Waveguide PIN photodiodes with different absorber thicknesses and lengths were fabricated and characterized for linearity. Device A has a thicker absorber and shorter length, resulting in a bandwidth of 20GHz while device B reduces the absorber by half while maintaining the intrinsic layer thickness and almost doubles the length, resulting in a smaller optical overlap factor and a bandwidth of 10GHz. Device B shows a significant enhancement in OIP3 with a record high maximum value for a PIN waveguide photodiode of 42.4dBm at 28mA and -4V bias compared to device A which has a maximum OIP3 of 32.7dBm at 10mA and -4V bias. The increased linearity in device B is attributed to the reduction in optical overlap factor and increase in device length resulting in an easing of the front facet photocurrent density and overall device heating. The DC saturation points are about 75mA and >160mA for device A and B at -2V bias.
©2009 Optical Society of America
High power and high spurious free dynamic range (SFDR) links require highly linear photodiodes. In particular, third order intermodulation distortion (IMD3), which is close to the fundamental, is important because it cannot be easily removed by filters . The waveguide photodiode (WGPD) is a good candidate for the frequency agile links. By using waveguiding principles to absorb light along the entire length of the photodiode, the PIN WGPD can offer benefits when compared to a surface normal style photodiode due to its thin intrinsic region which results in a fast transit time . Thermal runaway at the front part of the photodiode has been identified as a failure mechanism for current waveguide style devices . Surface normal photodiodes have measured a very high output third order intercept point (OIP3) of 52dBm at high photocurrents at frequencies less than 1GHz . Previous WGPDs have demonstrated a reduction in optical overlap factor which resulted in OIP3 measurements up to 40.9dBm at 80.6mA at 1GHz using a uni-traveling carrier (UTC) style photodiode . Additionally Jasmin et al. adjusted the active region in the intrinsic region to control confinement factor and increase efficiency . In this paper two PIN WGPDs are compared to demonstrate the benefits of reducing thermal heating by controlling the optical overlap factor and still maintaining a bandwidth of at least 10GHz. The significance of this work is to look at the interplay between bandwidth limitations, power handling capability and OIP3 performance.
Additionally, WGPDs have been analyzed over a range of frequencies for OIP3 . Previously reported is the flat response of OIP3 for a 20GHz PIN WGPD that will be revisited here and compared to a new design .
2. Device fabrication and design
The epitaxial structures can be seen in Fig. 2 where the variable X corresponds to the absorber thickness indicated in Fig. 1. There is a .01μm thick layer of p-InGaAs, followed by the p-InP barrier layers. Next there are three layers of intrinsic InGaAsP that are unintentionally doped (UID) and .007μm which provide bandgap smoothing. The absorber thickness is determined in Fig. 1. Next there are three more layers of UID InGaAsP to smooth out the bandgap and provide the remaining intrinsic region. Finally there is .59μm of n-type InGaAsP. The waveguiding layers are indicated in orange, where the lighter orange layer is also part of the intrinsic region. Both device mesas are 5μm wide with the length depending on the design. The devices are designed to have 20GHz and 10GHz bandwidth for device A and B respectively. The two factors that are modified are the absorption layer thickness which is reduced and the length which is increased from device A to device B. The result of the reduced absorption layer thickness is a smaller optical overlap factor for device B. Both devices are AR coated with Al2CO3.
The effects on OIP3 are studied using two devices with similar layer structures designed for 20GHz (device A) and 10GHz (device B) bandwidth. The factor of importance is the reduction in the optical overlap factor for the 10GHz device. For Device B, the absorber is made thinner, but the intrinsic layer thickness is maintained the same to examine the effects of reducing the heating at the front of the device and absorption along the length of the device. Additionally, the longer device is necessary to maintain similar responsivity for each device since the overlap factor (Γ) is significantly reduced in device B. Figure 1 shows the design parameters for each device. The layer structures are shown in Fig. 2 where X corresponds to the absorber layer thickness detailed in Fig. 1. The length of device B is almost twice as long as device A because of the capacitance requirements for the bandwidth design and to maintain responsivity.
Both devices were modeled using Silvaco International’s numerical device simulator ATLAS to solve for the DC responsivity curve. The result can be seen in Fig. 3. The simulator is 2D with top down illumination. Since both devices are waveguide style, the output is scaled to 3D using the device geometry and the optical overlap factor. The simulation does not take into account the thermal heating that occurs in the device and has not been calibrated with measured data. The results indicate a higher expected 1dB DC saturation for device B. Later results will demonstrate the beneficial behavior of the device in terms of linearity.
