We have been developing the monolithic active pixel detector XRPIX onboard the future x-ray astronomical satellite FORCE. XRPIX is composed of complementary metal-oxide-semiconductor pixel circuits, SiO2 insulator, and Si sensor by utilizing the silicon-on-insulator (SOI) technology. When the semiconductor detector is operated in orbit, it suffers from radiation damage due to x-rays emitted from celestial objects as well as cosmic rays. From previous studies, positive charges trapped in the SiO2 insulator are known to cause degradation of the detector performance. To improve the radiation hardness, we developed XRPIX equipped with a double-SOI (D-SOI) structure, introducing an additional silicon layer in the SiO2 insulator. This structure is aimed at compensating for the effect of the trapped positive charges. Although the radiation hardness of the D-SOI detectors to cosmic rays has been evaluated, the radiation effect due to x-ray irradiation has not been evaluated. Thus, we then conduct an x-ray irradiation experiment using an x-ray generator with a total dose of 10 krad at the SiO2 insulator, equivalent to 7 years in orbit. As a result of this experiment, the energy resolution in full-width half maximum for the 5.9 keV x-ray degrades by 17.8 % ± 2.8 % and the dark current increases by 89 % ± 13 % . We also investigate the physical mechanism of the increase in the dark current due to x-ray irradiation using technology computer-aided design simulation. It is found that the increase in the dark current can be explained by the increase in the interface state density at the Si / SiO2 interface. |
1.IntroductionSince the 1990s, the charge-coupled device (CCD) has been the standard detector used in x-ray astronomy satellites. The x-ray CCD has an excellent position and energy resolution; however, it has a poor time resolution of a few seconds and a narrow observable energy band of 0.3 to 10 keV. To realize broadband and high-sensitivity x-ray observation, we have been developing the monolithic active pixel detector XRPIX onboard the future x-ray astronomical satellite FORCE.1,2 FORCE will be equipped with two x-ray supermirrors, with angular resolution that will be better than 15″ in half-power diameter. The focal plane detector is composed of two stacks of Si sensors (XRPIX) and CdTe sensors, and these detectors cover the x-ray energy ranging from 1 to 79 keV. XRPIX is composed of complementary metal-oxide-semiconductor (CMOS) pixel circuits, the insulator called the buried oxide (BOX) layer, and the Si sensor by utilizing the silicon-on-insulator (SOI) technology.3 This makes it possible to implement Si with low and high resistivity in the circuit and sensor layers, respectively. Because the sensor layer has high resistivity, the depletion layer thickness can be a few hundreds of micrometers. In addition, the CMOS circuit in each pixel has a self-trigger function, so only the pixels for which an x-ray is incident can be read out, achieving a time resolution of . XRPIX will suffer from radiation damage due to the irradiation of bright x-rays from compact stars as well as high-energy cosmic rays. When the detector is irradiated by charged particles or x-rays, electron-hole pairs are produced in Si and in . In the former case, carriers are collected by electrodes. On the other hand, in the insulator, electrons are immediately collected to electrodes, but some parts of holes are trapped there because of their low mobility compared with that of electrons ( at 300 K).4 It is known that the trapped positive charges in the insulator cause a shift of threshold voltages of CMOS pixel circuits and degradation of the detector performance.5 To improve the radiation hardness, we introduce a double-SOI (D-SOI) structure, which has an additional Si layer called “middle-Si” in the insulator, as shown in Fig. 1. It is effective against radiation damage because the negative voltage applied on the middle-Si negates the effect of trapped positive charges in the insulator caused by radiation exposure.6 In a prior experiment, to evaluate the effect of cosmic rays mainly composed of high energy protons, we evaluated the radiation hardness of the D-SOI detector by 6-MeV proton beam irradiation.6 In this experiment, we found that, even after irradiation of , degradation of the energy resolution was as small as 7%. Moreover, we found that the gain degradation can be quantitatively explained by the sense-node capacitance increased by the trapped positive charges. On the other hand, because high-energy x-rays up to 79 keV emitted from celestial objects are focused with x-ray supermirrors at the focal point with a high angular resolution of , the focused high-energy x-rays can cause serious radiation damage against XRPIX.7 Thus, in this paper, we conduct an x-ray irradiation experiment to evaluate the x-ray radiation hardness of the D-SOI detector.8 The x-ray irradiation experiment is described in Sec. 2, and we show its results in Sec. 3. In Sec. 4, we discuss the possible cause of the degradation of detector performance, mainly regarding the dark current using device simulation. Section 5 provides the conclusions of this study. 2.X-Ray Irradiation ExperimentWe conducted an x-ray irradiation experiment on D-SOI detector, called XRPIX6C. Figure 1 and Table 1 show the schematic cross-sectional structure and chip design of XRPIX6C, respectively. The sensor layer, with a thickness of , is -type Si bulk, and its resistivity is . With this thickness, the back-bias voltage should be higher than for full depletion. In this experiment, we applied a back-bias voltage . In this device, each pixel is isolated from each other by a -stop and has a sense node surrounded by a buried -well (BNW). BNW was introduced to prevent interference between the sensor layer and the circuit layer.9 A buried -well (BPW) was introduced to generate a lateral electric field structure from the pixel boundary to the sense node so that electric charges collect at the sense node. It is also effective in suppressing dark current by covering the interface. We call this device D-SOI because an additional Si layer called middle-Si is introduced in the BOX layer. It compensates for the effect of positive charges trapped in the BOX layer by biasing negatively. We applied a negative voltage of to middle-Si during the experiment. Table 1The chip design of XRPIX6C.
