We have fabricated and characterized a germanium on silicon uni-traveling carrier photodetector for analog and coherent communications applications. The device has a bandwidth of 20GHz, a large-signal 1dB saturation photocurrent of 20mA at −3V, and a low thermal impedance of 520K/W.
© 2012 Optical Society of America
Coherent and analog fiber-optic links require high-power and high-speed photodetectors to achieve good signal-to-noise ratios over wide bandwidths. As coherent modulation formats become more common, the ability of a photodetector to handle currents above 1mA becomes increasingly important, and receiver packaging can be simplified if the photodiode can operate without active cooling. Power handling is even more important in radio-over-fiber applications, where the electronics driving the antenna can be simplified by increasing the photodiode output microwave power. In general, photodetector output power is limited by both space-charge  and thermal effects. Germanium/silicon detectors are attractive for high-power applications because the thermal conductivities of germanium and silicon are large compared to the thermal conductivities of InGaAs and InP. Ge/Si detectors have also gained popularity in recent years due to their compatibility with silicon photonics platforms. Efforts thus far have mostly focused on lowering dark current [2, 3], and increasing bandwidth [4–6]. Here, we present a germanium on silicon detector that has been designed for high output power by using a uni-traveling carrier (UTC) structure. The device has an output saturation photocurrent of 20mA at −3V bias, which is larger than a theoretical p-i-n detector with the same Ge thickness. It also has a responsivity, 0.12A/W at 1550nm, larger than a surface-normal p-i-n detector with the same transit-time limited bandwidth, while maintaining a low thermal impedance relative to comparable InP-based detectors. Overall, it is a promising candidate for use in analog and coherent photonic applications.
2. Device design and fabrication
The Ge/Si UTC structure consists of a highly p-doped germanium absorber and an unintentionally doped silicon collector on an n-type silicon substrate as shown in Fig. 1 . The absorber doping is graded from 1e19cm−3, which is near the solubility of boron in germanium, at the top of the device to 5e17cm−3 at the absorber-collector interface. This induces a small electric field that decreases the transit time in the absorber. There are substantial differences between the Ge/Si material system and the InP-based material system that affect the design of a UTC photodetector. First, there are fewer appropriate materials available for use as a diffusion-blocking layer. Without a diffusion block, carriers that are generated close to the p-contact will diffuse in the wrong direction, decreasing the collection efficiency of the device. Because the system is not lattice-matched, growing a layer with a wider bandgap on top of the germanium is challenging. Instead, the doping grade and the p++ contact layer are used to modify the electric field profile such that the electrons move toward the n-side of the diode. This may lead to some loss of internal quantum efficiency because it is still possible for electrons to diffuse into the p-contact. However, the measured responsivity (which is discussed in more detail in Section 3) is consistent with the trend in published literature for comparable detectors.
The germanium-silicon heterojunction also differs from the InGaAs-InP heterojunction, which affects the design near the absorber-collector interface. In InGaAs/InP detectors, the conduction band offset is significant, and conduction band smoothing or depleted absorber  layers are often used to help carriers get through the interface. In contrast, the conduction band offset in the Ge/Si system is very small , so good performance should be possible without these layers. Furthermore, the traps that form at the Ge/Si interface due to threading defects are acceptor-like , so decreasing the p-doping near the heterojunction may lead to an unwanted barrier there. The primary risk from pursuing this approach is that boron diffusion into the undoped silicon layer during Ge growth would move the band offset into the conduction band. However, germanium growth temperatures tend to be low enough to allow for very abrupt junctions (<800°C).
The detectors are single mesa structures with top ring contacts to allow for illumination from both the top side and the back side. The epitaxial structure is grown (by IQE Silicon), then mesas of varying diameters are patterned in a chlorine-based inductively coupled plasma etch. The mesa sidewalls are then passivated with SiO2 and vias to the Ge and Si are opened by a combination of dry and wet etching in order to minimize the amount of Ge lost during this step. The same contact metal stack, Ni/Ti/Au, is used on both the Ge and the Si. Before probe metal is deposited, SU8 polymer is patterned under the probe pads in order to minimize parasitic capacitance. Some of the devices are anti-reflection coated on the top side with Si3N4 and on others the Si substrate is thinned to 50μm to allow coupling from the back.
