1. Introduction

Monolithic optoelectronic integration is of great interest in semiconductor optoelectronic research [1–5]. III-V materials are essential for on-chip optoelectronic integration due to their exceptional electro-optical properties. In particular, III-nitride can cover a wide spectral range from ultraviolet to near infrared. III-nitride optoelectronic devices with various functions have been integrated into a small chip on different substrates monolithically, such as gallium nitride (GaN) [6,7], sapphire [8], silicon (Si) [9] and silicon carbide [10]. Monolithic optoelectronic integration offers unique advantages due to its ability to create different functional devices with the same quantum well structure on a single epitaxial wafer. It is popular in optical interconnect [11–15] and transceiver integrated sensing applications [16–20]. The feature of monolithic III-nitride on-chip integration is remarkable for its ability to increase efficiency and reduce costs. It’s special ability stems from the overlapping of the emission and the detection spectra, which is a fundamental physical phenomenon [21,22].

Based on quantum-confined Stark effect (QCSE) [23,24], the application of a vertical external electric field causes the energy band of the multiple-quantum-well (MQW) device to tilt, resulting in a deviation in the spatial distribution of electrons and holes within the well. This deviation leads to a change in the intensity of light absorption. Therefore, III-nitride MQW devices can be used as both light sources and different kinds of modulator under the action of different external electric fields. Combined with the characteristics of overlapping between luminescence and absorption spectra, III-nitride MQW devices with different functions can be manufactured from the same wafer using the same process and integrated natively on a monolithic chip. Actually, it has been demonstrated in recent studies [25,26]. However, the impact of different types of modulations on the spectrum in a monolithic III-nitride on-chip system has yet to be explored.

With the top-down approach, light source, electrical absorption modulator (EAM), waveguide and photodetector were integrated in a monolithic III-nitride chip. And detailed characterizations are presented in this paper. Further, we measured the spectra at the three trench locations and recorded the spectra when the EAM was applied with a modulation voltage. From the experimental data, it can be seen that the asymmetric absorption in a monolithic III-nitride on-chip system causes the spectrum to redshift. This suggests that the next monolithic GaN optoelectronic integration could be improved greatly in terms of spectrum utilization if some precise methods are used to fine-tune the spectrum in the future.

2. Device structure and fabrication

A monolithic on-chip optoelectronic system was fabricated from an as-grown wafer of III-nitride on Si substrate through top-down approach. Figure 1 illustrates the structure of each layer of the chip and the steps involved in the manufacturing process. As shown in Fig. 1(a), the Si substrate is at the bottom layer, and a buffer layer with graded aluminum components was gradually grown using metalorganic chemical vapor deposition (MOCVD) [27]. This step-graded AlGaN buffer is very useful for relaxing strain and reducing dislocations for crack-free growth of GaN on Si (111) [28]. Next, an unintentionally doped GaN buffer layer was grown. The total buffer layer is 2.1 $\mu m$ thick. Then, a Si-doped Al$_{0.03}$Ga$_{0.97}$N n-contact layer with a thickness of 2.45 $\mu m$ was grown. 750 $nm$ thick Al$_{0.1}$Ga$_{0.9}$N n-cladding layer, 80 $nm$ thick GaN n-waveguide, 4 cycles 3/10 $nm$ thick In$_{0.02}$Ga$_{0.98}$N /Al$_{0.08}$Ga$_{0.92}$N quantum wells and 7 $nm$ thick Al$_{0.08}$Ga$_{0.92}$N last quantum barrier, 60 $nm$ thick GaN p-waveguide, 20 $nm$ thick Al$_{0.25}$Ga$_{0.75}$N electron block layer (EBL), 500 $nm$ thick Al$_{0.1}$Ga$_{0.9}$N p-cladding layer were grown out one after another. Finally, magnesium (Mg)-doped GaN p-contact layer were grown on the top, forming the as-grown III-nitride-on-Si wafer. The silicon dioxide (SiO$_{2}$) and electrode layers are deposited on the chip in a subsequent fabrication process.