Figure 2 shows the schematic view of our experimental setup. XRPIX6C was installed in a vacuum chamber and cooled down to to reduce the shot noise of the dark current. We irradiated x-rays on the back side (sensor layer side; see Fig. 1) of XRPIX6C using an x-ray tube (Mini-X2, AMPTEK) attached to the vacuum chamber. The x-ray tube was operated at 20 kV with a target of Au. The energies of L-shell fluorescence lines of Au are 9.7 keV (), 11.4 keV (), and 13.4 keV (). XRPIX6C was irradiated with x-ray to a total dose of 10 krad at the BOX layer. Assuming that we observe the Crab Nebula, one of the observational targets of FORCE, for 100 ksec per month based on a previous study,7 10 krad corresponds to 7 years of in-orbit operation. Fig. 2Schematic view of experimental setup. We irradiated x-rays on the back side of XRPIX using an x-ray tube and irradiated x-rays of on the front side to evaluate the performance of XRPIX. ![]() The degradation of the detector performance was monitored by iterating the x-ray irradiation and data acquisition of . These evaluation data were taken after irradiations of 0.1, 0.5, 1, 2, 4, 7, and 10 krad. To avoid the large dark current after the irradiation, the evaluations were performed after the dark current settled down to steady values. 3.Results of Irradiation ExperimentWe evaluated the spectral performance of XRPIX by irradiating x-rays of the radioisotope from the front side (circuit layer side). We took frames of the x-ray data with a 1-ms integration time. Figure 3 shows the spectra of single-pixel events, which are extracted when a pulse height of one pixel exceeds the event threshold (), whereas the pulse heights of the surrounding 8 pixels are below the split threshold (). The horizontal axis is the uncorrected pulse heights in analog-to-digital unit (ADU), and the vertical axis is the number of counts. The tail structure on the low-energy side of the peak becomes noticeable as the dose increases. This is probably caused by the charge loss due to the increase in the interface trap at the interface by x-ray irradiation.10 Fig. 3energy spectra of XRPIX6C before and after x-ray irradiation. The horizontal axis is the uncorrected pulse heights in ADU, and the vertical axis is the number of counts. ![]() We also evaluated the conversion gain and energy resolution of XRPIX6C using the measured peak position and FWHM of the line derived by fitting with the Gaussian function. Figures 4(a) and 4(b) show the gain and energy resolution as functions of the total dose, respectively. Both of them were almost constant up to 2 krad, but after that, they degraded with increasing dose. Although an outlier of the gain at 2 krad indicates a possibility of a nonlinear relation to the dose, we assume a linear relation for simplicity. After 10-krad irradiation, the gain and energy resolution degraded by and compared with that of nonirradiation, respectively. Fig. 4(a) Conversion gain and (b) energy resolution of XRPIX6C as a function of dose level. Solid lines and shaded regions indicate the best fit linear functions and 95% confidence intervals, respectively. The confidence intervals are calculated on the assumption of Gaussian uncertainty of the data. ![]() Figures 5(a) and 5(b) show the dark current and readout noise as functions of the dose, respectively. To evaluate the dark current, we measured the pedestal levels as a function of the integration time by reading pulse heights from all of the pixels with integration times of 0.1, 1, 2, 4, 8, 16, and 32 ms. Although this measurement was performed under the irradiation of x-rays from , its count rate was much , even with the longest integration time of 32 ms. Thus, the effect of x-ray events on the pedestal measurement was negligible. In this measurement, the longer the integration time is, the more charge is accumulated and the higher pedestal level is output. Thus, we estimated the relationship between the integration time and pedestal level as a linear function and evaluated the dark current from the slope of the function. In addition, we evaluated the readout noise by measuring the pedestal width of each pixel. As shown in Fig. 5, both of them increased in proportion to the dose. After 10-krad irradiation, the dark current and readout noise increased by and compared with that of