Room-Temperature (RT) Extended Short-Wave Infrared (e-SWIR) Avalanche Photodiode (APD) with a 2.6 µm Cutoff Wavelength

Highly sensitive infrared photodetectors are needed in numerous sensing and imaging applications. In this paper, we report on extended short-wave infrared (e-SWIR) avalanche photodiodes (APDs) capable of operating at room temperature (RT). To extend the detection wavelength, the e-SWIR APD utilizes a higher indium (In) composition, specifically In0.3Ga0.7As0.25Sb0.75/GaSb heterostructures. The detection cut-off wavelength is successfully extended to 2.6 µm at RT, as verified by the Fourier Transform Infrared Spectrometer (FTIR) detection spectrum measurement at RT. The In0.3Ga0.7As0.25Sb0.75/GaSb heterostructures are lattice-matched to GaSb substrates, ensuring high material quality. The noise current at RT is analyzed and found to be the shot noise-limited at RT. The e-SWIR APD achieves a high multiplication gain of M~190 at a low bias of Vbias=− 2.5 V under illumination of a distributed feedback laser (DFB) with an emission wavelength of 2.3 µm. A high photoresponsivity of R>140 A/W is also achieved at the low bias of Vbias=−2.5 V. This type of highly sensitive e-SWIR APD, with a high internal gain capable of RT operation, provides enabling technology for e-SWIR sensing and imaging while significantly reducing size, weight, and power consumption (SWaP).

In this paper, we report on a new e-SWIR APD based on the In 0.3 Ga 0.7 As 0.25 Sb 0.7 5 /GaSb heterostructures lattice-matched to the GaSb subsrate.The lattice-matched heterostructure on the substrate offers high-quality materials with low strain-induced defects for low dark current.By increasing the indium (In) compostion to 0.3, the e-SWIR APD achieves a long detection cutoff wavelength of λ cuto f f = 2.6 µm.The new In 0.3 Ga 0.7 As 0.25 Sb 0.75 /GaSb heterostructure e-SWIR APD also features a separated absorption and multiplication (SAM) structure [30] to reduce the excess noise factor.By engineering the charge field of the In 0.3 Ga 0.7 As 0.25 Sb 0.75 /GaSb heterostructures, we also demonstrate a high multiplication gain of M ∼ 190 at a low bias of V bias = −2.5 V under the illumination of a distributed feedback laser (DFB) with an emission wavelength of 2.3 µm.The APD shows an excess noise factor of F ∼ 500, corresponding to a k-factor of k ∼ 0.003.Table 1 summarizes the comparison of this e-SWIR APD with previously reported APDs.

