High-power Flip-chip Mounted Photodiode Array References and Links

Four-element modified uni-traveling-carrier (MUTC) photodiode arrays (PDA) flip-chip bonded onto transmission lines on AlN substrates are demonstrated. High RF output powers of 26.2 dBm and 21.0 dBm are achieved at 35 GHz and 48 GHz, respectively, using a PDA with 28-μm diameter photodiodes. A systematic comparison between a PDA with four 20 μm-diameter elements and a discrete detector with the same active area (40-μm diameter) is presented. The PDA achieved higher output power and thermal dissipation compared to its discrete counterpart. InGaAs-InP photodiodes with high responsivity and high saturation power, " IEEE Photon. High-saturation-current modified uni-traveling-carrier photodiode with cliff layer, " IEEE J. Active and nonlinear wave propagation devices in ultrafast electronics and optoelectronics [and prolog], " Proc. High-power monolithically integrated traveling wave photodiode array, " IEEE Photon.


Introduction
As optical communication networks accumulate consumers with ever increasing demands, there will be a continuous need for greater capacity and functionality from these networks.A key factor toward this end is the development of high-performance photodiodes (PDs) for applications such as high-speed receivers, antenna remoting, optically-fed phased array antennas, and photonic oscillators.
Advancements in bandwidth, saturation-current, and power handling have been made in the past decade by transitioning to new materials and epitaxial structures.The development of the InP based uni-traveling-carrier (UTC) epitaxial structure has enabled high bandwidths by eliminating transport of slow moving holes across the depletion region [1].Although the carrier transit distance for UTC diodes is typically greater than that of PIN diodes, the total carrier transit time is less since electron mobility is an order of magnitude greater than hole mobility.Subsequent alterations to the UTC structure, such as the modified UTC (MUTC) [2] and charge-compensated MUTC [3] designs, further improved PD power performance.
Improvements in one performance parameter of a detector are often achieved at the expense of one or more equally significant figures of merit.This obstructive dynamic is particularly evident in regards to the physical dimensions of UTC-based PDs.For instance, increasing the device area typically results in higher saturation current and improved coupling efficiency.However, the detector bandwidth, which is determined by a combination of resistance-capacitance (RC) time-constant and carrier transit time limitations, has a more complicated but typically inverse relationship to the physical dimensions.In practice, increased bandwidth is typically achieved by decreasing device dimensions, sacrificing RF output capabilities [1].
While simply connecting detectors in parallel can produce higher saturation currents and greater RF output powers, the capacitances are cumulative and, once again, limit bandwidth.Alternatively, successively placed diodes, as lumped elements, can be integrated into a transmission line: while the currents are summed, the distributive nature of the diode capacitance and the characteristic parameters of the transmission line impose RC limitations on bandwidth equivalent to that of just a single diode [4,5].In this paper we study the performance of a flip-chip bonded photodiode array (PDA) with four normal-incidence backilluminated PD-elements and a travelling wave transmission line design.

Epitaxial structure
The MUTC detectors were designed with the motivation of maximizing the saturation current and bandwidth of the detector array.The included cliff layer reduces the space-charge buildup when the detector is illuminated at high optical powers [3].The thickness of the chargecompensated InP collection layer was increased from previous designs to 900 nm in order to reduce the capacitance and thereby increase the RC bandwidth.The In 0.53 Ga 0.47 As absorbing layer thickness, on the other hand, was decreased in order to reduce the transit-time.With the thinner absorbing layer the responsivity was 0.70 A/W.A more detailed description of this epitaxial structure is provided in [3] where it is referred to as MUTC2.

Chip fabrication
The back-illuminated double-mesa structures were formed by inductive coupled plasma (ICP) etching.The p-mesas for the three PDAs under test, designated as PDA-20, PDA-28, and PDA-40, have diameters of 20 μm, 28 μm, and 40 μm, respectively.Each four-element PDA was comprised of diodes PD1 through PD4 (Fig. 1) arranged in a line with a pitch, d, of 250 μm.On mesas, p and n, ohmic contacts were first deposited and then gold bumps were upplated to 2-μm thickness.The gold bumps, placed on each mesa-structure, served as both electrical contacts to the coplanar waveguide (CPW) located on the AlN sub-mount, and as thermal contacts to maximize heat transfer from the diode.As Joule heating is primarily generated in the high-field region in the drift layer, adding a gold bump (with a diameter 6 μm less than that of the diode) above each diode junction provides the most direct and efficient path of thermal dissipation.After fabrication, the back of the wafer was polished and an antireflection coating of 250-nm-thick SiO 2 was deposited for enhanced optical coupling.Finally the wafer was diced into 1.5-mm x 1.0-mm chips, each with four discrete detectors and one PDA array (Fig. 1).

