1. Introduction

Avalanche photodiodes (APDs) are widely used in fiber-optic communications due to their internal carrier multiplication mechanism. The traditional III-V APD receivers offer approximately 10dB improvement in sensitivity up to 10 Gb/s when compared with standard PIN receivers. However, due to the multiplication noise and limited gain-bandwidth product of InP-based APDs, they become less attractive as the bit rate increases. Considerable effort has been devoted to the development of APDs with high gain-bandwidth product and low excess noise [1-3]. Silicon (Si) is considered the best material for avalanche photodiodes among other bulk semiconductors because of its favorable ionization coefficient ratio [4]. In addition, when compared to other semiconductors silicon has a very low temperature dependence of the avalanche breakdown. The combination of these advantages could provide low-cost APD’s with very high gain bandwidth product, low noise and temperature independence.

Promising results obtained from InGaAs/Si APDs fabricated using wafer fusion technology indicate that it is possible to combine the low-noise properties of Si with the high absorption properties of III-V materials at telecom wavelengths [5, 6]. To allow scaling to 300mm diameter wafers, Ge is being considered as a viable candidate for absorption up to 1550nm. Both of these devices could theoretically achieve a gain bandwidth product greater than 300 GHz, which is much higher than that exhibited by currently commercial-available APDs (~100 GHz). The feasibility of Ge epitaxy on Si substrates has been demonstrated by the recent success of GeSi PIN photodiodes [7, 8]. In this paper, we demonstrate epitaxially-grown Ge/Si APDs with a gain-bandwidth product of 153GHz. The ability to monolithically produce Ge/Si-based APDs could have a dramatic impact not only on high data rate optical communications but also on new emerging areas such as imaging, sensing and biotechnology where cost has precluded extensive deployment of detectors with high sensitivity.

2. Device structure and fabrication

The Ge/Si APDs presented in this work are based on the conventional Separate Absorption, Charge and Multiplication (SACM) APD structure in which light absorption and carrier multiplication occur inside Ge and Si, respectively. A schematic cross-section of the mesa-type Ge/Si APD is illustrated in Fig. 1. The device fabrication started with epitaxial growth on low-doped Si substrates, with a typical resistance of 20 ohm-cm, in a commercial CVD chamber. All the Si epitaxial layers were grown at 850 °C, followed by a relaxed, unintentionally-doped Ge buffer layer grown at a lower temperature. The temperature was then increased again to complete the growth of the Ge layers.

After the epitaxial growth, circular mesas were first wet etched through the Ge using a mixture of H₂O₂:NH₄OH:H₂O and dry etched through the Si epitaxial films down to the Si substrate. The mesa diameters vary from 10µm to 200µm. The exposed mesa sidewall was passivated with amorphous Si and then annealed at elevated temperature in the range of 800 °C to 900 °C. This was done to reduce the threading dislocation density originated at the Ge-Si interface due to the 4% lattice mismatch between Ge and Si. A silicon nitride film was deposited for the planarization purpose and also served as an anti-reflection coating at 1310nm wavelength. Aluminium contact pads were sputtered on the top of the mesa and on the substrate. A coplanar waveguide (CPW) transmission line with characteristic impedance of 50Ω was designed for high-speed measurement probing. GeSi PIN detectors with the same layer structure except lack of p-type doping in the Si charge layer, i.e. P-Ge/i-Ge/i-Si/N-Si, were also fabricated in parallel with the APDs. These PIN detectors were used to calibrate the primary photoresponsivity of the APDs at unity gain so that the multiplication gain at every reverse bias can be precisely determined.

 

Fig. 1 Schematic cross-section of a GeSi SACM APD. The GeSi PIN PD used for primary photoresponsivity calibration has the same device cross section except for the p-type Si charge layer.

