EE Times
San Jose, Calif. ? A new generation of optical and copper components for 10-Gbit Ethernet is about to hit the market, promising vastly lower costs that could rev up demand for fast networks inside the data center. The parts emerge just as networking companies are laying the groundwork for a 100-Gbit Ethernet standard.
"It's show time. We have to stop showing PowerPoint slides and start delivering products to customers," said Mike McConnell, co-founder and director of strategic marketing at KeyEye Communications Inc. (Sacramento, Calif.). He tipped plans to sample early in 2006 copper transceivers for taking 10-Gbit Ethernet across 30 meters of Category 6 copper cabling.
Startup Solarflare Communications Inc. (Irvine, Calif.) will not be far behind, with transceivers expected before June that will carry 10-Gbit Ethernet up to 100 meters over CAT6 wiring.
"When users want structured cabling, it has to be 100 meters," said George Zimmerman, chief technology officer and founder of Solarflare.
Startup Teranetics (Santa Clara, Calif.) is also expected to ship copper transceivers based around the emerging 10GBase-T standard in the first half of next year. Established players Broadcom and Marvell are expected to follow with parts in the second half of 2006.
Another startup, Luxtera Inc. (Carlsbad, Calif.), hopes to trump the copper parts with a novel design using silicon optics to route 10-Gbit Ethernet over up to 2,000 meters of single-mode fiber at the same costs ? and significantly lower power and latency ? than the copper components. "We think 10 Gbit will be the tipping point between copper and optical media," said vice president of technology Cary Gunn.
Even the chairman of the IEEE 802.3an group that is setting the 10GBase-T standard agreed that the outlook for copper links faster than 10 Gbits is dim. The group expects to finish the 10GBase-T copper standard by July.
Meanwhile, OEMs including Cisco Systems and Force 10 have come to an informal consensus about starting work on a 100-Gbit Ethernet standard in 2006, leapfrogging what some had expected would be a move to 40 Gbits. "We just like to think in Base 10," joked Brad Booth, a strategic-marketing manager for Intel Corp. who chairs the 802.3an group.
"Most people believe we need to kick off [a 100-Gbit standard effort] next year. We need to get the optics people on board," Booth added.
"It's not too early to start talking about possible standards, but it's too early to pick specific technologies," said Bill Woodruff, vice president of marketing for new 10-Gbit startup Aquantia Corp. Many physical-layer standards will continue to be based on fiber by necessity, he said, but proponents of possible 100-Gbit copper solutions should not be excluded from discussions by preconceived notions of what is possible.
In the meantime, 10-Gbit copper approaches will battle in the market over trade-offs in distance, power consumption and latency.
KeyEye's single-chip KX1001 will consume less than 3.7 watts and provide a latency of less than 150 nanoseconds. However, it will reach only 30 meters over CAT6 wiring, said McConnell. The part, available in a 14-mm2 package requiring a heat sink and a 19-mm2 package without a heat sink, will sample in November and be in production early next year.
The company will offer a similar part for 15-meter stretches of Infiniband CX4 cabling. "We're down to a single piece of silicon for a 10-Gbit transceiver, and it's hard to compete with that," said McConnell, adding that 30 meters will cover 75 percent of the data center's needs.
The 10GBase-T standard calls for a 55-meter minimum distance. McConnell said some vendors are discussing the possibility of a standard for the shorter distance KeyEye targets.
McConnell estimated that OEMs may buy as many as 63 million 10-Gbit links over the next five years. Most of those interfaces will be on servers, switches and storage systems used inside corporate data centers.
For its part, Solarflare will trade off power consumption and latency to deliver at 10 Gbits the full 100-meter reach that is typically expected of Ethernet signaling. Zimmerman estimated initial adoption of 10-W parts made in 90-nanometer technology, followed by a broader uptake for parts that could hit 5 W in 65-nm technology.
Many 10G transceiver developers agree that reductions in power consumption made possible by a shrink in process feature size are absolutely critical for making 10-Gbit networks over Category 6 wiring more prevalent in the enterprise. Woodruff of Aquantia said that a single-chip CMOS implementation of a 10G transceiver is an important milestone, but that reasonable power dissipation is critical.