nonirradiation, respectively. Fig. 5(a) Dark current and (b) readout noise of XRPIX6C as a function of dose level. Best fit linear functions and 95% confidence intervals are overplotted similarly as in Fig. 4. ![]() We also evaluated the fraction of “noisy pixels” to all pixels. The pixels above in the histogram of readout noise is judged to be noisy pixels. As shown in Fig. 6, there was little change from nonirradiation to 10 krad. Because the noisy pixels are not remarkably different from the normal pixels, the noisy pixels at 0 krad were not exactly the same pixels as those at 10 krad. As shown in Fig. 7, distributions of the readout noises in each pixel were almost the same shape with a slight shift. Therefore, these results suggest that there is an increase in noise on average, but no pixels show any extreme increase in noise. 4.Discussion4.1.Energy ResolutionFigure 8 shows the energy spectra corrected for the gain degradation at 0 and 10 krad. The horizontal axis is the x-ray energy, and the vertical axis is the number of counts. In Fig. 8, the peak positions are aligned to focus on the change in spectral shape rather than the gain degradation. The increase in the tail structure of the x-ray spectra was obviously observed after 10-krad irradiation. It contributes to the degradation of spectral performance. Fig. 8energy spectra of XRPIX6C corrected for the gain degradation before and after x-ray irradiation. The horizontal axis is the x-ray energy, and the vertical axis is the number of counts. The tail structure increased at 10 krad compared with pre-irradiation. ![]() This tail structure can be seen even before the irradiation, and we investigated this issue in a previous study.10 This effect is likely caused by the charge loss at the interface. When carriers generated in the sensor layer drift toward the sense-node along the electric field, they pass through the interface, and some of them are captured in the trap level. As anticipated from prior studies, the interface state density increases due to radiation damage,11 increasing the effects of charge loss and leading to an enlargement of the tail structure, as observed in Fig. 8. 4.2.GainThe chip output gain is degraded by after 10-krad irradiation. According to a previous study, the gain degradation due to radiation damage is caused by the enlargement of BNW.6 This is explained by the effect of the positive charge trapped in the BOX layer due to x-ray irradiation. Its potential attracts electrons and enlarges the area of BNW. This phenomenon results in an increase in the sense node capacitance and the degradation of the gain. The relation between the inverse of the gain and BNW size is described in the previous study6 as According to this equation, the change in the inverse of the gain after 10-krad irradiation is equivalent to the enlargement of BNW by . As the BNW width is designed to be , at 10 krad. It is a reasonable value because the distance between BNW and BPW is designed to be . In addition, it is considered that the charge loss at the interface due to the increase in the interface state density contributed to the peak shift and caused the gain degradation. 4.3.Readout NoiseThe readout noise increased by after 10 krad irradiation. The gain degradation discussed in the previous section affects the increase of the readout noise. According to a previous study,12 the readout noise in XRPIX is related to the gain with an empirical relation of . Therefore, the gain degradation of after 10 krad contributes to the increase in the readout noise by . In addition, the increase in the shot noise due to the dark current increase also contributes to the increase in the readout noise. As the readout noise and dark current were evaluated using an integration time of 1 ms, the dark current increase of after 10 krad contributes to the increase in the readout noise by . Therefore, it is assumed that the gain degradation and shot noise increase due to the increased dark current do not contribute significantly to the increase in the readout noise. To solve the physical origin of readout noise increase, a more comprehensive analysis is needed. We are now formulating the readout noise due to the noise and thermal noise generated in the MOSFET in the main amplifier in each pixel circuit. In the radiation