Perovskite multifunctional logic gates via bipolar photoresponse of single photodetector

The explosive demand for a wide range of data processing has sparked interest towards a new logic gate platform as the existing electronic logic gates face limitations in accurate and fast computing. Accordingly, optoelectronic logic gates (OELGs) using photodiodes are of significant interest due to their broad bandwidth and fast data transmission, but complex configuration, power consumption, and low reliability issues are still inherent in these systems. Herein, we present a novel all-in-one OELG based on the bipolar spectral photoresponse characteristics of a self-powered perovskite photodetector (SPPD) having a back-to-back p+-i-n-p-p+ diode structure. Five representative logic gates (“AND”, “OR”, “NAND”, “NOR”, and “NOT”) are demonstrated with only a single SPPD via the photocurrent polarity control. For practical applications, we propose a universal OELG platform of integrated 8 × 8 SPPD pixels, demonstrating the 100% accuracy in five logic gate operations irrelevant to current variation between pixels.

1. First and foremost, I think the proposed architecture may have an intrinsic limitation for wide spectrum response. The logic functions in Figures 3 & 4 were realized by using monochromic 940 nm NIR and 625 nm visible light. In the multilayer structure, both the NIR and visible light were introduced from the bottom ITO side. What would be the result if different wavelengths were used? In figure 2, 530 nm was used for the visible light. Why was 625 nm used for later experiments? 2. In figure 1b, only the elemental mapping images for Sn, I, and Pb, without O and Au, which is in discrepancy with the figure caption. This confusion needs to be clarified. 3. I believe that the perovskite film thickness is critical for the device operation since it determines the light absorption and transmission. How was it optimized? 4. In page 3, for the statement, "The superior optoelectronic properties of organometal halide perovskite have been demonstrated in many studies 27-29", I think more research works on perovskite optoelectronics should be cited. An original work on the broadband perovskite photodetector is ACS Applied Materials & Interfaces 9, 37832-37838 (2017), and a more recent example of taking advantage of optoelectronic properties of perovskites is Applied Physics Reviews 7, 031401 (2020). There are certainly other notable examples the authors can cite and enrich the discussion, 5. What is the reason behind using a device structure for an array (figure 4a) from that for a single logic gate device (figure 1a)? The authors should provide more details. 6. The authors may need to tune down the statement of "ultrafast photoresponse". In figure S7, the authors used an oscilloscope and got a rise/decay time in ms level, and Table S1 compared the speed with other typical reports. However, most of the PDs in Table S1 adopted a planar structure, which has a much slower response speed compared to the vertical structure due to a much wider channel length. On the other hand, many vertical perovskite-based PDs have a speed at µs level or faster, such as one typical example: https://doi.org/10.1038/ncomms6404, among many other similar works.
Reviewer #3 (Remarks to the Author): In this manuscript, the authors demonstrated optoelectronic logic gates based on the bipolar spectral photoresponse characteristics of a perovskite photodiode. The photocurrent polarity of the photodetector was changed using the visible and near-infrared light, and the Boolean output state of "1" or "0" was determined by the positive or negative output photocurrent, respectively. Five representative logic gates ("AND", "OR", "NAND", "NOR", and "NOT") were successfully demonstrated by controlling the photocurrent polarity. In addition, 8x8 device arrays exhibited the high-accuracy operations in the five logic gates. It was impressive to demonstrate various logic circuits in one device with a unique device configuration. The experimental data were also well-organized. However, although the authors claimed that the photodiodes and logic gates would be useful for an artificial intelligence technology, they did not show the device characteristics related to the artificial intelligence, such as computation capability. In addition, the performance of some devices seems to be overstated, and I think such expressions should be toned down based on the actual measured data. I suggest the following revision of the manuscript before publication.
1. In the introduction, the authors mentioned that the explosive demand for artificial intelligence and big-data has sparked research interest in new logic gate systems. However, to claim that this work is a technology for artificial intelligence, it will be necessary to demonstrate the computing capability using 8x8 device arrays fabricated in Fig. 4. If not, please remove the overstated expressions about artificial intelligence.
2. The authors demonstrated the ultrafast/sensitive photoresponse and the substantial on-off ratio in the back-to-back self-powered perovskite photodetector (SPPD). However, in Supplementary  Fig. 7, the operating speed of milliseconds looks quite slow. Please specify the reason.
3. I strongly suggest that the authors should demonstrate the device array with a reduced pixel size of less than a few microns (Fig. 4), for the comparison with the current technology.
4. Perovskite-based devices typically have the weakness on retention and endurance. For the practical implementation, it will be important to show the reliable and repeated operation for a long period. 5. It is not clear why the optoelectronic logic gate platform can suggest an integrated chip for optical computing, optical communication, logic memory units, and photonic quantum computation (in the conclusion section). More detailed explanation is necessary even though the authors simply mentioned the future plan.