"The 1000Base-T realm has seen a 10:1 power reduction in the last decade. Why can't we expect the same in 10GBaseT?" Woodruff asked.
"I have to bet my business on this. We've been through detailed block-by-block estimates of power consumption validated by comparisons with actual working 65-nm chips from foundries," said Zimmerman. He said that the 2.5-microsecond latencies will not adversely impact most products that aim at application-level latencies of tens of microseconds. However, one board maker has expressed concerns that such latencies would be too high for his company's products, which are used in server clusters, where latency is a top priority.
Luxtera claims its devices will far surpass 10-Gbit copper transceivers in power, latency and reach while hitting the same price point of about $50 when they are selling in volume.
Luxtera's 10-Gbit parts, due to sample before June, consume just 1.7 W and have a latency of less than 10 ns, said Gunn.
The company has a proprietary subwavelength holographic lens that couples an MTRJ optical-fiber package directly to its die without the need for expensive free-space optics. The chip itself, made in a standard 130-nm CMOS silicon-on-insulator process at Freescale Semiconductor Inc., routes and modulates light through silicon paths on the die.
Luxtera has proven out the designs of the germanium-based optical photodiodes and modulators used on the chip. It is now finishing the digital design of the device, which could fit into a few square millimeters of silicon, Gunn said. The company is also in an advanced design stage for a 100-Gbit chip commissioned by the Defense Advanced Research Projects Agency. That part uses an array of 10G modulators and photodiodes in an architecture that can scale to terabit data rates, he added.
"We have the technical capability to handle the next four generations of Ethernet," said Gunn.
Intel's Booth said one attribute the Luxtera parts lack is the benefit of an industry standard. "A single-source supplier is not popular in the Ethernet world," Booth said.
"There are as many as 33 different standards in 10G Ethernet today ? it makes no sense," countered Gunn. "In the end, it will be the costs that drive the market."
San Jose, Calif. ? A new generation of optical and copper components for 10-Gbit Ethernet is about to hit the market, promising vastly lower costs that could rev up demand for fast networks inside the data center. The parts emerge just as networking companies are laying the groundwork for a 100-Gbit Ethernet standard.
"It's show time. We have to stop showing PowerPoint slides and start delivering products to customers," said Mike McConnell, co-founder and director of strategic marketing at KeyEye Communications Inc. (Sacramento, Calif.). He tipped plans to sample early in 2006 copper transceivers for taking 10-Gbit Ethernet across 30 meters of Category 6 copper cabling.
Startup Solarflare Communications Inc. (Irvine, Calif.) will not be far behind, with transceivers expected before June that will carry 10-Gbit Ethernet up to 100 meters over CAT6 wiring.
"When users want structured cabling, it has to be 100 meters," said George Zimmerman, chief technology officer and founder of Solarflare.
Startup Teranetics (Santa Clara, Calif.) is also expected to ship copper transceivers based around the emerging 10GBase-T standard in the first half of next year. Established players Broadcom and Marvell are expected to follow with parts in the second half of 2006.
Another startup, Luxtera Inc. (Carlsbad, Calif.), hopes to trump the copper parts with a novel design using silicon optics to route 10-Gbit Ethernet over up to 2,000 meters of single-mode fiber at the same costs ? and significantly lower power and latency ? than the copper components. "We think 10 Gbit will be the tipping point between copper and optical media," said vice president of technology Cary Gunn.
Even the chairman of the IEEE 802.3an group that is setting the 10GBase-T standard agreed that the outlook for copper links faster than 10 Gbits is dim. The group expects to finish the 10GBase-T copper standard by July.
Meanwhile, OEMs including Cisco Systems and Force 10 have come to an informal consensus about starting work on a 100-Gbit Ethernet standard in 2006, leapfrogging what some had expected would be a move to 40 Gbits. "We just like to think in Base 10," joked Brad Booth, a strategic-marketing manager for Intel Corp. who chairs the 802.3an group.
