Nitrogen‐Doped Carbon Quantum Dots on Graphene for Field‐Effect Transistor Optoelectronic Memories

The development of field‐effect transistor‐based (FET‐based) non‐volatile optoelectronic memories is vital toward innovations necessary to improve computer systems. In this work, for the first time, the unique charge‐trapping and charge‐retention properties of solution‐processed colloidal nitrogen‐doped carbon quantum dots (CQDs) are harnessed to achieve functional optoelectronic memories programmable by UV illumination with a multilevel writing possibility. Of particular note, long‐lasting memory function can be achieved thanks to the vast charge trapping sites provided by the N‐doped CQDs and the resultant photo‐gating effect is exercised on the graphene FET. The achieved memory can be erased by a positive gate bias which provides sufficient carriers to remove trapped charges through recombination. This study highlights the possibility to engineer high‐performance all‐carbon non‐volatile FET‐based optoelectronic memories through manipulating and coupling the charge‐trapping properties of colloidal CQDs and graphene.


Introduction
Innovations on memory components are essential to improve current computer systems, the performance of which is DOI: 10.1002/aelm.202300159 limited by the "von Neumann bottleneck." [1,2] As the developments of Internet of Things and big data processing are growing rapidly, there is a high demand for solutions toward ultrahighdensity non-volatile storage. Toward the development of high-performance memories, much research effort is currently being carried out, leading to the achievements on different types of nonvolatile memories such as capacitor-type memories, resistor-based memories, and transistor-based memories. [3] A major workhorse of transistor-based memories is composed of floating-gate field-effect transistors (FETs) which show advantages in terms of reliability, scalability, power consumption, non-volatility, and compatibility. They lead to the formation of non-volatile flash memories which are compatible with the complementary metal-oxide-semiconductor technology. In the field of nonvolatile flash memories, while silicon-based memories have played a significant role so far, [4] in order to overcome the limits of this technology, further research is still needed to identify new materials and devices alternative to silicon-based memories with new abilities to combine characteristics such as low material cost, absence of heavy-metal toxic elements, adaptability in large-area fabrication, mechanical flexibility, and low-power consumption. [3,[5][6][7][8] In particular, optoelectronic memories based on thin-film FETs, where the programming and/or erasing are attained by illumination instead of electrical bias, exhibit advantages in terms of power consumption and compatibility with optical circuits and artificial neuromorphic applications. [9][10][11][12][13] Concerning memory applications, colloidal nanocrystals or nanoparticles have aroused much interest to serve as the chargetrapping or charge-retaining material in non-volatile memory devices. [14][15][16][17][18] This is due either to their rich surface properties providing abundant charge traps [19][20][21][22][23][24] and/or to the variety of available synthesis/engineering approaches leading to formation of core-shell [21,25] or other nanoparticle/dielectric hybrid structures, [7,21] all of which capable to secure and store charges within the nanoparticle. In addition, many directbandgap semiconductor nanoparticles possess strong light absorption coefficients together with absorption/emission properties tunable by synthetic controls either over their dimension or surface properties. Their optical and charge-trapping properties were thus harnessed to achieve FET-based opto-www.advancedsciencenews.com www.advelectronicmat.de electronic memories over the past years, including attempts applying cadmium and zinc chalcogenide QDs and related core/shell structures, [21,25] rare-earth-doped upconversion fluoride nanocrystals, [26] perovskite quantum dots (QDs), [27] as well as TiO 2 dielectric nanoparticles. [19] For example, by embedding Yb 3+ /Er 3+ -co-doped NaYF 4 colloidal nanocrystals inside a layer of poly(3-hexylthiophene) within a FET architecture, Zhou et al. demonstrated flexible multilevel optoelectronic memories programmable by near-infrared illumination at = 980 nm. [26] Nevertheless, rare-earth materials are not sustainable and their low absorption coefficient limits the responsivity of the device.
