Monolithic germanium/silicon avalanche photodiodes with 340 GHz gain–bandwidth product

Significant progress has been made recently in demonstrating that silicon photonics is a promising technology for low-cost optical detectors, modulators and light sources1–12. It has often been assumed, however, that their performance is inferior to InP-based devices. Although this is true in most cases, one of the exceptions is the area of avalanche photodetectors, where silicon’s material properties allow for high gain with less excess noise than InP-based avalanche photodetectors and a theoretical sensitivity improvement of 3 dB or more. Here, we report a monolithically grown germanium/silicon avalanche photodetector with a gain–bandwidth product of 340 GHz, a keff of 0.09 and a sensitivity of 228 dB m at 10 Gb s . This is the highest reported gain–bandwidth product for any avalanche photodetector operating at 1,300 nm and a sensitivity that is equivalent to mature, commercially available III–V compound avalanche photodetectors. This work paves the way for the future development of low-cost, CMOS-based germanium/silicon avalanche photodetectors operating at data rates of 40 Gb s or higher. Avalanche photodetectors (APDs) are widely used in fibre-optic communications where high sensitivities are needed. Traditional InP-based APD receivers typically offer a 10-dB improvement in sensitivity at 10 Gb s21 when compared to standard p–i–n-based receivers. As the data rate increases above 10 Gb s21, however, the limited gain–bandwidth product of standard InP APDs ( 100 GHz) results in degraded receiver sensitivity. In order to overcome this limitation, an increasing amount of research is now focusing on alternative multiplication materials for APDs, and one of the most promising is silicon. The difficulty in realizing a silicon-based APD device at near-infrared wavelengths is that a compatible absorbing material is difficult to find. The leading candidate is germanium, and research on germanium/silicon p–i–n detectors has shown good responsivity at wavelengths up to 1,550 nm (ref. 7), although the 4% lattice mismatch can result in a high concentration of dislocations and dark current. However, careful processing and device design can minimize the impact of the dislocations. The work presented here extends the research on germanium/silicon p–i–n detectors to the development of a more sensitive APD structure. The key material property that determines the gain–bandwidth product and the excess noise of an APD is the effective k ratio of the ionization coefficients of electrons and holes. A low k value is desirable for high-performance APDs13,14. Conventional III–V APDs use InP (k 0.4–0.5) (refs 15,16) or InAlAs (k 0.1–0.2) (refs 17,18) for the multiplication region. Recent work has shown that by taking advantage of the dead space effect with a 150-nm-thick InAlAs multiplication layer, a gain–bandwidth product of 320 GHz can be achieved17, but at the expense of an impractically low quantum efficiency of 16%. Silicon has been studied for several decades and is well known for its low k value (k , 0.1). Promising results obtained from InGaAs/Si APDs fabricated using wafer bonding technology indicate that it is possible to combine the low-noise properties of silicon with the high absorption properties of III–V materials at telecom wavelengths. These devices had a gain–bandwidth product of 315 GHz (ref. 19) and a k-value of 0.02 (ref. 20). Although these were promising results, the wafer bonding technology would have the extra expense of III–V epitaxial wafers and the challenge of inserting III–V materials into a silicon CMOS fabrication facility. A more ideal approach would be to develop an APD based on a monolithic, CMOS-compatible process. 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 0 50 100 150 200 250 300 350 400 450