High Performance Predictable Quantum Efficient Detector Based on Induced-Junction Photodiodes Passivated with SiO2/SiNx

We performed a systematic study involving simulation and experimental techniques to develop induced-junction silicon photodetectors passivated with thermally grown SiO2 and plasma-enhanced chemical vapor deposited (PECVD) SiNx thin films that show a record high quantum efficiency. We investigated PECVD SiNx passivation and optimized the film deposition conditions to minimize the recombination losses at the silicon–dielectric interface as well as optical losses. Depositions with varied process parameters were carried out on test samples, followed by measurements of minority carrier lifetime, fixed charge density, and optical absorbance and reflectance. Subsequently, the surface recombination velocity, which is the limiting factor for internal quantum deficiency (IQD), was obtained for different film depositions via 2D simulations where the measured effective lifetime, fixed charge density, and substrate parameters were used as input. The quantum deficiency of induced-junction photodiodes that would be fabricated with a surface passivation of given characteristics was then estimated using improved 3D simulation models. A batch of induced-junction photodiodes was fabricated based on the passivation optimizations performed on test samples and predictions of simulations. Photodiodes passivated with PECVD SiNx film as well as with a stack of thermally grown SiO2 and PECVD SiNx films were fabricated. The photodiodes were assembled as light-trap detector with 7-reflections and their efficiency was tested with respect to a reference Predictable Quantum Efficient Detector (PQED) of known external quantum deficiency. The preliminary measurement results show that PQEDs based on our improved photodiodes passivated with stack of SiO2/SiNx have negligible quantum deficiencies with IQDs down to 1 ppm within 30 ppm measurement uncertainty.


Introduction
Silicon photodiodes based on inversion layer (induced-junction) have shown great promise for applications as a calibration standard due to their exceptionally high internal protect the silicon-dielectric interface from the degrading effects of moisture [24], which is an important asset for the stability of inversion-layer photodiodes. A dielectric stack consisting of a PECVD SiN x film deposited on a thin layer of thermally grown SiO 2 can potentially provide even superior passivation, leveraging both the excellent chemical passivation characteristics of thermal SiO 2 and excellent field-effect passivation characteristics of PECVD SiN x .
In our work, we have conducted a thorough study on the passivation and optical characteristics of different PECVD SiN x films and their stack with thermally grown SiO 2 . We employed 2D simulations to extract the surface recombination velocity and bulk lifetime and then 3D simulations to predict the response of an inversion-layer photodiode that would be fabricated with such passivation. Based on the predictions of the simulations, we have fabricated inversion-layer photodiodes using a passivation process that leads to unprecedently low IQDs and validated the results by testing photodiodes assembled in a light-trap PQED configuration.

Inversion Layer Photodiode Structure and Operation
The structure and operation of a p-type inversion-layer photodiode is illustrated in Figure 1. The surface of the active area is inverted by the positive fixed charges in the passivation dielectric. The electrical contact to the active area is made by a heavily phosphorous-doped n+ ring. Another n+ ring is implemented as a guard ring and biased at the same voltage as the active area. Due to the inversion of the surface by the positive field charges in the thermally grown field SiO 2 , the surface outside the active area is also inverted and therefore a p-type inter-electrode isolation structure, namely a p-stop ring, is incorporated into the structure, which remains floating. The physical edge of the photodiode is terminated with p+ doping in order to be able to apply all the biasing from the top side of the photodiode. The photodiode is operated in reverse-biased condition by applying the same positive voltage to the active area ring and the guard ring with respect to the p+ electrode. humidity barrier to protect the silicon-dielectric interface from the degrading e moisture [24], which is an important asset for the stability of inversion-layer phot A dielectric stack consisting of a PECVD SiNx film deposited on a thin layer of th grown SiO2 can potentially provide even superior passivation, leveraging b excellent chemical passivation characteristics of thermal SiO2 and excellent fie passivation characteristics of PECVD SiNx.
In our work, we have conducted a thorough study on the passivation and characteristics of different PECVD SiNx films and their stack with thermally grow We employed 2D simulations to extract the surface recombination velocity a lifetime and then 3D simulations to predict the response of an inversion-layer pho that would be fabricated with such passivation. Based on the predictions simulations, we have fabricated inversion-layer photodiodes using a passivation that leads to unprecedently low IQDs and validated the results by testing phot assembled in a light-trap PQED configuration.

Inversion Layer Photodiode Structure and Operation
The structure and operation of a p-type inversion-layer photodiode is illus Figure 1. The surface of the active area is inverted by the positive fixed charge passivation dielectric. The electrical contact to the active area is made by a phosphorous-doped n+ ring. Another n+ ring is implemented as a guard ring an at the same voltage as the active area. Due to the inversion of the surface by the field charges in the thermally grown field SiO2, the surface outside the active are inverted and therefore a p-type inter-electrode isolation structure, namely a p-st is incorporated into the structure, which remains floating. The physical edg photodiode is terminated with p+ doping in order to be able to apply all the biasi the top side of the photodiode. The photodiode is operated in reverse-biased cond applying the same positive voltage to the active area ring and the guard ring with to the p+ electrode.

