker for early silicon had suddenly fallen off Moore's Law.
It seems that the programmable-logic makers are still in the running, just not as quick to jump on a new process as they used to be. And it seems that Altera wasn't aware just how advanced its next process would be until TSMC decided it would lop another 5nm off - in name at least.
At 32 minutes into the call, Altera CEO John Daane corrected financial analyst Uche Orji about the next generation: "Our 40nm, and it is 40 not 45...we did call it 45 earlier but only because TSMC had not announced that it was really 40. They did that a few weeks ago, so I can tell you that we are the lead customer on the 40nm process technology...both software as well as components will be shipped this year."
Got that? Up until a few weeks ago, Altera thought it was working on a 45nm process. Until someone at TSMC decided that they were better off calling it a 40nm process. It's no big surprise, because the people talking about the self-same process at the International Electron Device Meeting (IEDM) last December also thought of it as a 45nm process. Which kind of makes you wonder just what TSMC claimed to be shipping last autumn when the company said it had shipped 45nm-process wafers to lead customers.
The issue of what constitutes a process node is something that has exercised Kevin Gibb of Chipworks, the Canadian electronics-analysis company. The number we have been using to describe a process node has pretty much lost all meaning. I have to agree with him, as well as Kaizad Mistry of Intel, who said in his talk about Intel's 45nm process at IEDM: "Contacted gate pitch is perhaps the most important [design] rule for density.”
Once upon a time, the number associated with a given process meant how long the transistor gate was. Then, thanks to some optical trickery, everybody worked out that they could reduce the size of the gate, and increase the speed of circuits, by much more than traditional scaling rules. The result was that, by the time the industry hit 90nm, gates were down to less than 60nm in length. At 65nm, you could find devices with gates not much longer than 40nm.
Then, because leakage power was going through the roof, manufacturers slammed the brakes on. Intel's latest devices seem to have gates no shorter than their 65nm predecessors, at around 40nm. TSMC's '40nm' process looks to be in the same ballpark. The company showed a picture of a 30nm p-channel device. However, these tend to be shorter than the n-channel transistors that people tend to base the overall measurement on. Certainly, in the case of both Intel's and TSMC's processes, leakage from n-channel transistors picks up steam from about 38nm down.
If you look at contacted gate pitch - this basically determines how tightly you can pack transistors that you have actually wired up to circuits - there is hardly any difference between Intel's 45nm process and what TSMC claims to be its 40nm process. That comes in at 160nm, with TSMC being very slightly wider at 162nm.
So, what we have now is the situation where the chipmakers are going to quote numbers in the hope that everyone thinks, because they chose a smaller number in the press release, they are in the lead. It's reminiscent of the situation in the late 1990s when LSI Logic thought it was clever to quote a different measurement to the one everyone was using at the time to make out it was one generation ahead. You would have to spend ten or fifteen minutes going through what their claim even meant before getting on with the important stuff. It seems we are going to be back to those days now as Altera and Xilinx limber up for the next round.
Altera will no doubt declare that it is in the lead with a '40nm' process. It's not clear whether Xilinx will follow suit - we will have to wait and see. I'm still wondering what we should really call TSMC's first 45nm process, which, according to the foundry has dimensions around 10 per cent larger than what it calls '40nm'. If we assume that Intel's published process is more or less dead centre in the 45nm zone, that would make TSMC's older '45nm' process something like a 50nm.
Posted by Chris at 11:04 PM | Comments (1) | TrackBack
April 16, 2008
Running the numbers on eMusic
The 200 millionth download at eMusic has provided an opportunity to take a stab at how many active subscribers the service now has. I did it the not so subtle way by plotting the cumulative downloads against days since eMusic went subscription only. The company conveniently provided three real data points and one implicit point in an arrangement that suggested some kind of power law was at work in the cumulative count.
For one, it implies that the growth of eMusic in the last couple of years has been pretty linear and got a bit of push sometime during 2006. Wasn't that roughly when AllofMP3 got its marching orders?
Using the graph it looks as though eMusic has surpassed six million downloads a month. The company claims 7 million a month now, which is kind of borne out by the fainter trend line in the graph. However, based on the figures provided, the launch of Amazon MP3 last September didn't seem to do a lot to download growth and it may even have tailed off a little. In either case, it's tough to square the downloads with the last known subscriber count of 350 000, released towards the end of November. It implies that the average user downloads way less than even the minimum subscription.