Submount integration
To improve thermal power dissipation and high power performance, the individual dies were then bonded to a circuit on an AlN submount via a flip-chip process.The bonding temperature was 285 °C with a constant pressure of 5 N.The submounts consist of 0.5-mm thick AlN with microwave contact pads fabricated for high-speed measurements and a CPW connecting the four photodiodes in parallel.The 75-Ω characteristic impedance, Z 0 , of the CPW transmission line before the chip was mounted was calculated using a signal-line width and signal-toground gap of 50 μm and 75 μm, respectively.After flip-chip bonding, the higher effective dielectric ϵ eff of the InP-AlN bi-layer resulted in a well-defined reduction from 75 Ω to 50 Ω, and the additional junction capacitance (dependent on diode dimensions) further decreased the impedence.The alignment accuracy of the flip-chip bonding was approximately 5-μm, which was sufficient for the p-mesa landing pads that are only slightly larger than the diodes themselves.Finally, for high-power measurements, the devices were cooled using a thermoelectric heat sink.

Time-domain
Time-domain measurements were carried out on the PDA in order to characterize the CPW and match the delay times of the diodes.A fiber-coupled femtosecond pulsed laser and a programmable optical attenuator were used as the optical source.A one-to-four beam-splitter was used to form the input beams.Each individual line traversed a free-space optical delay stage and was coupled into separate channels of the fiber array.Optical diaphragms placed in each delay path served as attenuators to balance the input signal of each diode.A 1x4 lensed fiber-array with ~250-μm pitch was aligned to the photodiode array.The fiber had a working distance of 2 mm and focal point diameter of 35 μm.The diodes were illuminated through the 350-μm thick InP substrate.RF probes were placed on either end of the symmetrical PDA; a bias-tee connected to the probe at port-1 (Fig. 1) served for both the application of the biasvoltage and the collection of the RF output, while the other probe at port-2 remained open ended.Lastly, the electrical signal was measured by a 50-GHz sampling oscilloscope.
Once the fiber-array was aligned and the signal collected, the responses from the individual diodes were monitored on the oscilloscope by blocking the inputs to three diodes at a time (Fig. 2).In order to achieve a phase match of all RF signals at the output, the electrical delay times from each photodiode along the transmission line were compensated for by changing the free-space delay path length.Monitoring the change in the arrival time of the reflection from the open-ended probe at port-2, the electrical group velocity, v 0 , could be determined.
The line characteristics of the CPW were then calculated from the measured v 0 .For the PDA-40 v 0 = 0.15c, where c is the speed of light in a vacuum, and the ϵ eff = 44.The PDA characteristic impedance, Z 0 , can be calculated by inserting the ϵ eff and the dimensions of the transmission line into the CPW design equations [6].The PDA can be considered as a capacitively loaded transmission line with [1]:  2 Ω, and 24.5 Ω, respectively.These calculations agree very well with the impedance values extracted from two-port S-parameter measurements (Fig. 3).Under ideal configurations, the Z 0 of the PDA would match the 50-Ω external load.For the devices under test, however, the mismatch should be noted.A final limiting bandwidth factor, the Bragg cut-off frequency, was estimated to be 60 GHz and above for all detectors [7].

Frequency domain characterization
Frequency domain measurements were performed to characterize the bandwidth and saturation current of the detectors.100% RF modulation of the optical input was achieved by heterodyning two balanced distributed feedback lasers operating near 1540 nm.The frequency of the modulation was controlled by varying the temperature, and subsequently the wavelength, of one of the lasers.To ensure constructive collection of RF signals at the output, the optical delays after the 1x4 fiber splitter were preserved from the previous time-domain measurements.Individual photodiode responsivities ranging from 0.66 A/W to 0.7 A/W were achieved using a single, lensed fiber with a focused beamspot much smaller than the diode diameters.During the normal operation of the array, the maximum coupling of the PDA is limited to 0.29 A/W for PDA-28 and 0.18 A/W for PDA-20 due to the large 35-μm spot-sizes of the lensed fiber array.