Methodology
Thanks to improved 2D and 3D simulation models, one can predict the photo response of an inversion layer photodiode made with surface passivation materials of given specific properties. The predictability of response through accurate modelling of losses is not only Sensors 2021, 21, 7807 4 of 18 essential for the use of the photodiodes as a primary standard, but also allows one to optimize the passivation to improve the quantum efficiency of the detector without having to fabricate photodiodes with all possible variations of passivation. The relevant material properties are measured on passivation material grown or deposited on dummy test wafers. The measured effective minority carrier lifetime and fixed charge density are used to extract the surface recombination velocity (SRV), which is the limiting factor for internal quantum deficiency, via 2D simulation models. The SRV and measured optical properties are then used to predict through 3D simulation models the response of a PQED of inversion-layer photodiodes that would be made with such a passivation material and process. This is an extremely powerful and efficient method that allows one to maximize the quantum efficiency of a PQED through optimization of passivation material and process on dummy test wafers in a time-and cost-effective manner.
In our study, we focused on PECVD SiN x films and their stack with a thin layer of thermally grown SiO 2 to develop a material system and process that can provide excellent surface passivation for inversion photodiodes with minimal optical absorption and reflection losses. Since the photodiodes will be assembled into a PQED with a 7-reflection light-trap configuration to minimize reflection losses, the optimization of the optical characteristics of the passivation layer is performed accordingly.

Lifetime Measurements
The effective minority carrier lifetime, or simply effective lifetime, (τ e f f ), in a semiconductor contains contributions from both bulk and surface recombination, and can be expressed as: 1 The lifetime measurements were carried out on double-side-polished, high resistivity, p-type, float zone (FZ) silicon wafers passivated identically on both sides. Such test wafers, which have a long bulk lifetime (typically in the order of 10 ms), are chosen so that the measured effective lifetime is dominated by surface recombination.
For measurement of injection-dependent lifetime curves, the quasi-steady-state photoconductance (QSSPC) method was adopted, using a Sinton WCT-120TS lifetime tester [25]. With this setup, the excess carrier density is calculated from the conductivity of the passivated silicon wafer under illumination, as measured by an inductively coupled coil.
A BT Imaging LIS-R1 unit with an excitation wavelength of 808 nm was used for recording photoluminescence (PL) images of 6-inch surface passivated silicon wafers. The PL intensity was calibrated to the effective minority carrier lifetime based on a QSSPC measurement carried out in the central region of the wafer.

Fixed Charge Measurements
The fixed charge is determined by capacitance-voltage (C-V) measurements on metal-insulator-semiconductor (MIS) structures made on 6-inch, high resistivity, p-type FZ silicon wafers. The measurements were carried out in TSK A-PM-90A automatic probe station using an HP 4284A LCR meter. The AC voltage used in the CV measurements is 100 mV.
The MIS structures are fabricated on a 6-inch wafer and the wafer includes~120 circular MIS capacitors of an area of 0.035 cm 2 , which also gives a good picture of the uniformity of the film properties across the wafer. The large amount of data obtained from each measurement is converted and sorted using a Python script. The script then also calculates the parameters of interest including dielectric thickness (t d ), flat-band voltage (V fb ), and fixed charge density (Q f ). The flat-band voltage is calculated automatically with an iterative procedure and linear interpolation. The fixed charge is then obtained by: where C acc is the capacitance in the accumulation mode, A is the area of the capacitor, e is the elementary charge, and ϕ ms is the metal-semiconductor work function. The high resistivity silicon substrate adds a series resistance, retarding the charging of the capacitor, which results in a measured C acc lower than the theoretical value. The accumulation capacitance should be equal to the dielectric capacitance, which is given by where d is the permittivity of the dielectric. For improving the accuracy of the results, the measurements were done at low frequencies, making sure that the measured C acc is reasonably close to C d which can be calculated by obtaining the dielectric thickness and permittivity from ellipsometry measurements. Hysteresis measurements were also done to make sure that the measured charge is primarily due to fixed charges and that the mobile charge contribution is negligible.

Ellipsometry Measurements and Reflectance Simulations
Thickness, refractive index (n), and extinction coefficient (k) of the passivating dielectric films are obtained with a non-invasive method applying ellipsometry, an optical technique based on the measurement of elliptically polarized light [26]. Ellipsometric data were collected with a variable-angle spectroscopic ellipsometer (VASE) from J.A. Woollam [27]. Measurements were taken from 400 nm to 850 nm with 10 nm steps at five angles around the Brewster angle.
In order to minimize reflection losses, the PQED assembly consists of two photodiodes mounted in light-trap configuration with an angle of 15 • between the diodes, as depicted in Figure 2. In this configuration, the light beam undergoes 7 reflections (one at 0 • degree and 2 at 15 • , 30 • , 45 • ) following the same incoming and outgoing paths [5,28,29]. Using the results of the ellipsometry measurements performed on passivated test wafers, the reflectance for this PQED configuration was simulated as a function of wavelength for different dielectric thicknesses to determine the optimal thickness for minimum reflectance.
urements were done at low frequencies, making sure that the me close to Cd which can be calculated by obtaining the dielectric th from ellipsometry measurements. Hysteresis measurements wer that the measured charge is primarily due to fixed charges and contribution is negligible.

Ellipsometry Measurements and Reflectance Simulations
Thickness, refractive index (n), and extinction coefficient (k) tric films are obtained with a non-invasive method applying ellip nique based on the measurement of elliptically polarized light were collected with a variable-angle spectroscopic ellipsom Woollam [27]. Measurements were taken from 400 nm to 850 nm angles around the Brewster angle.
In order to minimize reflection losses, the PQED assembly odes mounted in light-trap configuration with an angle of 15° be picted in Figure 2. In this configuration, the light beam undergo degree and 2 at 15°, 30°, 45°) following the same incoming and Using the results of the ellipsometry measurements performed o the reflectance for this PQED configuration was simulated as a fu different dielectric thicknesses to determine the optimal thickn tance.

