Founded in February 2000, the company is coming out of stealth mode as it starts a third round of venture funding in which it hopes to raise $20 million. President and founder Anil Gupta said he believes Foothill Ranch, Calif.-based Aristos could be profitable in the first half of 2004, based solely on two existing design wins with unidentified OEMs.

Aristos is currently sampling its FibreSlice processor and will ship production quantities in the first quarter of 2003 at prices that are still being negotiated with key OEMs. The chip offers 10 times the performance of general purpose processors used in Fibre Channel RAID controllers, but OEMs must be willing to move their software to the FibreSlice's applications programming interface. The part also offers hardware acceleration for so-called virtualization applications such as disk mirroring that are on the horizon for many storage systems makers.

"Our plans for profitability are all based around the market for Fibre Channel RAID controllers. We see virtualization as an upside possibility because it is still a very new market," Gupta said.

FibreSlice includes three 2-Gbit/second Fibre Channel physical and link layer interfaces to accept and transmit traffic running on one of today's most popular storage network interfaces. At the chip's heart are 10 proprietary cores geared to handle a range of storage applications such as RAID and virtualization applications.

Seven of the 10 cores are compute engines based on a generic "step and request core with programmable RAM and some specialized gates," said Marc Acosta, director of technical marketing for Aristos. Each of the seven is slightly different, geared for the various storage applications supported by the chip. The other three cores handle table and list management functions.

"This device is capable of 50,000 I/O transactions a second and that's faster than anything on the market," said Adam Zagorski, director of product marketing for Aristos.

OEMs can plumb the depths of the hardware architecture to fully exploit possible performance gains, or they can interact with a high level API based on calls to C language libraries. "That's not trivial but the APIs make the job [of moving off general purpose CPUs] a lot easier. We've been able to do that with two OEMs, and in a third case we were able to port their software in eight weeks," Gupta said.

The company was started in 2002 when it became clear that Western Digital's focus on desktop hard-disk drives would obsolete its VLSI engineering team. The company agreed to spin-out the 26-person team. Gupta presented management with a business plan and incubated the effort for four months while it sought external funding.

Infineon Technologies AG provided $18 million in a first round of funding, becoming Aristos' largest shareholder. The German chip maker saw Aristos as part of a broader plan to diversify away from memory, but never followed through on plans to build a portfolio of storage networking products, Aristos said.

"With the downturn, they did not fulfill their plans for a full suite of adapters and chips," Zagorski said.

Three venture capital groups, including Quantum Ventures, invested a total of $15.5 million in a second round of funding. Each investor took seats on the company's board, along with Western Digital.

Singular startup

Aristos stands apart from a host of other storage network startups by pursuing a demonstrated market for Fibre Channel RAID controllers. Others, including Trebia Networks Inc. and Silverback Systems Inc., are aiming for next-generation storage systems and their expected migration to Gigabit Ethernet and iSCSI.

But the iSCSI transition is going slower than once expected for two reasons. Gigabit Ethernet requires still expensive hardware acceleration for TCP offload and remote direct memory access features to hit its full performance capabilities. And the downturn has generally slowed business spending on storage networks, and has thus slowed the expected transition to Ethernet.

"A couple of our customers that once said Gigabit Ethernet support was a must have backed off and now say they will wait for 10-Gigabit Ethernet," said Acosta. "We don't see iSCSI as being a high volume market in the short term. But at 10G, it will be a whole new ball game."

The emerging field for storage-specific silicon is heating up just as the systems market has cooled with the downturn. Nevertheless, at least two other storage processor startups are expected to come out of the woodwork at the Network Processor Conference, to be held Oct. 22-24. Astute Networks and Seaway Networks Inc. are both expected to detail their first products at the conference.