Polymers speed optical interconnects

Polymers speed optical interconnects
Optical interconnects may provide the solution for the problem of bandwidth incompatibility between a CPU and printed-circuit board within a computer system. But physics and material sciences suggest this problem could become worse before it gets better, due to the nature of propagation of electrical signals in the very-large-scale integration of conductive wires. And as processor speeds rise, the problem becomes more challenging. Chip-level clock frequency for a mainstream PC has climbed from 100 MHz a few years ago to 800 MHz today.

Meanwhile, PC bus bandwidth for the same time frame grew from 33 MHz to only 100 MHz. Although wide adoption of 133 MHz is on the horizon, serious technical challenges remain. The large bandwidth discrepancy between the chip and board levels prompted Intel Corp. to invest aggressively in various technologies that will increase speed at the board level, to ensure that the faster and more powerful CPU chips of the future will perform effectively.

Rambus Inc.-with strong backing from Intel-has developed a faster memory bus technology that can increase the bus speed by several times. Sampling a data bus at both clock signal edges using the so-called double-data-rate technology is also showing promise. However, high-speed electrical signals in densely packed, conductive wiring structures inevitably generate electromagnetic interference. Clearly, tough challenges lie ahead for future generations of large-bandwidth interconnect technology at the board level.

Unlike traditional pc-board interconnects, optics deliver nonconductive signal propagation and thus potentially large-bandwidth, interference-free communications. The successful use of fiber optics in long-haul communications has provided a solid technological basis for solving other short-distance optical data-transmission bandwidth problems. And research and development efforts have accelerated in the area of optical interconnections, which typically tend to address intracomputer data-communication problems at the chip, board and backplane levels.

Various free-space and waveguide-oriented technologies have been researched and tested. The fundamental difference between those proposed optical-interconnect solutions and long-haul fiber communications solutions is that massively parallel spatial channels are preferred for interconnects. Vertical-cavity surface-emitting lasers (VCSELs), which are more cost-effective for large-array use, are being developed to meet the challenges that parallel optical-interconnect applications demand. Recent R&D shows that VCSELs with large bandwidth (greater than Gbits/second), low threshold current (under 100 milliwatts) and high power efficiency (above 50 percent) can be massively produced. Various array-oriented optical-detector technologies are also being developed to work compatibly with VCSELs.

Active optical components such as VCSELs and photo detectors are just one part of what's necessary to eliminate optical-interconnect problems. Another important part is finding manufacturable, packageable, durable and maintainable optical-channel solutions to interconnect in parallel optical transmission and receiving nodes. Free-space optics, using any combination of refractive, reflective and diffractive optics, offer large and reconfigurable connectivity, but packaging and maintenance issues remain as obstacles.

Silica and polymer-based planar waveguide technologies are potentially semiconductor-pro-cess compatible. The main problem with them is finding a way to reduce the insertion losses associated with multimode rectangular waveguides that could yield large power losses, typically a sizable fraction of a decibel (dB) per centimeter, which renders them suitable to only very short-distance applications.

To establish potentially low-cost optical channel solutions scalable from the subcentimeter to submeter application range that also are reliable, easily packaged and maintained, and that have low insertion loss, researchers at NEC Research Institute (NECI) have developed a polymer-optical-fiber embedding-based solution.

Flexible carrier

POFs are a high-loss transmission media for long distances and have a spectral-dependent loss of about 3 dB per meter for l = 850 nanometers, the dominant wavelength for VCSEL. A conventional optical-interconnect problem at the board and backplane levels uses channel lengths shorter than one meter.

The other great advantage POF offers is its robustness against harsh environments that can often cause damages to silica fibers. For example, POFs can be layered to run across each other without additional protective layers in between. They can be bent at a steep angle with a small bending radius. Their surfaces can even be mechanically etched to form various notches and grooves for various light-coupling purposes. None of those features can be produced cost effectively using silica-fiber-based solutions.

One possible application using a POF-based solution is to construct a high-bit-rate optical-clock distribution network to broadcast a multigigahertz optical pulse train to many optical receiving destinations on a printed-circuit board. Such a clock scheme has been extensively discussed in the past. Using polymer waveguides in an H-tree configuration, researchers at the University of Texas at Austin and Radiant Research Inc. have recently demonstrated a planar optical waveguide implementation of a 1-to-48 fan-out circuit with more than 10-dB excess power loss.

To aggressively reduce power loss, researchers at NECI have proposed using a single-end bundled POF bundle. The bundled end fuses together some 64 thin-cladding multimode POFs. The illumination efficiency of the bundled end can be as high as 75 percent. The other end of the bundle contains loose fibers, all of which have been terminated to an identical length to minimize the possibility of clock skews.

The bundled end is fixed into one end of a conventional fiber connector and the other end hosts either a clock laser source or an input fiber carrying a clock signal. A spacer is used to provide a free-space region between the two tips so that the incoming optical power can broadcast to all outgoing POF channels. The individual clock-carrying POFs are routed using unoccupied solder side space on the pc board to their destinations.

Next, each of these channels undergoes a 90 degrees bending to pass through a via hole on the pc board to the reverse side before interfacing to a photodetector chip. Finally, the POF channels are laminated and covered to finish the embedding. NECI's prototype offers one- to 64-port delivery with a per-port bandwidth limit of 10 Gbits/s. The overall excess loss was reported to be around 4 dB with a 3.5-dB nonuniformity among all 64 channels.

Aside from clock-distribution applications, the NEC group also worked on two-dimensional (2-D) parallel optical interconnects for VCSEL array-based board and backplane interconnect applications. Using a 3,500-element polymer-fiber image guide (PFIG), which is formed by pulling together the same number of close-packed polymer-fiber preforms, 2-D data-capable optical circuits can be formed on the surface or embedded into a conventional pc board.

Perfect shuffle

For point-to-point interconnect applications, the NECI group has fabricated several PFIG-based optical boards and backplanes that contain, for example, perfect shuffle, butterfly and other, similar interconnect patterns. VCSEL and photodetector arrays are butt-coupled to the end surfaces of each PFIG segment. The employed PFIGs have cable diameters of either 2 mm or 1 mm, and their corresponding channel-to-channel spacing is 30 mm or 15 mm, respectively. For VCSELs with a 250-mm pitch, such a PFIG-based optical circuit offers channel isolation of >30 dB among its nearest-neighbor optical channels. Because of the inherent flexibility that polymer materials can offer, large-scale and multilayer optical-circuit integration is possible.

In a separate project, NECI researchers have also successfully integrated 625 PFIG segments totaling >2.1 million independent optical channels for an optical pattern-matching application. Schott Fiber Optics has pursued a similar approach, but it used a glass-fiber image guide (GFIG).

A leached flexible GFIG is a coherent bundle of fibers with an additional acid-soluble glass-cladding layer. After the GFIG is drawn, all of the bundle, except the two end sections, is placed in acid to leach away the additional claddings so that the bundle's midsection becomes flexible. Such a bundle offers higher pixel resolution and lower transmission losses than its polymer counterparts. However, a GFIG is more fragile than a PFIG, and the additional fabrication steps make it more expensive to manufacture than the PFIG.

Looking ahead, managing bandwidth growth at the board and backplane levels will inevitably require help from alternative technologies such as optical interconnects. Researchers at NECI are optimistic about optical interconnects in general and their specific POF-based solutions in particular.