Modelling and Simulations of SRV and IQD
In order to predict the internal quantum deficiency of photodiode, two-step simulation was performed using the Genius Device Simulator from Cogenda [30]. The first step is to determine bulk lifetime τ bulk and surface recombination velocity (SRV) by using 2D simulation structure and the QSSPC lifetime measurement. This method has previously been presented by Stokkan et al. [31]. The 2D lifetime simulation structure is shown in Figure 3. The structure consists of a 500 µm thick, high resistivity silicon substrate passivated identically on both sides by SiO 2 and/or SiN x with thickness of two mesh elements. The width of the simulated structure is 7 mm. Excitation light with a wavelength of 808 nm is used for charge carrier generation in the simulation, which is the same wavelength as used in lifetime measurements. Figure 3. The structure consists of a 500 µm thick ated identically on both sides by SiO2 and/or SiN The width of the simulated structure is 7 mm. E nm is used for charge carrier generation in the si as used in lifetime measurements. where ∆ is the excess carrier density, an tron concentration at the surface, is the intri are the SRV of holes and electrons, respective as input simulation parameters, is calculate Various parameters from fabrication and tests of samples are used as input for the simulation such as doping type, doping concentration, wafer thickness, fixed charge density Q f , etc. The effective lifetime τ e f f can be calculated from the effective surface recombination velocity S e f f and τ bulk by Equation (1) which can be rewritten as [32]: 1 where W is the wafer thickness. The relation between S e f f and interface states is described by Shockley-Read-Hall theory and can be written as [31]: where ∆n bulk is the excess carrier density, p s and n s are respectively the hole and electron concentration at the surface, n i is the intrinsic carrier concentration, and S 0p and S 0n are the SRV of holes and electrons, respectively. With variation of S 0n , S 0p , and τ bulk as input simulation parameters, τ e f f is calculated by using Equation (3) and the carrier concentration is extracted from the simulation. The best fit of τ e f f as a function of carrier concentration between simulation and QSSPC lifetime measurement gives correct values of S 0n , S 0p , and τ bulk . The same values of S 0n and S 0p are used in the simulation. The second step is predicting the IQD of a photodiode, if manufactured by the given passivation recipe, by using a 3D simulation structure as shown in Figure 4 [9]. The structure consists of a doped silicon substrate, which is covered by a dielectric layer on the top and a doped layer on the bottom. Electrodes for electrical contacts are represented by n+ doping and p+ doping on top of the silicon substrate. Only 1/8 of the real device is used in the simulation due to computational limiting reasons, and the symmetric boundary conditions are applied to obtain the response of the whole photodiode. Besides the fabrication parameters which include fixed charge, doping concentration, etc., the fitted values S 0n , S 0p , and τ bulk from the 2D simulation are used as input parameters in the 3D simulation model.  From the 3D simulation, the total surface recombination , total bu tion , and total photon generation are extracted and the IQD can by the following equation: = +

Passivation Process and Sample Preparation
The SiNx films are deposited at SINTEF MiNaLab using a conventiona capacitively coupled plasma (CCP) type PECVD reactor from SPTS [33] placed on a chuck and SiNx films are deposited on the wafer by the react precursors SiH4, NH3, and N2. The plasma enhances the reaction rate, allow sition to be performed at low temperatures. A SiNx passivation process w by using a pressure of 2000 mTorr, RF power of 40 W, chuck temperature electrode spacing of 20 mm. A gas flow rate of 60 sccm was used for NH3 3600 sccm was used for N2. The SiNx deposition process developed origin gas flow ratio of 1:1) exhibits good passivation characteristics but too high o tion. The SiH4:NH3 gas flow ratio was then varied by changing the SiH4 fl keeping all other process parameters the same to achieve the desired optic tics without degrading the passivation performance. Samples with SiH4:NH tios of 1:3, 2:3, 2:1, and 3:1 were prepared for testing. Samples with SiNx fi of 150 nm and 65 nm were prepared.
For our experiments, 500 µm thick, double-side polished FZ silicon waf surface orientation from Topsil were used. The wafers were p-type with re fication of 5000-12,000 Ω cm. The wafers were cleaned using the standard (NH3+HCl), followed by a diluted HF treatment prior to oxidation or SiNx d wafers for lifetime measurements and ellipsometry measurements were pa tically on both sides. The wafers used for fixed charge measurements we planted with boron on the backside for a good ohmic contact and passivat front side. To make MIS structures, these wafers were then sputtered on b 1.2 µm thick aluminum and the aluminum on the dielectric-coated side with a photolithography process followed by wet etching. From the 3D simulation, the total surface recombination R sur f , total bulk recombination R bulk , and total photon generation G opt are extracted and the IQD can be calculated by the following equation:

