Reconfigurable two-dimensional optoelectronic devices enabled by local ferroelectric polarization

Ferroelectric engineered pn doping in two-dimensional (2D) semiconductors hold essential promise in realizing customized functional devices in a reconfigurable manner. Here, we report the successful pn doping in molybdenum disulfide (MoS2) optoelectronic device by local patterned ferroelectric polarization, and its configuration into lateral diode and npn bipolar phototransistors for photodetection from such a versatile playground. The lateral pn diode formed in this way manifests efficient self-powered detection by separating ~12% photo-generated electrons and holes. When polarized as bipolar phototransistor, the device is customized with a gain ~1000 by its transistor action, reaching the responsivity ~12 A W−1 and detectivity over 1013 Jones while keeping a fast response speed within 20 μs. A promising pathway toward high performance optoelectronics is thus opened up based on local ferroelectric polarization coupled 2D semiconductors.


/ 24
Manuscript no. NCOMMS-  Reply to reviewers' comments Firstly, we would like to thank all the reviewer's precious comments, which helped us to improve the work dramatically.
The major changes to manuscript in this revision were briefly listed as following: 1. Figure 3, 4, 5 were updated using the pn photodiode and npn bipolar transistor configured on the same device, therefore addressing the reconfigurability in devices by rewritable FE polarization.
2. The photoresponse speed of pn diode and npn phototransistor was measured again using fast switching light source and fast measure unit. The results showed 10-20 μs fast response of the devices, among the fastest in all kinds of 2D photodetectors. 3. Photocurrent map of the pn diode and npn transistors were added in Figure 3f and 4. Supplementary material was reorganized into section I-V for better understand. Fig. S2, S3, S4, S6, S10, Table S1 and S2 were added to address the concerns raised by reviewers. Figure S7, and S11 previously in main manuscript were now moved into supplementary material. 5. Table 1 was added to the manuscript to clearly present the device parameters in study. 6. A mistake was found in the calculated detectivity due to improper unit transformation.
This was now corrected in this revision with an optimal dark current limited detectivity of 10 13 Jones for npn phototransistor.

Some English and grammar mistakes have been corrected.
In the following pages, the reviewers' comments were replied point to point. Note that Reply to comments was marked in blue, while the related changes in manuscript were marked as purple.

/ 24
Reviewer #1 (Remarks to the Author): The paper reports on the formation of lateral diodes in multilayered MoS2 using different polarization states of the covering organic ferroelectric P(VDF-TrFE) copolymer. The ferroelectric nature of P(VDF-TrFE) in contact with MoS2 was use to manipulate the carrier doping in MoS2 based on their reversible polarization by external polling field provided by a biased AFM tip. The idea of using different polarization states to create a lateral diode is novel, and the experimental results of MoS2 optoelectronic device are convincing, however, there are several issues that should be addressed before publishing in Nature Communications.
We sincerely appreciate all the reviewer's comments that helped us to improve the manuscript.
1. Why the authors did not use the monolayer WSe2 in their optoelectronic device?

Reply:
We agree with the reviewer that WSe 2 was a good choice for reversible p/n doping by ferroelectric polarization for its bipolar characteristics. Unfortunately, in our initial experiments, the exfoliated WSe 2 thin flakes exhibited poor contact properties when transferred onto electrodes. We expected this issue came from the fast oxidation of WSe 2 in ambient conditions given that selenides are less stable than sulfides (Li et  To avoid the stability issue, we now spin-coated P(VDF-TrFE) layer right after transferring 2D materials onto electrodes. In this way, we were able to obtain reversible pn doping in WSe 2 (3 nm) as we have presented in MoS 2 . As indicated in the new Figure S6 in revised supplementary materials, the ON/OFF switching ratio in WSe 2 was >10 6 , and the initial bipolar characteristic of WSe 2 was tuned to P and N by a poling voltage ~±6V. To demonstrate the universality of the method, we have added the polarization results on WSe 2 as Figure S3  The following discussion is now included in Page 6 line 4-7 in the revised manuscript to reflect the change in Figure S6.
"The universality of present strategy was further demonstrated by its application in few layer WSe 2 (4.2 nm) (Supplementary material section II), which manifested nearly symmetric p/n doping transition at V p =±6V because of its bipolar characteristic." 2. Can the p-and n-doping concentrations of WSe2 be finetuned by modulating the amount of polarization induced in the ferroelectric copolymer? Since authors claimed that "early attempts to construct the kind of devices relied prominently on local-buried gates, lateral and vertical heterojunctions, the behavior of which was complicated to manipulate without finetuned pn doping." (line 49-52 on page2) If yes, what are the ranges of the p-and n-doping concentration tuning?