Next the devices were thermally modeled in COMSOL. The devices are setup in 3D based on the geometry and the bulk conductivities for each layer. In the simulation the power density is defined along the length of the device according to:
where Pin is the input optical power, W is the width of the mesa, D is the height of the intrinsic region, q is the electronic charge, h is Planck’s constant, v is the frequency, Vb is the bias voltage, Γ is the optical overlap factor, α is the absorption coefficient and η is the input optical coupling efficiency. The device is simulated over a range of input power and the maximum temperature in the device is recorded. The output photocurrent can be found by integrating Eq. (1) over the intrinsic region and dividing by the bias voltage:
where L is the length of the device. Using data from device A the output photocurrent is scaled to approximate that the device fails at 600K. From this, device B can be compared in measurement to assess whether the model predicts the failure point. The results of the simulation can be seen in Fig. 4. The simulation predicts that device B will fail at slightly twice the amount of photocurrent of device A.
4. Results and discussion
The devices were measured for bandwidth using an Agilent 86030A Lightwave Analyzer. The bandwidth measurements were made up to 50GHz at -4V bias and 1mW input optical power and can be seen in Fig. 5 and Fig. 6 for device A and B respectively. The graphs show measurements for many of the same devices to demonstrate the consistency of the bandwidth. The bandwidths are 20GHz and 10GHz for device A and B. Additionally, the device photocurrent failure point was measured at -4V bias with a result of 32mA and 49.3mA for device A and B. Device B exhibits almost two times the current capability of device A at -4V bias due to the reduction in optical overlap and the increased length of the device.
The devices were measured for DC saturation at -2V bias with a responsivity of .5A/W and can be seen in Fig. 7 along with the simulated DC saturation curves performed in Silvaco. The maximum recorded responsivity of each device is .75A/W and .74A/W for device A and B respectively. The saturation point was determined by recording output DC photocurrent versus input DC optical power and finding a linear line of best fit at low input power to determine the approximate photocurrent where DC saturation begins. Device A begins to saturate at about 75mA while device B did not saturate, but exhibited thermal runaway and failed at 160mA. The simulation was scaled according to device geometry parameters as detailed above and then calibrated to the results. The simulation does not predict saturation as in the 4V bias case for device B seen in Fig. 3. In Silvaco, the electric field is observed to collapse which should induce saturation. Instead a large increase in electron current density occurs in the p-region overtaking the hole current density, which leads to recombination near the intrinsic region to p-region interface. The recombination reduces the carrier densities and an induced electric field is observed in the intrinsic region. The field across the intrinsic region increases the carrier transport to their saturation velocities causing the runaway current observed in the measurement.
The device OIP3 was measured using the three laser two-tone setup in . Two distributed-feedback lasers were externally modulated and amplified with an erbium doped fiber amplifier (EDFA). The EDFA is held at constant power and the output at each is attenuated using a variable optical attenuator (VOA) and then combined with a 50/50 coupler. A third un-modulated distributed feedback laser is amplified with an EDFA and also controlled by a VOA. The two branches are combined with a wavelength division multiplexer (WDM) and input into the device. The measurements were made as discussed in  where the slope is three to ensure third order distortion at the given frequency is due to the detector alone.
IMD3 for each device was measured with frequency tones of 1GHz and 1.1GHz. The results can be seen in Fig. 8 with the fundamental and IMD3 plotted for both devices. The devices are biased at -4V and have a DC photocurrent of 10mA and 28mA for device A and B respectively. The OIP3 in Fig. 8 is 30.5dBm and 42.4dBm for device A and B. In the graph both IMD3 sets of data exhibit a slope of 2.97 based on the trend line which is within measurement error. The devices were measured for OIP3 versus a number of different variables for comparison.
In Fig. 9 the OIP3 of the devices are measured over a range of photocurrents with data for device A in blue and device B in red at input frequencies of 1GHz and 1.1GHz and bias voltage of -4V. Device A shows an increase in OIP3 up to 10mA and then is relatively flat until its failure point at -4V bias which is usually slightly more than 30mA. The coupling for device B was optimized for the best nonlinearity possible for the particular device. Two sets of data shown for device B are for two different devices to demonstrate the high reliance of OIP3 on fiber coupling. The circle data shows a significant increase in OIP3 over photocurrent and overall is between 5 to 10dBm higher than device A with a peak occurring at 28mA. The triangle data shows the same increase in OIP3 as photocurrent increases but a flatter response from 20mA to 30mA. During measurements a specific fiber position could be tweaked to see approximately a 10dBm reduction in IMD3 with little effect on the fundamental thereby significantly increasing OIP3. In device A, the fiber positioning was not observed to have a significant effect on OIP3 while maintaining a consistent responsivity. Because of the difference in device structures, there may be a difference in the excitation and propagation of the optical modes which would lead to different thermal distributions and possibly generated distortions that could make OIP3 of device B more sensitive to the fiber position.