environment, the noise increases due to the increase in the interface trap.13 Also, the increase in the sense-node capacitance6 must affect the propagation of these and thermal noise. These full noise analyses will be our future work. 4.4.Dark CurrentThe dark current increased by at 10 krad due to x-ray irradiation. We investigated its physical mechanism using technology computer-aided design (TCAD) device simulator HyDeLEOS, which is a part of the HyENEXSS.14 In the simulation, we implemented the device structure as shown in Fig. 1 and calculated the dark current flowing in a one-pixel region. Detailed profiles for -stops, sense nodes, BNWs, BPWs, and middle-Si layers were implemented based on the parameters provided by LAPIS Semiconductor Co. Ltd. In addition, it is generally known that the fixed positive charges are accumulated in the BOX layer during the wafer process. Therefore, we placed the fixed charge of uniformly between 1 and 3 nm above the interface in reference to a previous study.15 We also implemented the radiation damage effects in the simulation. We reproduced the accumulation of positive charges by placing positive fixed charges in the BOX layer. We assumed that the concentration of increases in proportion to the dose based on the experimental results of a previous study.5 In addition, to consider the carrier generation through the interface traps, we used the surface recombination model, expressed as16 where is the intrinsic carrier density, is the hole density, is the electron density, and is the surface recombination velocity. Subscripts and represent hole and electron, respectively. In this study, for the sake of simplification, we assume . The surface recombination velocity is expressed as , where is the carrier capture cross section, is the thermal velocity, and is the interface state density.16 We can calculate the carrier generation rate through the interface traps by applying this model to the interface. According to a previous study,17 increases due to radiation damage, and thus the surface recombination velocity must increase after x-ray irradiation. In Eq. (2), and are calculated by device simulation, and is a constant. Therefore, because only is an unknown parameter and depends on the dose, it is necessary to model as a function of the dose based on the experimental results shown in Fig. 5(a).To model as a function of the dose, we first need to reproduce the measured dark current in the simulation. Because the dark current is reproduced by the SUR [Eq. (2)] and Shockley–Read–Hall (SRH) models16 in the simulation, the unknown parameters, in the SUR model and carrier lifetime in the SRH model, are adjusted. In HyDeLEOS, it is possible to adjust the carrier lifetime in the Si bulk by the coefficients of the carrier lifetimes of electrons () and holes () as In this study, for the sake of simplification, we assumed and . and are the fiducial values of the carrier lifetime defined as default parameters in HyDeLEOS for the sensor layer of XRPIX6C composed of -type Si with a resistivity of , which corresponds to the doping concentration of . To evaluate the reproducibility of the measured dark current by simulation, we calculated , which is the average value of the differences between the measured and simulated dark currents at multiple back-bias voltages . This takes into account the dark current generated from the sensor layer depleted by . Then, we adjusted the parameters and to minimize . As shown in Fig. 9, the optimal values of do not change before and after radiation damage because the lattice defects in Si bulk, which shorten the carrier lifetime, do not increase due to x-ray irradiation in principle. The carrier lifetime remains at the same value (, ) after 10-krad irradiation. These values are reasonable compared with those measured by the microwave-detected photoconductance decay method in a previous study.18 Fig. 9The difference between the measured and simulated dark current as a function of the simulation parameters and . The white squares indicate the best parameters, which have a lowest for each dose. (a) 0 rad, (b) 2 krad, (c) 4 krad, (d) 7 krad, and (e) 10 krad. ![]() On the other hand, the optimal value of increased with increasing dose, as shown in Fig. 9. Then, we modeled the dependence of on the dose assuming a linear