All Change at Clearspeed
Another runner in the race to develop chips for next-generation telecom equipment emerged yesterday: ClearSpeed Technology Ltd., located in Bristol, U.K., is offering development kits for making high-performance network processors (see ClearSpeed Claims 40-Gig Processor).
Although ClearSpeed is a new name, the company itself has been in existence since 1997, operating under the name of PixelFusion. It started out developing chips for high-end graphics equipment, and when that didn’t work out it switched to its current strategy about 18 months ago.
So, is ClearSpeed going to have better luck this time around?
It might. For a kickoff, ClearSpeed has been developing its technology over a long period of time -- initially in a company called Division Group and then in PixelFusion, the result of a management buyout. “About 15 years of development has gone into this architecture,” notes Simon Stanley, ClearSpeed’s director of networking products.
This long experience is relevant to making network processors, according to Ajay Misra, ClearSpeed’s director of strategy and market development. That’s because graphics and network processor chips both have to process huge amounts of data on the fly, which plays to ClearSpeed’s core strength, in high-end parallel processing.
ClearSpeed has run simulations that indicate its technology will support sustained data rates of 40 Gbit/s and higher in network processors. “We’ve already been able to run it up to 50 Gbit/s,” says Misra, noting that the key concept is “sustained." Other developers can only support bursts of 40 Gbit/s, not continuous flows of traffic at this speed, he contends.
Still, if ClearSpeed’s technology is so amazing, why didn’t it succeed in the graphics market? Misra says it was a question of timing. The graphics market underwent a radical change before PixelFusion could complete its development work, leaving the company high and dry.
That's not going to happen again, says Misra. The company intends to be first to market with development kits to make 40-Gbit/s network processors. It’s offering three kits -- for developing hardware, software, and applications.
ClearSpeed, however, faces some serious competition. Plenty of other companies are also hoping to be first out of the gate with 40-gig technologies (see 40-Gig Forecast) -- and some of them also claim to have innovative network processor architectures capable of very high sustained throughputs.
Another concern is whether ClearSpeed has the necessary depth of networking knowledge. Its expertise in producing graphics chips might help it make ultra-fast network processors, but the processors need to be packaged with a software development environment that addresses the needs of telecom equipment vendors, say competitors. Many network processors have turned out to be total pigs to program, precisely because they were designed by ASIC (application-specific integrated circuit) engineers with little or no knowledge of the applications they would be called on to support.
ClearSpeed says it’s put a lot of effort, since it decided to switch strategy, into getting up to speed in telecom technology. The first step was to recruit Simon Stanley, who previously worked at Fujitsu Microelectronics and National Semiconductor Corp. (NYSE: NSM) on networking developments. Some of ClearSpeed’s staff already had telecom backgrounds, says Stanley. Others underwent retraining, and more telecom specialists have been recruited. ClearSpeed has also been talking intensively to potential customers about their requirements and has become an active member of the Network Processing Forum (NPF), an industry association developing standards. “It’s been a huge change that’s happened very fast,” says Stanley.
ClearSpeed appears to have plenty of cash to invest in its new vocation. It’s raised a total of $50 million. Some of this has come from institutions like Invesco, Deutsche Asset Management, and 3i Group PLC; but a large number of wealthy individuals have also put money in the company via private placements. Clearspeed has a total of 650 shareholders and had to get a majority of them to approve its switch in strategy before moving ahead. — Peter Heywood, Founding Editor, Light Reading