A new generation of networking chip startups is attempting to find a path into business networks by making 10G-bps networking gear more palatable to their IT budgets.
The companies, which include ClariPhy Communications and KeyEye Communications, are crafting optical physical layer interface chips and copper media transceivers, respectively, that they say network equipment makers can use to boost the bandwidth of their 10G-bit gear while also lowering prices.
Although they come at the market from different angles, the two companies aim to convince network gear makers that using their chips could cut the cost of equipment upgrades by hundreds per port, creating an opportunity to speed up the adoption of 10G-bit connections—still fairly rare—in network backbones and data centers. Drop the cost of 10G-bit gear enough, they argue, and it will sway businesses to upgrade.
ClariPhy, a 25-employee startup that emerged in February, plans to offer a chip that can save companies money by bumping the bandwidth of existing network cables. Its digital signal processor-based PHY (physical layer) for 10 Gigabit Ethernet, most likely to be built into line cards used by network switches, will make it possible for companies to continue using their existing 2.5G-bit multimode optical network cables, but it promises to bump the bandwidth of those cables to 10G bps.
The chip, which will conform to the forthcoming IEEE 10GBase-LRM standard for 10 Gigabit Ethernet over optical cables, is able to offer the increase because its built-in digital-signal processor can compensate for distortions created by sending the high-speed signals, said Paul Voois, ClariPhys CEO, in Irvine, Calif.
“Whats going to save them cost going to 10 Gigabit is not having to replace their cable,” Voois said.
Following the 10GBase-LRM standard means companies can cut costs by employing fewer lasers for sending networking signals via optical cables than current chips based on than todays 10GBase-LX4 standard. But ClariPhys secret sauce is in the way it employs its electronic dispersion compensation techniques, making use of the DSP to compensate for signal distortion, in addition to using CMOS, a standard method for manufacturing processors, memory and other chips, Voois said.
ClariPhys first PHY chip is still in development, allowing it to dovetail with the 10GBase-LRM standard, expected to be finalized later this year. The company expects to sample the chip to customers later this year. Thus it wont be available in equipment until 2007, Voois said.
ClariPhy ultimately aims to use its chips to help drive the per-port cost of 10 Gigabit Ethernet to less than $100 for equipment makers. Right now that cost is about $500, Voois said. Lowering manufacturers costs would allow them to, in turn, offer lower prices to business customers.
“For 10 Gigabit to really happen, you need to get to 10 times the performance at three times the cost,” Voois said. Historically, “when IT managers are faced with that equation, theyll pay” to upgrade, he said.
It will take time for prices to come down, however, and there are still strides to be made in port density. But, as the company sees it, theres plenty of opportunity in network backbones where there are about 100 million existing ports that use multimode fiber, which could be upgraded to 10G-bit speeds. Its chips will also work to enable data center computer interconnects, Voois said.
Next Page: Combining 10G-bit speeds and copper cables.
2
Meanwhile, KeyEye says its nearing full production of a family of transceiver chips designed to help combine 10G-bit speeds and inexpensive copper cables, a move that it says many companies would like to see happen in the data center.
The 40-employee network chip maker believes businesses want to combine 10G-bit bandwidth and low-cost Category 6 copper cables for connecting servers and storage, whereas today they use fiber or special copper InfiniBand cables, said Fred Lancia, vice president of sales and product marketing, in Santa Clara, Calif.
“The reality is theres growing demand for 10 Gigabit for a number of different reasons. Throughput is one,” he said. However, “Its still very expensive, so thats delaying implementation and demand for it.”
The company is offering three chips, which prepare signals for transmission over a network cable. Much of KeyEyes expertise, therefore, lies in encoding and decoding signals as well as in eco cancellation. Its KX1000 will enable using 10 Gigabit Ethernet to make short connections, such as between switches. Its KX1001 chip aims to enable the use of 10 Gigabit Ethernet over Cat. 6 cables to replace fiber cables. Meanwhile, its KX1003 chip aims to replace InfiniBand cables with Cat. 6 copper wire.
The KX1001 and KX1003 chips could be used to position Cat. 6 cables as a medium for Ethernet or InfiniBand for jobs such as high-performance computing clusters. Using Cat. 6, which helps cut costs, also makes Ethernet a more viable competitor to InfiniBand, Lancia argued.
KeyEyes chip family uses a specification called PAM-4, however, whose method of encoding differs from that of the forthcoming IEEE 10GBase-T Ethernet over twisted pair copper standard. However, the company believes that between pent-up demand and the time it will take gear that uses the standard, expected to be finalized this year, to hit the market, many companies will adopt equipment with its technology. Next year, the company plans to follow up with a line of 10GBase-T compliant chips, Lancia said.
KeyEyes pitch for moving to 10G-bit gear revolves around using copper versus optical cables. The company aims to sell its chips to companies that make network adapter cards, line cards or the modules to go on those cards. Modules using KeyEye chips would cost about a third of the price of 10G-bit optical modules for fiber cables used in adapter cards. Right now they cost manufacturers around $500, Lancia said.
“The demand is there. The thing thats preventing 10 Gigabit is cost. Everyones clamoring for lower cost solutions. So were happy to enable them,” he said.
Still, while both companies have the advantage of being early to an emerging space in the networking market, they face numerous challenges, not the least of which are the potential for larger competitors, such as Broadcom, to step in and the fact that equipment makers might not be as intent on lowering the price of 10G-bit gear as the two chip makers are, said Jag Bolaria, senior analyst at the Linley Group, in Mountain View, Calif.
“Its a cost story. Basically theyre all trying to reduce the cost of 10 Gigabit,” he said. “The argument theyre both working at is driving down the cost … will increase the available market for that.”
ClariPhys pitch for reusing cables would hold the most appeal to businesses whose cable installations are the most complicated—going through walls, for example. Meanwhile, KeyEye must convince its investors and its customers that its technology can move from pre-standard 10 G-bit over copper to the 10Gbase-T specification, Bolaria said.
But it might not be lower prices on gear so much as new applications, such as high-performance computing, that could crack open the 10G-bit space. A company with one foot in servers and another in networking—such as Hewlett-Packard, which already has 10 Gigabit Ethernet gear available—could get the ball rolling by introducing lower-priced gear, Bolaria said.
“As with any product, a little bit of luck has to come into it,” he said.