Device Structures, Material Growth, and the Device Fabrication
Figure 1 shows a cross-sectional, layer-by-layer diagram of the new In 0.3 Ga 0.7 As 0.25 Sb 0.75 / GaSb heterostructure e-SWIR APD.It consists of, from bottom (i.e., GaSb substrate) to top, n (5.0 × 10 18 /cm 3 ) tellurium (Te)-doped GaSb substrate, a 300 nanometer (nm) thick n + (2.0 × 10 19 /cm 3 ) Te-doped contacting layer, an undoped 500 nm In 0.3 Ga 0.7 As 0.25 Sb 0.75 active absorption layer, a p-1.5 × 10 17 /cm 3 beryllium (Be)-doped GaSb layer, an undoped GaSb layer (i-GaSb) as the avalanche region, and the p + 3.0 × 10 18 /cm 3 GaSb top contacting layer.The thickness of each layer is marked in the figure.The wider bandgap i-GaSb region was designed as the avalanche region to reduce dark current-and carrier generation/recombination (GR)-induced noise, thereby lowering both dark current and noise levels during the avalanche process [17,30,31].
In0.3Ga0.7As0.25Sb0.75active absorption layer, a p-1.5 10 /cm beryllium (Be)-doped GaSb layer, an undoped GaSb layer (i-GaSb) as the avalanche region, and the p + 3.0 10 /cm GaSb top contacting layer.The thickness of each layer is marked in the figure.The wider bandgap i-GaSb region was designed as the avalanche region to reduce dark current-and carrier generation/recombination (GR)-induced noise, thereby lowering both dark current and noise levels during the avalanche process [17,30,31].The In 0.3 Ga 0.7 As 0.25 Sb 0.75 /GaSb e-SWIR APD heterostructure was grown using a Veeco GEN Xplor Molecular Beam Epitaxy (MBE) at Tufts University Epitaxy Core facility (TEC).The GaSb wafer was ramped up to a substrate temperature of 620 • C and kept at this temperature for 10 min to remove the native oxide layer.After the de-oxidization, the substrate temperature was reduced to 542 • C. The growth temperature was kept at 542 • C throughout the growth of the material.The growth rate for the In 0.3 Ga 0.7 As 0.25 Sb 0.75 layer was 0.74 monolayers (ML) per second (i.e., 0.74 ML/s), and the growth rate for the GaSb layers was 0.52 ML/s.The growth rates were calibriated by reflection highenergy electron diffraction (RHEED) oscillation on the MBE machine right before the MBE growth, which ensured the accuracy of the layer thicknesses shown in Figure 1. Figure 2 shows the RHEED pattern after finishing the growth.The RHEED indicates high-quality lattice-matched In 0.3 Ga 0.7 As 0.25 Sb 0.75 /GaSb heterostructure growth on the GaSb substrate.The RHEED paterns were monitored throughout the material's growth period to ensure lattice-matching was achieved for all the layers.The In0.3Ga0.7As0.25Sb0.75/GaSbe-SWIR APD heterostructure was grown using a Veeco GEN Xplor Molecular Beam Epitaxy (MBE) at Tufts University Epitaxy Core facility (TEC).The GaSb wafer was ramped up to a substrate temperature of 620 °C and kept a this temperature for 10 minutes to remove the native oxide layer.After the de-oxidization the substrate temperature was reduced to 542 °C.The growth temperature was kept at 54 °C throughout the growth of the material.The growth rate for the In0.3Ga0.7As0.25Sb0.75laye was 0.74 monolayers (ML) per second (i.e., 0.74 ML/s), and the growth rate for the GaSb layers was 0.52 ML/s.The growth rates were calibriated by reflection high-energy electron diffraction (RHEED) oscillation on the MBE machine right before the MBE growth, which ensured the accuracy of the layer thicknesses shown in Figure 1. Figure 2 shows th RHEED pattern after finishing the growth.The RHEED indicates high-quality lattice matched In0.3Ga0.7As0.25Sb0.75/GaSbheterostructure growth on the GaSb substrate.Th RHEED paterns were monitored throughout the material's growth period to ensur lattice-matching was achieved for all the layers.After the MBE growth, the wafer was processed into 1.1 mm × 1.1 mm square mesa using standard photolithography, inductively coupled plasma (ICP) etching, electron beam (E-beam) metal deposition, and lift-off procedures.The ICP etching parameter were: BCl3, 10 standard cubic centimeters per minute (sccm), H2 5 sccm, 500 W ICP, 100 W RF, and a pressure of 2 millitorrs.The total etch depth was 850 nm, with a total etch tim of 9 min 30 s.The top and bottom contacts were 20 nm titanium (Ti) and 300 nm gold (Au) deposited using the E-beam metal deposition and lift-off processes.Figure 3 shows scanning electron microscope (SEM) image of the fabricated e-SWIR APD with a squar mesa and bonding wires on the top electrode.After the MBE growth, the wafer was processed into 1.1 mm × 1.1 mm square mesas using standard photolithography, inductively coupled plasma (ICP) etching, electron-beam (E-beam) metal deposition, and lift-off procedures.The ICP etching parameters were: BCl 3 , 10 standard cubic centimeters per minute (sccm), H 2 5 sccm, 500 W ICP, 100 W RF, and a pressure of 2 millitorrs.The total etch depth was 850 nm, with a total etch time of 9 min 30 s.The top and bottom contacts were 20 nm titanium (Ti) and 300 nm gold (Au), deposited using the E-beam metal deposition and lift-off processes.Figure 3 shows a scanning electron microscope (SEM) image of the fabricated e-SWIR APD with a square mesa and bonding wires on the top electrode.

Results and Discussions
The photocurrent spectra of the fabricated e-SWIR APD, referred to henceforth as the device, were measured using a Bruker INVENIO ® Fourier transform infrared (FTIR) spectrometer by Bruker Corporation at 40 Manning Rd, Billerica, MA, USA.The device replaced the internal DTGS detector of the FTIR, and the photocurrent signals were collected and transmitted to the FTIR through the equipment's external A/D converter unit.The device was tested through top-illumination from the surface-normal direction.The device was not polished on the backside.The spot size of the FTIR illumination light was estimated to be 2 mm in diameter.Figure 4 displays the measured FTIR photocurrent spectra at different bias voltages.At a low bias voltage  = −0.12(V) (dashed curve), the photocurrent spectra were primarily below the cutoff wavelength of a typical GaSb photodetector at 1.72 µm.Conversely, at a slightly higher bias voltage of  = −0.35(V) (solid curve), the device exhibited a longer detection wavelength with a cutoff wavelength of  = 2.6 μm.This corresponded to the collection of the photocurrent generated in the In0.3Ga0.7As0.25Sb0.75layer under the higher bias voltage.Note that the low-noise preamplifier used with the FTIR spectrometer had a low current overflow level, preventing the measurement of the photocurrent at higher biases.Nevertheless, the long cutoff wavelength of  = 2.6 μm at a low bias voltage of  = −0.35(V) still demonstrated the APD's capabilities in achieving the long cutoff wavelength of  = 2.6 μm at RT.

