"5 years out" - Challenges in optical backplanes, Part 1
For about two decades, the potential of optical interconnects across the backplane has been apparent: very high channel capacities, low-power Gigabit line rates, and potential weight savings are just a few of the potential advantages over copper. However, in the course of researching optical technologies, I came across a recurring theme:
- 2003: A conference paper entitled “VCSEL to waveguide coupling for optical backplanes” stated that “Hybrid OECB (Opto-Electrical Circuit Boards) are expected to make a significant impact in the telecomm switches arena within the next five years, creating optical backplanes with high-speed point-to-point optical interconnects.”
- 2011: At Interop Las Vegas, Charles Clarke, Distinguished Technologist at HP, said that optical backplane products were three to five years away.
- 2013: An article by Bob Hult, Director of Product Technology at market research firm Bishop & Associates, Inc., suggested, “it is possible that we may start seeing optical backplanes in high-end systems within the next five years.”
Needless to say, it has been (and could be) a long five years.
Market demand and the endurance of copper
Though technical issues like demanding alignment and temperature sensitivity still challenge optical backplanes, functioning systems based on optics already exist. Rather, the setbacks of optical technology are best considered in relation to copper-based interconnects and the interconnect market as a whole. According to Chuck Byers, Technical Leader and Platform Architect, Enterprise Networking Group, Cisco Systems, Inc. (), these challenges can be defined in terms of demand, cost, and legacy equipment.
“It’s uncertain there is market demand for optical interconnects in a reasonably sized system box,” Byers says. “In the 1990s, silicon connectivity rates per pin were purported to be at the point of plateau at some rate under 1 Gbps. This caused a flurry of activity for optical backplanes as a way to enable higher speed systems. Before long, however, the semiconductor suppliers came out with SERDES designs well over 1 Gbps and integrated a high quantity of channels onto various types of chips. This resulted in a low demand for optics.”
Current SERDES technology is capable of 25 Gbps per differential pair, which satisfies the needs of most board-level hardware available today. However, copper-based connector technologies have also evolved, and major suppliers offer high-density connector systems that can achieve 25 Gb line rates with pin densities of roughly 80 differential pairs per inch of backplane height. Assuming a theoretical interconnect bandwidth of 25 Gb and 6” of board edge connector, these connectors can yield as much as 6 Tbps of transmission capacity in each direction across the backplane per board, Byers explains (Equation 1). “This exceeds the bandwidth demand of any processor, storage, I/O, or optical line board. Unless some other factors come into play, or we abruptly reach a plateau of SERDES rate or connector density, metallic interconnects will suffice for at least several more cycles of Moore’s Law,” he adds.
(25 Gb/pair x 40 Transmit and Receive pairs/in.) x 6” = 6 Tbps
Despite the fact that optical interconnects are capable of transmission capacities far greater than 6 Tbps (1 Petabit per second (Pbps) fiber transmissions have been reported), they are still struggling to find a price/performance equilibrium. Although copper connectors that can achieve 25 Gb bandwidths per differential pair are considered “bleeding edge,” they still maintain significant cost benefits over comparable optical systems.
“The cost of an optical transmitter and receiver is currently at least an order of magnitude higher per unit of bandwidth carried than SERDES-based metallic interconnects,” Byers says. “The costs of connectors, waveguides, test equipment, repair procedures, and so on are also correspondingly higher for optics. There are some very promising technologies in low-cost parallel short reach optical interfaces, integrated waveguides, and free-space transmission that could eventually bring optics closer to parity, but they are slower to develop.”
An extension of the cost paradigm exists in the established legacy infrastructure. The vast majority of systems deployed today are not equipped for optics, so their implementation will result in a complete disruption of product lifecycles and company roadmaps in most cases.
“Many of today’s systems build upon legacy platforms. With few exceptions, provisions do not exist for optical backplanes in these platforms, so organizations may not be able to interoperate new optical interconnect enabled equipment with existing boards or chassis designs,” Byers says. “In general, optical interconnects will be brought in more of a revolutionary than evolutionary change to a platform, considering many other factors such as power, cooling, connector, or board area limitations will all conspire to demand a redesign that disrupts backward compatibility.”
- World Record One Petabit per Second Fiber Transmission over 50-km: Equivalent to Sending 5,000 HDTV Videos per Second over a Single Fiber.