InP- and GaAs-Based Photonic Power Converters Under O-Band Laser Illumination: Performance Analysis and Comparison

Photonic power converters (PPCs), which convert narrow-band light to electricity, are essential components in power-by-light systems. When designed for telecommunications wavelengths such as the O-band, near 1310 nm, the devices are well-suited to power-over-fiber applications. Despite the potential for very high power conversion efficiencies (<inline-formula><tex-math notation="LaTeX">$\mathbf {>50\%}$</tex-math></inline-formula>), PPCs can be adversely affected by high-intensity nonuniform illumination conditions. In this work, we characterized two O-band PPC designs based on: high-quality InGaAsP absorber material lattice-matched to an InP substrate, and metamorphic InGaAs absorber material lattice-mismatched to a GaAs substrate, a more cost-effective and scalable alternative. We measured each device under O-band laser illumination with five beam profiles having peak-to-average ratios ranging from 2 to 11. Both devices were insensitive to the beam uniformity for input illumination with average irradiance below 2 W/cm<inline-formula><tex-math notation="LaTeX">$\mathbf {^{2}}$</tex-math></inline-formula> over their 5.4-mm<inline-formula><tex-math notation="LaTeX">$\mathbf {^{2}}$</tex-math></inline-formula> active areas, but exhibited better open-circuit voltages under larger, more uniform illumination profiles at higher incident powers. Measured efficiencies reached 52.8% and 48.7% for the lattice-matched and mismatched devices, respectively. Distributed circuit modeling results suggested that both lateral conduction losses and localized heating effects were responsible for the measured dependence on beam-size. Our work demonstrates the potential for O-band PPCs, presenting two highly efficient designs suitable for powering devices requiring <inline-formula><tex-math notation="LaTeX">$\mathbf {\lesssim 250}$</tex-math></inline-formula> mW, with an appropriate illumination profile.


I. INTRODUCTION
P OWER-BY-LIGHT systems, also known as photonic or optical power systems, offer an alternative to conventional power delivery over conductive wires, with several advantages for sensitive electronics applications. Unlike the flow of electric current in wires, light neither produces nor is it affected by electromagnetic interference, and it does not spark. Power-bylight systems are well-suited to power sensitive devices that require electrical isolation, or to supply power in electrically noisy or hazardous environments [1], [2], [3]. They can also be integrated with optical communications systems for simultaneous power and data transmission [4], [5], [6], [7]. The application space for photonic power is diverse and rapidly growing. Examples include transmission line [8] and wind turbine monitoring [9], biomedical implants [10], [11], powering the Internet of Things [12], radio-over-fiber [7], [13], DC-DC power conversion [14], and many others [1], [2], [15].
Photonic power systems transmit power in the form of narrowband light, which is generated by a laser or LED. After transmission, the light is converted back to electrical power using a photonic power converter (PPC), which generates electricity through the photovoltaic effect [1], [2]. PPCs are similar to solar cells in their operation, however, they are designed to convert narrow-band light rather than the broad-band solar spectrum. The efficiency of a power-by-light system has three key components: the electrical-to-optical power conversion efficiency of the light source, the photonic power transmission efficiency, and the optical-to-electrical power conversion efficiency of the PPC. This article is concerned with the second and third efficiency components.
O-band PPCs have been less studied than their 850-nm-band counterparts. To date, only a handful of O-band PPC devices have been experimentally characterized and reported in the literature [31], [32], [33], [34], based on InP and GaAs substrates. Other studies have investigated pathways to increased power conversion efficiencies in O-band PPCs [35], [36]. For InPbased devices, high-quality lattice-matched quaternary materials such as InGaAsP or InAlGaAs can provide high absorptance in the O-band. For GaAs-based PPCs, there is no convenient lattice-matched absorber material for O-band operation and lattice-mismatched growth processes are required [32], [37]. Highly efficient GaAs-based devices represent a path toward large-scale manufacturing of O-band PPCs, as GaAs substrates are less expensive and available in larger sizes than InP. Noting that materials with smaller bandgaps yield lower efficiencies than their larger-bandgap counterparts, detailed-balance modeling of O-band devices predicts theoretical power conversion efficiencies exceeding 67% in the radiative limit [36]. This compares to the predicted efficiency of 73.7% for a comparable GaAs PPC operating at 830 nm in the radiative limit at room temperature [38].