Simply dividing the 30 downloads of a basic subscription into the six million downloads I estimate were made in the last month only gives you 200 000 subscribers. The subscriber numbers seem somewhat inflated based on this, but eMusic said at the time these were paid subscriber numbers so it seems unlikely that there is some kind of Second World effect coming in here. It seems, therefore, that a lot of people do not come close to downloading their full allocation of files every month.
I tend to make sure I use up almost all of my allocation - and I'm on the 65-download subscription - as download credits don't roll over. So, there must be quite a few people who come nowhere near their quota. That's a nice little earner for eMusic as it means royalties that the company does not have to pay out, or it means that users who eke every last credit out of their quote every month are getting music cheaper than they might if, for example, eMusic started rolling over credits each month.
Either way, it remains my favourite paid-music download site.
Posted by Chris at 9:35 PM | Comments (1) | TrackBack
DFM heads into the foundry
Sometimes, small deals can wind up changing the shape of a market. The deal between Blaze DFM and TSMC that has been gestating for close to a year is possibly one of them: it recalls the giant leap of faith that Artisan took when it came up with the "free library" idea.
Basically, Blaze and TSMC have cut a deal that will see the Taiwanese foundry use a version of the Blaze MO tool to alter transistors in a layout to make them less leaky just prior to manufacturing. The idea is not new and fairly simple: you make the transistor gate longer on logic paths that don't need to be fast. This typically shifts the threshold voltage up, which cuts leakage. STMicroelectronics has been offering the same sort of modifications using different logic cells. What is different is the nature of the deal between Blaze and TSMC and what it could mean for the whole DFM business.
Instead of trying to sell tools on a per-seat basis for something like a couple of hundred thousand dollars – the regular EDA business model – TSMC will host the tool. Although the companies will not talk about the money side of the deal, it does look broadly similar to the Artisan free-library model, where the foundry paid a royalty to Artisan for each chip made and charged a bit more for each chip to the customer.
For Jacob Jacobsson, CEO of Blaze DFM, this approach, in a way, opens up money that isn't available to the EDA tools vendors. "EDA has a had a more or less stagnant $3bn budget for as long as we can remember. It is more attractive for us to align with the manufacturing side of the business."
As Artisan found with the cell-libraries business, customers seemed happier to put a little bit more on the money paid to the foundry to make each chip than cough up for a large initial licensing fee. People argue whether Artisan's move was good for the libraries business, but it worked pretty well for Artisan, which ended up being bought by ARM. And it worked for TSMC.
Similar to the original free-library deal, Blaze's deal with TSMC is exclusive. Asked whether the company might look at deals with other foundries along similar lines, Jacobsson said it's far more likely it will concentrate on TSMC, because the foundry has such a dominant position in the market.
The changes to the design are applied during optical proximity correction (OPC). These are subtle changes to the layout that improve the chances of the drawn feature turning up on the die. Jacobsson claimed that applying the modifications during the OPC phase allows more aggressive optimisations: the tool can be more confident that a change will result in a transistor with a gate of a given length than if the alterations were applied to the original GDS II layout file. OPC is also a process that TSMC controls, so it's able to fold things like this into the flow quite easily.
Jacobsson said there is plenty of potential to push DFM into the foundry. The company's founder, Professor Andrew Kahng of the University of California at San Diego, has talked about the types of changes that are possible at conferences such as ESSCIRC. This deal does go some way to explain a rather opaque statement of Kahng's during the September conference in Munich. Talking about the guard bands applied to cell libraries to prevent designs failing due to vagaries in the manufacturing process, he said: "If you take canonical, TSMC-style libraries, you can see that some guard-band reduction results in higher yield because the die is smaller. There are some interesting trade-offs that are possible if you play with motivation. Dare I suggest to the foundries that they could provide some incentives for reducing guard bands?"
And there are other places where OPC-stage modifications would come in handy - "opportunistic, do no harm OPC", as Kahng termed it. These are things like dummy poly - effectively strips of polysilicon laid down between actual transistor gates to improve manufacturability. Intel's already doing this kind of thing big-style on the lower metal layers in Penyrn, as Chipworks has found.
Now, no-one is going to switch to TSMC just because the company can apply a bit of leakage reduction - although it is clearly an additional selling point. But, the idea that DFM analysis performed at the foundry using tools from a company such as Blaze DFM might result in smaller or faster integrated circuits, that stacks even more of the chips in TSMC favour. As it it didn't have enough already.
