Not Just For Backplanes Anymore

It’s been a long road for parallel fiber devices. When a new parallel optic array emerges that has promise of going beyond the limitations of copper, a lower-cost copper alternative tends to emerge. But, higher-fiber data rate parts at lower costs are pushing fiber to the forefront again; this includes both short- and long-haul applications.

While most consider fiber a long-haul solution between data centers, parallel fibers are finding higher uses within data centers and facilities. Whether connecting servers, switches, or racks, parallel and modular fiber connections provide high-signal densities with a modular nature. This makes it easier and more desirable when designing a new data hub, expanding an old one, or updating to provide more capacity in the same space.

An interesting place where parallel arrays are getting traction is with PCI Express. Since 2007, 2.5 Gbit/sec (Gen 1) and 5 Gbit/sec (Gen 2) versions of both copper short-haul and fiber medium-haul solutions were quick to vie for attention. Gen 3’s 8 Mbit/sec rate, though, is upping the ante as copper short-haul becomes increasingly more difficult to even span rack-to-rack. What’s more, visionaries are looking at 64 Gbit/sec single-channel links with 128 Gbit/sec bi-directional technology to follow.

The high-speed serial nature of PCI Express makes it an ideal interconnect medium for modern and next-generation high-speed peripherals, disk drives, CPUs, graphics engines, and other point-to-point high-speed lanes even within a box or chassis. As a plug-in card, it takes advantage of a high-speed CPU’s direct access through a wide parallel bus. This allows optimization of the CPU, local cache, local power, and direct I/O circuit board.

For example, storage arrays cannot reasonably take advantage of this parallel architecture; however, with ever-increasing serial speeds, it won’t matter. The drives will be fed and accessed in real time at longer distances without lag or latency.

Parts on the Trail

A partnership is often a great opportunity to establish expertise that spans different technological disciplines. In the same vein, a partnership between Avago Technologies and PLX Technologies has demonstrated the effectiveness of using 8 64 Gbit/sec cables in a Gen 3 PCI express applications (See Figure 1).

When Avago Technologies teamed up with PLX Technologies, they were able to demonstrate the effectiveness of fiber arrays to overcome distance, speed, and limitations with copper for an advanced PCI Express demo.

The PLX PCI Express engines are kept fed by the Avago MicroPOD and MiniPOD Transmitters and Receivers, such as the 120G/150G/168G. These 12-fiber bundled arrays feature programmable equalization and De-emphasis (per lane) and are data-rate agnostic; they don’t care if you pass 1 G or 14 Gbit/sec data through them. Each by 12 part is internally shielded and comes as either a by 12 transmitter array or a by 12 receiver array so they will always match up with a one-to-one correspondence. An online video shows this in action and a white paper describes this as well.

To help with rapid design and deployment, an FPGA partnership with Altera demonstrates how these high-speed serial drivers and receivers can mate with programmable logic and SERDES inside the high-speed and dense-logic arrays. 

Designers have done a good job providing lower-cost copper alternatives in some cases, but faster, cheaper fiber parts are offering more flexibility, reliability, and expandability. Is copper finally at the end of its road?

By: Jon Gabay