1. Introduction

High power and high linearity photodiodes are essential for high frequency analog optical links. Two types of photodiodes have been of interest for most applications, surface normal (SN) and waveguide (WG). Waveguide photodiodes allow the decrease of transit time while maintaining high efficiency, which results in bandwidth-efficiency products that can exceed surface normal devices [1]. Currently, one limiting factor for the WG photodiode is thermal heating at the front facet due to high current densities [2]. Additionally, the large current densities at the front facet cause space-charge screening, leading to saturation, which is a primary cause of nonlinear distortion [1]. Previous efforts to use alternate illumination techniques have shown linearity improvement for uni-traveling carrier (UTC) style photodiodes, as well as increased power handling without compromising bandwidth for surface normal structures [3]. By employing a design that is not limited by the front facet current density, the WG photodiode can overcome some of the current handling and saturation issues that limit power handling capability and linearity.

Monolithic integration capabilities for multiquantum-well phase modulators and couplers have been shown with UTC WG photodiodes, demonstrating the versatility provided by WG structures [4]. The use of evanescently coupled photodiodes to achieve high responsivity (>1A/W) and more uniform illumination as well as 0.5A/W responsivity at 50GHz with the use of a integrated spot converter, have demonstrated the ability to increase WG photodiodes performance [5,6]. In this work, a novel waveguide device is presented that utilizes a monolithically integrated directional coupled waveguide technique for guiding the optical power to the photodiode. The directional coupled photodiode (DCPD) with a UTC structure is designed to improve the absorption profile of a waveguide style photodiode. The front facet will no longer have the highest current density, or exponential absorption profile, because the coupler will provide control over the absorption profile. Previous efforts to reduce the coupling factor, and spread out the absorption have shown an increase in power handling capability and linearity but at the expense of device bandwidth [7]. The DCPD will benefit the device by spreading the absorption profile more uniformly, gaining the benefit of a more even current distribution.

2. Device design

Conventional waveguide directional couplers and multimode interference couplers are based on the interference in quasi-symmetrical waveguide structures that have a small number of modes. Although their power transfer and the interference process are well controlled, and the structure can be analyzed by techniques such as coupled mode or super-mode analyses, a UTC detector placed on top of low-mode coupler cannot be used to absorb a uniform and small fraction of incident power for three reasons. (1) The optical power in coupled low-mode waveguides is oscillatory, where even distribution is desired. (2) Only a limited total power can be handled in low-mode waveguides without saturation because of their small cross section. (3) The UTC detector is a major perturbation of the waveguide coupler. In order to detect a very large incident power without saturation in the UTC detector, only a very small uniform fraction of incident power should be absorbed in the absorber. For these reasons, a novel DCPD design which consists of a UTC structure on top of a very large multimode waveguide was conceived. Figure 1a shows the first design. It includes an input transitional waveguide, 50μm long, which transfers the incident optical radiation into desired modes. A 90μm propagation section of 8μm wide waveguide is used to control the input optical radiation pattern at the beginning of the DCPD. The active region (UTC layers) is 200μm long and 2μm wide. After the first 40μm, the width of the optical waveguide is tapered, so the distribution of absorbed power can be more uniform. The UTC on top of a highly multimode waveguide presents a highly asymmetrical total waveguide structure which cannot be analyzed analytically. Simulation techniques such as BeamProp and Fimmwave were used to first find the modes excited in the complex asymmetric structure. A few selected dominant modes were then used to analyze their interferences. These interference patterns will contribute to the actual radiation that may be excited by the incident radiation. Since the actual excited modes of the structure are much more complex, this preliminary result is used only as a guide to vary the dimensions and indices of the DCPD. Various proposed structures excited by incident radiation are then simulated to seek dimensional and index variation which yields uniform and low fractional absorbed power. Figure 2a illustrates the simulated absorbed power per μm length. Contrary to the conventional waveguide PD, the absorption is fairly uniform, with an average of less than 1% of the incident power along the device. Note that the simulated absorbed power per unit length will vary slightly, depending on the specific combination of modes excited by incident radiation determined by the relative position of the fiber. In order to assess the sensitivity of fiber position to the absorption profile in the present input coupling waveguide design, we simulated cases for ± 0.5µm in both the x and y directions from the ideal position that was shown in Fig. 2a. In Fig. 2b we see that changes of 0.5µm in either the x or y direction result in different absorption profiles that will lead to changes in the overall responsivity.