Download Full Size | PDF

3. Device characteristics and discussion

The measured dark current and photocurrent of a typical 30µm-diameter device at room temperature is shown in Fig. 2. All tested devices exhibit a current-voltage characteristic typical of SACM APDs, with a clear rectifying behavior. When the bias is lower than -12V, the depletion region is solely inside Si. The dark current is relatively constant and in the subnA level. As the reverse bias applied to the device increases, the depletion expands into the Ge region. As a result, the dark current increases due to the generation-recombination current inside the Ge absorption region. The punch-through in the absorption layer occurs between 12V to 20V. Its gradual nature in photocurrent is probably due to Ge-Si interdiffusion at the interface. The capacitance characteristics of the GeSi APD shows that the absorption layer is fully depleted and the device capacitance decreases to a constant value of ~70fF at bias of -21V, as plotted in Fig. 2. The measured avalanche breakdown voltage (defined at a dark current of 10 µA) for this device is -25.8V and was typically in the range of -24 to -26V for the devices on the same 6-inch wafer. The dark current density at -20V was 65mA/cm² and increased to 237mA/cm² at a bias equal to 90% of the breakdown voltage. This dark current density, even at a bias of 90% of the breakdown voltage, is comparable to the typical dark current density obtained from GeSi PINs [7, 8] indicating that the Ge-Si interface is good enough to support the high electric field intensities at the Ge-Si heterojunction. The dark current increases linearly with the device active area, which implies that the dark current is dominated by the bulk leakage current. Note that the dark current of the presented device is about two to three orders of magnitude higher than that of commercial III-V APDs. However, the high dark current of these Ge/Si APDs does not preclude their use in the high-speed fiberoptic communication applications, primarily because the large input-referred noise current of the high-speed transimpedance amplifier (TIA) limits the overall receiver noise. Despite the higher dark current level relative to III-V APDs, the high gain of the Ge/Si APD enabled by the low excess nose would result in superior receiver sensitivity [9].

 

Fig. 2. (a) Current- responsivity- and capacitance vs. voltage characteristics at 1310nm for a 30um-diameter APD at room temperature. (b) DC photoresponsivity and gain vs. voltage.

Download Full Size | PDF

The dark current density of the Ge/Si PIN PDs fabricated on the same chip as the APDs is 90mA/cm² at bias of -5V, about two orders of magnitude higher than that of commercial Ge APDs [10]. It appears that the generation-recombination current, which originates through threading dislocations at the Ge/Si interface and inside the Ge epi-layer, is the primary component of the dark current at high bias. It should be possible to reduce the dark current by optimizing epitaxial growth and the high-temperature treatment during the device fabrication to improve the Ge crystal quality [11].

The measured DC responsivity is shown in the Fig. 2b. The primary responsivity (gain =1) was obtained as 0.54A/W at 1310nm from the Ge/Si PIN PDs, corresponding to an external quantum efficiency of 51%. Using the value of 0.706µm^-1 for the absorption coefficient of Ge at 1310nm wavelength, the expected responsivity is 0.55A/W. The measured responsivity was close enough to the expected value so that the impact of the Ge-Si interdiffusion on the device photoresponsivity can be ignored [7]. The gain of the APDs was obtained by normalizing the responsivity by the primary responsivity measured from the PINs. Since the capacitance measurements have confirmed that the Ge absorption region is fully depleted after -21V, the method used to determine the gain should not introduce any error beyond -21V.

The Ge/Si APD breakdown voltage, V_bd, was measured over a temperature range of 200K to 340K and the results are depicted in Fig. 3. The breakdown voltage thermal coefficient, defined as δ=(ΔV_bd/V_bd)/ΔT, is 0.05%/°C. This value is one third to one half of that observed in typical InGaAs/InP APDs [12,13], consistent with theoretical predictions and experimental results for Si [14]. Among other semiconductors such as Ge, GaAs and InP, studies [15, 16] have shown that Si is the least temperature-sensitive material for breakdown voltage shifts. By using Si as the multiplication material, Ge/Si APD receivers therefore have a less stringent demand on thermal stability compared with other types of APD receivers.

 

Fig. 3 Temperature dependence of the breakdown voltage.