In this article, we analyze high-performance O-band PPCs based on both InP and GaAs substrates. We examine the impact of the illumination profile on device performance and employ a distributed circuit model (DCM) to better understand the mechanisms responsible for reduced performance of the 6.4-mm 2 PPCs under nonuniform illumination with average irradiance in the range of 2-10 W/cm 2 .

A. Design and Fabrication of PPC Devices
The design and fabrication of PPCs for operation under 1310-nm laser illumination was carried out at Fraunhofer ISE [32]. The devices were grown by metal-organic vapor phase epitaxy (MOVPE) using an AIXTRON G4 2800TM reactor. Two designs were considered, similar to existing literature [39], [40], [41], both using an n-p homojunction architecture for the active layers sandwiched between wider bandgap front and back surface field layers, which prevent the loss of minority carriers generated in the absorber region through electrical passivation. The first design, shown in Fig. 1(a), was grown on a 4-inch p-type InP substrate making use of lattice-matched In 0.32 Ga 0.68 As 0.69 P 0.31 (InGaAsP/InP) with a measured bandgap of E g = 0.890 eV for the absorber material. The second design, shown in Fig. 1(b), was grown on a 4-inch n-type GaAs substrate. Lattice-mismatched Ga 0.56 In 0.44 As (MM-GaInAs/GaAs) with a measured bandgap of E g = 0.855 eV was used for the absorber layers and the change in lattice constant was accommodated using an n-type GaInP-based step-graded metamorphic (MM) buffer [37], [39]. This polarity was chosen due to the superior mobility of majority carriers in n-type material. The change in doping polarity between the active n-p junction and the n-type MM buffer and substrate, which employ well-known growth techniques, required the addition of a tunnel diode (TD) between the active layers of the device and the MM buffer to swap the doping polarity.
Following epitaxy, ohmic contacts and an antireflection coating (ARC) were deposited. The front metallization consisted of two busbars on opposite sides of the cell and a comb grid design with parallel fingers. Wet-chemical mesa etching was used to separate individual PPC devices on the wafer. All devices were fabricated with a total area of A = 6.4 mm 2 , an active area between the busbars of 2.2 × 2.2 mm 2 , and a nominal designated area of A nom = 5.4 mm 2 , which is the total area excluding the busbars.

B. Illuminated Electrical Measurements
We characterized the performance of the fabricated PPC devices on a gold-plated temperature-controlled copper chuck at 25 • C. Current-voltage (I-V ) characteristics were measured with a Keithley 2420 source-meter using a four-wire configuration, ramping voltage in the forward direction. Two probes collected current from the top contact, with one for each busbar, and a third probe contacted one busbar to measure voltage. Two leads connected to the chuck conducted current and measured voltage at the back of the device.
The devices were illuminated using a fiber-coupled laser with a central wavelength of 1319 nm and a full width at half maximum of 9 nm. The laser output was coupled into an optical fiber with a core diameter of 400 µm and a numerical aperture of 0.22. The fiber output was collimated and then focused onto the device using a lens positioned directly above the measurement stage as shown in Fig. 1(c), forming a circular spot with radially decreasing light intensity.
The size of the laser spot on the device was controlled by manipulating the lens-to-device distance. We measured the beam profile at the device surface using the knife-edge profiling technique [42] revealing a symmetrical super-Gaussian beam shape where the intensity in the xy-plane was given by where I peak is the peak beam intensity, w is the 1/e 2 beam radius, and β is the beam shape parameter. For a perfectly Gaussian beam, β = 2. Knife-edge measurements and corresponding super-Gaussian profiles are shown in Fig. 1(d)-(g) for two beam sizes. We characterize the uniformity of these illumination profiles by the peak-to-average ratio (PAR), defined as the peak beam intensity divided by the average irradiance (I avg ) across the nominal designated area  I  ILLUMINATION PARAMETERS FOR THE FIVE SPOT SIZES   TABLE II  DCM PARAMETERS FOR INGAASP/INP PPC where P in is the power incident on the nominal designated area.