"Most people believe we need to kick off [a 100-Gbit standard effort] next year. We need to get the optics people on board," Booth added.
"It's not too early to start talking about possible standards, but it's too early to pick specific technologies," said Bill Woodruff, vice president of marketing for new 10-Gbit startup Aquantia Corp. Many physical-layer standards will continue to be based on fiber by necessity, he said, but proponents of possible 100-Gbit copper solutions should not be excluded from discussions by preconceived notions of what is possible.
In the meantime, 10-Gbit copper approaches will battle in the market over trade-offs in distance, power consumption and latency.
KeyEye's single-chip KX1001 will consume less than 3.7 watts and provide a latency of less than 150 nanoseconds. However, it will reach only 30 meters over CAT6 wiring, said McConnell. The part, available in a 14-mm2 package requiring a heat sink and a 19-mm2 package without a heat sink, will sample in November and be in production early next year.
The company will offer a similar part for 15-meter stretches of Infiniband CX4 cabling. "We're down to a single piece of silicon for a 10-Gbit transceiver, and it's hard to compete with that," said McConnell, adding that 30 meters will cover 75 percent of the data center's needs.
The 10GBase-T standard calls for a 55-meter minimum distance. McConnell said some vendors are discussing the possibility of a standard for the shorter distance KeyEye targets.
McConnell estimated that OEMs may buy as many as 63 million 10-Gbit links over the next five years. Most of those interfaces will be on servers, switches and storage systems used inside corporate data centers.
For its part, Solarflare will trade off power consumption and latency to deliver at 10 Gbits the full 100-meter reach that is typically expected of Ethernet signaling. Zimmerman estimated initial adoption of 10-W parts made in 90-nanometer technology, followed by a broader uptake for parts that could hit 5 W in 65-nm technology.
Many 10G transceiver developers agree that reductions in power consumption made possible by a shrink in process feature size are absolutely critical for making 10-Gbit networks over Category 6 wiring more prevalent in the enterprise. Woodruff of Aquantia said that a single-chip CMOS implementation of a 10G transceiver is an important milestone, but that reasonable power dissipation is critical.
"The 1000Base-T realm has seen a 10:1 power reduction in the last decade. Why can't we expect the same in 10GBaseT?" Woodruff asked.
"I have to bet my business on this. We've been through detailed block-by-block estimates of power consumption validated by comparisons with actual working 65-nm chips from foundries," said Zimmerman. He said that the 2.5-microsecond latencies will not adversely impact most products that aim at application-level latencies of tens of microseconds. However, one board maker has expressed concerns that such latencies would be too high for his company's products, which are used in server clusters, where latency is a top priority.
Luxtera claims its devices will far surpass 10-Gbit copper transceivers in power, latency and reach while hitting the same price point of about $50 when they are selling in volume.
Luxtera's 10-Gbit parts, due to sample before June, consume just 1.7 W and have a latency of less than 10 ns, said Gunn.
The company has a proprietary subwavelength holographic lens that couples an MTRJ optical-fiber package directly to its die without the need for expensive free-space optics. The chip itself, made in a standard 130-nm CMOS silicon-on-insulator process at Freescale Semiconductor Inc., routes and modulates light through silicon paths on the die.
Luxtera has proven out the designs of the germanium-based optical photodiodes and modulators used on the chip. It is now finishing the digital design of the device, which could fit into a few square millimeters of silicon, Gunn said. The company is also in an advanced design stage for a 100-Gbit chip commissioned by the Defense Advanced Research Projects Agency. That part uses an array of 10G modulators and photodiodes in an architecture that can scale to terabit data rates, he added.
"We have the technical capability to handle the next four generations of Ethernet," said Gunn.
Intel's Booth said one attribute the Luxtera parts lack is the benefit of an industry standard. "A single-source supplier is not popular in the Ethernet world," Booth said.
"There are as many as 33 different standards in 10G Ethernet today ? it makes no sense," countered Gunn. "In the end, it will be the costs that drive the market."
