Optics primed for box-level solutions

Optics primed for box-level solutions
Most of today's optical technologies are used to connect network end points outside the box over relatively long distances. But a number of companies, research institutes and universities are working on the next level of optical interconnects inside the box at the board-to-board and chip-to-chip levels.

Distributed processing, next-generation switch fabrics and processor-to-memory interfaces require higher bandwidths that traditional copper interconnect may no longer support.

The unique characteristic of board-to-board optical interconnects is the very short end-to-end distance, of 1 meter or less, which eases optical design constraints. This in turn allows a higher dB/m loss budget, lower power requirements for the use of cost-effective 850-nm vertical cavity surface-emitting lasers (VCSELs) and the use of lower-cost multimode fiber.

The three building-block pieces necessary for a board-to-board optical interconnect are optical transceivers, blind-mate optical connectors and an optical backplane. All three exist as commercial off-the-shelf products, and the cost and size of VCSEL transceivers continue to decrease while performance, speed and reliability continue to increase.

Although the devices are primarily intended for port-side or front-panel I/O applications, they can be optimized for board-to-board interconnect applications to further reduce cost, power and size. Innovative blind-mate optical connectors are currently produced by all top-tier connector vendors in single and parallel fibers of up to 24 fibers per module.

A number of vendors have created "fiber mesh" optical backplanes by laminating discrete fiber onto a backing material. The laminating process is automated with tight control for spacing and layout. Fiber mesh backplanes have a packaging advantage over discrete cable assemblies because they provide a single, tested, easy-to-handle assembly rather than a tangle of cables. The backplanes may range in size up to 24 x 24-inch panels, and they can accommodate hundreds of fibers. As the name implies, optical chip-to-chip interconnects replace the traditional electrical signals found on pc board assemblies with optical light signals.

Limited development
The technology is still primarily at the corporate and university research stages, with some limited development in the high-tech startup community. The problem at this point is more complex than the board-to-board interconnect problem, because the technology is at a relatively immature stage. The necessary building blocks required are waveguide technology, integration of semiconductors with optical interfaces and waveguide connectors.

Research on waveguide technology has been going on for the past decade, starting with avionics. The basic concept is to create polymer light pipes that interconnect the end points of an optical circuit on the scale of a pc board assembly. The challenges associated with producing a truly useful waveguide that can be integrated with today's products include:

• improvements in material science to minimize light loss due to material clarity obstacles;
• creating the necessary complex geometries for right-angle turns, multilayer and micromirrors;
• developing tools and processes for integrating the waveguide into the pc board design cycle; and
• achieving the appropriate yield, volume, quality and low-cost goals.

The largest barrier to a chip-to-chip optical interconnect solution is the integration of optical interfaces for semiconductor devices. The ideal answer would be to use the same semiconductor fabrication process to create the optical interface with a VCSEL, PIN diode receiver and associated circuitry as the base process of the device.

That would drive down the cost and size and address reliability issues, while exploiting the performance benefits to create an end-to-end optical interconnect. But there are technical barriers to integration, including the feature-size disparity between optical components and submicron VLSI devices, test problems and reliability concerns. Recent advances at Motorola have made it possible to overcome the difficulties in combining silicon base semiconductor material with gallium arsenide laser materials.

The final technological challenge is the physical connector that interfaces both the chip to the waveguide and one waveguide to other waveguides. Micrometer alignment tolerances are necessary for this interconnect, but flexibility is also needed for practical manufacturing tolerances.

Today, all inside-the-box optical board-to-board interconnects are proprietary; no open standards exist. But there are several trends that might make this a good time to pursue such an effort.

Most important is the new generation of applications, which require higher-speed interconnects (exceeding the capabilities of copper). As the demand for bandwidth continues to increase, optical interconnects may provide a solution .

But the time to start is now, because the standards process is usually a long and arduous one. The technology itself is mature enough to be developed into technological solutions.

By their very nature, standards are not a good vehicle to develop technology. Rather, they are a good vehicle for existing technology to create a framework for interoperability that benefits designers, manufacturers and vendors with a common platform.

This encourages the optical ecosystem needed to bootstrap and sustain the technology.