We measured each device under five beam sizes, listed in Table I. For the largest beam size (2w = 2.3 mm), the 1/e 2 diameter was not fully contained within the active area of the device. In this case, P in was adjusted to discount the portion of the beam that extended beyond the active area. The laser was allowed to stabilize before beginning measurements, which were completed in 1-2 s for each I-V curve. The laser beam was blocked between measurements for average irradiance > 1 W/cm 2 , where heating began to impact device behavior. For comparison, I-V measurements were also performed during the 1-ms irradiance plateau of a xenon flash lamp, providing broadband and uniform illumination.

C. Distributed Circuit Modeling
To better understand the impact of nonuniform illumination, we used a DCM [43] to simulate uniform and Gaussian (β = 2) illumination profiles on the lattice-matched InGaAsP/InP PPC. The simulated device area was 6.4 mm 2 with an active area of 2.2 × 2.2 mm 2 and busbars extending to the edges of the device on either side of the active region. The area was divided into discrete unit cells, each represented by an equivalent circuit described by parameters that were fit to or derived from experimental measurements. Simulation input parameters are given in Table II.
The n-p junction unit cells in the DCM were described by the two-diode model where J is current density, J ph is the photocurrent, J 01 and J 02 are the dark saturation current densities, k is the Boltzmann constant, T = 298 K is the temperature, q is the elementary charge, and R SH is the shunt resistance. The values of J 01 , J 02 , and R SH were found by fitting (3) to experimental J-V curves  (4), see Table III for fitting parameters.
for the InGaAsP/InP device, averaging over 20 measurements under average laser irradiance below 0.57 W/cm 2 , well below the threshold of ∼ 2 W/cm 2 above which nonuniform illumination effects were observed. The large value obtained for R SH signifies that shunting was insignificant for this device. The n-p junction unit cells were interconnected with resistive elements in the lateral and vertical directions to form a complete circuit model representing the device [43], [44], shown in Fig. 1(h). The sheet resistances and the top contact resistance listed in Table II were measured experimentally. The bottom contact resistance is an estimate based on other devices and had a negligible impact on the simulated I-V curve. Resistances in the metal busbars and gridlines were approximated using the resistivity of silver (ρ Ag ).

A. Current-Voltage Characteristics
Sample current-voltage (J-V ) characteristics are shown in Fig. 2(a) and (b) for the InGaAsP/InP and MM-GaInAs/GaAs PPCs, respectively, where the current density J avg is averaged over the total device area. We measured the devices under a range of incident laser powers between 7 and 526 mW, corresponding to average irradiance between 0.13 and 9.74 W/cm 2 . For clarity, only five incident powers and two beam diameters are shown in Fig. 2(a) and (b).
For both PPCs, the short-circuit current density (J SC ) scales linearly with increasing incident power, regardless of the beam diameter, with average responsivities of 0.912 and 0.957 A/W for the InGaAsP/InP and MM-GaInAs/GaAs PPCs, respectively. Smaller spot sizes are associated with reduced fill factors (FF) and open-circuit voltages (V OC ) compared to the larger spot sizes for both PPCs.
The larger responsivity of the lattice-mismatched device is due to its smaller bandgap and correspondingly larger optical absorption in the O-band. The smaller bandgap of the MM-GaInAs/GaAs PPC also gives rise to smaller (V OC ) values in general compared to the InGaAsP/InP PPC. By adjusting the alloy compositions, the bandgaps of both InGaAsP and MM-GaInAs can be tuned to optimize the trade-off between absorptance and voltage. In the case of the MM material, smaller bandgaps correspond to greater lattice mismatch between GaInAs and the GaAs substrate, increasing the difficulty in maintaining high material quality [39].