Passivation Process and Sample Preparation
The SiN x films are deposited at SINTEF MiNaLab using a conventional parallel plate capacitively coupled plasma (CCP) type PECVD reactor from SPTS [33]. The wafer is placed on a chuck and SiN x films are deposited on the wafer by the reaction of gaseous precursors SiH 4 , NH 3, and N 2 . The plasma enhances the reaction rate, allowing the deposition to be performed at low temperatures. A SiN x passivation process was established by using a pressure of 2000 mTorr, RF power of 40 W, chuck temperature of 350 • C, and electrode spacing of 20 mm. A gas flow rate of 60 sccm was used for NH 3 and SiH 4 , and 3600 sccm was used for N 2 . The SiN x deposition process developed originally (SiH 4 :NH 3 gas flow ratio of 1:1) exhibits good passivation characteristics but too high optical absorption. The SiH 4 :NH 3 gas flow ratio was then varied by changing the SiH 4 flow rate while keeping all other process parameters the same to achieve the desired optical characteristics without degrading the passivation performance. Samples with SiH 4 :NH 3 gas flow ratios of 1:3, 2:3, 2:1, and 3:1 were prepared for testing. Samples with SiN x film thicknesses of 150 nm and 65 nm were prepared.
For our experiments, 500 µm thick, double-side polished FZ silicon wafers with <100> surface orientation from Topsil were used. The wafers were p-type with resistivity specification of 5000-12,000 Ω cm. The wafers were cleaned using the standard RCA process (NH 3 +HCl), followed by a diluted HF treatment prior to oxidation or SiN x deposition. The wafers for lifetime measurements and ellipsometry measurements were passivated identically on both sides. The wafers used for fixed charge measurements were heavily implanted with boron on the backside for a good ohmic contact and passivated only on the front side. To make MIS structures, these wafers were then sputtered on both sides with 1.2 µm thick aluminum and the aluminum on the dielectric-coated side was patterned with a photolithography process followed by wet etching.
Some wafers were oxidized with a thin layer (6 nm) of thermally grown SiO 2 prior to PECVD SiN x deposition. Some of the oxidized samples were annealed prior to SiN x deposition. One of the annealing processes is a gettering process which is employed in actual photodiode manufacturing to remove impurities and defects from silicon bulk. The wafers were also annealed in forming gas at 350 • C for 30 min, which is known to reduce the interface traps at the Si/SiO 2 interface without affecting fixed charge density [34]. Annealing in forming gas was performed prior to SiN x deposition as SiN x is highly impermeable to H 2 at such low temperatures.

Results
The original SiN x deposition recipe developed using a SiH 4 :NH 3 gas flow ratio of 1:1 was tested for SiN x film thickness of 150 nm on unoxidized silicon. An effective lifetime of~4 ms at an injection level ∆n of 6.5 × 10 14 cm −3 and fixed charge density of 3 × 10 12 cm −2 were measured. However, ellipsometry measurements showed that this film has a high refractive index and absorbance with n and k values of 2.12 and 1.9 × 10 −3 at a wavelength of 632 nm, respectively. The gas flow ratio of SiH 4 :NH 3 was then varied to deposit SiN x with different stoichiometric ratios while keeping all other parameters the same. Both the refractive index and extinction coefficient are strongly dependent on the stoichiometric ratio of SiN x and increase with increasing Si content. The recipe with highest SiH 4 :NH 3 flow ratio of 3:1, yielded an n value of 2.67 and k value of 4.8 × 10 −2 at 632 nm. The recipe with SiH 4 :NH 3 = 1:3 yielded the best optical properties with n = 1.84 and k = 0 at 632 nm. The lifetime and fixed charge measurements showed that the passivation properties of the SiN x are not affected to any significant degree by the SiH 4 :NH 3 gas flow ratio within the range of investigation. One would expect better passivation characteristics with increasing SiH 4 flow in the deposition process due to the higher hydrogen content, which should provide more effective passivation of silicon dangling bonds [22]. However, the measurements showed that this was not the case, as probably even the recipe with lowest SiH 4 flow was still providing enough hydrogen. These investigations showed that the recipe with SiH 4 :NH 3 gas flow ratio of 1:3 leads to both optimal passivation and optical properties, with an effective lifetime of~4 ms at injection level ∆n of 7.1 × 10 14 cm −3 , a fixed charge density of~4 × 10 12 cm −2 , n = 1.84 and k = 0. This SiN x deposition recipe was then chosen for further investigation and development.