Now, I don't know if Mike Fister is a chair-kicking kind of person. But if he is, I imagine one or two are looking a bit worse for wear right now. This kind of arrangement, which breaks from the old per-seat pricing model is just the kind of thing the Cadence chief has been talking about since he joined the number-one EDA company. More deals like this would lock companies such as Cadence out of entire chunks of the DFM business as only the biggest fabless chipmakers are going to want to pay upfront instead of seeing that cost absorbed into that of the devices made by their foundry.
Posted by Chris at 8:10 PM | Comments (0) | TrackBack
A real embedded OS or just regular Windows in a corset?
Hold the front page. Major news from the Embedded Systems Conference in San Jose. It's a classic piece of deckchair rearrangement. And it's one that makes you wonder about the marketing acumen of a company that's supposed to be a lot better at it than its competitors.
Basically, the company couldn't give a better rod to companies such as Montavista and Wind River to beat it with than this one. The 'news' is that Microsoft is, once again, rebranding its embedded operating systems. The plural in the last sentence is a little misleading as Microsoft only really has one embedded operating system that was designed for the job. The other one is a 'componentised' version of XP that won't bitch and whine if you haven't plugged in a keyboard when it boots.
However, the XP you can put on a diet is now, apparently the 'standard' embedded operating system. Windows CE, which is a different piece of software altogether and was built more like a classic real-time operating system is now the 'compact' version. And there will be an 'enterprise' version which seems indistinguishable from regular Windows XP or Vista other than it will get some sort of extra embedded mojo over time, according to the company.
Now, you could argue that, with Linux making bigger and bigger inroads into the embedded business, it makes sense for Microsoft to focus on the larger OS. However, the Linux kernel has gradually been acquiring bits of technology that are useful for real-time work. To run Windows Embedded Standard in a real-time environment, you need to use some form of virtualisation. So, what tends to happen is that people use Windows XP for the pretty user interface part and something else for the real-time part. It works. But communications between the two parts is not quite as straightforward as doing everything on one OS.
Also, note the language used in the release about the next generation of software:
"The first product release under the new naming strategy will be Windows Embedded Standard, the next generation of Windows XP Embedded, and will be launched simultaneously at Tech•Ed North America and through a global webcast event on June 3. All presently available Windows Embedded products will be marketed under their current names until their next scheduled product release..."
The next generation of XP? Hmm, doesn't sound like Vista is going to go on the embedded diet anytime soon, which makes you wonder about the near-term prospects of a chopped-down Vista running on a mobile device.
Posted by Chris at 3:53 PM | Comments (0) | TrackBack
April 14, 2008
One more salvo in the 32nm war
Earlier today, IBM put out a release claiming a major "performance leap" for chips that use its forthcoming 32nm semiconductor process. Working out what's changed since the last release is a bit trickier. Basically, IBM and some of the companies in its group of chipmaking collaborators have made a bunch of test chips and are now confident enough to declare the 32nm process open for business.
Other than that, the content of today's missive is not broadly different from the one that IBM and its partners put out just ahead of the chipmaking industry's big conference on process technologies, the International Electron Device Meeting in Washington DC, held late last year. There really isn't a lot more detail, other than there is now a timetable: IBM will start running prototypes for customers of the companies in its Common Platform alliance in the third quarter of this year. The implication is that the company's in the Common Platform team will have a working 32nm process in the second half of 2009, about the same time as Intel and TSMC as long as they stay on schedule.
Where IBM, Intel and TSMC deviate is over how the transistors are made. IBM and Intel decided to with metal gates for their 45nm processes - although companies in the Common Platform alliance do not have access to the IBM metal-gate technology at 45nm. IBM has kept the 45nm version to itself for use by its computer operation. Intel, as it was keen to tell everybody, opted for metal gates and a hafnium dielectric for its latest generation of processors.
TSMC has, for 32nm, no publicly announced plans to shift to metal gates, despite the apparent benefits of the metal approach. The company continues to experiment with approaches but looks to be pressing ahead with a conventional gate structure. IBM has claimed the tests performed by its team show a speed boost of 40 per cent over conventional transistors with gates made using today's polysilicon. If the improvement is that good, why not do it? Cost, that's why.
One thing that's missing from anything that IBM has said so far about its, apparently, better process is what the developers have done to not push up the cost of making chips with metal gates. It's not unusual. Although Intel was only too keen to say that it was swapping out silicon dioxide in favour of a hafnium oxide for its metal gate stack, what was missing was how the world's largest chipmaker was going to manufacture devices without forcing up cost.