Reviewer #1
In this work, the authors reported multifunctional logic gates which was constructed based on bipolar photoresponse from single photodetector. This is an interesting and novel work that deserves publication in Nature Communications. However, there are still a couple of issues needed to be addressed and it may be published after major revision. The comments are as follows: We appreciate this comment. We gained crucial knowledge and clues from the references suggested by the reviewer #1.
As you commented, Yang et al., for the first time, reported the wavelength-induced dualpolarity property from a p-n heterojunction diode composed of n-type ZnO nanowires and ptype thermoelectric thin film (SnS or Sb2Se3) R1-R3 . According to the report, the dual-polarity occurs due to the coexistence of two photocurrent generation mechanisms: photovoltaic (PV) effect occurring by band-to-band transition and photo-thermoelectric (PTE) effect driven by photo-induced temperature-gradient R1-R3 . In their diodes, short-wavelength lights are mainly absorbed by the thermoelectric Sb2Se3 film. In contrast, long-wavelength lights are absorbed at the vicinity of the p-n junction. This finding indicates that the direction and amount of photocurrent can be controlled depending on the wavelength and power density. Therefore, we decided to importantly describe all the suggested references in the revised manuscript, as we think they would be very beneficial in solidifying our paper's theoretical basis. Thank you again for the great suggestion.

[Action]
We cited the main contents of the suggested references on page 3 of the revised manuscript as follows:

Q2) They need to provide the evidence for "the Vis-Perov layer was intrinsic (i-type), and
the NIR-Perov layer was the lightly doped p-type". [Answer] In the original manuscript, we provided the energy band diagram of the p + -i-n-p-p + structure configured with the results of UPS and UV-Vis measurements (Fig. 1d). The difference between the Fermi-level and the mid-gap state is 0.04 for the Vis-Perov (MAPbI3) and 0.1 eV for the NIR-Perov (MA0.5FA0.5Pb0.4Sn0.6I3). Therefore, in the original manuscript, we described that the NIR-Perov is a lightly doped p-type.
We agree that another strong evidence for the doping level is required. For this purpose, we conducted Hall measurement for ten samples of each perovskite film (Vis-Perov and NIR-Perov). Average Hall values were calculated as shown in the Table R1 below. The carrier concentration of Vis-Perov was measured as ~10 12 cm -3 , but the mobility was not reliably measured due to its low carrier concentration. We found in literature that intrinsic MAPbI3 thin films had a concentration of 10 11 -10 12 cm -3 R6-R7 . Therefore, the Vis-Perov film is regarded as an intrinsic semiconductor.

[Action]
We modified a sentence on page 6 of the revised manuscript and summarized Hall measurement results for the Vis-Perov and NIR-Perov films in Supplementary Table 1 (Table   R1) in the revised supplementary information.
Manuscript (page 6, paragraph 2): The Vis-Perov layer was intrinsic (i-type), and the NIR-Perov layer was lightly doped p-type, which were additionally supported by Hall measurements (Supplementary Table 1).