The p-polarization reflectance of a 7-reflection trap configuration PQED consisting of two photodiodes passivated with the optimized SiN x deposition recipe was simulated. Figure 5 shows the simulated reflectance as function of wavelength for different SiN x thicknesses. Figure 6 shows the mean and maximum reflectance versus SiN x thickness. These simulation results suggest an optimum SiN x thickness of~65 nm to minimize reflection losses. These simulations were also performed for a passivation dielectric stack including 6 nm thermally grown SiO 2 underneath PECVD SiN x . Figure 7 indicates that the SiN x thickness that leads to minimum reflectance becomes~60 nm in presence of 6 nm SiO 2 .
The original SiNx deposition recipe developed using a SiH4:NH3 gas flow ratio o was tested for SiNx film thickness of 150 nm on unoxidized silicon. An effective lifet of ~4 ms at an injection level Δn of 6.5 × 10 14 cm −3 and fixed charge density of ~3 × 10 12 c were measured. However, ellipsometry measurements showed that this film has a h refractive index and absorbance with n and k values of 2.12 and 1.9 × 10 −3 at a wavelen of 632 nm, respectively. The gas flow ratio of SiH4:NH3 was then varied to deposit S with different stoichiometric ratios while keeping all other parameters the same. Both refractive index and extinction coefficient are strongly dependent on the stoichiome ratio of SiNx and increase with increasing Si content. The recipe with highest SiH4:N flow ratio of 3:1, yielded an n value of 2.67 and k value of 4.8 × 10 −2 at 632 nm. The re with SiH4:NH3 = 1:3 yielded the best optical properties with n = 1.84 and k = 0 at 632 The lifetime and fixed charge measurements showed that the passivation properties of SiNx are not affected to any significant degree by the SiH4:NH3 gas flow ratio within range of investigation. One would expect better passivation characteristics with incr ing SiH4 flow in the deposition process due to the higher hydrogen content, which sho provide more effective passivation of silicon dangling bonds [22]. However, the meas ments showed that this was not the case, as probably even the recipe with lowest S flow was still providing enough hydrogen. These investigations showed that the re with SiH4:NH3 gas flow ratio of 1:3 leads to both optimal passivation and optical pro ties, with an effective lifetime of ~4 ms at injection level Δn of 7.1 × 10 14 cm −3 , a fixed cha density of ~4 × 10 12 cm −2 , n = 1.84 and k = 0. This SiNx deposition recipe was then cho for further investigation and development.
The p-polarization reflectance of a 7-reflection trap configuration PQED consistin two photodiodes passivated with the optimized SiNx deposition recipe was simula Figure 5 shows the simulated reflectance as function of wavelength for different S thicknesses. Figure 6 shows the mean and maximum reflectance versus SiNx thickn These simulation results suggest an optimum SiNx thickness of ~65 nm to minimize flection losses. These simulations were also performed for a passivation dielectric s including 6 nm thermally grown SiO2 underneath PECVD SiNx. Figure 7 indicates that SiNx thickness that leads to minimum reflectance becomes ~60 nm in presence of 6 SiO2.      The original SiNx tests were performed on a film thickness of 15 confirm that the reduction of the film's thickness to 60-65 nm does no sivation characteristics, new test samples with SiNx thickness of 65 n prepared for testing in addition to samples with the original SiNx thi This new batch also included test samples with SiNx deposited on 6 nm oxide and exposed to different annealing conditions. The process deta The original SiN x tests were performed on a film thickness of 150 nm. In order to confirm that the reduction of the film's thickness to 60-65 nm does not degrade the passivation characteristics, new test samples with SiN x thickness of 65 nm were then also prepared for testing in addition to samples with the original SiN x thickness of 150 nm. This new batch also included test samples with SiN x deposited on 6 nm thermally grown oxide and exposed to different annealing conditions. The process details of the samples are presented in Table 1. The QSSPC measurement results of all eight test samples that have gone through the processes described in Table 1 are shown in Figure 8. The wafers with a buffer layer of 6 nm thermally grown SiO 2 underneath SiN x clearly have a longer minority carrier lifetime, particularly at higher injection levels. The samples without SiO 2 (E5, E6, E7, and E8) show a similar trend among each other. The samples that have thicker SiN x and have gone through annealing in forming gas exhibit somewhat higher lifetime values over the entire range of injection density. This is in line with expectation, but the difference is quite small. The samples with a SiO 2 layer (E1, E2, E3, and E4) also exhibit similar lifetime values among each other, but the trend as a function of injection level is slightly different at very low and very high injection levels.