3. Responsivity and dark current
Figure 2(a) shows the responsivity as a function of wavelength for a top-illuminated large-area device at −1V bias and low photocurrent. The wavelength dependence is very strong due to the proximity of the bandgap energy to the photon energy. A theoretical curve assuming Eg,hh = 0.79eV and Eg,lh = 0.77eV  is also shown. The responsivity at 1310nm (not shown) is 0.29 A/W and the responsivity at 1550nm is 0.12 A/W. The responsivity under backside illumination is around 7dB lower due to free-carrier absorption in the substrate. Figure 2(b) shows the responsivities of various surface-normal p-i-n photodetectors in the literature at 1550nm as a function of germanium thickness. For photodetectors without anti-reflective coatings, the responsivity shown in Fig. 2(b) is greater than the reported value by a factor of 1.6 (1.4 in the case of detectors with top polysilicon layers) to make a fair comparison. Values that have been thus modified are indicated by crosses in the figure. Curves for absorption coefficients of 2000cm−1 and 5000cm−1 are also shown. The absorption coefficient of Ge at 1550nm varies with growth conditions [10, 11], but typically lies between these two values. The top-illuminated and anti-reflection coated Ge/Si UTC responsivity is in keeping with the trend of other reported values. Assuming an absorption coefficient of 7060cm−1 at 1310nm and perfect antireflective coating at this wavelength, the collection efficiency is 75%. The dark current of this large-area device at −1V is 3.8μA, and the dark current of the smaller device discussed below is 280nA at −1V and increases to 8.9μA at −3V. In order to extract the contributions of sidewall passivation and junction leakage, the dark current of diodes of varying diameters was measured. The dark current is dominated by the area, rather than the sidewall, component and is 35mA/cm2 at −1V bias.
The small-signal bandwidth of a backside-illuminated 14μm diameter device is shown in Fig. 3 . The bandwidth of topside-illuminated detectors with the same diameter (not shown) is the same at low photocurrents. The 3dB bandwidth at 2V and larger biases is 20GHz below 5mA of photocurrent. The bandwidth is limited by both the resistance-capacitance (RC) charging time and the transit time. The diode series resistance and capacitance were extracted by fitting scattering parameter (S11) data to a small-signal model similar to that in . The capacitance of the device shown in Fig. 3 is around 80fF, which includes a parasitic capacitance around 20fF, and the series resistance is around 1Ω. The transit time can be inferred, again using the model in , from the frequency response and the S11 data to be 5.5ps. The transit time limited 3dB frequency (36GHz) is greater than the transit time limited frequency for a p-i-n detector with the same germanium thickness, 29GHz. The frequency response as a whole is in good agreement with the model presented in . The dominant contribution to the transit time is the response of the absorber, which implies that making the silicon collector thicker will increase the overall bandwidth of the device by decreasing the capacitance. This would, however, come at the expense of lower saturation output power because the saturation current is inversely proportional to the thickness of the intrinsic region. It is worth noting that the doping grade in the absorber significantly enhances the bandwidth. The predicted transit-time limited bandwidth for the same structure without the grade is only 6GHz.
The bandwidth of the same device at −3V is shown as a function of photocurrent in Fig. 3(b) and decreases steadily to 18GHz at 15mA. The line in the curve is a fit using series resistance and capacitance values from S11 data and assuming a constant transit time of 5.5ps. Often, UTC PDs exhibit an increase in bandwidth at moderate photocurrents that is not evident in Fig. 3(b). This is generally attributed to a decrease in transit time due to electron velocity overshoot in the collector  and the space-charge induced field in the absorber . Since there is no velocity overshoot in silicon and the absorber transit time is much longer than the collector transit time, it is not expected that the first effect would occur in these detectors. The second effect, where a favorable potential drop is created by the photocurrent and the resistivity of the absorber, on the other hand, could occur. It most likely does, but is not noticeable because the resistivity of p-type Ge is very low (compared to most semiconductors), which results in a low induced potential: approximately 2mV at 15mA, which is much smaller than the 80mV potential induced by the doping grade. Instead, the decrease in bandwidth is due to increasing series resistance and capacitance caused by the space-charge effect in the collector. Both the resistance and the capacitance increase by about 10% as the current is increased from 200μA to 15mA.