For the MM-GaInAs/GaAs PPC, its TD [see Fig. 1(b)] impacts the shape of the J-V curves and each maximum power point. TDs are characterized by a peak tunneling current density, above which tunneling is no longer possible and the TD switches from low-resistance tunneling to higher-resistance thermal diffusion (the dominant electrical transport mechanism in standard pn-junction diodes) [45]. When embedded in a photovoltaic device, the switch from tunneling to thermal diffusion significantly increases the device's total series resistance [46], [47], [48], [49], [50], [51], [52]. Fig. 2(b) shows this effect for the MM-GaInAs/GaAs PPC, where a dip in total current is observed for bias voltages > 0.1 V. Note that the J-V curves were measured in the forward direction from J SC to V OC , and hysteresis is often observed in the J-V curves of photovoltaic cells that contain TDs [47], [50], [51]. Measuring in the forward direction gives a more accurate assessment of the threshold beyond, which the TD begins to influence the cell performance [47]. In Fig. 2(b), we observe that the dip in the J-V curve becomes more pronounced with increasing illumination power and is more significant for smaller beam sizes, which are characterized by higher PAR values and correspondingly higher peak illumination intensities (and peak current densities) at the center of the laser spot. The TD begins to impact the J-V curves for the lattice-mismatched MM-GaInAs PPC on GaAs when the peak illumination intensity I peak 20 W/cm 2 . Whereas the maximum power is strongly impacted by TD and resistive effects, the open-circuit voltage is determined largely by the bandgap and material quality. One measure of quality is the bandgap-voltage offset, defined as W OC = E g /q − V OC . Lower values of W OC indicate better device performance [53]. W OC for both PPCs is plotted in Fig. 2(c) as a function of the short-circuit current density averaged over the total device area (J SC,avg ). Data are shown for three of the five laser spot sizes. For comparison, data measured under broad spectrum uniform illumination is also shown for the exact same InGaAsP/InP device [32] and for a substantially similar MM-GaInAs/GaAs device with the same layer structure.
In general, V OC increases logarithmically with increasing J SC,avg because of the larger carrier concentrations in photovoltaic devices under higher irradiance. Correspondingly, the obtained W OC decreases logarithmically with increasing J SC,avg for J SC,avg < 1 A/cm 2 . When J SC,avg > 1 A/cm 2 , the measurements deviate from the logarithmic trend. Fits were performed for the laser-illuminated measurements in the region J SC,avg ≤ 1 A/cm 2 using the nonideal one-diode equation where n is the ideality factor and J 0 is the average dark saturation current density over the total area. We used the onediode equation here to allow for direct comparison between the two PPC designs, because the experimental J-V curve of the MM-GaInAs/GaAs PPC with an embedded TD was not well-represented by the two-diode equation (equation 3) across the fitting regime. The one-diode fit with n = 1, however, was highly representative of the J-V characteristics throughout this range. For the InGaAsP/InP PPC, the one-diode and two-diode models were found to agree with each other within the fitting regime.
Fitting parameters for each device are given in Table III, including average values for each device. The fits are consistent with the measured J-V characteristics in the power quadrant for both PPCs within the fitting regime J SC,avg < 1 A/cm 2 . For comparison, ideal parameters in the radiative limit are also shown, where n ideal = 1 and J 0,ideal is calculated from the measured bandgaps assuming a cell temperature of T = 298 K [53] The lattice-matched InGaAsP/InP PPC exhibits an average ideality factor of 1.06 and an average dark saturation current density of 2.4 × 10 −11 A/cm 2 . The smaller bandgap of the MM-GaInAs/GaAs PPC is expected to produce a larger dark saturation current [see (5)], however, comparison between the corresponding experimental and ideal parameters in Table III shows that its behavior is less ideal compared to the latticematched device. The average ideality factor is 1.35 and the average dark saturation current density is 3.2 × 10 −8 A/cm 2 . Even though a direct quantitative comparison of dark saturation current densities is complicated by the significant difference in ideality factor, this data shows more pronounced nonradiative recombination for the MM-GaInAs/GaAs PPC, as expected for a lattice-mismatched device [54].

