Silicon photonics (Si-Ph) has been identified as a key enabling technology [1] for integrated artificial intelligence (AI) circuits, on-chip microwave photonics systems, light detection and ranging (LIDAR), high-performance computers, quantum computing/communication, gyroscopes, coherent communication, and data centers [2–11]. One of the critical components of Si-Ph is a photodetector (PD) that converts received optical signals to electrical signals [3,5,6,11]. An ideal PD should possess high quantum efficiency (QE), low dark current, wide input power dynamic range, and detect high-speed optical signals of up to 100 Gbit/s for single lane. The indirect band-gap structure of the Si material with 1.1 eV band-gap energy, makes it suffer from low photodetection QE at 1.3–1.55 µm communication wavelengths. However, the germanium (Ge) material, possessing large absorption coefficients of up to 1550 nm and even 2000 nm, has exhibited excellent photodetection characteristics. Various high-speed Ge-on-Si photodetection structures have been extensively investigated and demonstrated [12–21]. Remarkably, the 3-dB cutoff frequency larger than 67 GHz and even estimated up to 120 GHz has been reported [14,21,22], which is very impressive. So far, the light coupling configurations of a Ge-based PD can be summarized as follows: vertical illumination [23–26], butt-coupling [21,27], and evanescent-coupling [13,14,17–19]. In our previous work [28], we proposed and demonstrated the concept of a lateral silicon nitride (${{\rm Si}_{3}}{\rm N}_4$)-waveguide coupled Ge photodetector, which possesses uniform photogenerated carrier distribution in the Ge region. It is helpful for improving the operation speed for high-power signal detection. However, the demonstrated internal QE is only about 32% at 1550 nm, which is mainly caused by the optical losses resulting from the top copper and the inefficient coupling between the tapered lateral ${{\rm Si}_{3}}{\rm N}_4$ waveguides and the Ge region. The QE of the lateral ${{\rm Si}_{3}}{\rm N}_4$-waveguide coupled Ge-on-Si PD is highly dependent on the spacing between the ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the bottom Si, the gap between the tapered ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the Ge region, the width of the top Ge region, and the ${{\rm Si}_{3}}{\rm N}_4$ layer thickness. Therefore, it is desirable to further improve the QE and the operation speed of the ${{\rm Si}_{3}}{\rm N}_4$-waveguide coupled Ge-PD in order to fulfill the application requirements.

In this Letter, first, the QE of the ${{\rm Si}_{3}}{\rm N}_4$-waveguide coupled Ge-on-Si PD is extensively researched. By decreasing the spacing between the tapered ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the bottom Si to 200 nm and decreasing the ${{\rm Si}_{3}}{\rm N}_4$ thickness to 300 nm, the QE is significantly improved. Second, the 3 dB optoelectrical (OE) bandwidth measurements for the ${{\rm Si}_{3}}{\rm N}_4$-waveguide coupled Ge-on-Si PD are executed under different bias voltages. Finally, the eye diagram measurements, including the non-return-to-zero (NRZ) and the four-level pulse amplitude (PAM-4) modulation formats are executed.

Figure 1 depicts the three-dimensional (3D) and cross-sectional schematic of the proposed ${{\rm Si}_{3}}{\rm N}_4$-waveguide coupled Ge-on-Si PD. Different from our previous work [28], the thickness of the ${{\rm Si}_{3}}{\rm N}_4$ layer is changed to be 300 nm, and the lateral ${{\rm Si}_{3}}{\rm N}_4$ waveguides are tapered from 1000 to 400 nm with a 30 µm length. The length of Ge-on-Si PD is 30 µm, which is equal to the tapered length of the ${{\rm Si}_{3}}{\rm N}_4$ waveguide. The light from the standard single-mode fiber was coupled by the ${{\rm Si}_{3}}{\rm N}_4$ edge coupler, then split through a ${{\rm Si}_{3}}{\rm N}_4$ $1 \times 2$ multi-mode interferometer (MMI). The ${{\rm Si}_{3}}{\rm N}_4$ edge-coupler was fabricated based on a suspended spot-size converter (SSC) structure with a $-{2.5}\;{\rm dB}/{\rm facet}$ coupling loss [29], which was subtracted in the analysis of the internal responsivity. Here, the actual center wavelength of the $1 \times 2$ ${{\rm Si}_{3}}{\rm N}_4$ MMI deviated from its designed value. The MMI suffers from $-{1.1}\;{\rm dB}$ insertion loss at 1550 nm. The typical propagation loss of 1.0 µm width and 300 nm thickness in the ${{\rm Si}_{3}}{\rm N}_4$ waveguide is less than 3 dB/cm at 1.55 µm wavelength. As the lengths of routing the ${{\rm Si}_{3}}{\rm N}_4$ waveguide are less than 500 µm, the propagation loss is approximately equal to 0.15 dB. Here, the propagation loss is not subtracted when calculating the responsivity. Then the light was injected into the double lateral ${{\rm Si}_{3}}{\rm N}_4$ waveguides with the same direction and evanescently coupled to the Ge region. The device is fabricated on a commercial 90 nm Si-Ph process platform. The distance from the metal electrode to the ${{\rm Si}_{3}}{\rm N}_4$ waveguide is about 900 nm. The high-quality Ge film with 500 nm thickness was epitaxially grown on the 0.22 µm Si through a low-pressure chemical vapour deposition (LPCVD) process. The spacing between the ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the bottom Si layer was customized to be 200 nm. The following will sequentially investigate the QE dependence of the spacing between the ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the bottom Si, ${{\rm Si}_{3}}{\rm N}_4$ thickness, width of the top Ge region, and the gap between the ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the Ge region.