Reply:
We thank the reviewer's careful examination. The answer is yes, given that the conductance in p-type and n-type MoS 2 and WSe 2 were switched with large ON/OFF ratio (10 3 -10 7 ) by adopting different FE polarization voltage.

/ 24
To demonstrate this, we attempted to extract the p/n doping concentration under FE polarization from the measured transfer curves of MoS 2 and WSe 2 using back Si gate after each FE polarization, with n, or p=σ/μe calculated from the extracted conductance (σ) at V bg =0 and the estimated carrier mobility from µ = (L/W) V ds -1 C gate -1 (dI ds /dV g ), where C gate was the gate coupling capacitance with 2D channel. Correct mobility evaluation is then essential for the estimation of carrier concentration.
It should be mentioned that C gate was usually approximated using the oxide capacitance (in our case that for 300 nm SiO 2 ), this was however based on the assumption of highly conductive semiconductor channel, e.g. the degenerately doped one in our manuscript. In the case of depleted channel, the small semiconductor capacitance start to determine the overall gate coupling. As a result, the mobility can be underestimated, which then leads to the overestimation of carrier concentration in depletion. Here, we approached the band edge mobility in MoS 2 and WSe 2 by the extracted maximum carrier mobility from measured transfer curves under back gate modulation at each polarization.
To be specific, for n-doped (p-doped) samples, μ n,0 (μ p,0 ) was approximated by the maximum mobility extracted at positive (negative) gate bias (V bg =+30 or -30V) that raises the Fermi level close to the conduction (valance) band. The estimated free carrier concentrations in p and n doped MoS 2 and WSe 2 at different FE polarization voltages were now supplied in supplementary material Figure 3. A summary of the tuned carrier concentration range was also given in Table S1.  In the revised manuscript, the discussion on doping range is supplied in Page 7 line 8-14, as following: "By using the carrier mobility extracted from transfer curves at each FE polarization state, the free carrier concentration tuned by FE polarization was estimated ~10 9 -10 12 cm -2 in MoS 2 for both electrons and holes, and in WSe 2 ~10 7 -10 11 cm -2 (Supplementary material section II). The reversibly and significantly tuned p/n doping and large ON/OFF switch ratio covering metallic, semiconductor and insulate behaviors will promote their potential applications in various optoelectronic devices with reprogrammable functions." 3. The following question: The P-V hysteresis is very symmetrical (Supplementary Fig. 1). Can the authors estimate the transition region of doping from Pup to Pdown?

Reply:
We thank the reviewer's suggestion to estimate the transition region of doping by P↑ and P↓.
This shall be determined by the transition region of polarization strength for P↑ and P↓ by AFM polling. In experiments, because of the electric field distribution near AFM tip, the tip induced

Carrier Concentration (cm -3 )
MoS 2 (e) MoS 2 (h) WSe 2 (e) WSe 2 (h) Maximum 5x10 12 10 12 2x10 10 3x10 11 Minimum 10 9 10 9 10 7 10 7 6 / 24 polarization switch decays with the distance from tip position. Also, the previously polarized region may be partially switched by adjacent line scan with reversed bias if in its influence region. Thus, the transition region of doping shall be related to the polarization switch area induced by AFM tip polling, which shall be influenced by the spatial resolution of tip scan or the domain size in FE thin film.
In our experiments, to define FE polarization pattern in device, we adopted scan resolutions of 256x256 or 512x512 for 10x10 μm 2 to 20x20 μm 2 device area, meaning a fine spatial resolution of ~40 nm between adjacent line scans.
On the other hand, to examine the domain size in P(VDF-TrFE) thin film, we performed direct PFM imaging of the domain size formed by applying local polarization using an AFM tip set at 10 V. The results were now supplied in supplementary materials as Figure 4.