E3
6 The QSSPC measurement results of all eight test samples that h processes described in Table 1 are shown in Figure 8. The wafers w nm thermally grown SiO2 underneath SiNx clearly have a longer mi particularly at higher injection levels. The samples without SiO2 (E5 a similar trend among each other. The samples that have thicke through annealing in forming gas exhibit somewhat higher lifetime range of injection density. This is in line with expectation, but the di The samples with a SiO2 layer (E1, E2, E3, and E4) also exhibit s among each other, but the trend as a function of injection level is sli low and very high injection levels.  Table 1.
The PL images showing the spatial variation of effective life were taken for all eight wafers. The PL images were calibrated using lifetime at injection levels shown in Table 2. The calibrated PL im Figure 9. The samples with SiO2/SiNx passivation exhibit ~3 times h compared with the samples with SiNx passivation but inferior uni this insignificant non-uniformity is not understood.  Table 1.
The PL images showing the spatial variation of effective lifetime across the wafer were taken for all eight wafers. The PL images were calibrated using the measured QSSPC lifetime at injection levels shown in Table 2. The calibrated PL images are presented in Figure 9. The samples with SiO 2 /SiN x passivation exhibit~3 times higher effective lifetime compared with the samples with SiN x passivation but inferior uniformity. The origin of this insignificant non-uniformity is not understood.    Table 1.
The color bar shows τeff in µs.
The C-V measurements have shown that reducing the SiNx film thickness from 150 nm to 65 nm to minimize the reflectance losses does not affect the fixed charge density to any significant degree, eliminating processes E1, E3, E5, and E7. The gettering process after oxidation, which is an annealing process normally implemented to reduce silicon bulk impurities, has been proven to have no negative effect on the fixed charge density by comparing the C-V characteristics of samples E2 and E4. Annealing in forming gas after SiNx deposition is a process normally carried out in the end of photodiode fabrication for metal sintering, and skipping this process (namely, E8) has not shown any advantage, either. Therefore, passivation processes E2 and E6 were overall proven to be the most viable candidates among the group with thermal oxide and the group without thermal oxide, respectively. Figure 10 shows the measured C-V characteristics of MIS structures fabricated using passivation processes E2 and E6 at a frequency of 1 kHz. As mentioned in Section 3.1.2, the high resistivity of the silicon substrate retards the capacitance charging at high frequencies, leading to lower measured Cacc than the theoretical value. Therefore, the measurements were performed at different frequencies to determine the optimum measurement frequency that yields reasonably accurate Cacc with minimal noise, which was found to be 1 kHz. The fixed charge densities were determined from the C-V characteristics to be 1.3 × 10 12 cm −2 and 4.0 × 10 12 cm −2 for E2 and E6, respectively. The passivation is expected to become gradually more dominated by SiO2 properties than SiNx with increasing buffer  Table 1. The color bar shows τ eff in µs.
The C-V measurements have shown that reducing the SiN x film thickness from 150 nm to 65 nm to minimize the reflectance losses does not affect the fixed charge density to any significant degree, eliminating processes E1, E3, E5, and E7. The gettering process after oxidation, which is an annealing process normally implemented to reduce silicon bulk impurities, has been proven to have no negative effect on the fixed charge density by comparing the C-V characteristics of samples E2 and E4. Annealing in forming gas after SiN x deposition is a process normally carried out in the end of photodiode fabrication for metal sintering, and skipping this process (namely, E8) has not shown any advantage, either. Therefore, passivation processes E2 and E6 were overall proven to be the most viable candidates among the group with thermal oxide and the group without thermal oxide, respectively. Figure 10 shows the measured C-V characteristics of MIS structures fabricated using passivation processes E2 and E6 at a frequency of 1 kHz. As mentioned in Section 3.1.2, the high resistivity of the silicon substrate retards the capacitance charging at high frequencies, leading to lower measured C acc than the theoretical value. Therefore, the measurements were performed at different frequencies to determine the optimum measurement frequency that yields reasonably accurate C acc with minimal noise, which was found to be 1 kHz. The fixed charge densities were determined from the C-V characteristics to be 1.3 × 10 12 cm −2 and 4.0 × 10 12 cm −2 for E2 and E6, respectively. The passivation is expected to become gradually more dominated by SiO 2 properties than SiN x with increasing buffer SiO 2 layer thickness. Therefore, a higher measured fixed charge density in bare SiN x film as compared with a stack of SiO 2 and SiN x shouldn't come as a surprise. Indeed, it has been reported that the overall fixed charge density in stack of Al 2 O 3 (which is known to have negative fixed charge) with SiO 2 turns positive as the interfacial SiO 2 thickness becomes more than~5 nm [35].
Sensors 2021, 21, x FOR PEER REVIEW 12 SiO2 layer thickness. Therefore, a higher measured fixed charge density in bare SiNx as compared with a stack of SiO2 and SiNx shouldn't come as a surprise. Indeed, it been reported that the overall fixed charge density in stack of Al2O3 (which is know have negative fixed charge) with SiO2 turns positive as the interfacial SiO2 thicknes comes more than ~5 nm [35]. Figure 10. Capacitance-voltage (C-V) measurement results of MIS capacitors prepared with passivation processes E2 (6 nm SiO2 + 65 nm SiNx) and E6 (65 nm SiNx) as described in Table 1 frequency of 1 kHz. Figure 11 shows the fit of the 2D simulation with the measured injection-depen effective lifetime for passivation processes E2 and E6. The simulated lifetime curves obtained using the following fabrication and test parameters as inputs: substrate do type (p-type), substrate doping concentration (2 × 10 12 cm −3 ), substrate thickness (500 µ and measured fixed charges (1.3 × 10 12 cm −2 for E2 and 4.0 × 10 12 cm −2 for E6). The simul lifetime is in an excellent agreement with the measured one up to an injection level o 10 15 cm −3 for E2. For E6, on the other hand, the simulated lifetime fits the measured life well only at lower injection levels and diverges from the measured one at injection le above 2 × 10 14 cm −3 . The bulk lifetime and surface recombination velocities obtained f these fits are reported in Table 3. Figure 11. Injection dependent effective minority carrier lifetime τeff (Δn) of test samples passiv with processes E2 (6 nm SiO2 + 65 nm SiNx) and E6 (65 nm SiNx) as described in Table 1 with s lation fits to extract SRV and τbulk.  Table 1, at a frequency of 1 kHz. Figure 11 shows the fit of the 2D simulation with the measured injection-dependent effective lifetime for passivation processes E2 and E6. The simulated lifetime curves are obtained using the following fabrication and test parameters as inputs: substrate doping type (p-type), substrate doping concentration (2 × 10 12 cm −3 ), substrate thickness (500 µm), and measured fixed charges (1.3 × 10 12 cm −2 for E2 and 4.0 × 10 12 cm −2 for E6). The simulated lifetime is in an excellent agreement with the measured one up to an injection level of 2 × 10 15 cm −3 for E2. For E6, on the other hand, the simulated lifetime fits the measured lifetime well only at lower injection levels and diverges from the measured one at injection levels above 2 × 10 14 cm −3 . The bulk lifetime and surface recombination velocities obtained from these fits are reported in Table 3.
SiO2 layer thickness. Therefore, a higher measured fixed charge density in bare SiNx as compared with a stack of SiO2 and SiNx shouldn't come as a surprise. Indeed, i been reported that the overall fixed charge density in stack of Al2O3 (which is know have negative fixed charge) with SiO2 turns positive as the interfacial SiO2 thicknes comes more than ~5 nm [35].  Table 1 frequency of 1 kHz. Figure 11 shows the fit of the 2D simulation with the measured injection-depen effective lifetime for passivation processes E2 and E6. The simulated lifetime curve obtained using the following fabrication and test parameters as inputs: substrate do type (p-type), substrate doping concentration (2 × 10 12 cm −3 ), substrate thickness (500 and measured fixed charges (1.3 × 10 12 cm −2 for E2 and 4.0 × 10 12 cm −2 for E6). The simu lifetime is in an excellent agreement with the measured one up to an injection level o 10 15 cm −3 for E2. For E6, on the other hand, the simulated lifetime fits the measured life well only at lower injection levels and diverges from the measured one at injection l above 2 × 10 14 cm −3 . The bulk lifetime and surface recombination velocities obtained these fits are reported in Table 3. Figure 11. Injection dependent effective minority carrier lifetime τeff (Δn) of test samples passi with processes E2 (6 nm SiO2 + 65 nm SiNx) and E6 (65 nm SiNx) as described in Table 1 with s lation fits to extract SRV and τbulk. Figure 11. Injection dependent effective minority carrier lifetime τ eff (∆n) of test samples passivated with processes E2 (6 nm SiO 2 + 65 nm SiN x ) and E6 (65 nm SiN x ) as described in Table 1 with simulation fits to extract SRV and τ bulk .

