1. Introduction

High-power, high-linearity photodiodes are essential components for photonic microwave applications since they have the potential to improve many aspects of the link performance such as link gain, noise figure, and spurious free dynamic range. With respect to linearity, the output third-order intercept point (OIP3) is widely accepted as a key figure of merit to characterize nonlinear distortions in microwave and photonic devices [1]. At present the RF output power is limited by two primary factors, saturation originating from the space-charge effect [2] and heat-induced catastrophic failure. Uni-traveling carrier (UTC) photodiodes [3] have mitigated the space charge effect and demonstrated improved high-power performance relative to p-i-n photodiodes while maintaining high speed and good linearity. The advantage of the UTC structure is that only electrons, which have higher saturation velocity than holes, are employed as active carriers in the drift/collection region. Modified uni-traveling carrier (MUTC) photodiodes have been developed to further enhance space charge tolerance by incorporating a “cliff” layer to control the relative electric field strength in the absorber and collector regions [4]. This approach achieved a maximum photocurrent of 152 mA at 6-V reverse bias and 18-GHz bandwidth with a 40-μm-diameter backside-illuminated MUTC photodiode. In this device heat generated in the active-region mesa is mostly dissipated into the substrate and, to some extent, into air. The light is coupled through the substrate, which has an anti-reflection coating on the input interface. The maximum output power achieved by MUTC photodiodes with a cliff layer was not limited by saturation but by thermal failure. Higher RF output power necessitates improved thermal management. Among the various techniques to improve thermal dissipation, flip-chip bonding has achieved the best results. Kuo et al. reported a cascaded two-diode photodetector flip-chip bonded onto AlN with an output power of 63 mW at 95 GHz [5]. Itakura et al. have demonstrated a maximum output power of 790 mW at 5 GHz using a flip-chip bonded 4-diode array with a monolithically integrated Wilkinson power combiner circuit on AlN [6]; the measured output third order intercept point at 4.95 GHz was 32.5 dBm.

In this paper, a thermal-reflectance imaging method was used to characterize the surface temperature of conventional backside-illuminated MUTC photodiodes. These measurements enabled calibration of simulation parameters, which were used to generate two-dimensional plots of the temperature distribution inside the photodiode. The thermal modeling predicted that flip-chip bonding to AlN would result in ~70% improvement in the thermal limit. This was verified experimentally. A 40-μm-diameter photodiode achieved 0.7 A/W responsivity (1540 nm), 15 GHz 3-dB bandwidth, and 0.75 W RF output power. The OIP3 at 330 MHz was 59 dBm at 140 mA photocurrent and remained as high as 40 dBm at 15 GHz and 160 mA.

2. Thermal imaging and simulation

To investigate the thermal characteristics of the conventional backside-illuminated photodiode [4], the temperature profile on the surface of the Au contact area was measured by thermo-reflectance imaging [7]. A top-view photograph and a reconstructed thermal image of a 34-µm photodiode are shown in Figs. 1(a) and 1(b), respectively. The experimental setup measures the change in surface reflectivity using pulses from a green LED array (λ = 530 nm), timed to coincide with the end of the device heating cycle. The reflectivity of the surface is a function of temperature and, thus, the change in reflectivity can be used to obtain the change in temperature. In our measurement, the diode was placed on a thermoelectric cooler maintained at 15°C. A 2-mm-diameter hole in the center of the cooler permitted back-illumination through the InP substrate. The optical input signal was modulated at 400 Hz with 10% duty cycle and the optical illumination was carefully adjusted to ensure a stable outputphotocurrent of 40 mA for each bias level. Adjusting the bias voltage varied the amount of heat generated in the active area. Figure 2(a) shows the measured surface temperature of 34- and 40-μm-diameter backside-illuminated photodiodes versus power dissipation. The imaging measurements were carried out with increasing power dissipation until the devices failed; catastrophic failure occurred when the surface temperature was ~500 K. The relatively low surface temperature at failure can be attributed to lateral inhomogeneities, which create “hot spots”, as shown in Fig. 1(b). A simulation model was created using the finite element analysis tool ANSYS. The model structure follows the epitaxial structure layout of MUTC2 in [4]. The backside of the diode was set at 15°C and the environment temperature was set at 23°C to simulate the same environmental conditions as the measurement. As shown in Fig. 2(a) good agreement with the measurements was achieved. The parameters obtained from thermal imaging were incorporated into the two-dimensional model.

 

Fig. 1 (a) The layout image and (b) thermal image of 34-μm backside-illuminated MUTC photodiode.

Download Full Size | PDF

 

Fig. 2 (a) Measured and simulated surface temperatures of 34- and 40-μm-diameter backside-illuminated photodiodes under different heat generation levels. (b) Simulated surface temperature distribution of 40-μm-diameter photodiode with 680 mW heat flow.

Download Full Size | PDF

It was found that catastrophic thermal failure occurs when the core temperature of a 40-μm photodiode is ~470 K with dissipated power of 680 mW as shown in Fig. 3(a) . For flip-chip bonding to an AlN substrate through 8-μm thick gold vias the core temperature was reduced to 370 K for the same operating conditions, Fig. 3(b). In order to reach the same failure core temperature of 470 K, the dissipated power of the flip-chip device can increase to 1.15 W, which means the flip-chip bonded photodiode shows a thermal limit enhancement of 70%.

 

Fig. 3 Simulated vertical cross section temperature distribution of a 40-μm MUTC photodiode: (a) conventional back-illuminated structure, (b) flip-chip bonded on AlN substrate with 680 mW power dissipation.