Reply:
We appreciate the reviewer's careful examination and pointing out this. As the reviewer noticed, we previous adopted two devices throughout the manuscript, one for p, n and pn diode Since the discussion in Figure 1, 2 focused on the discussion of p/n doping in MoS 2 enabled by FE polarization switching, and Figure 6 intended to validate the high gain observed in npn bipolar transistors, they were kept unchanged in this revision. To avoid confusion, we have added a Table 1 listing all the devices presented in the manuscript by including MoS 2 thickness, width and length in channel and their defined functions.  3, 4, 5 in manuscript were changed as following, the corresponding discussion were also updated in the main manuscript.
In Figure 3, b-e were replaced with data collected from a new Dev. 2, and the photocurrent map from Dev. 3 in the revised manuscript.

/ 24
In Figure 4b, PFM image for the device configured into npn phototransistor was replaced with that for Dev. 2 in revised manuscript.
In Figure 5. a-e were all replaced with data collected from new Dev. 2 and the photocurrent map from Dev. 3 in the revised manuscript.

/ 24
The related discussion of the above figures were all updated in the revised manuscript. Fig. 3b, asymmetric I-V characteristics are frequently seen in 2D materials such as WSe2, MoS2, etc, due to the Schottky barriers at the contacts. Could the authors comment on how the effects of the contacts are accounted for?

Reply:
We appreciate the reviewer's comment by mentioning the possibility of asymmetric contact in forming the rectification.
To clarify this issue, we performed photocurrent mapping on a pn diode made by FE polarization in the self-driven mode without applying external bias. Since the photocurrent generation relies on successful separation of photogenerated electron-hole pairs in MoS 2 , the spatial distribution of photocurrent intensity reflects the local electric field. As indicated in Figure   3f, To address the same concern from readers, we have provided the photocurrent map in  channel, compared to its optical microscope image.
The following discussion was included in the revised manuscript at Page 9 line 21-Page 10 line 7.
"To elucidate the self-powered photocurrent generation, a spatial resolved photocurrent distribution has been characterized for a short circuited pn diode defined in Dev. 3. This was achieved by locally illuminating the device using a fine laser spot (λ=532 nm) in a confocal microscope. The pn junction was defined in the middle of MoS 2 channel. Figure 3f displays the optical microscopy image of the device and the associated photocurrent map. It was clear that most of the photocurrent was generated near the defined junction. We note that self-driven photocurrent may also appear in Schottky contacted devices, but usually with reversed polarity near the source and drain electrodes due to opposite charge separation. 55-57 However, in all the devices we studied, photocurrent barely appeared near the contact electrodes and there was no change on the photocurrent polarity across the device area. These results thus validated the role of FE polarization defined pn diode in bringing the high rectification in IV characteristics and self-driven photocurrent." 6. I am also missing a justification for the current-voltage characteristics found in Supplementary Fig. 2. If that was for gate modulated current in pn diodes at the dark and illuminated condition for different configurations to get qualitative comparisons, it should be mentioned in the paper.

Reply:
We are sorry for the confusion to Supplementary Fig. 2. They were not gate modulated current in pn diode, but corresponded to the current in MoS 2 device under varied FE polarization conditions, with V p from -20V to 25V.
The intention of the figure was to reflect the contact characteristics in differently doped MoS 2 , including the initial n-type, and FE polarization induced heavily n-doped, depleted and reversely p-doped states. In the figure, logarithm scale was adopted in I ds or both I ds and V ds axis to clearly distinguish the different current range at each state, and in Supplementary Fig.   2b to reveal the space charge limited current (SCLC, with I ds~Vds 2 ) in p-doped MoS 2 at large bias.
The SCLC behavior was usually observed in insulators or semiconductors with rich trap defects.
In present case, both SCLC behavior (supplementary Fig. 3) and Ohmic contact ( Figure 3b) were found in p-doped MoS 2 by FE polarization, because of the different hole doping state. In lightly p-doped device I (G=0.1 nS at V ds =0.1V), SCLC appeared because of the presence of hole trapping centers within the bandgap, which induced space charge by capturing holes injected from electrodes under large bias. In comparison, for heavily p-doped MoS 2 (device II), the contact was nearly Ohmic because of the above hole trap states were prone to be occupied already given that E F is closer to valance band.
To avoid this confusion, the discussion on supplementary Fig. 2  The authors present lateral p-n and n-p-n junctions based on a two-dimensional semiconductor (MoS2), enabled by local patterning of the ferroelectric polarization in P(VD-TrFE). Using this technique, they realize reconfigurable devices, such as diodes and bipolar transistors, and demonstrate optoelectronic applications such as photovoltaic energy conversion and photodetection.
Although the idea of using ferroelectrics for such purpose is not entirely new, this work constitutes an important advance in this emerging field as the device performance appears to be dominated by the properties of the junctions, rather than the contacts. I thus would like to recommend publication of the manuscript in Nature Communications, provided that the authors can satisfactorily address the following comments/issues: We sincerely thank the reviewer's comments. All the concerns were addressed as following: 1. In Fig. 3c the authors present photovoltaic properties of their p-n junction. (It is not clear how the data were taken; I assume the whole device was illuminated?) The authors should present the same measurements for n-n and/or p-p configuration to exclude the possibility that the photoresponse stems from the (possibly asymmetric) metal/MoS2 Schottky junctions.
Alternatively, the authors may choose to locally illuminate the p-n junction.