The problem with metal gates is that, in most processes described to date, you need two different metals for the two types of transistor used in a CMOS process. And that means double the number of steps during the most expensive part of the chipmaking process. Intel worked around it by adopting a so-called gate-last process - make dummy versions of the gates and then, at the last minute, etch them out and drop the right materials down the hole. It sounds messy but Intel claimed at IEDM that it was little more complex than previous processes, and the company was careful not to give too much away. Officially, we don't know what they used. Unofficially, companies such as Chipworks have provided clues as to what Intel used.
With IBM's process we don't even have that much: even the gate dielectric is a closely guarded secret although it looks as though the VLSI Technology Symposium in Honolulu may provide some of the answers. What we do know is that IBM is working with just one type of metal and using tweaks in the process to tune the metal's properties for the two types of transistor needed.
A number of groups are working on this kind of single-metal process, something that most mainstream chipmakers believe is essential if metal gates are to be cheap enough for them to use. Intel and IBM had little choice but to go with metal gates for 45nm as their high-end processors would suffer with the polysilicon approach. But most companies's designs don't need the speed or ability to cope with tens of watts of power that a Penyryn does. For many of them, the only thing that counts is transistor density: a lot of these processors should not chew up more than 3W of power, let alone the 130W of a high-end processor. This is why TSMC does not seem overly concerned about delivering metal gates: many of its biggest customers just don't need them.
Update: As I'm increasing the amount of blogging that I'm doing on chip-related matters, I'm gradually bringing up a second blog – shrinkingviolence.com. Right now, a lot of the material has been crossposted from here. But I'm now adding posts that you can only find there: such as this one going into more detail on Common Platform versus TSMC.
Posted by Chris at 10:35 PM | Comments (0) | TrackBack
April 13, 2008
Hidden comments
A bunch of people are up in arms about yet another social site that hoovers up newsfeeds so that people can collect all their comments into one place. The two big problems that some blog owners have are these: it's an infringement of copyright as content is being sucked into another site wholesale; and it encourages people to comment on posts away from the source blog, so that the blog owner can't get to see them without subscribing to this new site.
The first point is a tricky one. You could argue that it is an infringement of copyright. However, if you are providing full feeds then Shyftr is really only acting like an online newsreader. The name Shyftr doesn't really help the service's image but, if you don't want copy hoovered up in this way, don't provide full feeds. As this blog isn't ad-supported, it is not that big a deal where the material is read as long as it's attributed to me. Sure, I'd like to know how many people are reading. Owners of sites such as Shyftr would buy themselves a bit more slack if they ponied up readership stats to the people who provide the actual content. But it's not in evil country yet. Anyway, if you're that worried about content leeching, just used a bit of Apache mod_rewriting to serve up partial feeds, or a list of links to Rick Astley videos, to those service's spiders – assuming they've been good and announced themselves.
The second 'problem' is an indication of how misguided some bloggers are when it comes to the subject of The Conversation, although I think there is a small, subtle issue with a site like Shyftr. Because comments appear on blogs, it is easy to be misled into thinking that is where all the action is happening.
Take Scoble, for example, who can be relied upon in these circumstances to come out with this sort of line: "The era when bloggers could control where the discussion of their stuff took place is totally over."
And bloggers had control before? How so? Is that like how nobody discussed what appeared in the papers before blogs came along? Pubs and cafés were eerily quiet as people digested their daily news in total silence, fearing to talk about it because the nasty media had all that control?
The ratio of comments on blogs to page views is very low: I don't think 1 per cent is an unreasonable number and that's after weeding out obvious spiders. The referrer logs often show up links from webmail accounts, forums and sometimes intranets. Have the people asking for some kind of comment aggregation system so that "the blogger can see all the conversation" actually thought about the problem for a moment? You can't see it all. Because, even if you designed a superspider able to track down and parse links to blogs so that it could extract relevant comments, it still wouldn't be enough.
Are you going to bug people's email accounts? The water cooler? Even if you did have such a piece of spider software, the thing would just be the most powerful spam magnet ever invented. It would be stuffed full of poker and pills ads in minutes.
And does it matter? If people want to make a point to the blogger, they go to the blog and post a comment there. This is where I can see something like Shyftr being a problem: the site makes it look as though the home for comments for a blog is there. It doesn't even attempt to locate comments on the source blog because it's working from the RSS or Atom feed. So, I can see people posting points on Shyftr thinking they are being made to the blog's writer when those comments are, in fact, invisible to them.
Shyftr might do itself some good by coming up with some kind of widget - assuming its programmers haven't done this already - to at least show where comments on a post might be located. Perhaps they could even feed the comments back to the site. However, the second option sounds dangerous even just typing it in now. It has cross-site scripting vulnerability written all over it. After all, who is going to weed the spam out of Shyftr comments? The blog owner can't do it. It doesn't look as though Shyftr users can do it. The man and his dog at Shyfter's owner Upshot Interactive (and I'm not convinced there's a dog)? Good luck with that.