 

Fig. 1 (a) Tapered DCPD device geometry. (b) DCPD layer structure, with bandgap, wavelength (in μm), dopant, carrier concentration (in cm⁻³) and thickness.

Download Full Size | PDF

 

Fig. 2 (a) Simulation of DCPD and WGPD power absorption along the device and (b) simulation of varying fiber position in x or y direction in units of microns, with the original case from (a) at coordinates (0,0).

Download Full Size | PDF

The device structure was grown on an InP substrate by metal-organic chemical vapor deposition. A cross section of the device is shown in Fig. 1b employing a modified UTC style design [8]. Previous reported UTC style designs have demonstrated high linearity capabilities, due to the ability to have both a thin depleted absorber thickness but large responsivity by incorporating a p-type absorber [8].

3. Measurement and results

The responsivity was measured at 1550nm versus voltage and plotted as a normalized responsivity versus electric field in Fig. 3 . The different ranges of electric field are due to the power dissipation limitation, where at higher photocurrents the device will burn out at a lower electric field. The device exhibits some responsivity variation due to coupling, which is expected since different alignments of incident radiation to the input transition waveguide will excite the modes differently, so that the profile in Fig. 2 might have a stronger peak or peak at different positions along the device. The responsivity can vary up to 0.2A/W with small changes in fiber position (~0.5µm), which is a decrease in align tolerance from previous waveguide photodiodes measured in [7]; however once the fiber is aligned the responsivity is very stable. The data from Fig. 2b supports this change in responsivity from small alignment adjustments. Additionally, the change in alignment along with responsivity resulted in large changes (~10dBm) in third order distortions, meaning the linearity measurements, which are discussed later, are also more sensitive to fiber positioning.

 

Fig. 3 Normalized responsivity versus electric field at various photocurrents and dark current versus bias (inset).

Download Full Size | PDF

In measuring the responsivity, the highest value was first obtained, which was 0.88A/W at 4V and 10mA. Each photocurrent case is normalized using the responsivity value at 2V, 0.8A/W, 0.81A/W and 0.88A/W for 0.1mA, 1mA and 10mA cases respectively. The responsivity has less variation versus electric field as current increases. The largest increase in normalized responsivity occurs at very low current (0.1mA). It has been suggested that the low electric field impact ionization coefficient leads to this electric field or voltage dependent responsivity [9]. The DCPD contains InGaAsP in the depletion region which has an electron ionization coefficient relevant at fields greater than 200kV/cm [10]. At high fields, impact ionization causes excess carriers to be created resulting in an increase in responsivity and unwanted nonlinearities [9]. In Fig. 3 the effect is seen at higher electric fields for the 0.1mA and 1mA case, where there should be little device heating. As the photocurrent increases, so does the device temperature. Investigations into nonlinearities of responsivity as a function of voltage have suggested the presence of impact ionization and Franz Keldysh effects, both of which are dependent on temperature [11]. At 10mA we were unable to measure past 175kV/cm as the device would fail due to power dissipation limitations. Additionally, the dark current was measured as a function of reverse bias voltage in Fig. 3 (inset) to determine if this contributed to the increase in responsivity at 0.1mA. The device begins to break down around 14V (350kV/cm electric field), but the dark current is ~225nA, which is more than an order of magnitude less than the change in photocurrent measured at 0.1mA as bias is increased. We conclude that up to 175kV/cm the device shows less dependence on bias voltage at higher photocurrent, where this behavior may be due to the ionization coefficient effects that have been suggested before [9].

Next the device OIP3 was measured using the setup outlined in [12]. In measuring OIP3 we noticed significant changes in the third order distortion with positioning. We believe the increased sensitivity in distortion changes is due to the complicated absorption profile that changes with the positioning of the fiber. The changes can cause photocurrent peaks in different areas of the device that may negatively or positively affect the distortion in certain cases by compensating or adding to other nonlinearities which are present. Figure 4 shows OIP3 versus bias at various photocurrents for 1GHz and 1.1GHz tones. OIP3 increases sharply from 0 to 2V; similar behavior has been observed before in [7], due to the changing capacitance with increasing electric field. Additionally, the OIP3 increases with photocurrent (~13dB), particularly at 4V from 5 to 25mA. Increases in OIP3 with photocurrent have been observed previously in UTC style surface normal devices [11] and PIN waveguide devices [7]. The increase in OIP3 with photocurrent may be a result of the responsivity as a function of bias voltage being more linear (up to 175kV/cm) at higher photocurrents; however, since the devices are measured at 1GHz, the thermal heating may not be the dominant issue. In order to confirm that R(V) is the dominant nonlinearity we plan to continue measurements by using a bias modulation measurement technique presented in [13]. Additionally, the increase in OIP3 may be due to the self-induced field that has been seen before in UTC photodiodes [14].