Predicted Photodiode Responsivity
Using the extracted SRV and bulk lifetime (τ bulk ), the IQD of a photodiode that would be made with the corresponding passivation was simulated using 3D simulation models. Figure 12 shows the simulated IQD as a function of reverse bias voltage for a photodiode passivated with the process E2 or E6. Figure 13 shows the simulated IQD as a function of wavelength at reverse bias voltage of 5 V. The passivation process E2 (SiO 2 + SiN x ) is estimated to lead to IQDs as low as~2 ppm at 488 nm and to more than a factor of 10 lower IQD compared with the passivation process E6 (SiN x ).  Simulation SRV (S0n, S0p) 1.5 × 10 3 cm/s 6.0 × 10 4 cm/s

Predicted Photodiode Responsivity
Using the extracted SRV and bulk lifetime (τbulk), the IQD of a photodiode that would be made with the corresponding passivation was simulated using 3D simulation models. Figure 12 shows the simulated IQD as a function of reverse bias voltage for a photodiode passivated with the process E2 or E6. Figure 13 shows the simulated IQD as a function of wavelength at reverse bias voltage of 5 V. The passivation process E2 (SiO2 + SiNx) is estimated to lead to IQDs as low as ~2 ppm at 488 nm and to more than a factor of 10 lower IQD compared with the passivation process E6 (SiNx). Figure 12. Simulated IQD as a function of reverse bias voltage for p-type inversion-layer photodiode that would be fabricated with passivation E2 (6 nm SiO2 + 65 nm SiNx) and E6 (65 nm SiNx) as described in Table 1. The simulations were performed at a wavelength of 488 nm. Figure 12. Simulated IQD as a function of reverse bias voltage for p-type inversion-layer photodiode that would be fabricated with passivation E2 (6 nm SiO 2 + 65 nm SiN x ) and E6 (65 nm SiN x ) as described in Table 1. The simulations were performed at a wavelength of 488 nm.
Sensors 2021, 21, x FOR PEER REVIEW 14 of 18 Figure 13. Simulated IQD as a function of wavelength for p-type inversion-layer photodiode that would be fabricated with passivation E2 (6 nm SiO2 + 65 nm SiNx) and E6 (65 nm SiNx) as described in Table 1. The simulations were performed at a reverse bias voltage of 5 V.

Photodiode Design and Fabrication
Photodiodes of active area size of 11 mm × 11 mm, 11 mm × 14 mm, and 11 mm × 22 mm were included in the wafer layout. The photodiode design matches the cross-sectional Figure 13. Simulated IQD as a function of wavelength for p-type inversion-layer photodiode that would be fabricated with passivation E2 (6 nm SiO 2 + 65 nm SiN x ) and E6 (65 nm SiN x ) as described in Table 1. The simulations were performed at a reverse bias voltage of 5 V.

Photodiode Design and Fabrication
Photodiodes of active area size of 11 mm × 11 mm, 11 mm × 14 mm, and 11 mm × 22 mm were included in the wafer layout. The photodiode design matches the cross-sectional sketch depicted in Figure 1 and includes n+ ring for electrical contact to the active area, p-stop ring to isolate n+ electrodes, n+ guard ring to shield the active area from the sensor edge effects, and p+ ring around the physical edge of the photodiode for electrical contact to the p+ electrode from the top side of the photodiode.
The photodiode fabrication was carried out in SINTEF MiNaLab cleanroom facilities. The same wafer material as the one used for the passivation tests discussed in the previous section was used for photodiode fabrication (500 µm thick, p-type FZ wafers with resistivity range of 5000-12,000 Ωcm and <100> surface orientation). The wafers were first thermally oxidized to grow thick SiO 2 that is used as field oxide throughout the processing. The wafers were processed with 5 photomask layers that were used to define the areas of p+ implantation, n+ implantation, active area passivation dielectric, contact holes through SiO 2 , and metallization. The p+ and n+ electrodes were made with boron and phosphorous implantation, respectively. On most wafers, the active area passivation was done using the process optimized in the previous section to minimize reflection losses and IQD, namely, process E2 (6 nm SiO 2 + 65 nm SiN x ). On a couple of wafers, process E6 (65 nm SiN x ) was implemented for comparison and validation of the test and simulation results. The wafers were metallized with 1.2 µm aluminum on both sides. Figure 14 shows a picture of a completed wafer of photodiodes.

PQED Assembly
The photodiodes with physical dimensions of 13 mm × 24 mm (acti sions of 11 mm × 22 mm) were adhesively bonded to silicon carriers havi The I-V characteristics of all photodiodes were measured at room temperature at wafer level. The measurements were carried out by using the biasing scheme shown in Figure 1. The measurement results show some variation across the wafer and between wafers, with typical active area leakage current values in the range of 1-5 nA/cm 2 at reverse bias voltage of −20 V.

PQED Assembly
The photodiodes with physical dimensions of 13 mm × 24 mm (active area dimensions of 11 mm × 22 mm) were adhesively bonded to silicon carriers having dimensions of 15 mm × 38 mm (Figure 15). On the remaining space on each carrier, a 12 mm × 12 mm printed circuit board (PCB) was glued. The PCB and chip carrier have 3 mm diameter holes for screw assembly into PQED mechanics. Wire bonds were made between the photodiode and the PCB using an Au ball-wedge bonder with 17.5 µm wire diameter. U.FL connectors soldered to PCB enables electrical connections to external instruments. todiode and the PCB using an Au ball-wedge bonder w connectors soldered to PCB enables electrical connection Both the PCB and photodiode were aligned and place pico die bonder. Stycast 1266 epoxy was used for all adhe viscosity, easily spreads out during bonding and forms th todiode is parallel to the carrier. The PCB metallization w nickel immersion gold (ENIG) coating in order to facilita  Both the PCB and photodiode were aligned and placed on the carrier using a Finetech pico die bonder. Stycast 1266 epoxy was used for all adhesive bonds, which, due to its low viscosity, easily spreads out during bonding and forms thin layers, ensuring that the photodiode is parallel to the carrier. The PCB metallization was 35 µm copper with electroless nickel immersion gold (ENIG) coating in order to facilitate wire bonding.
The packaged photodiodes were mounted in a mechanical wedge trap structure to form windowless seven-reflection PQEDs ( Figure 2). The mechanical housings were purged with a constant flow of N 2 to prevent dust particles from entering inside the detector. Alignment of the PQEDs was carried out with a laser beam at 405 nm on the same optical path as in the measurements at 488 nm. Reflection from the PQED is visible to the eye at 405 nm but not at 488 nm for laser beam power of less than 1 mW.