Posted by Chris at 11:16 AM | Comments (0) | TrackBack
April 12, 2008
EDA's acceleration option
John Busco at John's Semi-Blog has pointed to the launch by Nascentric of an analogue-circuit simulator accelerated by nVidia's graphics processors, and wondered: "Will general-purpose GPU computing become the acceleration platform for EDA?"
I was sitting at the Many-core and Reconfigurable Supercomputing (MRSC) conference in Belfast the other week wondering the same thing. In recent years, hardware-specific EDA has been a dirty word. Mentor Graphics, which made its name selling proprietary workstations before it became a software-only company made a foray back into hardware in a deal with Mercury Computer Systems in late 2006. Mercury used the IBM Cell processor – the same one used in the Sony Playstation 3 – to speed up the job of checking chip designs before they go to fab. Mercury sells the hardware and Mentor provides a special version of Calibre.
It's not clear how well hardware acceleration has gone for Mentor and Mercury. However, in its 2007 annual report, Mercury declared that it saw a "slight rebound" in its semiconductor business, partly due to the sale of one accelerator for chip-mask inspection – which is not related to Calibre – and its deal with Mentor. The number-three EDA company has been busy showing off the hardware at events like the SPIE lithography conference, so the company must have some faith in the idea of speciality accelerators.
The algorithm in Calibre is probably a good candidate for acceleration by GPUs as well as the Cell. One thing that was noticeable from MRSC was that users in the academic environment there were not making that much use of Cell but they were very keen to look at GPUs as well as field-programmble gate arrays (FPGAs) – the latter just happening to be the EDA acceleration technology that nobody really notices.
People have been using either emulators made from hundreds of FPGAs - OK, not that many people – or FPGA breadboards to simulate digital chips for years. Synplicity made such a good business out of doing tools for FPGA-based prototyping that Synopsys dismissed the fact it killed off a tool to do the same thing a couple of years ago and bought the company. (Actually Synopsys has gone through three different FPGA synthesis tools in recent years - FPGA Compiler, FPGA Compiler II and DC FGPA - we will wait and see how it does with the stuff it buys in.)
When Mentor unveiled its deal with Mercury, Joe Sawicki, the head of Mentor's Calibre operation, said they had changed the way the tool worked in such a way that it would better suit an accelerator like Mercury's. In the past, tools like Calibre used a sparse representation of the chip's surface to perform their analysis. Mentor's argument was that, at 45nm, the features on the surface of a chip are so densely packed that you might as well just chop it up into a regular grid and have at it with fast Fourier transform operations.
If there are two things that run well on accelerators, it's regular grids and FFTs. And I can't see a reason why a GPU would not be a potential candidate for the nmDRC software. I'd be surprised if Mentor wasn't looking at a GPU option.
But, not everyone is convinced that accelerators are the future. Srini Raghvendra of Synopsys made the point at the time that, with general-purpose multicore processors on the way from AMD and Intel, optimising for a dedicated accelerator from a single hardware vendor was unlikely to be a long-term option: "We believe we can be comfortable riding the general-purpose processor horse."
One thing that hits a lot of EDA software is bandwidth: between processor and memory and from memory to disk. It can take hours just to read a design in and hours to write it back out again. Your best bet might not be a funky accelerator but a half-decent storage area network with a bunch of fat pipes into the back of your server farm.
Then there is the issue of how much EDA software has actually been multi-threaded to run across multiple processors. With the kind of job that a tool that Calibre does, multi-threading is commonplace. If design teams don't want the Mercury accelerator they can just buy a bunch of Calibre licences and run them across a server farm. With layout checks, the shape of one logic gate does not affect another one just micrometres away. You can, with some limitations, chop up the grid into little chunks and distribute them to many processors without worrying too much.
A lot of EDA software is not so lucky. It is only recently that Spice simulators, such as those from Nascentric, have gone multi-threaded. As recently as last year, people were arguing how useful multi-threading would be in that environment. Regular Spice is all about solving regular matrices. Fast-Spice simulators typically use sneaky mathematical tricks to avoid having to crunch through massive regular matrices. The name of the game, as with a lot of EDA, is to convert a problem that scales with square or cube of the number of elements to something much more linear, or even logarithmic.