 

Fig. 4 OIP3 versus bias voltage at various photocurrents for 1.0GHz and 1.1GHz tones.

Download Full Size | PDF

Since this is the first time that a DCPD has been demonstrated, it is as important to understand its behavior as to show its performance. We were unable to reach the high photocurrent predicted from simulations due to a fabrication defect that unintentionally reduced the p-mesa width by 0.5µm due to undercutting from the wet etch process. In order to avoid this in the future, making the mask dimensions slightly wider would avoid the undercutting in the p-mesa. We believe the undercutting lead to self heating at a much lower power than anticipated. To investigate this issue we compared previous waveguide photodiodes from [7] and the DCPD to see where failure occurred in the device. In Fig. 5a the device is 5µm wide and the catastrophic failure clearly occurs at the front of the device, with damage to both the p-mesa and p-metal. In contrast, in Fig. 5b, the DCPD shows about 10µm of metal at the front end before the failure occurs. In the case of the DCPD, we were able to avoid catastrophic failure at the very front of the device. The damage is mostly in the metal, rather than the mesa. From Fig. 5b we show that the DCPD is capable of relieving some of the front end current density compared to a traditional WGPD. Reducing the p-absorber thickness and widening the p-mesa should help avoid the p-metal damage we see in Fig. 5b. Alternatively, other detector structures such as the partially depleted absorber (PDA) [15] may perform better. The PDA structure has currently demonstrated RF output of 26.5dBm, over 700mA of compression current and a very high linearity figure of merit [16]. The device utilizes an InGaAs absorber in the intrinsic layer with additional p-doped InGaAs layers. The DCPD design may benefit from this, as a typical waveguide photodiode with a large InGaAs layer would have very high absorption and consequently high current density at the front of the device. Since the previous work has shown that the catastrophic failure occurs at the front due to the high current density [2], we may be able to use the DCPD design to ease this complication for waveguide style devices.

 

Fig. 5 Images of catastrophic failure for (a) 5µm by 55µm PIN WGPD and (b) MUTC DCPD, note zoom levels are not equal.

Download Full Size | PDF

Finally, OIP3 was measured versus frequency from 1 to 10GHz at 10mA and 4V bias voltage as shown in Fig. 6 . OIP3 remains relatively flat up to 10GHz. Note that from the electrical point of view, the device length is much less than the microwave wavelength up to 10 GHz. Therefore the bandwidth is limited by the RC constant of lumped element circuits, which was measured to be 10GHz. The capacitance was measured as a function of voltage and plotted in the inset of Fig. 6 at 5MHz. Using the model outlined in [17], we plotted the calculated capacitance limited OIP3 for the device, which is plotted in Fig. 6. We can see from this that the data follows the calculated limit at frequencies above 6GHz.

 

Fig. 6 OIP3 versus frequency at 20mA and 4V bias voltage with modeled OIP3 curve (black) based on C(V) measurement at 5MHz (inset).

Download Full Size | PDF

4. Conclusion

A novel DCPD device has been presented. The device exhibits a maximum responsivity of 0.88A/W and OIP3 of 35dBm at 4V and 25mA. The device linearity was measured over a range of frequencies and photocurrents. The design allows for coupling that reduces the front facet current density. Simulation results indicate that absorption per unit length in UTC detector could even be much lower than the 1% incident power reported here. The smaller the fraction of absorbed power per unit length, the larger is the total power that can be detected without saturation. A smaller fraction of the absorbed power implies a much longer length will be required to achieve a responsivity 0.8A/W and higher. Higher bandwidth (>10GHz) can still be achieved for a longer DCPD, if the UTC detector is a microwave transmission line. This implies that, with proper design, the power handling capacity of DCPD can be very large, without limiting bandwidth.