Download Full Size | PDF

The excess noise was also measured as a function of the gain. The device under test was biased using a stable voltage source. An optical CW laser source operating at 1310nm was used to illuminate the APD. The total noise power density was measured at 130MHz, a frequency well above 1/f noise regime, by an HP8970B noise figure meter. Careful system calibration has been carried out with a standard noise source. The noise of the APD was then measured at various biases. With known total current and gain at every bias point, the excess noise factor, F, was then extracted as a function of the gain.

 

Fig. 4 Excess noise factors versus gain. The symbols are the measurement results of a Ge/Si APD; and the solid lines are calculated values based on McIntyre’s theory [17].

Download Full Size | PDF

Figure 4 shows the excess noise factors extracted from a room temperature noise measurement of a 30 µm-diameter Ge/Si APD. As a reference, the calculated excess noise factor, F (M), based on McIntyre’s local field model [17] is also shown as solid lines for various effective k values, k_eff. An excess noise factor smaller than 4 was obtained at a gain of 15. Compared with McIntyre’s model the estimated k_eff is ~0.1. The low k_eff value clearly indicates that the carrier impact ionization events are confined well inside Si multiplication area; otherwise any participation to the avalanche process in the Ge layer would degrade the noise performance. The measured k_eff value is much smaller than the typical k_eff of ~0.5 and ~0.3 obtained from InP APDs or InAlAs APDs, respectively [18, 19], and is an indication of the promise of this approach.

The electrical 3dB bandwidth of Ge/Si APDs was measured using an Agilent 8703A Lightwave Network Analyser with its internal laser and modulator at 1310nm. Figure 5 plots the normalized radio frequency (RF) response obtained from a 30 µm-diameter device for two different gain values. For an APD operated at low gain, the bandwidth is limited by RC and transit time constants, which is similar to PIN photodiodes except that the depletion region includes both depleted Ge absorption and Si multiplication layers. The maximum 3dB bandwidth at a gain up to 10 measured on 30 µm-diameter devices was 10GHz. Theoretical calculation shows that the bandwidth at a low gain for a 30 µm-diameter device is 13GHz, and it is transit time dominated, assuming that both the series resistance and the parasitic capacitance are small and can be ignored. The discrepancy between the measurement and the theoretical expectation is mainly due to the parasitic effects like the insufficient contact layer doping concentrations and capacitance between metal pads. Most importantly, the close-to-theoretical bandwidth at low bias voltage implies that the Ge/Si interface has very limited impact on the device performance. Any remaining potential barrier due to the bandgap discontinuity, interface states or defect traps at the heterojunction does not prevent carriers from crossing the hetero-interface. When operated in the high gain region, the APD bandwidth is dominated by the avalanche build-up time effect [20]. As a result, the 3dB bandwidth for 30µm-diameter devices decreases to 9GHz when the gain increases to 17. The resulting gain-bandwidth product is 153GHz as shown in Fig. 6.

 

Fig. 5. RF response at 1300 nm for a 30 µm diameter device. The 3dB bandwidth is 10GHz and 9GHz for a gain of 6.5 and 17, respectively.

Download Full Size | PDF

 

Fig, 6. Frequency response versus multiplication gain showing a 153GHz gain-bandwidth product. Symbols (diamond, square and triangle) are experimental results of three 30µm-diameter APDs.

Download Full Size | PDF

4. Summary

In this paper, we have experimentally demonstrated mesa-type Ge/Si SACM APDs. A pure Ge absorption layer was epitaxially grown on Si substrates. These devices exhibit a high responsivity of 0.54A/W at unity gain for 1310nm wavelength. Dark current densities as low as 237mA/cm² at 90% of the breakdown voltage were obtained. The breakdown voltage thermal coefficient is only 0.05%/°C. The 3dB bandwidth for 30µm-diameter device is 10GHz at a gain up to 10 and 9GHz at gain of 17, resulting in a large gain-bandwidth product of 153GHz. The measured excess noise exhibit effective k value is ~0.1 at a gain of 15. These results allow the realisation of high sensitivity 10Gbps long wavelength photoreceivers built on Ge-Si based APDs. By utilizing side-illuminated waveguide structure, we believe future devices could achieve even higher gain-bandwidth products.