Unfortunately, these more optimised algorithms don't necessarily divide well. If you're not careful, you can spend so much time sifting and sorting data that the speedup you get from multiprocessing gets almost wiped out. So, something like Spice, which looks like a great candidate for multiprocessing, doesn't fare so well.
The results from academia on sparse-matrix acceleration using GPUs are good but not spectacular so far. People tend to report the same problems: memory bandwidth issues, limited cache memory on the GPUs themselves and the need to run thousands of threads in parallel to get any meaningful acceleration. People have reported speedups of maybe 2x, sometimes 10x, but not the 100x you might expect from taking something that runs on one processor to a graphics chip with 128 processors inside.
The release from Nascentric is a masterpiece of legerdemain in that respect - leading you to assume that you will see a 100:1 or even 500:1 level of acceleration, based on the number of SIMD processors you get in the hardware. And then there's the claim from John Croix, Nascentric founder and CTO: "Using nVidia's Tesla platform we can perform circuit simulations in minutes to hours that would previously have taken hours, days and weeks."
However, there are no numbers in the release to back this up. You may have the situation where regular Spice code gets a big speed boost but does that happen for some of the Fast-Spice algorithms. Bear in mind there is a lot of tweaking inside Fast-Spice. Designers are invariably turning things off in the hope that their circuit doesn't depend on those elements. The detail on the numbers for each of those cases would be pretty revealing.
A second issue is one of numerical precision. I'm not sure how much codes like Spice depend on double-precision floating point maths but, on GPUs, you really only get single-precision. You are more vulnerable to underflow and overflow issues in code that iterates a lot. Scientists tend to worry about this a lot, although numerical analysis work by computer scientists is doing a lot to assuage those concerns. However, accuracy could be an issue with GPU acceleration in EDA.
On the other hand, any company that looks at GPU acceleration is set up for the future. With OmegaSim GX, today you have to buy a separate accelerator. A few years from now that same code may simply harness the integrated GPU sitting on the AMD or Intel processor. It gives floating-point intensive software an source of Glops in addition to extensions such as SSE2. Why not make use of it?
Personally, I reckon that hardware acceleration in EDA will rise to the surface for a while then disappear again as the general-purpose processor and PC-blade makers absorb those technologies - who knows, they might be a lot better at running technical codes than games. EDA companies would be wise to explore the FPGA and GPU options if only because elements of those products are likely to wind up inside the workstation and the blade. But take the speedup claims with a large dose of salt.
Posted by Chris at 10:43 AM | Comments (5) | TrackBack
"We're sorry. This album is unavailable for download in your country"
So there I was typing away as an email from eMusic came in. In a bout of continuous partial attention, I clicked on it and thought: "Ooh, new Black Francis album. Glad I saved some downloads instead of splurging them when the subscription rolled over at the end of last month."
But, thanks to the byzantine nature of distribution deals in the music industry, it is, naturally, not available for download in the UK from eMusic. Then I remembered the Breeders album was meant to be coming out this week and that one, thankfully, is available here. And now downloading.
As is Black Francis' Svn Fingers as I decided to see if it was on iTunes. I could have waited to get the CD but as my mind normally goes blank the moment I walk into a record shop, decided to lay out a fiver on the mini-album there and then. Looks like it's going to a kind of ex-Pixies day.
But this is something that the music industry needs to get a grip on. There are people who do pay for music - when they can find it. It's crazy to have a situation where it's easier to find pirated versions than the paid-for recordings, particularly when it comes to back catalogues of minor artists.
Posted by Chris at 9:24 AM | Comments (0) | TrackBack
April 11, 2008
"If it's wrong, it's not our fault - the client made us do it"
I've seen plenty of releases with disclaimers - mostly about forward-looking statements. Or as the CFO of one big analogue chipmaker put it at a financial conference some years back as he put up the obligatory safe-harbour statement: "This basically says that everything I am about to tell you may be a lie."
This disclaimer from Webit PR, however, is a new one on me:
"Disclaimer:Whilst WebitPR Ltd endeavour to ensure the accuracy of the information contained in this Release, WebitPR Ltd cannot accept any liability for:-
• the inaccuracy or otherwise of any information contained in this Release; or
• any loss liability or expense which may be suffered by any party in consequence of acting or omitting to act as a result of any information contained in or omitted from this Release; or
• any loss or suffering which may be caused by or to any party either as a result of the information contained in this Release or such information contained in this Release being inaccurate or otherwise misleading."
I guess this is one of the consequences of more releases being turned up directly by search engines. But it only serves to confirm what we already know: everything in the release may be a lie.