Q3) The authors should add more descriptions about the role of PCBM layer. Contrast
tests of photodetectors with/without a PCBM layer need to be supplied. [Answer] The PCBM layer serves as an intermediate n-type semiconductor playing a key role for realizing the bipolar photoresponse of p + -i-n-p-p + device. To confirm the role of PCBM layer, we fabricated a PCBM-less device (the inset of Fig. R1a). Both NIR and visible lights drove photocurrents in the same direction (Vis-PerovNIR-Perov), as shown in Fig. R1a. The photocurrent was markedly suppressed to < 0.1 μA, 20 times smaller than that of the PCBMinterposed device.
The unidirectional photocurrent generation in the PCBM-less device can be explicated based on the energy band diagram depicted in Fig. R1b. The large valence band offset (∆Ev) between the two perovskites can influence hole transport. The deep HOMO level of the Vis-Perov, located at 0.64 eV lower than that of the NIR-Perov, facilitates hole transport from Vis-Perov to NIR-Perov, simultaneously hindering the reverse flow. Therefore, photo-generated holes are allowed to flow in only one direction regardless of incident wavelength.
In contrast, the PCBM with a high electron density can block the hole transport, as depicted in Fig. 1d and 2e. Thus, only photo-generated electrons can move through the PCBM layer and be accumulated in the counter perovskite. Therefore, the photocurrent direction can be readily controlled by changing the incident light wavelength when the PCBM is interposed.

[Action]
We added a sentence for the role of PCBM layer on page 9 of the revised manuscript. The optoelectronic properties of the PCBM-less device can be found in Supplementary Fig. 15. [Answer]

Manuscript
We calculated photoconductive gain (Gλ) and specific detectivity (Dλ) using the following formulas adopted in the suggested papers R9,R10 ,where Rλ is a responsivity of photodetector, recorded under a wavelength of light of λ, Id and A are a dark current and active area of the device, respectively; h, c, and q are physical constants representing Planck constant, speed of light, and unit charge, respectively. Note that we employed specific detectivity (Dλ), with the assumption that a shot noise dominantly determines a noise current of photodetector.

Q5) They need to illustrate the reason why the energy bands are inclined across the whole
Vis-Perov layer and NIR-Perov layer (Fig. 2e).

[Answer]
A built-in potential caused by a Fermi-level difference leads to band-bending at material interfaces, as shown in Fig. R3a R11-R13 . Herein, we may ignore the band-bending at the interfaces between the perovskites and hole transport layers (PEDOT:PSS or Spiro-OMeTAD) because of the minute Fermi-level difference. Then, we can simply consider the heterojunction between the PCBM and perovskite regarding on this matter. The built-in potential across the perovskite can be estimated as ,where Vi is the Fermi-level difference between the PCBM and perovskite; εPerov and Na are the dielectric constant and doping concentration of perovskite, respectively; q is the unit charge; x is the perovskite depth from the interface R11-R13 ; and xp is the thickness of depletion zone (energy-bending region) which is defined as From the formula above, we confirmed that the theoretical depletion width (xp) of Vis-Perov is above 10 μm which is much thicker than the actual thickness (~200 nm). It implies that the film was fully depleted, rendering the inclined energy band through the entire film. This estimation can be well accepted considering the intrinsic Vis-Perov. The xp of NIR-Perov was calculated as ~220 nm, which is more than two-thirds of the NIR-Perov thickness. Thus, the energy band of NIR-Perov is also expected to be bent across the film. Inherent parameters for each layers in the SPPD: Vi is the relative Fermi-level difference with the PCBM; Na (Nd) is the hole (electron) concentration; ε is the dielectric constant (ε0 is the vacuum permittivity); and xp is the theoretical depletion width.

[Action]
We added a sentence on page 9 of the revised manuscript to provide information on the built-in potential and energy band-bending. The above discussion (for Fig. R3) was added in [Action] We made actions in Fig. 3d

Reviewer #2
In this manuscript, Kim et al. reported a self-powered perovskite photodetector (SPPD) with a back-to-back multilayer structure, which achieves optoelectronic logic gates beyond "AND" and "OR" with only one single device, enabling a possible improved integration for the future Thank you for this comment. We agree that the data using different wavelengths in Fig. 2 and 3 may confuse readers.
Delicate tuning of the logic functions via the light intensity modulation for the opposite photocurrent-offset can be most efficient when the combination of two wavelengths having a similar photoconductive gain is applied as the input and modulator. As shown in Supplementary   Fig. 9a, the photoconductive gain strongly depends on the wavelength of light. The photoconductive gain for the 625 nm light was 0.007, almost similar to that (0.008) for the 940 nm light. The photoconductive gain for 530 nm light is 0.02, much larger than that for the 625 nm. It means that the delicate tuning of the logic functions becomes much difficult. In this case, the irradiance of 530 nm, equivalent to a third of that of the 625 nm, should be used to achieve optimal logic outputs. Then, the five logics were faithfully implemented in the same SPPD device under 530 and 940 nm illumination (Fig. R4).