Reply:
We thank the reviewer's suggesting in improving the discussion. Further, we also attempted to locally illuminate the p-n junction and collected a photocurrent map in short circuit mode, which is now included in the main manuscript as Figure 3f. The results clearly demonstrate that the self-driven photocurrent stemmed from the junction area defined in the middle of MoS 2 channel. Since there were barely photocurrent observed near the contacts, the potential Schottky contact effects can be explicitly excluded. In this revision, we have improved the discussion as following, to account for the potential Schottky contact effect in the observed rectification and self-powered photodetection: 2. Device operation as bipolar transistor (Fig. 6) is convincingly demonstrated. When operated as phototransistor, though, the device shows some behaviors that resemble those commonly seen in pristine MoS 2 devices: namely, the drop of the photoresponsivity with illumination intensity and the rather slow response time (4 ms). The authors argue by comparison with a pristine MoS2 device (Fig. 3d) that the response times observed in their junctions are shorter, but 4 ms still seems very long to me. Can the authors provide further evidence that the gain stems from the transistor operation, rather than charge trapping in (short-lived) defects?

Reply:
We thank the reviewer's careful examination and comments. The bipolar transistor is known  In response to the reviewer's comments, we have included the following discussion in the revised manuscript.
At Page 11 line 21-24 "By using a fast switching 365 nm LED source, the photoresponse speed of the bipolar transistor was estimated ~20 μs (inset of Fig. 5a), making it one of the fastest MoS 2 photodetectors but with high gain characteristics." At Page 12 line 9-23 "It was noticed R slightly decreases with the negative shift of V th at higher light intensity, which was also usually found in other type phototransistors with photogate effect. However, we emphasize that the origin of such dependence was different from the usual saturated charge trapping or separation in phototransistor, but due to the increasing recombination losses at the forward biased E-B junction under large injection. 28 In Figure 5d, we present the photocurrent map for the present npn bipolar phototransistor under V CE =0 and 1V to validate its operation principle. It was seen that photocurrent of reversed polarity appeared near C and E terminal at V CE =0V, while at V CE =1V the photocurrent was more efficiently generated near the reversely biased B-C junction. This was consistent with the expected electric field strength in device that eventually separate the photogenerated electron-hole pairs. The present photocurrent map also differed from other type phototransistors that usually displayed uniform photocurrent distribution within the photogate area. 60-62 Such difference in the two kinds of phototransistors was indeed ascribed to their different gain generation mechanism in device, i.e. via the lateral in-plane charge injection and the out-of-plane photovoltaic effects, respectively." -It would also be helpful if the authors could comment on the stability of the junctions (versus time, bias voltage, etc.)

Reply:
We thank the reviewer's comments. Though  In experiments, we observe that the device after AFM polling tended to degrade in 1-2 hours, without apparent dependence on the applied bias but can be accelerated under thermal effect.
In the following figure, MoS 2 conductance after P/N doping by FE polarization was seen to degrade, but after 1h still maintains the defined P/N doping type. The study reported here were generally conducted within a short period before significant degradation occurred. An example of the degradation for p and n-doped MoS 2 was shown in following.