5. 1dB compression measurements
The large-signal compression characteristics of the backside-illuminated device are shown in Fig. 4(a) and 4(b) along with theoretical curves in 4(b). An 80% modulation depth tone fixed at 20GHz was generated using the standard heterodyne technique with two free-running lasers at 1537nm. The saturation of the response is caused by both the voltage drop across the load resistance and the space-charge effect in the collector. The voltage across the photodetector, VPD, can be written as:Eq. (1) and Eq. (2) simultaneously assuming uniform illumination and current density yields the maximum current as a function of bias voltage.
The maximum current calculated by solving Eqs. (1) and (2) simultaneously is used to calculate the 1dB compression current, which is shown as the upper theoretical line in Fig. 4(b). The 1dB compression current is defined as the time-average photocurrent at which the RF output power is lower than the output power predicted by the time-average photocurrent by 1dB. In order to relate the maximum current to the 1dB compression current, some assumptions have to be made about the shape of the waveform. When the photodiode is operating in compression, the output is approximately a clipped sinusoid with the maximum output limited to the maximum current that can be sustained by the device. As the portion of the sinusoid’s period (T) during which time the output is clipped (Tc) increases, both the time-average and the RF photocurrent decrease. The DC photocurrent decreases by
It should be noted that although Eq. (4) can be used to accurately predict the RF power of the waveform, the Fourier series converges very slowly, so this equation should not be used to calculate the current in the time domain. The 1dB compression current can be found by comparing the RF power predicted by Eq. (3) to the power given by Eq. (4). The result is that the 1dB compression current (Idc in Eq. (1)) is the maximum photocurrent multiplied by 0.60 for the case of 100% modulation depth and 0.65 for 80% modulation depth. Since Eq. (1) assumes that the maximum current through the device is the sum of Idc and Iac, the appropriate value for Iac in that equation is the maximum photocurrent multiplied by 0.40 for 100% modulation depth and 0.35 for 80% modulation depth. The theoretical line (for 80% modulation depth) is also plotted in black in Fig. 4(b).
For the purpose of comparison, the same model can be used to estimate the 1dB compression current of a p-i-n detector with the same germanium thickness and total device area by modifying Eq. (2). The calculated compression current for such a device at −3V is 8mA, which is smaller than the compression current of the Ge/Si UTC at the same voltage, 20mA. At high levels of photocurrent, the measured 1dB compression current drops below the value predicted by the model. This is due to heating in the device, which decreases the saturated electron velocity and thus decreases the maximum current. The lower theoretical line in Fig. 4(b) includes this effect, using the empirical fit from  for the electron saturation velocity in Si, measured thermal impedance of 520°C/W for electrical power dissipation, and an estimated thermal impedance of 550°C/W for incident optical power. As will be discussed below, most of the heating in the backside-illuminated device is due to free-carrier absorption in the doped substrate, which implies that better output RF power could be obtained by using a semi-insulating substrate.
6. Thermal impedance
The thermal impedance of the Ge/Si UTC is low due to the high thermal conductivities of Ge and Si relative to InGaAs and InP, respectively. Figure 5(a) shows the simulated peak junction temperature as a function of dissipated power for the Ge/Si UTC and a comparable InGaAs/InP device. The simulation is performed using Comsol software. It is a 2D finite-element model with radial symmetry. The heat source is assumed to be uniformly distributed in the collector, and the bottom of the chip is assumed to be held at a constant temperature by the heat sink. The thermal conductivity of silicon (at room temperature) is 1.5 W/cm·K, 2.2 times higher than the thermal conductivity of InP, 0.68 W/cm·K. The thermal conductivity of Ge, 0.56W/cm·K is similarly 11 times higher than the thermal conductivity of InGaAs (0.05W/cm·K), and the net effect is that the device thermal impedance is 1.7 times lower than the thermal impedance of the comparable III-V based device. Thermal conductivities tend to decrease with temperature, and this is taken into account in the model using the data from  for Si and Ge , for InGaAs, and  for InP.