B. Impact of Nonuniform Illumination on V OC
For all laser-illuminated measurements, the W OC values deviated from the logarithmic trend when J SC,avg > 1 A/cm 2 . More significant deviations were observed for smaller beam diameters with larger PARs. We used a DCM to estimate the impact of lateral conduction losses on V OC for the InGaAsP/InP PPC. Under nonuniform illumination, charge carriers diffuse from regions of high to low irradiance, producing lateral current flows that increase with PAR, even under external open-circuit conditions [44], [55]. This causes resistive losses that are more significant for small beam sizes and high incident powers.
The model results for Gaussian (2w = 1.35 mm) and uniform illumination profiles are shown in Fig. 3(a) with experimental data for comparison. The DCM shows good agreement with uniform illumination measurements, within 1% for all measured data points. For the Gaussian beam profile, the DCM predicts a deviation in V OC away from the logarithmic trend at higher currents, similar to experimental measurements. However, this deviation was much more pronounced in the measured data than in the DCM. This discrepancy suggests that localized heating, which was not accounted for in the DCM, contributed in part to the deviation in measured V OC . As described in Section II-B, the uniform illumination measurements were collected using a flash measurement to mitigate heating. As such, the uniform data are directly comparable to and agree well with the model, which does not account for temperature effects.
Temperature dependence in the parameter J 0 leads to an approximately linear reduction in V OC with increasing temperature [56], [57], [58], [59], [60]. From Dupré et al. [57], we assume that where γ ≈ 3 describes the temperature sensitivity when bulk and surface recombination processes are dominant such that the ideality factor n ≈ 1, which is satisfied for the lattice-matched PPC device (see Table III). We estimated dV OC dT by taking T = 298 K, E g = 0.890 eV, and using the simulated values of V OC at 298 K from the DCM. For V OC,298 K = 0.7 V, corresponding to J SC ∼ 4 A/cm 2 , we obtain dV OC dT ≈ −0.9 mV/K, which is approximately −0.13 %/K. This is in the range of what was observed experimentally for GaInP and GaAs solar cells [56]. Considering the difference in voltage between the model and experiment, ΔV OC , and using dV OC dT from (6), we found an effective difference in average cell temperature (ΔT eff ) that increased with J SC,avg . The result is shown in Fig. 3(b). Given the significant nonuniformity in the illumination profile, localized heating from the laser is expected to produce a similar nonuniformity in the temperature profile across the device, with more significant heating at the center of the illumination spot. ΔT eff does not account for this nonuniformity, which depends on the thermal and electrical conductivities of the cell materials and the thermal contact efficiency between the cell and the temperature-controlled chuck.
In Fig. 3 Fig. 3(e) and (f). Under uniform illumination, small dips in voltage are found beneath each gridline, where the cell is shaded. Across the entire device, a voltage deviation of only 15.3 mV is calculated, with the peak voltage in the center and the lowest voltage in the shaded regions beneath the busbars. A more significant voltage gradient is simulated under Gaussian illumination. The peak voltage occurs at the center of the device where the illumination intensity is largest. The lowest voltage is found at the edges of the device, which are not illuminated, and beneath the busbars. Under this highly nonuniform illumination condition, the total difference in voltage between the central peak and V MP is 61.7 mV, a factor of four larger than the maximum voltage gradient under uniform illumination.
Despite the significantly larger peak voltage simulated under Gaussian illumination, the value of V MP is largest under uniform illumination [see Fig. 3(c) and (d)]. This is because lateral conduction losses are more significant under nonuniform illumination, and the darker regions at the cell's periphery bring down V MP . Given that the DCM does not account for localized heating, the temperature-induced voltage drop [see Fig. 3(a) and (b)] may be even more significant than lateral conduction losses in a physical device, resulting in a larger drop in V MP under nonuniform illumination than predicted by the model. Fig. 2(c) shows that deviations from the logarithmic trend in W OC at higher illumination powers are more significant for the lattice-mismatched MM-GaInAs/GaAs PPC compared to the lattice-matched InGaAsP/InP PPC. This suggests that the MM-GaInAs material has lower lateral conductivity compared to the lattice-matched InGaAsP, resulting in more significant resistive losses. This observation is consistent with the sheet resistances measured in the window layers of each device, which are 7.7 − 8 Ω/ and 39 − 49 Ω/ for the InGaAsP/InP and the MM-GaInAs/GaAs PPCs, respectively. Improving the lateral conductivity within the metamorphic device may help to enhance its voltage under nonuniform illumination. Lower thermal conductivity within the MM-GaInAs/GaAs PPC compared to the lattice-matched device may also contribute to its increased W OC , but this effect has not been quantified.