IQD Measurements and Comparison with Simulated Response
Spatial uniformity of the PQEDs was measured at 488 nm with a 1/e 2 beam diameter of 2.4 mm and power of 100 µW to 150 µW ( Figure 16). All PQEDs were measured with 5 V reverse bias. The responsivity of the PQEDs is constant within approximately 50 ppm peak-to-peak over the central area with a diameter of 1 mm to 3 mm.
The absolute responsivity was determined at the center of each PQED against a reference PQED with a predicted external quantum deficiency of 37 −9 +140 ppm [5,7,36], where the asymmetric uncertainty limits correspond to 95% confidence level. The responsivity estimates of the reference PQED are validated via another PQED with measurements against a cryogenic radiometer [6]. The temporal stability of the reference PQED was confirmed by responsivity measurements over a period of almost ten years [37].

IQD Measurements and Comparison with Simulated Response
Spatial uniformity of the PQEDs was measured at 488 nm with a 1/e 2 beam diameter of 2.4 mm and power of 100 µW to 150 µW ( Figure 16). All PQEDs were measured with 5 V reverse bias. The responsivity of the PQEDs is constant within approximately 50 ppm peak-to-peak over the central area with a diameter of 1 mm to 3 mm. The absolute responsivity was determined at the center of each PQED against a reference PQED with a predicted external quantum deficiency of 37−9 +140 ppm [5,7,36], where the asymmetric uncertainty limits correspond to 95% confidence level. The responsivity estimates of the reference PQED are validated via another PQED with measurements against a cryogenic radiometer [6]. The temporal stability of the reference PQED was confirmed by responsivity measurements over a period of almost ten years [37].
The PQED with purely SiNx passivation layer had 9 ppm higher responsivity than the reference PQED, while the PQEDs made of SiO2/SiNx stack photodiodes P18-55-45 and P18-54-44 had 36 ppm and 16 ppm higher responsivity than the reference PQED, respectively. The standard uncertainty of 30 ppm of the above results is determined by the photocurrent ratio measurement and spatial uniformity of the detectors (Figure 16). After taking into account the negligible reflectance loss of SiNx and SiO2/SiNx stack PQEDs, these results translate into measured IQD values of 28 ppm (SiNx), 1 ppm (P18-55-45), and 21 ppm (P18-54-44). This confirms that the manufactured photodiodes with the SiO2/SiNx stack have a very low IQD.
The measured IQD values of the PQEDs are consistent with the predicted responsivity from the lifetime measurements of the passivation test samples given in Figures 12 and  13. The measured IQD values are so small that their deviation from zero is beyond the usual capabilities of radiometric measurements and substantially more work is needed to validate the predicted charge carrier losses of the SiO2/SiNx stack PQEDs at such uncertainty levels. The evidence we have managed to achieve supports the lifetime curve fit prediction method and shows that the photodiodes have a record low external quantum deficiency. The PQED with purely SiN x passivation layer had 9 ppm higher responsivity than the reference PQED, while the PQEDs made of SiO 2 /SiN x stack photodiodes P18-55-45 and P18-54-44 had 36 ppm and 16 ppm higher responsivity than the reference PQED, respectively. The standard uncertainty of 30 ppm of the above results is determined by the photocurrent ratio measurement and spatial uniformity of the detectors (Figure 16). After taking into account the negligible reflectance loss of SiN x and SiO 2 /SiN x stack PQEDs, these results translate into measured IQD values of 28 ppm (SiN x ), 1 ppm (P18-55-45), and 21 ppm (P18-54-44). This confirms that the manufactured photodiodes with the SiO 2 /SiN x stack have a very low IQD.
The measured IQD values of the PQEDs are consistent with the predicted responsivity from the lifetime measurements of the passivation test samples given in Figures 12 and 13. The measured IQD values are so small that their deviation from zero is beyond the usual capabilities of radiometric measurements and substantially more work is needed to validate the predicted charge carrier losses of the SiO 2 /SiN x stack PQEDs at such uncertainty levels. The evidence we have managed to achieve supports the lifetime curve fit prediction method and shows that the photodiodes have a record low external quantum deficiency.

Conclusions
We thoroughly investigated and optimized PECVD SiN x passivation with and without a buffer layer of 6 nm thermally grown SiO 2 for surface recombination and optical losses to develop PQEDs with negligible quantum deficiency based on p-type-induced-junction photodiodes. We demonstrated excellent passivation characteristics (Q f > 1 × 10 12 cm −2 and SRV ≈ 1500 cm/s) and optical characteristics (n = 1.84 at wavelength of 632 nm and k = 0 for the whole range between 400 nm to 850 nm) with an optimized stack of thermal SiO 2 and PECVD SiN x . PQEDs assembled of photodiodes with optimized passivation were tested against a reference PQED and showed record high quantum efficiencies with IQDs down to 1 ppm with 30 ppm measurement uncertainty. Within the measurement uncertainties, the efficiency results are in line with the predictions of the simulations based on the passivation and optical parameters extracted from test samples.