Fig. R4
Logic gate operation of the 50×50 μm 2 pixel array using a pair of 530 and 940 nm light sources. Three-dimensional bar charts for all the outputs ("OR", "AND", "NAND", "NOR", and "NOT") obtained from the 16 pixels in SPPD-OELG array.
In addition, we investigated I-V characteristics of the SPPD using the 625 nm light. As expected, the sign of VOC (Fig. R5a) was still negative. It means that the red (625 nm) light can replace the green (530 nm) light in our logic device. However, as discussed above, the irradiance should be considered for the efficient current-offset.

[Action]
We added additional sentences on page 10 and 12 of the revised manuscript to explain the use of 625 nm for logic operations rather than 530 nm. The optoelectronic properties using the 625 nm light (Fig. R5) were added in Supplementary Fig. 16, and the logic gate operation using the combination of 530 and 940 nm lights (Fig. R4) were discussed in Supplementary  9a), hereafter, we utilized these two light sources for the following logic gate operations. Prior to the use of 625 nm, we confirmed that the 625 nm irradiation generated the opposite photocurrents compared to the 940 nm case (Supplementary Fig. 16).

Manuscript (page 12, paragraph 2):
In addition, we also confirmed logic gate operations using a pair of 530 and 940 nm light sources after precisely adjusting the irradiance of 530 nm light ( Supplementary Fig. 17).

Q2) In figure 1b, only the elemental mapping images for Sn, I, and Pb, without O and Au,
which is in discrepancy with the figure caption. This confusion needs to be clarified. [Action] We are sorry for the confusion. O and Au mapping images were added in revised Fig. 1b.

Q3) I believe that the perovskite film thickness is critical for the device operation since it determines the light absorption and transmission. How was it optimized?
[Answer] Before device fabrication, we simulated theoretical photoconductive gain (Gλ) by varying the thickness of perovskite films to explore an optimized condition. As shown in Fig. R6a, the color and tone represent the gain ratio (G940/G625), which is distributed on the plane as a function of perovskite film thickness ranging from 50 to 400 nm. The gain ratio can be obtained   R6b and c). Under the two light sources, the photoresponses of both 'Device A' and 'Device C', having a perovskite layer thickness less than 150 nm, are not as balanced as that of 'Device B', which is not good for the logic gate operation (Fig. 6c). The combination of 300 nm NIR-Perov / 200 nm Vis-Perov has the gain ratio of 1, optimal for the efficient logic gate operation.

[Action]
We added an additional sentence on page 5 of the revised manuscript to show how we optimized the perovskite thickness for the efficient logic gate operation. The simulation results discussed above were added in Supplementary Fig. 2.

Manuscript (page 5, paragraph 1):
The thickness of each perovskite layer (300 nm NIR-Perov and 200 nm Vis-Perov) was decided via a theoretical simulation, aiming a photoconductive gain ratio (G940/G625) of 1 for the efficient current offset, which is beneficial for the tuning of logic gate operation ( Supplementary Fig. 2).

[Action]
We cited the suggested references on page 3 of the revised manuscript as follows: [Answer]

Manuscript
We appreciate this comment as it can significantly help to enrich the manuscript. The single device (from Fig. 1 to 3) was made to investigate device characterization and working principle for the logic gate operation. We desired to make a more practical platform being urged in realworld applications for data processing. In this perspective, it is imperative to construct an integrated logic gate platform and investigate its accuracy to ensure the practical feasibility for a chip-level system. Herein, we succeeded in constructing an 8×8 crossbar-type SPPD array in which five different logics were faithfully implementable by individual pixel in an array at a 100% precision. The integrated SPPD-OELG array exhibits a superior reproducibility for tunable logic gate operations.

