In an electronic integrated circuit, tiny conductors connect the various components, and in a nanophotonic integrated circuit, photonic wires — slim submicron waveguides millimeters or centimeters in length — can serve the same function. But whereas electronic integrated circuits are fabricated in multiple layers so conductors can cross each other on different layers without making contact, photonic integrated circuits typically are fabricated on a single layer. Thus, photonic wires cannot duck under each other as they zigzag across the device and instead must intersect directly. The problem becomes designing intersections such that light scattered into the crossing wire is minimized, and such that the light transmitted straight through the intersection is maximized.[Not all photonic integrated circuits are fabricated on a single layer. An experimental technique developed at the University of California, Los Angeles, called “separation by implantation of oxygen (SIMOX) 3-D sculpting,” allows three-dimensional placement of photonic and electronic components on a single monolithic silicon chip. (See “Photonics Goes Underground in Silicon,” Photonics Spectra, October 2005, p. 18).]To solve the problem of intersecting wires, Wim Bogaerts and his colleagues from IMEC and Ghent University, both in Belgium, studied a variety of intersection designs both theoretically and experimentally. A straightforward intersection of silicon-on-insulator wires introduces about 30 percent loss in the signal transmitted straight across the intersection, and the coupling into the crossing wire is –9 dB (Figure 1a). Figure 1. Scientists experimented with several designs for the intersection between two photonic wires, looking for the one that would minimize crosstalk and maximize throughput. The numbers shown above are the throughput — the transmission directly across the intersection — for the various designs. Images reprinted with permission of Optics Letters.The primary reason for the high crosstalk and low throughput in this case is the narrowness of the waveguide, which results in many high-order wide-angle components in the propagating light. Ideally, the wavefront incident on the intersection should resemble more closely a plane wave. With this in mind, the scientists designed a parabolic expansion section around the intersection (Figure 1b). Although this allowed the propagating mode to expand and reduced the crosstalk significantly to –17 dB, the lack of confinement in the waveguide resulted in a straight-through transmission of only 62 percent. When the same technique was applied to both intersecting wires, the throughput dropped to 32 percent, whereas crosstalk increased again (Figure 1c).So the scientists found a middle ground between the straightforward intersection and the parabolic expansion section (Figure 1d). Employing a double-etch technique, they fabricated an intersection in which the parabolic expansion sections were etched to a thickness of 150 nm, while the intersecting wires retained their original pre-etch thickness of 220 nm. In a further refinement, the wires themselves expanded linearly from 500 to 800 nm across the expansion section (Figure 2), which, in its entirety, on both sides of the intersection, had a length of only 6 μm. Figure 2. The double-etched design, a compromise between maximizing the lateral confinement and allowing the beam to expand freely at the intersection, produced the best experimental results.It was this design that produced by far the best experimental results. The transmission across the intersection was more than 96 percent, and the crosstalk was merely –40 dB. To ensure a realistic measurement of the transmission, the scientists coupled as many as 21 intersections in series with each other and measured the loss across the ensemble. These measurements were consistent with a loss of 0.16 ±0.01 dB per crossing, or greater than 96 percent transmission.Optics Letters, Oct. 1, 2007, pp. 2801-2803.