To verify the simulation result, thermoreflectance imaging  was used to measure the temperature of the device under operation. A thermal image of the device when it is dissipating 40mW of electrical power is shown in Fig. 5(b). This technique measures surface temperature by comparing images of the device when it is dissipating a large amount of power to images taken when it is dissipating a very small (typically 10μW) amount of power. The change in temperature, ΔT, is linearly proportional to the change in surface reflectivity, ΔR:Fig. 5(a). Though the simulation result shown is for maximum temperature rather than surface temperature, these were always within 0.2K in the simulation. The temperatures predicted by the model are well within the margin of error of the experiment.
Because the laser was on for both the high-power and low-power images, heating due to optical absorption in the substrate does not appear in the images. This heating was measured by moving the fiber far from the device and pulsing the optical input rather than the voltage, and is typically larger than the heating due to normal device operation. For example, when the device is dissipating 5mW of power (1.5mA photocurrent and 3.3V bias), the total temperature rise is 16K, but 13.5K can be attributed to heat generation in the substrate. The effect of substrate heating is also evident in Fig. 4(b), as the 1dB compression current falls below the predicted value at higher powers.
We have presented a Si/Ge uni-traveling carrier photodetector designed for high-power operation. The UTC design increases the saturation current at a given voltage by allowing for a thin intrinsic region without compromising on responsivity (germanium thickness). Despite being thicker in total than a p-i-n detector with the same germanium thickness, the transit-time limited bandwidth is faster because of the graded doping profile in the absorber. The thermal impedance of the device is low, and it shows promise for use in high-power applications.
This work was supported by the DARPA CIPhER program under contract HR0011-10-1-0079. The authors thank Anand Ramaswamy, Jin-Wei Shi, Keith Williams, and Joe Campbell for useful discussions.
References and links
1. P.-L. Liu, K. J. Williams, M. Y. Frankel, and R. D. Esman, “Saturation characteristics of fast photodetectors,” IEEE Trans. Microw. Theory Tech. 47, 297–1303 (1999).
2. H.-C. Luan, D. R. Lim, K. K. Lee, K. M. Chen, J. G. Sandland, K. Wada, and L. C. Kimerling, “High-quality Ge epilayers on Si with low threading-dislocation densities,” Appl. Phys. Lett. 75(19), 2909 (1999). [CrossRef]
3. L. Colace, P. Ferrara, G. Assanto, D. Fulgoni, and L. Nash, “Low dark-current germanium-on-silicon near-infrared detectors,” IEEE Photon. Technol. Lett. 19(22), 1813–1815 (2007). [CrossRef]
4. S. Klinger, M. Berroth, M. Kaschel, M. Oehme, and E. Kasper, “Ge-on-Si p-i-n photodiodes with a 3-dB bandwidth of 49 GHz,” IEEE Photon. Technol. Lett. 21(13), 920–922 (2009). [CrossRef]
5. M. Jutzi, M. Berroth, G. Wohl, M. Oehme, and E. Kasper, “Ge-on-Si vertical incidence photodiodes with 39-GHz bandwidth,” IEEE Photon. Technol. Lett. 17(7), 1510–1512 (2005). [CrossRef]
6. Y. Ishikawa and K. Wada, “Near-infrared Ge photodiodes for Si photonics: operation frequency and an approach for the future,” IEEE Photon. J. 2(3), 306–320 (2010). [CrossRef]
7. X. Wang, N. Duan, H. Chen, and J. C. Campbell, “InGaAs-InP photodiodes with high responsivity and high saturation Power,” IEEE Photon. Technol. Lett. 19(16), 1272–1274 (2007). [CrossRef]
9. C. Masini, L. Calace, G. Assanto, H.-C. Luan, and L. C. Kimerling, “High-performance p-i-n Ge on Si photodetectors for the near infrared: from model to demonstration,” IEEE Trans. Electron. Dev. 48(6), 1092–1096 (2001). [CrossRef]
10. J. Liu, D. Cannon, K. Wada, Y. Ishikawa, D. T. Danielson, S. Jongthammanurak, J. Michel, and L. C. Kimerling, “Deformation potential constants of biaxially tensile stressed Ge epitaxial films on Si(100),” Phys. Rev. B 70(15), 155309 (2004). [CrossRef]
11. J. Liu, D. D. Cannon, K. Wada, Y. Ishikawa, S. Jongthammanurak, D. T. Danielson, J. Michel, and L. C. Kimerling, “Tensile strained Ge p-i-n photodetectors on Si platform for C and L band telecommunications,” Appl. Phys. Lett. 87(1), 011110 (2005). [CrossRef]
12. T. H. Loh, H. S. Nguyen, R. Murthy, M. B. Yu, W. Y. Lohl, G. Q. Lo, N. Balasubramanian, D. L. Kwong, J. Wang, and S. J. Lee, “Selective epitaxial germanium on silicon-on-insulator high speed photodetectors using low-temperature ultrathin Si0.8Ge0.2 buffer,” Appl. Phys. Lett. 91, 073503 (2007).