[Action]
We made modification on page 13 of the revised manuscript as follows: Manuscript (page 13, paragraph 1): For real-world applications, it is imperative to construct integrated SPPD-OELG platform and investigate its accuracy to ensure the practical chip-level feasibility.
Q6) The authors may need to tune down the statement of "ultrafast photoresponse". In figure S7, the authors used an oscilloscope and got a rise/decay time in ms level, and Table   S1 compared the speed with other typical reports. However, most of the PDs in Table S1 adopted a planar structure, which has a much slower response speed compared to the vertical structure due to a much wider channel length. On the other hand, many vertical perovskite-based PDs have a speed at µs level or faster, such as one typical example: https://doi.org/10.1038/ncomms6404, among many other similar works. [Answer] We agree with the reviewer's opinion. In the revised manuscript, we toned down the overexpression of 'ultrafast'.
As shown in Fig. R7, we confirmed that the rise/decay times of perovskite PD could be shortened to tens of microseconds by reducing the active area by 100 times (from 4.64 mm 2 to 0.04 mm 2 ), even though there is still an apparent gap compared to the state-of-art levels (e.g., hundreds of ns) R17-R19 .
We believe that it can be enhanced by sophisticated pixel-downsizing. However, at the current lithographic and vacuum technology, it is difficult to downsize to a few micron size without perovskite degradation. In other way, producing a high-crystalline perovskite film to depress charge-trapping can be a realistic approach to enhance the response speed R17-R19 .
Thank you again for this constructive comment. [Action] We changed with new results on page 7 of the revised manuscript. Details on the rise/decay times of the SPPD with active area of 0.04 mm 2 were described in Supplementary Fig. 8. The reference lists in Supplementary Table S2 were edited with only vertical perovskite-based photodetectors.

Manuscript (page 7, paragraph 2):
The rise/decay times of SPPD with an active area of 0.04 mm 2 were 62/680 μs for 530 nm and 40/72 μs for 940 nm (Supplementary Fig. 8). Even though there is still an apparent gap compared to the state-of-art level (e.g., hundreds of ns, Supplementary Table 2), it can be enhanced using a high-crystalline perovskite film with further downsizing the active area. In this manuscript, the authors demonstrated optoelectronic logic gates based on the bipolar spectral photoresponse characteristics of a perovskite photodiode. The photocurrent polarity of the photodetector was changed using the visible and near-infrared light, and the Boolean output state of "1" or "0" was determined by the positive or negative output photocurrent, respectively.
Five representative logic gates ("AND", "OR", "NAND", "NOR", and "NOT") were successfully demonstrated by controlling the photocurrent polarity. In addition, 8x8 device arrays exhibited the high-accuracy operations in the five logic gates.
It was impressive to demonstrate various logic circuits in one device with a unique device configuration. The experimental data were also well-organized. However, although the authors claimed that the photodiodes and logic gates would be useful for an artificial intelligence technology, they did not show the device characteristics related to the artificial intelligence, such as computation capability. In addition, the performance of some devices seems to be overstated, and I think such expressions should be toned down based on the actual measured data. I suggest the following revision of the manuscript before publication.

Q1) In the introduction, the authors mentioned that the explosive demand for artificial intelligence and big-data has sparked research interest in new logic gate systems.
However, to claim that this work is a technology for artificial intelligence, it will be necessary to demonstrate the computing capability using 8x8 device arrays fabricated in [Answer] Humbly accepting the reviewer's comment, we deleted the overstated terminologies and descriptions to prevent confusing potential readers.