 

Fig. 1. (a) Three-dimensional (3D) schematic and (b) cross-sectional view of the ${{\rm Si}_{3}}{\rm N}_4$-waveguide coupled PD. The width of the bottom Ge is fixed at 2.5 µm.

Download Full Size | PDF

The QE dependences of the ${{\rm Si}_{3}}{\rm N}_4$ thickness and spacing between the ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the bottom Si are investigated for types A, B, C, and D, as shown in Fig. 2. These structure distinctions are defined as follows: type A, ${{\rm Si}_{3}}{\rm N}_4$ thickness 400 nm, spacing between the ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the bottom Si 400 nm; type B, ${{\rm Si}_{3}}{\rm N}_4$ thickness 400 nm, spacing between the ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the bottom Si 200 nm; type C, ${{\rm Si}_{3}}{\rm N}_4$ thickness 300 nm, spacing between the ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the bottom Si 200 nm; type D, ${{\rm Si}_{3}}{\rm N}_4$ thickness 300 nm, spacing between the ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the bottom Si 100 nm. Here, the gap between the Ge layer and the ${{\rm Si}_{3}}{\rm N}_4$ waveguide is 150 nm. The length is 30 µm. As provided, process data from the fabrication and the thickness of the Ge film is modeled as 500 nm with a 2.5 µm bottom width.

 

Fig. 2. Quantum efficiency of the ${{\rm Si}_{3}}{\rm N}_4$-waveguide coupled PD as a function of the wavelength for types A, B, C, and D.

Download Full Size | PDF

For type A, which is proposed in our previous work [28], the experimentally demonstrated internal QE is estimated to be 32% at 1550 nm wavelength. However, as shown in Fig. 2, the theoretical QE of type A can reach to 39%. The difference of the theoretical and experimental QE is ascribed to the optical losses resulting from the metal copper. Increasing the overlap of the ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the Ge absorption region is proved to be an effective way to improve the QE, as shown in Fig. 2. For type B, the internal QE is about 63% at 1550 nm with the spacing reducing from 400 nm to 200 nm. The maximum internal QE is about 95% for 1400 nm wavelength. Compared to type B, the ${{\rm Si}_{3}}{\rm N}_4$ thickness of type C is 300 nm. The overall QE of type C can be further enhanced for wavelength ranging from 1300 to 1530 nm. When the spacing between the ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the bottom Si is set to be 100 nm (type D), the theoretical QE improves to 94.3% and 87.8% for 1300 nm and 1550 nm, respectively. For longer wavelengths, the light also couples with the ${{\rm Si}_{3}}{\rm N}_4$ waveguide to the bottom Si, as depicted in Fig. 3(c). So in the Ge-on-Si PD fabrication process, the spacing between the ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the bottom Si and the ${{\rm Si}_{3}}{\rm N}_4$ thickness are set to be 200 nm and 300 nm, respectively.

 

Fig. 3. (a) Quantum efficiency of the ${{\rm Si}_{3}}{\rm N}_4$-waveguide coupled PD as a function of the wavelength for the top Ge width of 1, 1.5, and 2 µm. (b) The side view optical field distribution of the Ge region (along the $ X $ axis) with the top Ge 1.5 µm width. (c) Cross-sectional view of the Ge and the bottom Si optical field distribution with the top Ge width 1.5 µm. The length of the Ge and the ${{\rm Si}_{3}}{\rm N}_4$ is 30 µm.