13. H.-Y. Xue, C.-L. Xue, B.-W. Cheng, Y.-D. Yu, and Q.-M. Wang, “High-saturation-power and high-wpeed Ge-on-SOI p-i-n photodetectors,” IEEE Electron Device Lett. 31(7), 701–703 (2010). [CrossRef]
14. Y. Kang, H.-D. Liu, M. Morse, M. J. Paniccia, M. Zadka, S. Litski, G. Sarid, A. Pauchard, Y.-H. Kuo, H.-W. Chen, W. Sfar Zaoui, J. E. Bowers, A. Beling, D. C. McIntosh, X. Zheng, and J. C. Campbell, “Monolithic germanium/silicon avalanche photodiodes with 340 GHz gain–bandwidth product,” Nat. Photonics 3(1), 59–63 (2009). [CrossRef]
16. Z. Huang, N. Kong, X. Guo, M. Liu, N. Duan, A. L. Beck, S. K. Banerjee, and J. C. Campbell, “21-GHz-bandwidth germanium-on-silicon photodiode using thin SiGe buffer layers,” IEEE J. Sel. Top. Quantum Electron. 12(6), 1450–1454 (2006). [CrossRef]
17. S. Famà, L. Colace, G. Masini, G. Assanto, and H.-C. Luan, “High performance germanium-on-silicon detectors for optical communications,” Appl. Phys. Lett. 81(4), 586 (2002). [CrossRef]
18. Y.-S. Wu, J.-W. Shi, and P.-H. Chiu, “Analytical modeling of a high-performance near-ballistic uni-traveling-carrier photodiode at a 1.55-μm wavelength,” IEEE Photon. Technol. Lett. 18(8), 938–940 (2006). [CrossRef]
19. T. Ishibashi, S. Kodama, N. Shimizu, and T. Furuta, “High-speed response of uni-traveling-carrier photodiodes,” Jpn. J. Appl. Phys. 36(Part 1, No. 10), 6263–6268 (1997). [CrossRef]
20. M. Chtioui, A. Enard, D. Carpentier, S. Bernard, B. Rousseau, F. Lelarge, F. Pommereau, and M. Achouche, “High-performance uni-traveling-carrier photodiodes with a new collector design,” IEEE Photon. Technol. Lett. 20(13), 1163–1165 (2008). [CrossRef]
21. H. Ito, S. Kodama, Y. Muramoto, T. Furuta, T. Nagatsuma, and T. Ishibashi, “High-speed and high-output InP-InGaAs unitraveling-carrier photodiodes,” IEEE J. Sel. Top. Quantum Electron. 10(4), 709–727 (2004). [CrossRef]
22. C. Jacoboni, C. Canali, G. Ottaviani, and A. Alberigi Quaranta, “A review of some charge transport properties of silicon,” Solid-State Electron. 20(2), 77–89 (1977). [CrossRef]
23. C. J. Glassbrenner and G. A. Slack, “Thermal conductivity of silicon and germanium from 3°K to the melting point,” Phys. Rev. 134(4A), A1058–A1069 (1964). [CrossRef]
24. P. Bhattacharya, Properties of Lattice-Matched and Strained Indium Gallium Arsenide (Institution of Engineering and Technology 1993), Chap. 2.
25. S. Adachi, Handbook on Physical Properties of Semiconductors, vol. 2 (Springer-Verlag 2004), Chap. 16.
26. J. Christofferson, K. Maize, Y. Ezzahri, J. Shabani, X. Wang, and A. Shakouri, “Microscale and nanoscale thermal characterization techniques,” J. Electron. Packag. 130(4), 041101 (2008). [CrossRef]