 

Fig. 1. (a) Layered structure of the on-chip optoelectronic system. (b) Diagram of the manufacturing process.

Download Full Size | PDF

The process of manufacturing, from the as-grown wafer to the monolithic chip, is depicted in Fig. 1(b). In step I, the active pattern is transferred to a photoresist using a standard photolithography process. The wafer was then etched for 200 $nm$ precisely using an inductive coupling plasma (ICP) to remove the excess p-contact. In step II, the p-mesa of 1.6 $\mu m$ height was etched out and the n-contact layer was exposed. And three trenches were formed at the same time. In step III, a 200 $nm$ thick SiO$_{2}$ passivation layer was deposited on the chip by plasma enhanced chemical vapor deposition (PECVD). In step IV, The SiO$_{2}$ layer is treated by the buffered oxide etching (BOE) solution and used as the insulation layer. In step V, 20/200 $nm$ thick Ni/Au electrode films were deposited by electron beam evaporation (EBE) and the metal electrode was patterned by lift-off technology. A rapid thermal annealing (RTA) process was then used to improve the ohmic contact performance in a pure nitrogen environment at 550 °C for 60 seconds.

The morphology and layer structure of the on-chip system were characterized by scanning electron microscopy (SEM), as shown in Fig. 2. A monolithic system consisting of four relatively independent devices united by a common substrate. Figure 2(a) shows that the active regions of the four devices are separated by three isolation trenches. Figure 2(b) zooms in on the position of trench 3 and shows that the trench is approximately 5 $\mu m$ wide and 50 $\mu m$ long. The other two trenches are designed to be the same width and length as trench 3. A test wafer chip was used to characterize the layer structure of the active area in cross-section. In particular, we focus on the edge of the ridge type area because its characterization not only reflects the layer structure but also reveals some details of the manufacturing process. Figure 2(c), (d) show the layer structure composed with Si substrate, the graded buffer, undoped GaN buffer, n-contact, cladding, active layer containing waveguide structure, SiO$_{2}$ passivation and electrode. The two distinct steps and the dip of the SiO$_{2}$ layer for electrode contact are well characterized, in accordance with the fabrication process described above.

 

Fig. 2. (a) An aerial view of the chip under SEM. (b) A partial zoom of the trench 3 position. (c) SEM image of the cross-section of the device. (d) Enlarged view of a specific area around the active layers.

Download Full Size | PDF

3. Experimental results and discussions

In order to obtain a more comprehensive understanding of the on-chip system, we have carried out detailed testing of the electro-optical and photoelectric characteristics of the devices in the system, as well as the interaction between the components. Our on-chip optoelectronic system is composed by a light source, two cascade EAMs and a photodetector. Waveguides are natively integrated into all devices to make a planar photoelectric system. Figure 3 displays the electro-optical and photoelectric characterization results for individual device and the interaction between different devices. The I-V curve of light source was characterized by a semiconductor device analyzer (Agilent SMU B1511A), with high precision mode. When scanned from −8 V to +10 V in 20 $mV$ steps, the I-V curve of the light source shows a typical p-n junction characteristic. Results are shown in Fig. 3(a). The inset of Fig. 3(a) is an electroluminescence (EL) photograph of the light source with an injection current of 10 $mA$. At the same time, the voltage was stable at 7.83 V. From the image, two separate light spots are visible at trenches 2 and 3, suggesting that light is propagating along the waveguide within the chip. Investigating the spectral differences among the three light spots is worthwhile as it can reflect the influence of MQW EAMs on the light source spectrum. We will describe the test methods and experimental results of this part with Fig. 5 in the following.

 

Fig. 3. (a) Volt-ampere characteristic curve and electroluminescence photograph of the light source. (b) The emission spectrum of the light source at trench 1 with a 10 $mA$ injection current and the spectral responsivity of the devices on the chip under different bias voltages. (c) Photocurrent of the PD versus bias voltages on PD at different light source injection currents. (d) Photocurrent of the PD versus bias voltages on EAM$_{1}$ at different light source injection currents.