Posted by Chris at 11:54 AM | Comments (6) | TrackBack
The 200 million amp 'low-power' memory
The way IBM describes its racetrack memory – yet another candidate for "memory of the future" – it's easy to be left with the impression that Big Blue is out on its own with this one. Om Malik* Stacey Higginbotham breathlessly opines: "IBM sure has some seriously crazy semiconductor researchers locked in its basement. These guys question everything when it comes to advancing chip technology."
Maybe IBM does. But it's not alone. What IBM claimed in the press release was that a memory 100 times denser than today's flash devices is on its way:
"The devices would not only store vastly more information in the same space, but also require much less power and generate much less heat, and be practically unbreakable; the result: massive amounts of personal storage that could run on a single battery for weeks at a time and last for decades."
Sounds great. When can I buy one? Not any time soon if you look more closely at what IBM's release is based on. The journal Science has published a paper on the work of Stuart Parkin's group at its Almaden lab in San Jose that describes a tweak to a type of magnetic memory. It's a bit like a solid-state disk state. You store bits magnetically: the state depends on which way the stored field points, either forwards or backwards along a metal wire.
With a disk drive, you read the bits by passing them at high speed under a magnetic head. With domain-wall memory, you push the bits along until they pass by a circuit that can read the state of the field. This is where IBM came up with its moniker of 'racetrack' memory. In reality, it's a bit more like a sushi counter: you have to wait until the bits you want slide into view. In that respect, it works a bit like a NAND flash memory, albeit with much longer chains of bits. And so, the main application, if and when this type of memory appears, is going to be in media players where you simply want to suck bits from it in a more or less continuous stream.
In principle, domain-wall devices sound great, and there are other groups pushing ahead on the research in addition to Parkin's, although he has a key patent on the technology. When Professor Russell Cowburn's group at Imperial College, London reported making logic switches using this approach, they explained why the world might want to switch from silicon semiconductors to this new way of working. Because you make the circuits out of conductors instead of semiconductors, you get a lot more electrons into a small space. That makes it, in principle, easier to scale down in size. On top of that, you don't have to mess with all the complicated steps that silicon technology needs: you simply lay down tracks of a simple magnetic alloy.
Now the bad news. There are still some big problems with this type of memory. The one that is likely to trip it up is power. Parkin's team describes in the Science paper a way of pushing the bits along a short wire – just three bits in a row – using current. The previous devices use magnetic fields. By using pulses of electric current to push the domains along the wire instead of magnetic fields, it should be easier to make the devices - chipmakers have been dealing with that kind of circuit for years; the idea trying to control magnetic fields on a chip is far less well developed. But it's not just IBM doing it.
Professor Teruo Ono's team in Kyoto published in Applied Physics Express in January a description of their work on current-driven domain walls, and they name-check quite a few groups doing work in the same area. What Ono's team points out clearly, and you find people with similar concerns in the Science news item as well, is that using current to push the bits along demands a lot of juice: something like 100 million amps per square centimetre. The wires are tiny, so you don't have to put the entire generating capacity of a small nation into the chip, but this is two orders of magnitude more juice than the magnetic memories that are going into production today. And they consume a lot more power than electronics engineers want, which is why Freescale Semiconductor's magnetic memories are going into satellites and not handsets.
Where the IBM team has gone further is in building a small memory; Ono's team was just demonstrating the ability to use current to push the magnetic domains around. Parkin's memory is even more power hungry: 200 million amps per square centimetre. Which kind of makes a nonsense of IBM PR's claim that this type of memory is going to consume less power than today's devices, or even disk drives. And the bits are big. The wires are only a couple of hundred nanometres across, but the bits have to have 2000nm separating them to stop them running into each other and disappearing – another knotty problem with this kind of memory. That makes the size of each bit considerably larger than today's flash memories.
There is a way round the density problem: you can stack layers of wires on top of each other – the wires are are less than 10nm thick. That is the design that Cowburn favours. Not only does it improve density, if you read the layers in parallel, you could potentially get very high datarates out of these devices even if you read the bits out quite slowly – a technique that should reduce power consumption. IBM seems to favour a vertical configuration so the read elements are all on the bottom of the chip. Strings like this have yet to be made but Toshiba has pursued a similar idea with a flash memory it unveiled at last year's VLSI Symposium.
Ono's team has worked on the energy problem, identifying an alternative to the regular nickel-iron wires normally used in this kind of research that can work at room temperature – another Japanese team slashed the energy needed by three orders of magnitude a few years back, but the device needs to be cooled artificially. The cobalt-platinum alloy Ono's group used has a built-in magnetic field which should improve on the way these nanowires behave. However, according the paper, the current needed is still in that 100 million range.
