1. Introduction

The development of high performance optical modulators is essential for silicon photonics integrated circuits. One potential option for long distance transmission regimes is the silicon Mach-Zehnder interferometer (MZI) modulator which employs phase-shift modulation through carrier dispersion. However, its long active region inevitably leads to high energy consumption per bit and a large on-chip footprint [1,2]. In contrast, the electro-absorption (EA) modulator which uses intensity modulation has a shorter active length, making it particularly attractive for short reach links [3]. Furthermore, a Group IV-based (i.e. silicon or germanium) EA modulator allows compatibility with standard complementary metal-oxide-semiconductor (CMOS) processing. The electro-absorption of light by germanium (Ge) -based materials through either the quantum-confined stark effect (QCSE) [4–6], or the Franz-Keldysh (FK) effect [7,8] has been achieved. Leveraging on these results, a butt-coupled GeSi EA modulator was recently reported [8]. Alternatively, an evanescent-coupled EA modulator would facilitate ease of integration for CMOS processing. This work reports a novel design for an evanescently-coupled Ge EA modulator. Through this design, monolithic integration of the EA modulator with an evanescent Ge p-i-n photodetector is demonstrated as well.

2. Design and fabrication

The evanescent EA modulator design comprises of a Ge rib that sits on a Si slab. The optical signal is evanescently-coupled into the Ge active region via a crystalline silicon waveguide (WG). The schematic and cross-section of the Ge EA modulator is shown in Fig. 1(a) and 1(b), respectively. 8-inch silicon-on-insulator (SOI) substrates with Si and buried oxide (BOX) thicknesses of 220 nm and 2 μm, respectively, were used. Firstly, arsenic and boron were implanted into the Si substrate to form the n and p regions. Substrate and contact implantation using these two species were done with a dosage of 5 × 10¹⁴ cm² and 1 × 10¹⁵ cm², respectively. After implantation and activation, a silicon etch step was done for Si slab and Si WG formation. The Si WG width was 500 nm with a nanotaper width of 200 nm at the ends to allow efficient fibre-to-WG coupling and vice versa. Next, Ge selective epitaxy growth (SEG) was achieved using a single wafer cold wall ultra-high vacuum chemical vapour deposition (UHV-CVD) system. The Ge SEG process consisted of growing a SiGe buffer layer and Ge seed at 370°C, followed by a cyclic Ge growth at 550°C [9,10]. For this work, a 400 nm thick intrinsic Ge was grown on both the photodetector and modulator regions. Reactive ion etching (RIE) with a Cl₂-based chemistry was used to form the Ge rib at the modulator region with the photodetector region was masked. Subsequently, the Ge modulator received phosphorous and boron ion implantation with a dosage of 1 × 10¹⁵ cm² and tilt 15° at the sidewalls, while the top of the Ge photodetector was boron-doped with a doping level of 4 × 10¹⁵ cm². Finally, interlayer dielectric deposition, contact hole formation and metallization was done to complete the fabrication. Metal electrodes were located at the sides of the EA modulator to apply a lateral electric field for modulation. The key process steps are shown schematically in Fig. 1(c).

 

Fig. 1 (a) Schematic and (b) cross-section cut of the evanescent-coupled Ge EA modulator structure. The metal electrodes are not shown in the schematic, but are included in the cross-sectional cut. (c) Key process steps for monolithic integration of EA modulator and photodetector are illustrated (Dimensions are not to scale).

Download Full Size | PDF

The modulator dimensions included lengths (L_Ge) of 20, 40 and 60 μm, and effective widths (W_Ge) of 0.6, 0.8 and 1 μm. The vertical p-i-n photodetector had a width and length of 5 μm and 25 μm, respectively. Figure 2(a) shows the scanning electron microscope (SEM) images of both the EA modulator and photodetector after Ge epitaxy and etch steps. Transmission electron microscope (TEM) cross-sectional image [Fig. 2(b)] confirmed that vertical sidewalls were obtained by dry etching. This enabled a uniform electric field to be applied laterally across the Ge rib. The optical measurement setup uses a Photonic Dispersion and Loss Analyzer (PDLA) with a Tunable Laser Source (TLS) module as the optical input. Light is coupled from the PDLA, into and out of the EA modulator using single-mode lensed fibers. When the optical signal is input from the Si WG into the Si slab, it is evanescently-coupled up into the Ge rib because of refractive index difference between the two materials. The optical output from the EA modulator is then fed back to the PDLA to measure the insertion loss at different reverse biases. Before the actual device was measured, a normalization fiber-to-fiber measurement was conducted to capture the loss of the measurement system and test path. Ultimately, the EA modulator insertion loss calculated by the PLDA was obtained through subtraction of the normalization loss.