[Action]
We modified the overstated expression on page 1 and 2 of the revised manuscript and added paragraphs as follows: Manuscript (page 1, paragraph 1): The explosive demand for a wide range of data processing has sparked interest towards a new logic gate platform as the existing electronic logic gates face limitations in accurate and fast computing.
(In the original manuscript) The explosive demand for artificial intelligence and big-data has sparked interest towards a new logic gate platform as the existing electronic or all-optical logic gates face limitations in accurate and fast data processing.
Manuscript (page 2, paragraph 1): Optoelectronic logic gates (OELGs) are receiving significant attention as crucial building block of future integrated circuits for accurate and fast data processing 1-4 . Existing circuits or processors based on electronic logic gates would face limitations in computing extensive data sets which are expected to markedly increase in the fourth industrial revolution age, due to performance shortfalls in switching, operation, computing, and decision making/regenerating 5,6 . Thus, developing an innovative logic gate platform that can implement faster computation with less power consumption is imperative to fulfill upcoming new computing trends.
(In the original manuscript) Optoelectronic logic gates (OELGs) are a crucial building block of the future integrated circuits because the fourth industrial revolution requires a novel computing system that enables unprecedentedly accurate and fast data processing 1-3 . The explosive demand for artificial intelligence and big-data has sparked research interest in new logic gate systems 4 . Existing circuits or processors based on electronic logic gates face limitations in the sophisticated analysis and rapid extraction of information due to performance shortfalls in switching, computing, and decision making/regenerating 5,6 . Thus, it is imperative to develop an innovative logic gate platform that can be applied to efficiently perceive hidden values, such as patterns, trends, and associations from extensive data sets that seem meaningless at first glance.
Q2) The authors demonstrated the ultrafast/sensitive photoresponse and the substantial on-off ratio in the back-to-back self-powered perovskite photodetector (SPPD). However, in Supplementary Fig. 7, the operating speed of milliseconds looks quite slow. Please specify the reason. [Answer] We agree with the reviewer's opinion. In the revised manuscript, we toned down the overexpression of 'ultrafast'.
As shown in Fig. R7, we confirmed that the rise/decay times of perovskite PD could be shortened to tens of microseconds by reducing the active area by 100 times (from 4.64 mm 2 to 0.04 mm 2 ), even though there is still an apparent gap compared to the state-of-art levels (e.g., hundreds of ns) R17-R19 .
We believe that it can be enhanced with by sophisticated pixel-downsizing. However, at the current lithographic and vacuum technology, it is difficult to downsize to a few micron size without perovskite degradation. In other way, producing a high-crystalline perovskite film to depress charge-trapping can be a realistic approach to enhance the response speed R17-R19 .
Thank you again for this constructive comment. [Action] We changed with new results on page 7 of the revised manuscript. Details on the rise/decay times of the SPPD with active area of 0.04 mm 2 were described in Supplementary Fig. 8 Q3) I strongly suggest that the authors should demonstrate the device array with a reduced pixel size of less than a few microns (Fig. 4), for the comparison with the current technology. [Answer] We strongly agree that a perovskite PD array at submicron scale is critical for practical applications. To fabricate such a sub-micron scaled pattern array, highly resolved photo-or ebeam lithography should be used but, access to those tools and severe degradations in organic layers (perovskite, PCBM, PEDOT:PSS, and Spiro-OMeTAD) R20,R21 are problematic during the lithography process.
As per the reviewer's request, we tried our best to reduce the pixel size by using shadowmask-assisted thermal evaporation at ~50 μm width (200 μm in the original manuscript). We appreciate it if the reviewer understands the difficulty of making submicron-scale pixels. Fig. R8 shows optical microscope images of the 4×4 array at a pixel size of 100×100 and 50×50 μm 2 . Fig. R9 and R10 present the logic gate operations of the 100×100 and 50×50 μm 2 pixel array, respectively. The photocurrent was reduced with decreasing the active pixel size, but the output polarity was maintained in every pixel, faithfully yielding all the five logic operations.

[Action]
We added a new sentence regarding the pixel down-sizing effect on page 13 of the revised manuscript, and the new results (logic gate operations of 100×100 and 50×50 μm 2 pixel arrays) were added in Supplementary Fig. 18-20.
Manuscript (page 13, paragraph 2): Even after down-sizing the pixel to 50×50 μm 2 , the SPPD array succeeded in yielding all the logic gate operations with 100% accuracy. The photocurrent was reduced with decreasing the active pixel size, but the output polarity was maintained in every pixel ( Supplementary Fig. 18-20).

