Integrated Germanium-on-Silicon Detector Opens the Eye at 40 Gb/s

Breck Hitz

While one company’s approach to integrated silicon-germanium photodetectors is described in detail in the feature on page 74 of this issue, Mario J. Paniccia and his Intel Corp. colleagues in Santa Clara, Calif., and in Jerusalem have taken a different approach, fabricating germanium-on-silicon photodetectors with 29.4-GHz bandwidth and 93 percent quantum efficiency. The product of these two parameters is a meaningful measure of a detector’s merit, and the Intel researchers believe that their number — 27.3 GHz — is the highest yet reported for a Ge photodetector at 1550 nm.

Figure 1. Light traveling down the silicon waveguide is evanescently coupled into the germanium photodetector at the far end (a). The cross-sectional view shows the vertical p-i-n junction that forms the photodetector at the end of the waveguide (b). SM = single mode.

The result is important because, as Cary Gunn and Gianlorenzo Masini of Luxtera Inc. explain in the feature article, the feat they call “photonic large-scale integration” will bring vast improvement to photonics communication technology. A crucial step in achieving this total integration is incorporating the optical sensitivity of germanium photodetectors into the electronic capabilities of silicon. Researchers at Intel, at Luxtera and elsewhere are pursuing and achieving that goal, but the efficiency and bandwidth of the Intel device seem to hold the world’s record … for now.

Figure 2. The responsivity of the longer waveguide photodetector is better because the light can be absorbed over a longer distance in the germanium.

The photodetector is at the end of a 1.4-μm-wide silicon waveguide patterned onto a silicon-on-insulator chip (Figure 1). Lateral tapers on the ends of the waveguide improve the coupling to the incoming fiber on one end, and to the germanium waveguide detector at the other. Conceptually, as the light propagates into the photodetector region at the far end of the waveguide, it is evanescently coupled into the germanium, where it is absorbed to produce carriers (holes and electrons). There is an inherent advantage to a waveguide detector such as this, in comparison with a normal surface photodetector, because the light can interact with the detector over the whole length of the waveguide.

But once all the light is absorbed, it doesn’t do any good to make the germanium waveguide any longer. The Intel scientists fabricated some waveguide photodetectors that were 100 μm long and others that were 250 μm long but saw no increase in sensitivity in the longer ones. Evidentially, all the light was absorbed in the first 100 μm of germanium.

Figure 3. Because the two detectors have nearly the same areas, their frequency characteristics are very similar. Both had a bandwidth of ∼30 GHz. The inset shows the bandwidth for various bias values.

Further tests compared two other waveguide detectors, one 7.4 μm wide and 50 μm long, the other 4.4 μm wide and 100 μm long. The longer absorption path in germanium gave the 100-μm-long detector a better responsivity at wavelengths below the cutoff of germanium’s spectral sensitivity (Figure 2). But because the two detectors had similar areas, they displayed similar frequency responses. Both exhibited a bandwidth of about 30 GHz when biased at –2 V (Figure 3).

Figure 4. The open-eye pattern indicates successful performance at 40 Gb/s. These data are for the 7.4 × 50-μm photodetector, biased at –5 V. The scientists believe that fairly straightforward design improvements could boost performance.

To assess how well the photodetectors would work in a 40-Gb/s system, the Intel scientists analyzed the eye pattern that resulted when they fed a 40-Gb/s signal, generated at 1550 nm with a commercial LiNbO₃ modulator, into the detector. The result was an open eye, despite the impedance mismatch between detector and a radio-frequency amplifier that boosted the detector’s signal by 16 dB (Figure 4). This relatively low gain required the scientists to illuminate the detector with as much as 1.2 mW to obtain an acceptable signal.

The investigators believe that packaging the photodetector with an impedance-matched, high-speed transimpedance amplifier would improve significantly the 40-Gb/s eye pattern.

Optics Express, Oct. 17, 2007, pp. 13965-13971.