Frequency dependence
Figure 4 compares the relative frequency responses of a PDA-20 with the optional 50-Ω termination at port-2 (solid squares), and the discrete 40-μm detector (DPD-40) (empty squares).The detectors were operated with 5-V reverse bias at 140 mA.The comparison shows only a slight difference in 3-dB bandwidth, but a significant improvement for the PDA-20 at higher frequencies due to the combination of an added parallel resistor and the distributive nature of the device.While terminating the CPW with matched impedance preserves the wide-band aspect of the detector, the RF output is reduced by 6 dB at low frequency.To maximize the output power (forfeiting broadband capabilities) and ensure peaks at 35 GHz and 48 GHz the probe at port-2 was left open (Fig. 5).The reflection from the open terminal resulted in an interference pattern with a period proportional to the distance between diode and reflection point (~20 to ~30 mm).The detector was biased at −5 V and each diode operated at 30 mA.Initial measurements were performed on individual diodes (Fig. 5 solid lines) by blocking the three remaining optical inputs.The PDA was measured with the simultaneous illumination of all four diodes (Fig. 5 solid squares) and the response agrees well with the cumulative output (Fig. 5 dotted line) calculated by summing the currents from the isolated diode measurements.
In the cumulative response a pronounced envelope is present with a valley around 30 GHz.Below 10 GHz the periodic output patterns of all four diodes overlap well; however, due to the incremental difference in distance between diode and the open terminal at port-2, a mismatch of the four patterns occurs around 30 GHz.A second overlapping of PD1 and PD4 is evident at 48 GHz corresponding to a peak in the overall response.In the current PDA configuration, the four diodes will all constructively overlap again far above the Bragg limit.operating conditions.PDA-28, achieved two and half the saturation current of a discrete 28-μm device.We expect that with better impedance matching and spatial uniformity of the optical inputs, the PDA-28 can achieve three to four times the saturation current of its discrete counterpart.

Current dependence of RF output
Selecting frequencies corresponding to peaks in the spectral interference pattern, the PDA-28 demonstrates record output powers at 35-GHz and 48-GHz [Fig.6(b)].At 9.5-V bias and 35 GHz, the saturation current and RF output are 320 mA and 26.2 dBm, respectively.With 8-V reverse bias at 48 GHz, the saturation current is 350 mA and the RF output is 21.0 dBm.This is a marked improvement over previous results (Fig. 7) obtained from discrete detectors for this MUTC structure [9].Thermal dissipation improved in the array structures as well.PDA-28 dissipated 3.25 W of power before thermal failure: a record for high-speed detectors.PDA-20 dissipated a total of 2.14 W before thermal failure, an 8% improvement over DPD-40.

Conclusion
This paper describes back-illuminated four-element MUTC PDAs flip-chip bonded to AlN sub-mounts with an integrated transmission line.To the best of the authors' knowledge, PDA-28 produced the highest recorded RF output powers at 35 GHz and 48 GHz (26.2 dBm and 21.0 dBm, respectively).This was achieved by phase matching the combined outputs of each 28 μm-diameter element.Furthermore, this detector dissipated a total of 3.25 W (34.8 dBm) before thermal failure.

Fig. 1 .
Fig. 1.Superimposed photo of a segment of the MUTC chip (dash-line) with four diodes (PD1 -PD4) and the AlN submount with coplanar transmission line.

Fig. 2 .
Fig. 2. Individual element (PD1 -PD4) time-domain measurements of an open-ended PDA with an additional 2-cm long transmission line.Delay time from initial pulse to reflected pulse is given for each diode.where L' cpw (Z 0 /V 0 for a lossless transmission line) and C' cpw are per unit length inductance and capacitance of the transmission line, and C pd is the photodiode junction capacitance.Utilizing C pd values extracted from S-parameter measurement of discrete detectors and a ϵ eff of 10 (calculated for a transmission line sandwiched between AlN and InP), the resulting C cpw was 220 pF/m.The corresponding impedance of just the transmission line without diodes, ' ' cpw p cp w w c Z L C = , was 49 Ω, and was in good agreement with the designed value.For PDAs with d = 250 μm and element diameters of 20-, 28-, and 40-μm, the characteristic impedances were 37.2 Ω, 31.2Ω, and 24.5 Ω, respectively.These calculations agree very well with the impedance values extracted from two-port S-parameter measurements (Fig.3).Under ideal configurations, the Z 0 of the PDA would match the 50-Ω external load.For the devices under test, however, the mismatch should be noted.A final limiting bandwidth factor, the Bragg cut-off frequency, was estimated to be 60 GHz and above for all detectors[7].

Figure 6 (
Figure 6(a) shows the current dependence of two detectors with open terminations at port-2 and with equal active areas (PDA-20 and DPD-40).While they have similar saturation currents when biased with −5 V, the PDA shows a more pronounced super-linear region.Subsequently, PDA-20 demonstrates a significant 2.6-dB improvement over DPD-40 in RF output power at high currents.

Fig. 6 .
Fig. 6.(a) Comparison of current dependence of power output for PDA-20, DPD-40, and DPD-20 taken at 48 GHz.(b) PDA-28 RF output power versus current at both 38 GHz and 48 GHz.Dotted lines are ideal output powers for a 50-Ω load.Previously, a two-element PDA with 40-μm diameter elements and an integrated termination resistor demonstrated bandwidth saturation currents twice that of a similarly loaded 40-μm discrete detector[8].At 48 GHz the PDA-20, with its four elements, achieved saturation currents more than three times that of a discrete 20-μm detector under the same