relationship between them. Figure 10 shows the optimal values of as a function of the dose, the best fit linear function, and its confidence intervals for 95% and 99%. As a result of linear fitting, slope and intercept were obtained as the best-fit parameters. According to the obtained linear model, the surface recombination velocity is after 10-krad irradiation. This value of the surface recombination velocity is consistent with a previous study19 for an irradiation of 10 krad. Fig. 10The optimal values of as a function of dose level. The solid line indicates the best fit linear function. The shaded regions represent its confidence intervals of 95% and 99%. ![]() Using the linear model of against the dose shown in Fig. 10, we compare the dose dependence of the measured dark current with the simulated dark current. Figure 11(a) shows the comparison between the measured dark current and simulated dark current using the assumed linear model. The shaded regions show the simulation results corresponding to the confidence intervals for each dose in Fig. 10. However, because never takes negative values, at 0 rad of the linear model and the lower limits of the confidence intervals at 0 rad and 2 krad are set to . As shown in Fig. 11(a), TCAD simulation successfully reproduced the experimental result by taking into account two radiation damage effects, i.e., accumulation of BOX charges and increase of interface traps. Fig. 11Comparison of the experimental dark current degradation with the simulation in the cases (a) with and (b) without . The solid lines show the simulated dark current using the assumed linear model. The shaded regions show the simulated dark current using at the upper and lower limits of the confidence intervals for each dose in Fig. 10. ![]() To understand how the accumulated positive charge and the interface traps contribute to the dark current, we discriminate these effects. In Fig. 11(b), is not added for any doses. In both Figs. 11(a) and 11(b), the parameter increases approximately in proportion to the dose as shown in Fig. 10, and the simulated dark current tends to increase as the dose increases. Therefore, the increase in the interface traps contributes to the dark current increase. Comparing Figs. 11(a) and 11(b), the simulated dark current in Fig. 11(a), which contains with different concentrations at each dose, is slightly lower than that in Fig. 11(b), which contains no . To reveal the physical mechanism of this difference, we focus on the electron density distribution near the interface. Figures 12(a) and 12(b) show the electron density map around the BNW without and with , respectively. In the case without of Fig. 12(a), both ends of the BNW are depleted and the interface is bare. On the other hand, in the case with of Fig. 12(b), the region of high electron density extends horizontally because the attracts electrons near the interface. These electrons fill the interface, making it difficult for carriers to be generated, resulting in a lower dark current. Therefore, does not increase the dark current, but slightly decreases it. 5.ConclusionWe performed an irradiation experiment on D-SOI XRPIX using x-rays with a total dose of 10 krad and investigated the physical mechanism of the degradation of detector performance. As the results, we found that the energy resolution at 5.9 keV x-ray degraded by and the dark current increased by . Especially regarding the dark current, we found that the increase in the interface trap density predominantly contributes to the increase in it. Moreover, the accumulated positive charge in the BOX layer does not increase the dark current. Thus, in the case of XRPIX and possibly the other SOI pixel sensors as well, it is important to reduce the dark current due to the interface traps to suppress the increase in dark current under the radiation environment. AcknowledgmentsWe acknowledge the relevant advice and manufacture of the XRPIXs by the personnel of LAPIS Semiconductor Co., Ltd. This study was supported by MEXT/JSPS KAKENHI Grant-in-Aid for Scientific Research on Innovative Areas 25109002 (Y.A.) and 25109004 (T.G.T., T.T., K.M., A.T., and T.K.), Grant-in-Aid for Scientific Research (B) 25287042 (T.K.), Grant-in-Aid for Young Scientists (B) 15K17648 (A.T.), Grant-in-Aid for Challenging Exploratory Research 26610047 (T.G.T.), and Grant-in-Aid for Early Career Scientists 19K14742 (A.T.). This study was also supported by the VLSI Design and Education Center (VDEC), Japan, and the University of Tokyo, Japan, in collaboration with Cadence Design Systems, Inc., USA; Mentor Graphics, Inc., USA; and Synopsys, Inc., USA. We also thank Kazuya Matsuzawa, Yutaka Akiyama, and Nobutoshi Aoki (Kioxia Corp.) for fruitful discussions on the device simulation. ReferencesK. Mori et al.,
