Lexra clones Mips core, tailors features to boost network processing performance

Lexra clones Mips core, tailors features to boost network processing performance
By J. Robert Lineback, Semiconductor Business News
June 13, 2000 (12:41 p.m. EST)
URL: http://www.eetimes.com/story/OEG20000613S0006

SAN JOSE -- Lexra Inc. here this week will announce it has created a "cleanroom" clone of Mips Technologies Inc.'s RISC core to offer embedded processor capability for high-speed networking applications. The LX8000 network engine is fully programmable and uses the Mips instruction set, but the IC design core has been tailored for network processor applications, said Charlie Cheng, president and CEO of Lexra.

During this week's Embedded Processor Forum, the San Jose company plans to describe details of its new NetVortex network processor architecture, which consists of two components: the LX8000 processor core and the VortexBus. The Mips-compatible processor core has been optimized to handle networking operations--such as packet processing and fast switching between multiple packets in communications.

Cheng said the LX8000 is 50 times faster than traditional processor in switching between packets, and about five times faster overall in the networ king application.

Up to 16 LX8000 cores can be combined with the VortexBus. Total of four VortexBus structures can be used to create a "two-dimensional" system design (see related story of the architecture).

The NetVortex architecture is aimed at a range of network processing systems applications--from single-processor home office applications or cable modems to high-end OC192 routers. The first offering can support systems running at 10 times OC192 speeds, according to Cheng.

At first, Lexra wasn't interested in this exploding processor application, recalled Cheng. "Last year, a top network communications systems company came to us and said they needed a processor for this specific function. They wanted us to work it out and license it to them," he explained. "Originally, we were somewhat resistant because we were afraid of the low volume [potential of serving one customer with a new processor core]."

But after checking with ven ture capitalists and a crop of new startups in the networking arena, Cheng and his management team got excited about the potential for a new programmable CPU core and architecture tailored for communications. Lexra is now aiming to give systems makers an alternative to ASICs and traditional embedded processors from larger chip competitors are pursuing networking applications.

"The chip companies [like Motorola, IBM, and Intel] want you to think that eventually communications and networking will be like the PC market, where semiconductor suppliers dominate the innovations and 'box' companies just to the packaging," Cheng stated. Because of the intense time-to-market pressures in networking, chip companies will play a key role in turning around new designs and capabilities, he admits, but he sees the need for more programmability and flexibility in creating new IC solutions.

"As the world becomes more complex and speeds get higher, network processing is probably one of the most difficult problems in co mmunications gear right now," he said. "The original concept of doing 1 million-gate ASICs doesn't work any more because of the time-to-market pressure and changes in communications protocols."

So, Lexra decided to leverage off the existing Mips development tools and support, but create its own core from the ground up. "The Mips instruction set is just the starting point. We have added a lot of instructions to achieve the five-times the performance [of existing solutions]," said the Lexra chief executive. About 95% of the tools support Mips chip development and applications can be used with the Lexra core, he added.

In addition to higher performance, the LX8000 core is a quarter the size of computer-targeted CPU cores. It does not include floating-point unit and other functions traditionally found in workstation central processor units, Cheng said. The smaller die area and processing speeds in networking applications results in 20 times the price/performance of other solutions, he said.

Lexra ex pects to ship the design elements of the NetVortex architecture to customers in the third quarter. Production of ICs is expected to begin by the end of the year, with systems based on the processor core appearing at the end of 2001.

Another unusual aspect of Lexra's embedded processor offering--the company has set a public price for the design technology. (Most core suppliers won't release pricing information). A regiater-transfer level (RTL) version of NetVortex is priced at $695,000 for one project.

Lexra's original system customer requesting development of the processors core expects to move its chip into production at a large North American foundry, using 0.13-micron process technology. Lexra said it is working with pure-play foundry suppliers for production of ICs for other customers with 0.15-micron and below processing technologies.

The San Jose company said it will license NetVortex in two forms. A portable RTL version for any silicon manufacturer (which will run at 250 MHz worst case) in 0.15-micron technology. In this version, the die area for each LX8000 is 2 mm², and power consumption is 100 mW, said the company. A second version is a complete hard macro optimized for a specific foundry process. This will run at 427 MHz on a typical 0.15-micron process, said Lexra.

Test chips are now under development at several large pure-play foundries, Cheng said.