Results and Discussions
The photocurrent spectra of the fabricated e-SWIR APD, referred to henceforth as the device, were measured using a Bruker INVENIO ® Fourier transform infrared (FTIR) spectrometer by Bruker Corporation at 40 Manning Rd, Billerica, MA, USA.The device replaced the internal DTGS detector of the FTIR, and the photocurrent signals were collected and transmitted to the FTIR through the equipment's external A/D converter unit.The device was tested through top-illumination from the surface-normal direction.The device was not polished on the backside.The spot size of the FTIR illumination light was estimated to be 2 mm in diameter.Figure 4 displays the measured FTIR photocurrent spectra at different bias voltages.At a low bias voltage V bias = −0.12(V) (dashed curve), the photocurrent spectra were primarily below the cutoff wavelength of a typical GaSb photodetector at 1.72 µm.Conversely, at a slightly higher bias voltage of V bias = −0.35(V) (solid curve), the device exhibited a longer detection wavelength with a cutoff wavelength of λ cuto f f = 2.6 µm.This corresponded to the collection of the photocurrent generated in the In 0.3 Ga 0.7 As 0.25 Sb 0.75 layer under the higher bias voltage.Note that the low-noise preamplifier used with the FTIR spectrometer had a low current overflow level, preventing the measurement of the photocurrent at higher biases.Nevertheless, the long cutoff wavelength of λ cuto f f = 2.6 µm at a low bias voltage of V bias = −0.35(V) still demonstrated the APD's capabilities in achieving the long cutoff wavelength of λ cuto f f = 2.6 µm at RT.
Note that the photocurrent was nearly zero at a wavelength of 2.6 µm, with a bias of V bias = −0.35V.This occurred due to the electron band filling effect, also known as the Moss-Burstein effect [32][33][34], where the bias voltage shifts the Fermi-level and thus changes the occupations of the conduction bands and valence band.Similar phenomena have previously been reported and analyzed [35].
To obtain the photoresponse at higher bias voltages, the current versus bias voltage characteristics (i.e., I-V curves) of the APD were measured under the illumination of a distributed feedback (DFB) semiconductor laser (Eblana Photonics ® EP2327) by Eblana Photonics at 3 West Pier Campus, Dun Laoghaire, Co. Dublin, A96 A621, Ireland with an emission wavelength of λ laser = 2327 nm (referred to henceforth as the 2.3 µm laser) from the top illumination.Figure 5 shows the measured I-V curve under laser illumination (dashed curve) compared to the dark I-V curve (solid curve) without laser illumination.The APD-received laser power was estimated to be P r = 41 µW.The photocurrent under 2.3 µm laser illumination confirmed the APD's ability to detect IR light wavelengths exceeding 2.0 µm.Additionally, the steep slopes of the I-V curves at higher bias voltages than V bias < −1.5 (V) indicated the avalanche gain-induced high current.The avalanche process at the low bias voltage was attributed to the stepwise heterojunction design [29], which reduced the depletion width and thus led to a low breakdown voltage.It may have also been due to the breakdown at the sharp edges of the device.Note that the photocurrent was nearly zero at a wavelength of 2.6 µm, with a bias of  = −0.35V.This occurred due to the electron band filling effect, also known as the Moss-Burstein effect [32][33][34], where the bias voltage shifts the Fermi-level and thus changes the occupations of the conduction bands and valence band.Similar phenomena have previously been reported and analyzed [35].To obtain the photoresponse at higher bias voltages, the current versus bias voltage characteristics (i.e., I-V curves) of the APD were measured under the illumination of a distributed feedback (DFB) semiconductor laser (Eblana Photonics ® EP2327) by Eblana Photonics at 3 West Pier Campus, Dun Laoghaire, Co. Dublin, A96 A621, Ireland with an emission wavelength of  = 2327 nm (referred to henceforth as the 2.3 µm laser) from the top illumination.Figure 5 shows the measured I-V curve under laser illumination (dashed curve) compared to the dark I-V curve (solid curve) without laser illumination.The APD-received laser power was estimated to be  = 41 μW.The photocurrent under 2.3 µm laser illumination confirmed the APD's ability to detect IR light wavelengths exceeding 2.0 µm.Additionally, the steep slopes of the I-V curves at higher bias voltages than  −1.5 (V) indicated the avalanche gain-induced high current.The avalanche process at the low bias voltage was attributed to the stepwise heterojunction design [29], which reduced the depletion width and thus led to a low breakdown voltage.It may have also been due to the breakdown at the sharp edges of the device.To obtain the avalanche gains and the excess noise factor, we analyzed the noise components and the measured noise of the APD under different biases.The thermal noise current spectral density  /√ in A/√Hz can be expressed as: where  is the Boltzmann's constant,  is the absolution temperature in kelvin (°K), and  is the differential resistance of the APD, which can be calculated from the I-V charac-  The temporal response was determined by the RC constant of the e-SWIR APD, which was calculated to be ~ 40 nanoseconds (ns).
The noise spectral density  in dBV/√Hz of the device was measured using an SR770 FFT spectrum analyzer from Stanford Research Systems at 1290-D Reamwood Ave.Sunnyvale, CA, USA after converting the current signal to a voltage signal through  The VSD in dBV/ √ Hz was related to the noise spectral density i sh / √ B by: Combining Equations ( 2)-( 5), it was possible to calculate the VSD at different bias voltages and obtain the k factor value through curve-fitting with the measured VSD values.Figure 7 shows the calculated VSD at different bias voltages.The measured VSD values are marked in Figure 8   The photoresponsivity ℜ was defined as [37]: where  is the photocurrent and  is the laser power received by the device.Figure shows the photoresponsivity ℜ at different bias voltages.The noise equivalent power (NEP) in W/√Hz was defined as [37]: is the calculated current noise using the  factor value of  = 0.003 .Figure 1 shows the NEP at different bias voltages.The photoresponsivity R was defined as [37]: where I ph is the photocurrent and P r is the laser power received by the device.Figure 9 shows the photoresponsivity R at different bias voltages.The noise equivalent power (NEP) in W/ √ Hz was defined as [37]: i noise is the calculated current noise using the k factor value of k = 0.003.Figure 10 shows the NEP at different bias voltages.The specific photodetectivity (D*) in cm √ Hz/W was defined as [37]: where A is the area of the APD. Figure 11 shows the D* at different biases.The reverse bias increases caused a rise in noise level, resulting in a decrease in D*.
The noise equivalent power (NEP) in W/√Hz was defined as [37]: is the calculated current noise using the  factor value of  = 0.003 .Figure 1 shows the NEP at different bias voltages.The specific photodetectivity (D*) in cm√Hz/W was defined as [37]: where  is the area of the APD.

