It is time for an optical architecture road map
At the risk of irritating all the copper-based creatures in the embedded backplane and board business, it is time for us to start thinking about optical architectures. The semiconductor folks keep upping the data transfer frequencies with every next-generation chip: The 2.5G is now old technology, 5G is being deployed, and 10G is developing rapidly. And we are heading to 100G connections in a few years. Every time the frequency goes up, the backplane folks pull their hair out trying to make it work consistently and reliably on copper. We are near the end for copper-based links running more than a few inches, so it is time to develop an optical road map.
And, as many of my readers know, I have been advocating the transition to optical architectures and hexagonal board formats in high-performance computing machines at the VSO for about four years now. Figure 1, a futuristic concept of an optically connected computing cluster, shows how both these concepts would look.
An attempt at a road map
The primary way to look at optical architectures is the domain perspective: chip-to-chip, board-to-board, and box-to-box. IBM and others have been working on optical chip-to-chip interconnect technology, but it is currently in the nascent phase (see). The market driving this semiconductor research and component development is the data center where bandwidth requirements rise every day. But many embedded market segments can benefit from these new opto chips aimed at data centers.
Knitting together an optical backplane
Boards are already being interconnected in backplanes using fibers. Just look at the backplane as a patch panel of holes where the fibers pass through to the connectors (see ). So there are many ways to interconnect boards in backplanes using fibers, with different connectors and layouts. We have already seen numerous programs implementing OpenVPX (VITA 65) boards, and they are using discrete fiber cables (with ferules) plugged into the optical connector block on the backplane defined in the VITA 66 specification.
Here is a better way
These backplane patch panels (using fiber cables) work, but they are fairly low tech. Many companies are now developing polymer reflective optical waveguides embedded in the backplane PCB material (. This concept holds the most promise for board-to-board optical interconnects and for the introduction of true optical backplanes in the near future. Imagine building an eight-slot mesh with fiber cable connections through a backplane [that is n(n-1) or 56 fiber cables and 112 connectors]. For a 10-board optical mesh, there are about 90 fibers and 180 connectors. As one can see, building large-scale centralized systems with optical fiber patch panel backplanes is expensive and squirrelly. Optical waveguides in the PCB material make much more sense and are cheaper and more reliable.
Optical networks between boxes
Box-to-box optical connections have been used for quite some time. These cabled connections are used heavily in the data centers now, but those applications are already moving from 10G optical Network Interface Cards (NICs) to 40G optical connections. And they want to move to 100G optical as quickly as possible. In the embedded mil/aero space, we have seen a number of optical connections used in data recorder and sensor interfaces. The components (fibers, connectors, and silicon) are mature but very slow compared to the newer developments in the data center. The increased interest in small “cubes” of electronics – Line Replaceable Units (LRUs) in Unmanned Aerial Vehicles (UAVs) and ground vehicle applications – will probably move to high-speed optical interconnections instead of copper. All distributed system interconnects will migrate to optical connections in the near future and eliminate the copper cables.
So a lot of the basics are already in place. Now the silicon engineers need to move their developments into mass production, the PCB makers need to establish a bulletproof process for embedding optical waveguides in the laminate for backplanes and daughtercards, and the connector folks need to integrate the silicon into highly reliable connectors. That should all happen in the next 18 to 24 months.
For more information, contact Ray at firstname.lastname@example.org.