/ 24
We note that such instability could be feasibly avoided by 1. It is not clear why such reconfigurability is needed. We can easily apply an electric bias to tune the device operational condition. Moreover, there are many different approaches to make photodetectors and the introduction of ferroelectric materials into optoelectronic devices do not seem to be well-justified, at least in this case. You can make a pn junction or directly make a phototransistor. You can also make an APD. Sometimes you do need reconfigurable photonic devices (e.g. in optical networks) but I do not think here making a reconfigurable photodetector has intrinsic advantages.

Reply:
We thank the reviewer's critical comments, which helps us to improve the work. It is undoubted that the reconfigurability renders much freedom in defining the device function or performance after manufacturing. For photodetectors, the essential requirements include fast response, high gain, and low energy consumption in arrayed image sensors. This was usually fulfilled in market by different kinds of photodetectors, e.g. diodes, transistors, or APD, depending on the target application in sensing, imaging or optical communication. It was however difficult to balance the response speed, gain and energy consumption in complicated scenes, e.g. the one perceived by human vision system with dramatically varied light conditions.
The photodiode is the most energy efficient type of detector because of its self-driven operation, whereas its low gain <1 make it more suitable for high light levels. In the revised manuscript, the following sentences have been included to elucidate the potential merits of having reconfigurability in optoelectronic devices.
Page 2 line 10-11 "An ultimate pursuit to this end would be however a reconfigurable function device that can be customized on demand, so that a universal device architecture can be deployed in various application scenes."

Page 3 line 22-24
"Such reconfigurable device characteristics may promote the evolvement of smart image sensors that reflect to external light environments for the balanced photoresponse gain and energy efficiency." 2. The performance of the photodetector is not very impressive. Indeed the responsivity is high due to the gain. However the response time is long (4 mS) and as a result, the speed is very low (below kHz). It is very easy to achieve high responsivity if you do not care about the speed.
The difficult part is high responsivity, high speed and low noise simultaneously. In fact, such a high gain can be easily achieved in a simple silicon photoconductor if high speed is not needed.

Reply:
We thank the reviewer's critical comments. The npn transistor could be potentially operated at a fast response speed and high gain. In our case, the optical gain >1000 was so far the highest in various kinds of bipolar transistors based on 2D materials, whereas due to the speed limitation of light source and measurement unit, the photoresponse speed in the configured npn transistor was previously underestimated. In this revision, we have performed another measurement of the photoresponse speed by using a fast switching LED (M365FP1, Thorlabs) and a fast measurement unit (B1530, Agilent). We found the npn phototransistor exhibited a fast photoresponse <20 μs, making it one of the fastest photodetectors based on 2D materials while offering the high gain factor, as indicated in the following Table S2. The results could therefore demonstrate the high performance of the present npn phototransistor, and the 23 / 24 potential of exploiting FE polarization enabled p, n doping in 2D materials and further patterned doping for function device applications.
In this revision, an inset figure was added in Figure 5a showing the fast transient response, and in Figure 5f the responsivity and speed were updated with the new data collected from the npn phototransistor on device II, thereby demonstrating the fast photoresponse capability of npn transistors.

Reviewer #1 (Remarks to the Author):
The new manuscript has been well revised according to the reviewers' comments. In particular, they have detailed the rewritable capability that was lacking in the original manuscript and have provided experimental evidence by adding different devices. As a result, the paper is in much better shape now. Therefore, I suggest that this revised manuscript is now suitable for publication.
Reply: We sincerely thank all the comments from the reviewer that helped to greatly improve our manuscript.

Reviewer #2 (Remarks to the Author):
The authors have fully addressed my remarks and have significantly improved the quality of the manuscript. I recommend publication of the manuscript in Nature Communications.
Reply: We really appreciate all the reviewer's comments that instructed us to improve the discussion and quality of the manuscript.