Other similar examples, programmable by visible illumination, were also demonstrated by organic conjugated molecules and polymers. [28,29] Yet, typically, a relatively large lateral source-drain bias (e.g., >10 V) is necessary for their programming. In addition to optoelectronic memories that utilize visible illumination, ultraviolet (UV)-triggered memories can exhibit advantages avoiding interference from solar irradiation and indoor lighting. [30] Toward this direction, Han et al. sandwiched charge-trapping CdSe/ZnS core-shell QDs between dielectric layers to achieve flexible UV-controlled FET-based photonic memories. [21] Star et al. applied CVD-grown carbon nanotubes for charge storage in a polymer thin film FET leading to UV-writable optoelectronic memories. [7] On both examples, besides the source-drain voltage, an important gate bias is also necessary to work together with the UV illumination to achieve programming. Recently, Sun et al. decorated the surface of graphene FETs by nitrogen-doped (N-doped)-TiO 2 nanocrystals to achieve highly promising UVwritable multilevel optoelectronic memories which exhibit negligible current loss 10 h after UV programming. [19] The synthesis of such nanocrystals requires nevertheless flammable and dangerous precursor (titanium (IV) butoxide) due to its high sensitivity to air and moisture. Toward the further development of this field, alternative solution-processed material candidates synthesized from environmentally benign and/or biosourceable reactants with earth-abundant compositions can potentially lead to further progresses.
To this aim, colloidal carbon quantum dots (CQDs), [31][32][33] highly luminescent nanoparticles of carbon which is one of the most abundant elements in our planet, could be excellent candidates. They can be readily synthesized in water by various biosourced materials such as citric acid, [34] orange peels, [35] water hyacinth leaves, [36] and pomegranate [37] under relatively mild synthesis temperature (≤200°C) by hydrothermal, solvothermal, or microwave routes, giving nanoparticles with dimensions ranging from a few to a few tens of nanometers. Instead of being controlled by the nanoparticle dimension, their optical properties are controlled by their rich surface chemical groups and dopants which can be tuned from the synthetic routes and the choice of reactants. [38][39][40] Attracted by their favorable properties such as high optical absorbance and fluorescence, chemical stability, biocompatibility, and low toxicity, [41,42] previous works have successfully applied them in bio-imaging, [43] drug delivery, [44] photocatalytic conversion, [45] and photovoltaic devices. [46][47][48] Toward FET-base memories, Zhang et al. has embedded CQDs inside the active layer of polyvinyl pyrrolidone (PPV)-based FET. [49] Nevertheless, such CQD/PPV FET-memories are not optically enabled and its programming necessitates a high electrical bias of 60 V. To our best knowledge, there is not yet any report of applying colloidal CQDs in optically enabled optoelectronic memories.
Herein, for the first time, we harnessed the unique chargetrapping and retention properties of solution-processed N-doped CQDs to achieve functional UV-enable optoelectronic memories with multilevel writing possibilities. These memories are achieved by exploiting the optical and charging-trapping properties of N-doped CQDs to result in a strong photo-gating effect on graphene FETs. After UV-programming, prolonged charge retention can be obtained exhibiting >96% of its initial programmed photocurrent value for a period of 10 h. Such a programmed state can be subsequently erased by a positive gate bias. Together with their facile colloidal synthesis, abundant nature, environmentalfriendly compositions, and their excellent memory functionalities, these fully carbon-base non-volatile optoelectronic memories thus provide bright perspectives for the further development of this field.

Results and Discussion
The N-doped CQDs under study were synthesized by a reported hydrothermal route with mild synthesis temperature (200°C) by condensing citric acid together with ethylenediamine (EDA). [41] Transmission electron microscopy (TEM) images of assynthesized N-doped CQDs exhibit nanoparticles with a diameter mainly between 2 and 10 nm (Figure 1a). When characterized by high-resolution TEM, some of them exhibit clear lattice fringes as the one shown in the inset of Figure 1a, where a 0.21nm interplane distance can be measured corresponding well to the (100) plane distance in graphitized carbon (JCPDS no. . When deposited onto a clean Si substrate, these N-doped CQDs reveal a typical height of 2-5 nm by Atomic force microscopy (AFM) characterizations (Figure 1c,d), confirming the dimension of the nanoparticles. Additional TEM and AFM images can be found in the Figure S1, Supporting Information. In terms of optical properties, N-doped CQDs exhibit two major absorption peaks in the UV spectrum with one located at ≈ 256 nm and another strong and broad peak located at ≈ 356 nm ( Figure 1b). According to previous assignments on similar carbon or graphene QDs, [40,50,51] the high-energy absorbance band corresponds to -* transitions in the sp 2 carbon