* Note to self: check byline before citing article. This might be time to edit the stylesheets in NetNewsWire.
Posted by Chris at 10:17 AM | Comments (0) | TrackBack
April 10, 2008
Size is everything (sometimes)
When former STMicroelectronics R&D director Jo Borel tried to convince the French government that it should try to convince Europe's three largest chipmakers to merge, he almost certainly didn't have in mind what ST and NXP Semiconductors plan to do. They are not merging the entire companies but taking the wireless business units and glueing them together.
The argument used for the merger is not all that dissimilar to Borel's: it's all about scale. Borel wanted Infineon, NXP and ST to team up to be big enough to build and operate a leading-edge fab - it is something that is only worth doing if you are selling billions of dollars' worth of chips every year out of that facility. Not able to do that on their own, the three companies expect to buy wafers made using the latest processors from foundries such as TSMC.
The availability of foundry-made silicon is one reason why Infineon chief Wolfgang Ziebart has said that there is not all that much point in trying to be big for the sake of being able to keep building fabs. His view is that companies will specialise and do whatever they can to be in the top three of their chosen market. Infineon has been bulking up in wireless recently, thanks to its purchase of a business unit that was only briefly part of LSI when that company bought Agere Systems.
The move by NXP and ST is on a larger scale, creating an as-yet unnamed joint venture that is comfortably in the top-three wireless silicon makers and around twice as big as the next largest supplier. According to iSuppli, that will be Infineon once the deal is done. The German company is at the head of a line of $500m to $1bn suppliers. The ranking switches a little if you look at it from the perspective of baseband processors - the single most important segment in cellular wireless silicon. ST lies at number three, NXP at five.
According to Francis Sideco, senior analyst for wireless communications at iSuppli, Mediatek is currently number three behind Qualcomm and TI. ST is at four and NXP at sixth: separated by Freescale Semiconductor.
Sideco agrees with the bosses of NXP and ST that scale matters in this business. For Sideco, the turnover for a long-term survivor in the wireless space is in the $3bn to $4bn range. JV had sales of a little under $3bn in 2007, according to ST president Carlo Bozotti. TI and Qualcomm are turning over more than $5bn, according to figures from iSuppli quoted by ST in its analyst call. The merger will put some distance between JV and Infineon, according to iSuppli's number.
Just glueing business units together is not necessarily going to keep JV at number three -all too often these deals are less than the parts, let alone the sum of the parts. But the overlap between NXP and ST's wireless units is surprisingly low. NXP is good at standard products and basebands; ST has specialised in doing custom jobs for major handset makers.
Assuming the merger is successful, the deal could be the catalyst for a wave of similar deals in this business. Or it could presage a wave of price cutting as the mid-sized players try to make sure they can maintain a toehold in the market while the big fish attempt to tie up all-you-can-eat deals with the handset makers?
One thing that counts against there being a series of consolidation deals is the credit crunch. Private equity's foray into the chip business did not last as long as the private equity firms expected. And the few that are in the hands of the financiers are too heavily loaded with debt for comfort. That helps explain why the deal between ST and NXP looks a little odd.
If you just took the two wireless groups and put them together, you would expect ST to have the larger share, but not by much. Instead, ST is going to own 80 per cent of the JV and will pay $1.5bn for the privilege, all out of the company's own cash reserves. The company does not seem keen to repeat the Numonyx experience of trying to raise loans to fund the deal.
It begs the question of why NXP did not just sell the whole wireless business to to ST: it would raise more cash, and cut NXP's heavy debt burden. But, NXP chief Frans van Houten claims the company is in the deal for the long run, even though it has a piece of paper that explains how NXP might cash-out of the JV. NXP is presumably hoping that the value of its investment will rise as the supplier count in the wireless-silicon business falls. Margins in the wireless segment are lower than the rest of NXP's operations, so the company may feel that it can make a better return by hanging onto a small part of that business rather than selling it all now.
It may be that, without the pressure of the debt, NXP would have soldiered on alone -and try to stay in the top five by continuing with the kinds of small deals the company did with Silicon Labs and Glonav.
Most of the other companies in the top ten are not in that position. The wildcard is Freescale, which is saddled with the same kind of debt-heavy position as NXP. But who would try to buy Freescale's handset silicon group? The company is not in the position to do the reverse. It might take the kind of chaos that has forced the memory makers to consider shotgun weddings to sort out what happens in the remainder of the top ten.
Posted by Chris at 11:28 PM | Comments (0) | TrackBack