Both devices were measured over a range of bias voltages which can be seen in Fig. 10. The purpose of this measurement was to determine the optimal bias point of the device when considering linearity and thermal heating tradeoffs. At a higher bias the device may have better linearity, but will dissipate more power. For device A, the measurements were taken at 15mA photocurrent from -2V to -6V. The device failed at -6V and at about 90mW of power. For device B, the measurements were taken at 20mA from -2V to -9.5V, where the device failed at about 190mW of power, which is almost double that of device A. In Fig. 10, the OIP3 shows an increase from -2V to -4V and then a leveling off for both devices, with device B about 8dBm higher for OIP3. Device B however shows an increase in OIP3 from -7V to -9.5V with a peak of 40.5dBm for OIP3 at -9.5V. Both devices were designed to operate at -4V, in order to maximize linearity by maintaining a particular electric field across the device that is below the breakdown voltage but greater than the point at which carrier velocities are no longer constant. From the results is the desired bias operation is confirmed with the leveling off of OIP3 at voltages higher than -4V.
Next OIP3 is measured over a range of frequencies. The results can be seen in Fig. 11 for device A and B. Device A is measured from 1–18GHz at 10mA output photocurrent and -4V bias with an initial OIP3 of 32.7 . The device exhibits a flat response up until about 16GHz where OIP3 begins to roll off, which interestingly looks similar to the responsivity roll off although the two may not necessarily have any correlation. The OIP3 roll off may be influenced by both thermal and transit time effect. The model to detail this behavior is currently under investigation. Device B is measured from 1-10GHz at 25mA output photocurrent and -4V bias with an initial value of 34.6dBm. The device has a flat response up until 8GHz where there is a slight roll off.
Two WGPD devices were fabricated and characterized demonstrating the effects of reducing the optical overlap factor and increasing thermal capacity through lengthening the device, with the tradeoff of bandwidth. The devices were characterized first through two simulation programs for output DC saturation and thermally for device maximum photocurrent. The 1dB DC saturation points were measured as 80mA and >150mA for device A and B. Device B showed a high dependence of OIP3 on fiber positioning due to the decreased absorber thickness. A significant enhancement of OIP3 and power capability is observed for device B with a record high maximum OIP3 of 42.4dBm at 28mA and 1GHz frequency for a PIN waveguide photodiode.
The authors would like to acknowledge Prof. Sadik Esener for the loan of RF spectrum analyzer. This work was supported by DARPA STTR program and DARPA/SPAWAR program N66001-03-8938 TDL46, both under Dr. Ron Esman.
References and links
1. H. Jiang, D. S. Shin, G. L. Li, T. A. Vang, D. C. Scott, and P. K. L. Yu, “The frequency behavior of the third-order intercept point in a waveguide photodetector,” IEEE Photon. Technol. Lett. 12, 540–542 (2000). [CrossRef]
2. D. C. Scott, T. A. Vang, J. Elliott, D. Forbes, J. Lacey, K. Everett, F. Alvarez, R. Johnson, A. Krispin, J. Brock, L. Lembo, H. Jiang, D. S. Shin, J. T. Zhu, and P. K. L. Yu, “Measurement of IP3 in p-i-n photodetectors and proposed performance requirements for RF fiber-optic links,” IEEE Photon. Technol. Lett. 12, 422–424 (2000). [CrossRef]
3. H. Jiang and P. K. L. Yu, “Waveguide integrated photodiode for analog fiber-optic links,” IEEE Trans. Microwave Theory Tech. 48, 2604–2610 (2000). [CrossRef]
4. A. Beling, H. Pan, H. Chen, and J. C. Campbell, “Linearity of modified uni-traveling carrier photodiodes,” J. Lightw. Technol. 26, 2373–2378 (2008). [CrossRef]
5. J. Klamkin, A. Ramaswamy, N. Nunoya, L. A. Johansson, J. E. Bowers, S. P. DenBaars, and L. A. Coldren, “Uni-traveling-carrier waveguide photodiodes with >40dBm OIP3 for up to 80mA of photocurrent,” IEEE J. Quantum Technol. 44, 354–359 (2008). [CrossRef]
6. S. Jasmin, N. Vodjdani, J. Renaud, and A. Enard, “Diluted- and distributed-absorption microwave waveguide photodiodes for high efficiency and high power,” IEEE Trans. Microwave Theory Tech. 45, 1337–1341 (1997). [CrossRef]
07. M. Draa, J. Ren, D. C. Scott, W. S. C. Chang, and P. K. L. Yu, “Frequency behaviors of the third order intercept point for a waveguide photodiode using a three laser two-tone setup,” [Conference Paper] LEOS 2008, 284–285 (2008).