“A broadband x-ray imaging spectroscopy with high-angular resolution: the FORCE mission,”
Proc. SPIE, 9905 990510
(2016). https://doi.org/10.1117/12.2231262 PSISDG 0277-786X Google Scholar
K. Nakazawa et al.,
“The FORCE mission: science aim and instrument parameter for broadband x-ray imaging spectroscopy with good angular resolution,”
Proc. SPIE, 10699 106992D
(2018). https://doi.org/10.1117/12.2309344 PSISDG 0277-786X Google Scholar
T. G. Tsuru et al.,
“Kyoto’s event-driven X-ray astronomy SOI pixel sensor for the FORCE mission,”
Proc. SPIE, 10709 107090H
(2018). https://doi.org/10.1117/12.2312098 PSISDG 0277-786X Google Scholar
R. C. Hughes,
“Hole mobility and transport in thin films,”
Appl. Phys. Lett., 26 436
(1975). https://doi.org/10.1063/1.88200 APPLAB 0003-6951 Google Scholar
K. Hara et al.,
“Radiation hardness of silicon-on-insulator pixel devices,”
Nucl. Instrum. Methods Phys. Res. Sect. A: Accel., Spectrom., Detect. Assoc. Equip., 924 426
–430
(2019). https://doi.org/10.1016/j.nima.2018.05.077 Google Scholar
K. Hagino et al.,
“Radiation damage effects on double-SOI pixel sensors for X-ray astronomy,”
Nucl. Instrum. Methods Phys. Res. Sect. A: Accel., Spectrom., Detect. Assoc. Equip., 978 164435
(2020). https://doi.org/10.1016/j.nima.2020.164435 Google Scholar
K. Mori et al.,
“Total ionizing dose effects on the SOI pixel sensor for X-ray astronomical use,”
Nucl. Instrum. Methods Phys. Res. Sect. A: Accel., Spectrom., Detect. Assoc. Equip., 924 473
–479
(2019). https://doi.org/10.1016/j.nima.2018.07.057 Google Scholar
M. Kitajima et al.,
“X-ray radiation damage effects on double-SOI pixel detectors for the future astronomical satellite,”
in IEEE NSS/MIC Conf. Rec.,
(2021). Google Scholar
Y. Arai et al.,
“Development of SOI pixel process technology,”
Nucl. Instrum. Methods Phys. Res. Sect. A: Accel., Spectrom., Detect. Assoc. Equip., 636 S31
–S36
(2011). https://doi.org/10.1016/j.nima.2010.04.081 Google Scholar
K. Hagino et al.,
“Sub-pixel response of double-SOI pixel sensors for X-ray astronomy,”
J. Instrum., 14 C10023
(2019). https://doi.org/10.1088/1748-0221/14/10/C10023 Google Scholar
J. R. Schwank et al.,
“Radiation effects in MOS oxides,”
IEEE Trans. Nucl. Sci., 55 1833
–1853
(2008). https://doi.org/10.1109/TNS.2008.2001040 IETNAE 0018-9499 Google Scholar
S. Harada et al.,
“Performance of the silicon-on-insulator pixel sensor for X-ray astronomy, XRPIX6E, equipped with pinned depleted diode structure,”
Nucl. Instrum. Methods Phys. Res. Sect. A: Accel., Spectrom., Detect. Assoc. Equip., 924 468
–472
(2019). https://doi.org/10.1016/j.nima.2018.09.127 Google Scholar
Y. Nemirovsky, I. Brouk and C. Jakobson,
“1/f noise in CMOS transistors for analog applications,”
IEEE Trans. Electron Devices, 48
(5), 921
–927
(2001). https://doi.org/10.1109/16.918240 IETDAI 0018-9383 Google Scholar
“3D TCAD Simulator HyENEXSS ver. 8.5K,”
(2019). Google Scholar
V. V. Afanas’ev and A. Stesmans,
“Positive charging of thermal interface by hydrogen annealing,”
Appl. Phys. Lett., 72 79
–81
(1998). https://doi.org/10.1063/1.120650 APPLAB 0003-6951 Google Scholar
S. M. Sze, Physics of Semiconductor Devices, 2nd ed.Wiley, New York
(1981). Google Scholar
J. Shi et al.,
“Radiation-induced charge trapping in Si-MOS capacitors with HfO2/SiO2 gate dielectrics,”
Nuclear Instrum. Methods Phys. Res. Sec. B: Beam Interact. Mater. At., 479 150
–156
(2020). https://doi.org/10.1016/j.nimb.2020.06.039 Google Scholar
D. Schroder,
“Carrier lifetimes in silicon,”
IEEE Trans. Electron Devices, 44 160
–170
(1997). https://doi.org/10.1109/16.554806 IETDAI 0018-9383 Google Scholar
A. M. Tonigan et al.,
“Impact of surface recombination on single-event charge collection in an SOI technology,”
IEEE Trans. Nucl. Sci., 68 305
–311
(2021). https://doi.org/10.1109/TNS.2021.3056898 IETNAE 0018-9499 Google Scholar
Biography |