Download Full Size | PDF

The EL spectrum of the light source with an injection current of 10 $mA$ was recorded using a high-resolution spectrometer (HR4D1557, Ocean Insight Corp.). We measured the spectral responsivity of the on-chip PD using a professional test tool (Oriel IQE200B, Newport, Corp.) after recalibration. The PD was biased at 1.5, 0, −5 and −10 $V$ respectively. These observations were then normalized, including the EL spectrum. They share the same wavelength axis and are combined as shown in Fig. 3(b). It shows that there is an overlap between the shorter part of the EL spectrum and the longer part of the spectral responsivity. Since the light source and the PD have the same structure, the EL spectrum and the spectral responsivity can be considered as properties of each other. In fact, we have measured four independent devices respectively, and the results are almost the same. Figure 3(c) depicts the photocurrents of the PD at various biased voltages when the light source is injected with 0, 5, 10, 15 and 20 $mA$, respectively. Under different light intensities, the photocurrents exhibit obvious distinctions. Figure 3(d) shows the photocurrents of the PD versus different negative voltages biased on the EAM$_{1}$ while the light source is injected with 0, 10, 20 and 30 $mA$, respectively. The alteration of the reverse bias on EAM$_{1}$ results in the energy band tilt of MQW, accompanied by a change in the quasi-two-dimensional exciton concentration and its energy state within the quantum well. This affects the light absorption rate, which subsequently alters the intensity of the transmitted light. Ultimately, this leads to a variation in the photocurrent on PD. These test results demonstrate that light emission, transmission, modulation, and detection can be realized simultaneously in our monolithic III-nitride microsystem.

An on-chip hybrid modulation experiment combining light emission modulation and two-stage absorption modulation is shown in Fig. 4. Figure 4(a) illustrates the composition of the test system and the connection of the circuit. The light source was driven by a programmable power supply (Keithley 2636B), and the modulation signals were generated by an arbitrary waveform generator (Keysight 33622A). The produced photocurrent signals were transmitted to a digital oscilloscope with an impedance of 1$M\Omega$. The connections consist of bonding golden wire, PCB onboard line, SMA connector, and coaxial cables. The light source is alternately injected with 10 $mA$ and 30 $mA$ currents at a 20Hz rate through the program. The modulated signals of the two channels (6Vpp, −6V offset, 100Hz & 500Hz) are loaded onto the two EAMs respectively. Figure 4(b) presents the results shown on the oscilloscope screen, and the enlarged views show the variation in signal intensity resulting from different pre-modulation outcomes.

 

Fig. 4. (a) Test scheme diagram of hybrid modulations on the chip. The polarity of the signal loaded on EAM$_{1}$ and EAM$_{2}$ is in opposition to that of the light source. (b) The detection signal captured on the oscilloscope.

Download Full Size | PDF

Figure 5 shows the diagram and results of the spectral change test. A stable current of 10 $mA$ from the power supply is injected into the light source through tungsten steel probes and coaxial cables. The power supply’s other channel is configured as a voltage source mode, and the output negative bias is applied to the EAM. The UV reinforced fiber used in this experiment was 2 meters in length and had a core diameter of 105 $\mu m$ and a cladding diameter of 125 $\mu m$. Both ends of the fiber were cleaved using an optical fiber cleaver (FITEL, S326) and then cleaned with anhydrous ethanol. The spectrometer and upper computer software are the main acquisition equipment here. A visual diagram of the test device is shown in Fig. 5(a). With the assistance of a probe station system containing a microscope and several three-axis controlling probe holders, fiber optic and tungsten probes are carefully positioned. The spectra at trench 3 are displayed in Fig. 5(b) when the light source is injected at 10 $mA$, and EAM$_{1}$ is subjected to various bias voltages while EAM$_{2}$ is inactive. Figure 5(c) shows the spectra at trench 3 when EAM$_{2}$ is applied at 0, −5, −10, −15, and −20V, respectively, while EAM$_{1}$ remains idle. As shown in Fig. 5(b) and 5(c), the MQW diode’s asymmetric absorption of their corresponding EL spectra results in a difference in intensity and a slight spectral redshift caused by EAM’s influence on the spectrum.