Figure 2 .
Figure 2. RHEED pattern after the growth was finished.

Figure 2 .
Figure 2. RHEED pattern after the growth was finished.

Figure 3 .
Figure 3. SEM image of the e-SWIR APD with bonding wires on the top electrode.

Figure 3 .
Figure 3. SEM image of the e-SWIR APD with bonding wires on the top electrode.

Figure 4 .
Figure 4. Measured FTIR photocurrent spectra of the APD at different reverse biases.At  = −0.12(V) (dashed curve), the photocurrent spectrum was mainly below the cutoff wavelength of 2.0 µm, whereas at a higher bias of  = −0.35(V) (solid curve), the photocurrent spectrum was extended to a longer cutoff wavelength of 2.6 µm.

Figure 4 . 12 Figure 5 .
Figure 4. Measured FTIR photocurrent spectra of the APD at different reverse biases.AtV bias = −0.12(V) (dashed curve), the photocurrent spectrum was mainly below the cutoff wavelength of 2.0 µm, whereas at a higher bias of V bias = −0.35(V) (solid curve), the photocurrent spectrum was extended to a longer cutoff wavelength of 2.6 µm.Micromachines 2023, 14, x FOR PEER REVIEW 6 of 12

Figure 5 .
Figure 5. Measured I-V curve (dashed curve) of the APD under 2.3 µm laser illumination compared to the dark I-V curve (solid curve).

Figure 6 .
Figure 6.Calculated avalanche gain of the device under different voltage biases.

Figure 6 .
Figure 6.Calculated avalanche gain of the device under different voltage biases.

Figure 7 .
Figure 7. Measured VSD of the APD under different voltage biases.

Figure 7 .
Figure 7. Measured VSD of the APD under different voltage biases.

Figure 8 .
Figure 8. Calculated VSD values at different bias voltages (solid curve).The circles are the measure VSD values after noise floor adjustment.

Figure 8 .
Figure 8. Calculated VSD values at different bias voltages (solid curve).The circles are the measured VSD values after noise floor adjustment.

Figure 9 . 1 Figure 10 .
Figure 9. Photoresponsivity R at different voltage biases.Micromachines 2023, 14, x FOR PEER REVIEW 10 of 1 Figure 11 shows the D* at different biases.The revers bias increases caused a rise in noise level, resulting in a decrease in D*.

Table 1 .
Comparison of this e-SWIR APD with previously reported APDs.
N/A: Not reported in the reference.