 

Fig. 2 SEM images showing monolithic integration of evanescent-coupled Ge EA modulator (right) and Ge p-i-n photodetector (left) after Ge epitaxy and etch steps. (b) TEM cross-sectional image of the Ge rib whereby vertical sidewalls are clearly seen.

Download Full Size | PDF

3. Results and discussion

Direct bandgap absorption for bulk Ge occurs for wavelengths less than 1550 nm, and the absorption coefficient drops significantly for wavelengths higher than 1550 nm. However, tensile strain present in the Ge-on-Si grown layers results in bandgap shrinkage, and shifts the absorption edge by ~50 nm [11]. The bandgap energy of intrinsic Ge can further be decreased via the FK effect by applying an electric field to enhance absorption for larger wavelengths [12]. The transmittance spectra of a 40 μm long, 0.6 μm wide Ge EA modulator for a range of wavelengths at different reverse biases is shown in Fig. 3(a) . At 0 V reverse bias, the lowest level of transmittance was observed for wavelengths less than 1560 nm due to direct bandgap absorption (not shown). For wavelengths beyond 1560 nm at 0V bias, the transmittance increases abruptly when Ge absorption coefficient starts to roll-off. When a reverse bias is applied, the transmittance decreases for wavelengths beyond 1560 nm. This decrease in transmittance was attributed to the FK effect whereby bandgap shrinkage enhances optical absorption beyond the absorption edge. It should be noted that the applied electric field across the Ge rib should be less than 100 kV/cm to prevent breakdown of the Ge intrinsic region. The absorption edge of Ge can easily be shifted towards the C-band by using a Ge-rich Ge_{1- x}Si_x alloy film to tune the modulation wavelength [8]. The series resistance of the device was small (~100 Ω) and optical modulation was observed even at 1 V reverse bias. The extinction ratio (ER) was calculated using Eq. (1), where IL(0) and IL(V) is the insertion loss in decibel (dB) at zero bias and a reverse bias of V, respectively.

 

Fig. 3 (a) Transmission spectra of Ge EA modulator (L _Ge = 40 μm, W _Ge = 0.6 μm) for a range of wavelengths at different reverse bias voltages. (b) The corresponding extinction ratio as a function of wavelength at different reverse biases.

Download Full Size | PDF

The ER for the range of wavelengths from 1560 to 1640 nm was extracted at different reverse bias voltages using Eq. (1) and plotted in Fig. 3(b). For a reverse bias of 5 V, the ER of 10 dB or more was achieved for the wavelength range of 1580 – 1610 nm.

The impact of device dimensions on optical modulation was also investigated. Figure 4(a) shows that the ER increases from 3 to 9 dB linearly at 3 V bias for 1600 nm as the modulator length increases from 20 to 60 μm (at a fixed W _Ge of 0.8 μm) due to a larger active area for optical modulation. Correspondingly, the ER decreases from 6 to 4 dB with increasing modulator width from 0.6 to 1 μm (at a fixed L _Ge of 40 μm) because the EA effect reduces as the electric field across the modulator gets smaller at the same reverse bias [Fig. 4(b)]. The IL of the EA modulator was obtained by excluding contributions of fiber-to-WG coupling loss and Si WG propagation loss from the transmission spectra by using a reference Si WG [Fig. 4(c)]. The IL consists of the Si WG-to-EA modulator coupling and Ge absorption loss. At wavelengths smaller than 1560 nm, the IL was the largest due to direct bandgap absorption by Ge. As the Ge absorption coefficient rolls off at the absorption edge, the IL reduces at larger wavelengths when optical absorption weakens. An IL of less than 10 dB was obtained for wavelengths larger than 1600 nm for a 40 μm long, 0.6 μm wide EA modulator. A modulator with a larger width (W _Ge = 0.8 μm) was found to give a lower insertion loss, indicating that the IL can be lowered by optimizing device design. The ER/IL ratio is one of the parameters to determine the EA modulator’s operating range. This ratio is representative of the modulator’s signal-to-noise performance. Therefore, a high ER/IL ratio should be chosen to determine the modulator’s operating range. For the EA modulator with L _Ge = 40 μm, and W _Ge = 0.6 μm, the ER/IL ratio at 5 V is > 1 for wavelength range of 1593 – 1638 nm with the optimum point at ~1625 nm (Note: Maximum λ was 1638 nm). For a wider modulator (W _Ge = 0.8 μm) of the same length, the ER/IL ratio was similar in trend, albeit lower due to a weaker FK effect in the modulator at the same bias. In addition to a high ER/IL ratio, the absolute ER of the modulator at its operating wavelength should also be large. In this work, we decide on the operating range by ensuring the ER is >10 dB and the ER/IL ratio is > 1. As such, the operating wavelength range for the device is defined from 1593 to 1610 nm.