Q4) Perovskite-based devices typically have the weakness on retention and endurance.
For the practical implementation, it will be important to show the reliable and repeated operation for a long period. [Answer] As requested, we conducted the repeatability and long-term stability test of the SPPD under 530 nm (Fig. R11a) and 940 nm (Fig. R11b) light pulses (5s on / 5s off). The photocurrent values are normalized with respect to the photocurrent recorded at the first pulse. Each magnified photocurrent behavior recorded after 4 hrs clearly shows the on-off current behavior without degradation. In this test, the SPPD retained its initial responsivity even after 8 hrs (approximately 3000 pulses) under ambient conditions.
For long-term stability testing, we analyzed a SPPD that had been stored in laboratory for one year under dry-air condition (Fig. R12). The saturated on-current, under 530 and 940 nm lights, was reduced to 90% and 50% of its initial value, respectively. The relatively significant reduction under the NIR light can be attributed to the instability of Sn cations of the NIR-Perov in the air R22 .
The relatively severe degradation in the NIR-Perov indicates that the output of logic gate would not be correct or constant any longer. Encapsulating the SPPD array with hydrophobic polymers (Polyisobutylene R23 , photocurable fluoropolymers R24 , Teflon R25 , or Parylene C R26 ) or compact insulating oxides (Al2O3) R27-R29 can effectively retard the degradation by preventing the access of oxidizing gas molecules (O2 or H2O).  [Action] We added new sentences regarding the repeatability and long-term stability results on page 8 of the revised manuscript. Detailed discussion on the new results can be seen in Supplementary Fig. 10 and 11.
Manuscript (page 8, paragraph 2): The SPPD retained its initial responsivity for 3000 pulses (5s on / 5s off) under ambient conditions as shown in Supplementary Fig. 10. For long-term stability testing, the SPPD was stored in dry-air condition for one year. The saturated on-current was reduced to 90% and 50% of the initial value under 530 and 940 nm, respectively

Q5) It is not clear why the optoelectronic logic gate platform can suggest an integrated chip for optical computing, optical communication, logic memory units, and photonic
quantum computation (in the conclusion section). More detailed explanation is necessary even though the authors simply mentioned the future plan. [Answer] We agree with the reviewer's comment that a more detailed explanation for the possible application is necessary.
This study would contribute to designing a highly integrated multifunctional perovskite optical devices in short-term. Moreover, concerning the capability of delicately modulating optoelectronic states by controlling the photocurrent polarity, the proposed array platform will pave the way for developing and advancing other applications such as light-fidelity (Li-Fi) transmission, security circuits, data processors, and healthcare sensors in the future.
1. The Li-Fi is emerging as a future wireless system beyond the conventional wireless fidelity (Wi-Fi) technology owing to high energy efficiency, compatibility within most households, fast data transmission, and enhanced security R20 . As a pivotal component for the system, photodetectors (PDs) merely serve as a recorder that converts photonic signals to electrical information and transfers it to exterior processing units. In the Li-Fi system, however, the SPPD-OELG platform can take over functions of the PD and execute the role of processor simultaneously, thus enabling efficient energy consumption and high-speed computing.
2. The capability of running multiple logics can realize a novel encryption technology. For instance, Wu et al. contrived the security circuit using a polymorphic logic device executing NAND and NOR R30 . In the same manner, assuming that the NIR and visible lights are perceived as data input (ciphertext) and gate modulator (encryption key), respectively, the SPPD-OELG can serve as a code reader to output a plaintext.
3. The data processor carrying out actual data processing under specific instructions (e.g., add, multiply, or count) can be implementable with a combinational circuit of multiple SPPD-OELG devices in a single chip, which is more spatially efficient compared to the conventional logic circuit based on electronic transistors.
4. The role of SPPD can be expanded to healthcare systems, monitoring bio-information using light sources. For example, blood oxygen saturation (SO2), defined as the percent of oxyhemoglobin in the total amount of hemoglobin in the blood, has been regarded as a significant biomarker for real-time monitoring of respiratory diseases. One recent tactic to measure the SO2 is to track the reflectance change in the blood under the illumination of two different light sources (e.g., red and NIR range) R31 . Compared to existing systems requiring two separated photodetectors, a SPPD-based system can be more spatially efficient. [Action] We added new sentences in the discussion part on page 15 of the revised manuscript as follows: Manuscript (page 15, paragraph 2): The data processing under specific instructions (e.g., add, multiply, or count) could be implementable with a combinational circuit of multiple SPPD-OELGs in a single chip, which is much spatially and costly efficient compared to the conventional logic circuit based on electronic transistors, potentially advancing to future applications for optical computing, optical communication, and logic memory. In short-term view, this development can be applicable to light-fidelity (Li-Fi) transmission 20 , security circuits 44 , and healthcare sensors 45 , utilizing the distinguished optoelectronic output states based on the photocurrent polarity.