Download Full Size | PDF

3. Photodiode structure

The epitaxial layer structure of the active diode structure was grown by MOCVD on semi-insulating InP substrate. The detailed diode structure can be found in [4] labeled as MUTC2. The diameter of the diode is 40 µm and the total depletion thickness is 1.18 µm. The wafer was fabricated into back-illuminated double-mesa structures via ICP etching. A 250 nm-thick SiO₂ film was deposited on the backside as anti-reflection coating. Gold contact vias were up-plated on p- and n- mesas to serve as electrical contacts and heat dissipation paths. The wafer was cut into 2.8 mm × 1.2 mm chips and flip-chip bonded onto a gold contact pad circuit on 0.5-mm thick AlN. The finished flip-chip photodiode cross section and SEM picture are shown in Figs. 4(a) and 4(b), respectively.

 

Fig. 4 (a) Cross-sectional schematic view of the photodiode. (b) SEM picture of the bonded chip on probe pads.

Download Full Size | PDF

4. Measurement results

The frequency tunable optical input was obtained from two equal power heterodyned DFB lasers operating near 1540 nm. The modulation depth was 100% and the frequency was swept by tuning the temperature of one of the lasers. The measured responsivity was 0.7 A/W at 1-mA photocurrent and 5-V reverse bias. Figure 5(a) shows the frequency responses of a 40-µm-diameter flip-chip bonded photodiode under 5 V reverse bias and various photocurrent levels. The 3-dB bandwidth was >15 GHz when the photocurrent was greater than 40 mA. The bandwidth increased with increasing photocurrent, which can be attributed to carrier acceleration by the enhanced self-induced field in the gradated p-type absorber [8]. The S-parameter S₁₁ of the 40-μm flip-chip bonded photodiode at various bias levels is plotted in Fig. 5(b). The extracted capacitance of 210 fF at 5-V reverse bias is 50 fF higher than a backside-illuminated photodiode of the same diameter. The extra capacitance is due to the relatively large pad surface area in the bonding region that was employed for alignment tolerance. The series resistance from S₁₁ measurement is only 2 Ω.

 

Fig. 5 (a) Frequency responses of 40-μm flip-chip bonded photodiode under various photocurrent conditions at 5-V reverse bias. (b) S-parameter S₁₁ of 40-μm flip-chip MUTC2 photodiode at various bias levels.

Download Full Size | PDF

Figure 6(a) shows the measured output RF power versus average photocurrent at 15 GHz under various reverse bias conditions. The space-charge-limited saturation effect has been greatly mitigated through the cliff layer structure [4], and with improved thermal dissipation, the 40-μm diameter photodiode was able to operate under a high reverse bias of 11 V and 180 mA photocurrent without saturation. The RF power from a single photodiode at 15 GHz was 0.75 W (28.8 dBm). Figure 6(b) shows the maximum output power versus frequency for different operating conditions, 5V/130mA to 11V/180mA.

 

Fig. 6 (a) Measured RF power versus average photocurrent at 15 GHz, under various bias levels. (b) Maximum output power at various bias levels versus frequency.

Download Full Size | PDF

We define the dissipated power in the photodiode as the product (V*I_avg), where V is the applied reverse bias voltage and I_avg is the average photocurrent. With V = 11 V and I_avg = 180 mA it follows that the power dissipated by the photodiode is 1.98 W. Recall that the maximum power dissipated by the backside-illuminated photodiode of equal size was only 0.91 W [4].

The OIP3 of the photodiode was measured with a 3-tone apparatus [9]. Figure 7 shows the dependence of OIP3 on photocurrent under various bias conditions at 330 MHz and 15 GHz. It has become standard to report OIP3 in terms of 2-tone measurements. Hence, the data in Fig. 7 has been converted to equivalent 2-tone OIP3 [9]. Two significant OIP3 peaks were observed with bias > 7 V. The OIP3 peaks can be explained by the third-order intermodulation distortion products (IMD3) trough caused by the Franz-Keldysh effect [10], which impacts the voltage-dependent responsivity [11]. OIP3 peaks higher than 59 dBm are present at photocurrents of 55 mA and 141 mA at 9V, the higher current peak corresponds to an output RF power of more than 400 mW at 15 GHz. It has previously been shown that at higher frequencies the OIP3 degrades primarily as a result of the nonlinear voltage-dependent capacitance [12]. However, since we were able to operate the photodiode at a high reverse bias of 9 V, the minimum OIP3 at 15 GHz remained above 30 dBm up to average photocurrents of 170 mA. The OIP3 maximum of 40 dBm near 160 mA suggests a partial compensation of the nonlinear capacitive effect by the aforementioned Franz-Keldysh effect and/or impact ionization. The high OIP3 values up to photocurrents of 170 mA at both low and high frequencies compare favorably with previous reports of 39.5 dBm at 1 GHz (100 mA) [13], 40 dBm at 5.8 GHz (22 mA) [14], and 35 dBm at 20 GHz (40 mA) [15].

 

Fig. 7 OIP3 of 40-μm MUTC PD versus photocurrent at various bias levels measured (a) at 330 MHz and (b) at 15 GHz.

Download Full Size | PDF

5. Conclusions

A MUTC photodiode flip-chip bonded on AlN substrate has been demonstrated. Thermal imaging and simulation were utilized to characterize thermal failure and implement improved heat dissipation. The responsivity was 0.7 A/W and the 3-dB bandwidth was 15 GHz. Saturation current > 180 mA was measured for reverse bias greater than 9 V leading to an output power of 0.75 W at 15 GHz. At 330 MHz, a high OIP3 > 59 dBm was obtained at 140 mA, while the OIP3 at 15 GHz was 40 dBm at a photocurrent of 160 mA.