Reviewer #3 (Remarks to the Author):
The authors addressed some of the concerns but the referee here still feels this work does not represent an important breakthrough in nanophotonics.
1. First, the speed is still very low. 20 us response time, in fact, indicates that the device can only operate in tens of kilohertz. Such a speed is not impressive at all even with the gain of 1000. If we use the gain-bandwidth product as the performance metric, an APD can easily operate at a gain of 100 and speed in GHz range (ns response time). The gain-bandwidth product is at least 3 orders of magnitude better for APD (it could even be 4 orders of magnitude better). Reply: We appreciate the reviewer's critical comments to the manuscript, which have pushed us to improve the performance of detector. By improving the sampling rate in measurements, we observe the best photodetection speed for the npn photodetector was ~3-5 μs, which at present stage shall be the switching limit in our experiments setup (Supplementary Figure 11).
However, in principle, the speed of bipolar transistor could be optimized by suppressing extrinsic defects in MoS 2 and by refining the width and length of base and collector region, so to minimizing the effect of charge trapping and parasitic capacitance in response speed. In terms of the response speed and gain, the present bipolar phototransistors yield competitive performances at considerably lower operation voltages, which may benefit its application in imaging and wearable devices.
To clearly explain their difference, the following discussion have been included in the main manuscript: On Page 10 line 10-12 "Compared to the avalanche photodetector (APDs), the bipolar phototransistor could work at considerably lower operation voltages (~150 V for commercial Si APDs) while yielding the similar photodetection gain." On Page 10 line 25-Page 11 line 5 "Faster response within as short as ~3-5 μs is also achieved in experiments (Supplementary Figure 11), which is close to the switching limit of the adopted light source. It is believed that the ultimate device response speed depends on both material characteristics and device geometries. Further improved speed is likely attainable given higher carrier mobility in MoS 2 and improved design on the width of base and collector, as they directly determine the overall carrier transit time in device.
2. Second, modern optoelectronic devices are usually made from semiconductor heterostructures for optimal performance. The referee here does not think such gate-induced, homojunction pn diodes and pnp (or npn) transistors can have potential for practical applications. Reply: We appreciate the reviewer's comments. Heterojunction bipolar transistors (HBT) principally yield higher gains than homojunctions by having a wide band gap semiconductor as the emitter compared to base with narrower bandgap. However, the high performance comes at the price of the efforts in optimizing the junction interface, as good lattice matching and highquality interface are always necessary to reduce electron-hole recombination loss there.
Therefore, the present HBT markets are dominated by III-V semiconductors (InP, GaAs, AlGaAs, GaN) and SiGe heterojunctions. However, it should be mentioned that the homojunction bipolar transistor could offer well balanced sensitive and speed with considerably reduced material and fabrication cost. Again, it shall have position in photodetector market when cost is concerned.
In addition, the present bipolar transistors based on the ultrathin 2D van der Waals materials may benefit flexible detectors compared to the traditional devices based on vertical epitaxy layers by avoiding strain issues. [Nature Communications, 2018, 5266] It is therefore believed that 2D homojunction bipolar transistor can be practical by fulfilling specified several but not all requirements.
As the reviewer has mentioned, hetero-structured bipolar transistors are promising for better performances. This can also be done based on artificially assembled 2D semiconductors, whereas at present the limited control on the exfoliation and stacking of 2D materials, and also doping state in each region (emitter, base and collector) make it challenging to achieve optimal device performances. With the ultrathin thickness (<10 nm) of 2D materials comparable to the Debye length, gate modulations were often employed to enhance the junction characteristics and optimize the performance of various kinds of 2D devices, the same however cannot be applied for conventional 3D semiconductors with large characteristic size. In present work, the adopted ferroelectric gate modulation not only enables widely tuned doping polarity and carrier concentration over conventional gate oxides, the reconfigurable polarization pattern here also greatly facilitates the exploration of high-performance photodetectors by feasibly changing the design, which have not been possible before. The device configuration thus is also not limited to the already demonstrated pn junction and bipolar transistors, but includes APDs mentioned by the reviewer. Further exploration of ferroelectric coupled photodetector in different designs would therefore undoubtedly promote the evolution of high performance 2D photodetection.
Accordingly, we have included the following discussion that outlooks the possible future efforts toward higher performance photodetection.
On page 15 line 1-4 "The gate-free yet reconfigurable methodology introduced the great potential of exploiting locally coupled FE polarization in customizing high performance optoelectronic devices based on the thriving 2D semiconductors and in the future their van der Waals heterojunctions, which in principle could offer even larger speed and gain product than present homojunctions.
Finally, we sincerely thank all the reviewer's comments that pushed us to improve the discussion and quality of the manuscript.