Optical backplanes - Route 66
This detour looks at what's being done with optical backplane systems and when we can expect them to arrive.
Backplanes can be a data path bottleneck, and designers are increasingly concerned about data rates. Recent product introductions with 40 G Ethernet have turned up the heat even more. Talk about optical backplanes as a solution has been bantered about for decades. Using optics in backplanes has been seen as a way to overcome the limitations of copper-based implementations for EMI and bandwidth. Much research has been conducted over the years, but why has optical not gained any traction in commercial backplanes? How close are we to seeing widespread deployment of optical backplanes?
While the VME product manager at the Motorola Computer Group, some of my earliest discussions involving optical backplanes were with Ray Alderman, Executive Director of VITA, back in the early 1990s. We discussed how great it would be to have a combination of copper and optical traces, copper for the low-speed control signals and optical for the high-speed data paths. We knew that we wanted to see an optical pathway for the data streams, but we were not sure which serial protocols would be used. Serial switched fabrics were virtually nonexistent except in very high-cost, proprietary implementations in specialized supercomputing. A VITA committee even developed a “photonic” architecture sometime around 1993 (Figure 1).
But a lot has changed since then. Serial switch fabrics are implemented in computing systems of all types, becoming ubiquitous in embedded computing and replacing parallel buses in many new applications. The cost and miniaturization of components have been reduced. What has not changed is the physics of bending light through the backplane/blade transition zone.
Types of implementations
Two camps of thinking exist on how optical interconnects should be implemented on a backplane: 1) waveguides embedded in the backplane; or 2) optical fiber meshes that connect via an overlay on the back.
Waveguides are much more difficult to implement. Embedding waveguides in the backplane itself is a costly challenge, and then the connection to the blades is a problem for which a cost-effective solution has not been found. Embedded waveguides have the advantage of being a much cleaner installation – no fiber cables to deal with in the chassis – but at the same time, you give up a substantial amount of flexibility in how the optical paths are routed from slot to slot. It is a fixed configuration that cannot be easily modified to accommodate changing or varying needs. The HP Innovates sidebar examines an interesting proposal for embedded waveguides (see Sidebar 1).
Meshes eliminate a lot of the waveguide challenges. Meshes can be constructed to route signals in very specific configurations that meet application needs. They present some cabling issues for the rear of a chassis, for instance, routing I/O through the backplane, but system architects have defined specific regions within the backplane that are reserved for optical signal routing.
“ANSI/VITA 66, VPX: Fiber Optic Interconnect” is an example of a mesh implementation. ANSI/VITA 66.0 is the base specification that defines the mounting interface characteristics for all VPX optical interconnects used as independent or stand-alone connectors (Figure 2). The mating interface characteristics for each optical interface variant are contained in a separate VITA 66 dot specification. Currently specifications exist for MT, ARINC 801, and expanded beam style connectors. The VITA 66 working group quietly made great progress, having reached ANSI ratification of VITA 66.0 as well as VITA 66.1, which defines an MT variant blind mate fiber optic interconnect for use with VPX backplanes and plug-in modules. Other variants are moving through the standards process.
Beyond the physical limitations has been the cost trade-off between copper and optical fiber. For years, optical implementations have been hundreds of times more costly for the gain in performance. But now that is changing. Recent implementations are within an order of magnitude of copper, making the cost closer to being acceptable.
VITA Architectures for Optical
The mission of the VITA Architectures for Optical (VAO) Study Group is to research and determine the feasibility of developing a standard architecture for optical interconnects suitable for deployment in critical embedded systems. The study group is focused on high-density options for backplanes and connections between line-replaceable units, mezzanines, and daughtercards. This group is researching high-density optical interconnect technology and developing a proposal for next-generation architectures for critical embedded systems.
The call for participation in the study group was opened to non-VITA members and other Standards Development Organizations (SDOs) to make presentations and participate in discussions with the study group. Existing standards and those under development by SDOs – in addition to appropriate industry alliances, community collaboration efforts, and other groups – will be used whenever practical. The study group will proactively reach out to such groups to facilitate their early involvement. Several organizations have taken VITA up on this offer since that time.
Bookmark the VAO website to monitor the efforts: www.vita.com/vao. Individuals and companies interested in participating should contact email@example.com, Subject: VITA Architectures for Optical Study Group.
Interconnects between systems using fiber optics are already common in data centers around the world. I suggest that the first real systems using optical backplane technology will be in high-end blade-based servers, followed in the not-too-distant future in critical embedded systems more commonly found using VITA technologies. The work on VPX and the user-driven VITA 66 shows that the user community is anxiously awaiting the arrival of optical technology to address their performance needs. The day of reckoning is finally within sight. Based on the amount of information available today versus the last time I researched this topic 8 to 10 years ago, something is about to happen.
The transition point will happen when the cost and complexity of making backplane electronics faster intersect with the decreasing costs of the optical equivalent, perhaps sometime in the next 2 to 4 years.