Download Full Size | PDF

For the ${{\rm Si}_{3}}{\rm N}_4$-waveguide coupled Ge-PD, the sidewall of the Ge film is very important for theoretically analyzing the optical absorption in Ge. According to the technology handbook afforded by the foundry, the epitaxial high-quality Ge-on-Si growth is performed through a two-step Ge growth technique. The Ge mesa is mainly surrounded by inclined {111} and {113} facet sidewalls. Generally, there are obvious width differences of the bottom and top Ge. Here, to comprehensively study the QE dependence of the top Ge width, the top Ge width is assumed to be 1, 1.5, and 2 µm with the bottom Ge width fixed at 2.5 µm in the simulation process, as shown in Fig. 3(a). The gap between the Ge layer and the ${{\rm Si}_{3}}{\rm N}_4$ waveguide is 100 nm. For the top Ge width of 1 µm, the high QE only occurs in the special wavelength range, such as 1300 nm to 1400 nm and 1500 nm to 1560 nm. The overall QE is relatively low, which results from the weak interaction between the ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the Ge region. Increasing the top Ge width to 2 µm will significantly enhance the QE, which is larger than 80% in a 280 nm optical bandwidth (1300 nm to 1580 nm). The subvertical sidewall of the Ge film can be further structured with a dry/wet etch process [30–32], which is beneficial to improving the QE. However, in this work, depending on the standard Si-Ph platform, the fabricated top Ge width is about 1.5 µm with the bottom Ge width of 2.5 µm and a 500 nm thickness. Its QE, as a function of the wavelength, is also simulated and depicted in Fig. 3(a). The maximum QE can reach to 98% at 1410 nm wavelength. Figure 3(b) shows the side view along the $ x $-axis of the Ge region optical field distribution with the top 1.5 µm width. The light spreads uniformly in the whole Ge region. Figure 3(c) shows the cross-sectional view of the optical field distribution with the top Ge width of 1.5 µm at a typical wavelength of 1.31, 1.42, 1.51, and 1.58 µm. The inserted color bar of Fig. 3(c) is normalized. The overall theoretical QE decreases with the increasing of the wavelength. At shorter wavelengths (1.31 and 1.42 µm), the light is mainly absorbed by Ge. So the optical field distribution in the Ge region is very small. However, at longer wavelengths (1.51 and 1.58 µm), there are still some energy distributions in the Ge and the bottom Si region because of the low absorption efficiency.

The gap between the linear tapered ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the Ge absorption region, as shown in Fig. 1(b), also affects the QE. In general, the smaller gap benefits from a higher coupling efficiency, as shown in Fig. 4(a). However, in practice, the precise control of a gap less than 50 nm might greatly increase the cost of the manufacturing process. The selected gap 100 nm is a preferable compromise between performance and fabrication. Therefore, the final parameters for fabrication and simulation are as follows: the spacing between the ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the bottom Si is 200 nm, the gap between the ${{\rm Si}_{3}}{\rm N}_4$ and the Ge region is 100 nm, the width of the top Ge region is 1.5 µm, and the thickness of the tapered lateral ${{\rm Si}_{3}}{\rm N}_4$ waveguide is 300 nm. Figure 4(b) shows the top view of optical field distribution in the lateral ${{\rm Si}_{3}}{\rm N}_4$ waveguides and the Ge region at typical wavelengths of 1.31, 1.42, 1.51, and 1.58 µm. The light is evanescently coupled to the Ge absorption region.

 

Fig. 4. (a) Quantum efficiency of the ${{\rm Si}_{3}}{\rm N}_4$-waveguide coupled PD versus the wavelength for gap 50, 100, and 200 nm. (b) Top view of the optical field distribution of the lateral ${{\rm Si}_{3}}{\rm N}_4$ waveguides and the Ge region at typical wavelengths of 1.31, 1.42, 1.51, and 1.58 µm. The light propagates along the $ x $-axis. The profile monitor is located at the height of 300 nm in the Ge region.

Download Full Size | PDF

To evaluate the optical absorption efficiency of the proposed device, the responsivity was simulated and measured versus the optical wavelength at $-{1}\;{\rm V}$ bias, as depicted in Fig. 5(a). The measured responsivity is evaluated to be approximately 0.61 A/W at 1550 nm, which is much larger than our previous demonstrated value 0.36 A/W [28]. However, the theoretical responsivity can reach to 0.92 A/W. The significant responsivity reducing is mainly ascribed to the following: (1) the optical loss caused by the $1 \times 2\,{{\rm Si}_{3}}{\rm N}_4$ MMI waveguide, (2) the threading dislocation densities of the fabricated Ge [33,34], and (3) the absorption efficiency mismatch of the modeled and fabricated Ge. At shorter wavelengths, the theoretical and experimental responsivity matched relatively well. The maximum responsivity is about 0.9 A/W at 1480 and 1485 nm.