 

Fig. 4 The variation of extinction ratio with (a) modulator length and (b) modulator width is shown at 3 V bias for λ = 1600 nm. (c) The insertion loss for the Ge EA modulator after subtracting contributions from fiber-to-Si WG and Si WG propagation losses by using a reference Si WG.

Download Full Size | PDF

Figure 5 shows that the relationship of ER versus reverse bias (i.e. modulation efficiency) for a range of wavelengths is linear which is desirable for analog applications. The modulation efficiency was observed to be ~2 dB/V. This linearity is similarly observed in Ref. [8] when the EA modulator is operating in the pseudo-linear regime. The highest ER was observed to be around wavelengths of 1590 to 1600 nm. The inset gives the modulated optical signal from the output of the EA modulator (L _Ge = 20 μm, W _Ge = 0.8 μm) at a wavelength of 1600 nm for a bit rate of 1.25 Gbps. The RF signal was from a pseudorandom bit sequence (PRBS) source with a pattern length of [2⁷ − 1] bits. Due to larger insertion loss and sensitivity of our measurement set-up, we were unable to obtain RF measurements for the longer EA modulators. In our future work, optimization to the design will be done to reduce insertion loss for such measurements.

 

Fig. 5 A linear relationship was observed between applied reverse bias and extinction ratio giving a modulation efficiency of 2 dB/V. Inset shows a 1.25 Gbps modulated waveform from a pseudorandom bit sequence (PRBS) source with a pattern length of [2⁷ − 1] bits.

Download Full Size | PDF

One of the advantages of the EA modulator is the low power consumption during operation. This is crucial as integrated circuits are becoming denser which results in increased power consumption and on-chip heating. An energy consumption of less than 100 fJ per bit is achievable at 3V reverse bias with our current EA modulator design (assuming a 1Gbps bit rate) [8]. However, at bias of larger than 3 V, the reverse leakage current increases and contributes to higher energy consumption. This leakage current at higher reverse bias was attributed to Ge film quality and/or implant conditions which can be minimized by process optimization [10]. Even so, the energy consumption is still less than 1 pJ/bit at 5 V operation bias. Figure 6 shows the responsivity of the integrated photodetector for wavelength range of 1580 – 1610 nm at a reverse bias of 1 – 3 V. The lower responsivity (~0.4 A/W) obtained for this range of wavelengths at 1 V, as compared to a wavelength of 1550 nm (~0.9 A/W) can be improved by strain engineering of the Ge layer [13]. We also observe that the responsivity increases with reverse bias which is a clear indication of the FK effect is also occurring in the Ge p-i-n photodetector. This is regardless of whether the electric field is applied laterally in the Ge EA modulator, or vertically in the Ge p-i-n photodetector. In addition, this could be leveraged to improve the responsivity at the weaker absorption wavelength range of the Ge p-i-n photodetector. Finally, a comparison of the key results in this work with other Ge-based EA modulators in literature is presented in Table 1 .

 

Fig. 6 The responsivity of the monolithically-integrated Ge photodetector (L _Ge = 25 μm, W _Ge = 5 μm) at modulation wavelength range of 1580 – 1610 nm.

Download Full Size | PDF

Table 1. Key Results of this Work are Compared with Other Ge-Based EA Modulators ^a

View Table

4. Summary

For the first time, an evanescent-coupled Ge EA modulator was monolithically integrated with Ge photodetector. High modulation efficiency, RF signal modulation and low power consumption are demonstrated in this work. This work provides a route to realize high speed low power electronic-photonic integrated circuits with Ge-based modulators and photodetectors.