 

Fig. 5. (a) Schematic diagram of spectral change test. (b) Spectra obtained at trench 3 with the light source injected with 20 $mA$ while modulating EAM$_{1}$. (c) Spectra obtained at trench 3 while modulating EAM$_{2}$ with EAM$_{1}$ in idle. (d) The observed spectral redshift of the light source as the injection current is gradually increased. (e) Spectra at trench 2 when the light passes through EAM$_{1}$. (f) Spectra at trench 3 when the light passes through EAM$_{1}$ and EAM$_{2}$.

Download Full Size | PDF

Figure 5(d), (e) and (f) display the spectra at three trench locations when the light source is continuously injected with a current ranging from 5 to 40 $mA$ at 5 $mA$ intervals, while both EAMs remain idle. The broadening of the spectral lines and the redshift, as shown in Fig. 5(d), may be attributed to the multibody effects of carrier scattering, bandgap renormalization, and Coulomb enhancement caused by the large injection current [29]. The spectral redshift at trench 2 and 3, relative to that at trench 1, is likely due to asymmetric absorption and self-absorption. It should be note that no isolated system is unaffected by the gravitational field, which creats the irreversibility. Theoretically, even if a perfect light-emitting device exists, it will absorb higher photon energy than it emits because gravitational fields would cause the irreversibility between the photon emission and absorption process [30,31]. The object at different positions has different quantized states in a gravitation field and thus, its mass related to its total internal energy is different at different energy states, because the total internal energy is equal to its mass times the square of the speed of light, $E=mc^{2}$. According to the law of conservation of energy, the total energy of the system is conserved. Therefore, the work done during the process from one position to another position is not symmetrical for that done during the return trip, which creates the irreversibility. The frequency difference can be expressed as $\omega _{det}-\omega _{emi} =(E_{gap}gH)/hc^2$, where $\omega _{det}$ is the frequency of the detected light, $\omega _{emi}$ is the frequency of the emitted light, $E_{gap}$ is the energy gap between the conduction and valence bands in a gravitational field, $h$ is the Planck constant, $c$ is the velocity of light, $g$ is the normalized acceleration and $H$ is related to the geometrical height between the conduction and valence bands in a gravitational field. The frequency difference is tiny because the physical height $H$ of the energy gap is small. However, both EL spectra and responsivity spectra are broad. Moreover, either loss of energy in the excited state to lattice modes or changes in molecular configuration and vibrational modes also cause the shift in EL versus responsivity spectra in reality. Therefore, there is an asymmetric overlap between the emission and the detection spectra. The experimental results indicate that the asymmetric absorption would cause spectral redshift in a quantum well diode, and self-absorption is a fundamental phenomenon in quantum wells. This suggests that the next monolithic GaN optoelectronic integration could be improved though some spectral fine-tuning techniques, such as selective area growth [32] or transfer processes [33].

4. Conclusion

In summary, we designed and fabricated a specific on-chip system from an as-grown III-nitride on Si wafer with the monolithic top-down approach. It includes light emission, transmission, two-stage absorption modulation and light detection components. The article presents detailed information on morphology and cross-section characterization, basic properties, spectral overlap, and comprehensive signal testing. In particular, the effect of EAM on the spectrum was characterized by measuring the local spectrum using a fiber probe. During the observation of the on-chip system’s spectrum, three distinct spectral redshifts were detected. The analysis of these phenomena indicates that quantum wells exhibit inherent asymmetric absorption and self-absorption. It provides significant reference for future III-nitride optoelectronic integration.