 

Fig. 5. (a) Theoretical and experimental responsivity of the ${{\rm Si}_{3}}{\rm N}_4$-waveguide coupled PD versus the input wavelength under $-{1}\;{\rm V}$ bias. (b) Radio frequency (RF) response of the ${{\rm Si}_{3}}{\rm N}_4$-waveguide coupled PD response versus the frequency under different bias conditions.

Download Full Size | PDF

To evaluate the bandwidth of our proposed Ge-on-Si PD under different bias voltgaes, the small-signal (S21) measurements are implemented. The OE bandwidth test experiments were accomplished by collecting the radio frequency (RF) response as a function of frequency, as shown in Fig. 5(b). The bandwidth test setup is similar to previousely reported works [15,20]. The 3 dB OE bandwidth exhibits an increase from 27 GHz up to 54 GHz when the bias voltage varies from 1.1 V to 3.3 V.

The feasibility analysis of the Ge-on-Si PD was further carried out by measuring the large-signal eye diagrams under $-{3.3}\;{\rm V}$ bias. The high-speed optical signal was generated in OOK and PAM-4 modulation formats. The (${{2}^{15}} - {1}$) long optical NRZ pseudo-random bit sequence (PRBS) pattern at 80, 90, 100, and 105 Gbit/s was generated by a commercial Mach–Zehnder lithium niobite (${{\rm LiNbO}_{3}}$) modulator at 1540 nm. The output electrical data was received with a Keysight oscilloscope. From Fig. 6, we can see that the clear opening of the eye diagrams is obtained for a 80, 90, 100, and 105 Gbit/s optical reception signal with a $-{3.3}\;{\rm V}$ bias. Additionally, the clear opening of the eye diagrams up to 150, 160, 170, and 180 Gbit/s PAM-4 are also achieved based on the lateral ${{\rm Si}_{3}}{\rm N}_4$-waveguide coupled Ge-on-Si PD, as shown in Fig. 7. The experimentally obtained clear open electrical eye diagrams in Figs. 6 and 7 exhibit a favorable high-speed signal reception performance of the lateral ${{\rm Si}_{3}}{\rm N}_4$-waveguide coupled Ge-on-Si PD with improved QE.

 

Fig. 6. Measured 80, 90, 100, and 105 Gbit/s NRZ eye diagram under a $-{3.3}\;{\rm V}$ bias condition.

Download Full Size | PDF

 

Fig. 7. Measured 150, 160, 170, and 180 Gbit/s PAM-4 eye diagram under a $-{3.3}\;{\rm V}$ bias condition.

Download Full Size | PDF

In summary, a high QE and high-speed Ge-on-Si PD with lateral ${{\rm Si}_{3}}{\rm N}_4$ waveguides fabricated using a 90 nm commercial Si-Ph process platform is experimentally reported. The QE dependences of the spacing between the ${{\rm Si}_{3}}{\rm N}_4$ waveguide and the bottom Si, the gap between the ${{\rm Si}_{3}}{\rm N}_4$ and the Ge region, the width of the top Ge region, and the ${{\rm Si}_{3}}{\rm N}_4$ thickness are comprehensively investigated. The theoretical responsivity can reach to 0.92 A/W at 1550 nm, while the measured maximum responsivity is about 0.9 A/W at 1480 and 1485 nm. The RF (S21) measurements are implemented to characterize the dynamic properties under different bias voltages. The 3-dB bandwidth of up to 54 GHz is demonstrated at a $-{3.3}\;{\rm V}$ bias condition. Additionally, the 80, 90, 100, and 105 Gbit/s NRZ and 150, 160, 170, and 180 Gbit/s PAM-4 eye diagrams are also obtained. Overall, compared to earlier demonstrated results [28], the ${{\rm Si}_{3}}{\rm N}_4$-waveguide coupled Ge-on-Si PD in this work possesses a higher QE and operates at the highest data-rates reported so far. These attractive results enable the lateral ${{\rm Si}_{3}}{\rm N}_4$-waveguide coupled Ge-on-Si PD to realize high-bit rate, high QE, and high-power detection for silicon photonic integrated circuits in the future.