C. Optical-to-Electrical Conversion Efficiency
Though analysis of the V OC is critical to understanding PPC performance, it is not the only factor influencing device efficiency. The optical-to-electrical conversion efficiency is plotted in Fig. 4 as a function of average irradiance for the five laser beam diameters. The largest efficiency for the InGaAsP/InP PPC was 52.8% ± 1.3% (abs.), measured at an average irradiance of 5.94 ± 0.14 W/cm 2 . The largest efficiency for the MM-GaInAs/GaAs PPC was 48.7% ± 1.1% (abs.) at 8.84 ± 0.20 W/cm 2 . The primary source of error is uncertainty in the incident beam power (see Appendix). The maximum efficiency measurements for both devices were recorded for the largest beam diameter of 2w = 2.3 mm with the correspondingly smallest nonuniformity with PAR = 2. Further efficiency enhancements could be obtained through design optimization and the integration of a back-reflector to enhance the output voltage through photon recycling and optical resonance effects [16], [35].
In general, both PPCs were tolerant to nonuniform illumination profiles up to an average irradiance of ∼ 2 W/cm 2 . Efficiency losses observed for smaller beam diameters at higher incident powers can be explained by the reduction in V OC and fill factor observed with increasing PAR. Note that, like V OC , fill factor is also impacted by localized heating effects and series resistance losses. The observed trends indicate that more uniform laser illumination profiles would improve PPC performance under high illumination powers. Transitioning to multijunction device structures would also improve the tolerance to nonuniform illumination by lowering the current density through the device structure, mitigating resistive losses [43]. These measures have the potential to enhance the performance of PPCs at any operation wavelength, but are especially important for long-wavelength PPCs that operate in the O-or C-bands because of the larger number of incident photons per watt of power compared to shorter-wavelength PPCs.

IV. CONCLUSION
We measured two 6.4-mm 2 O-band PPCs under laser illumination. The lattice-matched InGaAsP/InP device exhibited a record power conversion efficiency for the O-band of 52.8% ± 1.2% (abs). A robust efficiency of 48.7% ± 1.1% (abs.) was measured for the MM-GaInAs/GaAs PPC, demonstrating the potential for GaAs as a more cost-effective, scalable alternative to InP for O-band PPCs. Even higher efficiencies could likely be achieved for both material systems through design optimization and the integration of a back-reflector.
For both PPCs, the nonuniformity of the illumination profiles had minimal impact on device performance for I avg < 2 W/cm 2 and, for J SC,avg < 1 A/cm 2 , a logarithmic relation was observed between the W OC and J SC,avg in accordance with the nonideal diode equation. A deviation from this trend under nonuniform illumination occurred for J SC,avg > 1 A/cm 2 , which can be explained by: 1) lateral conduction losses, as shown by the DCM, and 2) localized heating. Overall, both PPCs are suitable for powering sensors and other devices requiring 250 mW, with an appropriate illumination profile. Better performance at higher illumination powers could be achieved by engineering a more uniform illumination profile, optimizing the front metallization, or adopting multi-junction device architectures.

APPENDIX UNCERTAINTY IN MEASURED EFFICIENCY
The dominant source of error in the determination of PPC optical-to-electrical conversion efficiencies is uncertainty in the incident illumination power. During the experiment, the incident laser power was determined from the laser diode current, which was controlled directly. The relationship between optical power and diode current was established through a calibration experiment in which the current was stepped up in discrete intervals from the threshold current and the resultant laser power was measured using a Newport 918D-IR-OD2 photodiode detector. Neutral density (ND) filters were used to attenuate the beam power and their transparencies were measured experimentally. The measured efficiencies are shown in Fig. 5 along with the associated error bars, which account for the following sources of error. 1) ±2% calibration uncertainty for the photodiode detector.