core while the low-energy broad absorbance band contains multiple contributions from both n-* transitions (e.g., n O2p -* and n N2p -*) and -* charge transfer transitions (between the inner and the outer part of the sp 2 domains). Analyzing the on-set of optical absorption by a Tauc plot allows us to determine the optical bandgap as ≈ 3.1 eV ( Figure S2a, Supporting Information). The obtained photoluminescence excitation (PLE) spectrum is in agreement with the absorbance with a more resolved view on the transitions contributing to the PL (fitting on the broad peak shown in Figure S2b, Supporting Information). A peak PL intensity was observed when the excitation wavelength is at ≈380 nm, presumably corresponding to n-* transitions which was considered as the major radiative decay pathway for CQDs. Strong blue PL can be observed when these N-doped CQDs are excited at different wavelengths ( Figure 1b). Under an excitation wavelength ( ex. ) of 405 nm, time-resolved PL trace reveals a bi-exponential decay behavior with an average PL lifetime of about 3.4 ns ( Figure S2c, Supporting Information), which is in agreement with previous works [52,53] on this type of CQDs. Near identical PL peak shape was observed at different ex when it is shorter than ≈400 nm ( Figure S3d, Supporting Information). At a longer ex , the PL spectrum starts to red-shift with a significant reduced intensity likely due to the increased contribution of the surface states in the PL [40] and the limited absorbance at these wavelengths. The photoluminescence quantum yield (PLQE) of the N-doped CQDs in water was measured to be about 40% under ex = 365 nm. The presence of EDA in the reactants of the present synthesis leads to the formation of N-doping, which was found to contribute to a higher PLQE in comparison to the CQDs synthesized without EDA. [41] For the purpose of comparison in the memory devices which will be discussed below, undoped CQDs (without any nitrogen-containing compounds in the reactants) were also synthesized. Coherent with the absence of heteroatom doping, the UV-vis absorbance of undoped CQDs exhibited a strong -* transition band and a very weak low-energy broad absorption band which could be assigned as n-* and -* charge transfer transitions ( Figure S3, Supporting Information). By comparison to N-doped CQDs, a much-diminished PL was observed on undoped CQDs at ex = 365 nm ( Figure S3, Supporting Information) with a very small PLQE reaching only about 1%. This is likely due to both the absence of the nitrogen doping and the very small absorption coefficient at this ex .
Both X-ray diffraction (XRD) and Raman scattering experiments suggest the presence of a high degree of disordering in the present N-doped CQDs likely due to defects and dopants.
The XRD spectrum of the N-doped CQDs shows only a broad diffraction peak at 2ϴ ≈ 21.5°which is at a lower diffraction angle than the (001) diffraction peak expected for crystalline graphite (2ϴ = 26.4°, JCPDS no. 41-1487) (Figure 2a). This is likely due to the presence of a high disorder alone the C-axis and/or the presence of surface/dopant groups enlarging the interplane distance along this axis. The above analysis is coherent with the Raman spectrum obtained (Figure 2b), where two Raman bands, located at the Raman shift of 1355 and 1645 cm −1 , can be observed and assigned as the D and G bands, respectively. The ratio between these two Raman bands (I D /I G ), typically considered as an indicator of disorder or defect density and as the ratio between sp 3 /sp 2 carbons, [54,55] was measured to be ≈3.2 on the current sample, suggesting a breakage of the hexagonal symmetry of the graphene due to defects and/or heteroatom dopants. From the obtained I D /I G ratio and the laser wavelength, by the relationship proposed by Matthews et al., [56] we estimated the (average) size of the sp 2 domains of the sample to be about 4 nm, which is coherent with the nanoparticle dimension characterized above. Fourier-transform infrared (FT-IR) spectroscopy was further performed on both the N-doped and undoped CQDs in the present work (Figure 2c). On both samples, different vibrations can be observed and attributed to O-H, C-H, C=O, and C-O stretching according to previous reports. [34,[57][58][59][60] In comparison to undoped CQDs, N-doped CQDs exhibit additional vibrations which can be assigned to C-NH-C asymmetric stretching, N-H bending, and/or C=N stretching, and possibly also to the  CN stretching, confirming the nitrogen-doping in the N-doped CQDs. X-ray photoelectron spectroscopy (XPS) results confirm well that the present N-CQDs mainly composed of C, N, and O (Figure 3a). The C 1S spectrum can be fitted into four peaks centered at 284.5, 286.1, 287.7, and 288.3 eV, which can be attributed to bonds of C-C, C-N, C-O, and C=N/C=O, respectively, according to previous reports on CQDs (Figure 3b). [61] The N 1S spectrum reveals two peaks centered at 399.8 and 401.1 eV, which can be assigned from pyrrolic and graphitic N contributions, [62] respectively ( Figure 3c). Finally, the O 1S spectrum shows two peaks centered at 531.0 and 532.3 eV, which are attributed to bonds of C=O and C-OH/C-O-C, [61] respectively.
The device architecture of the N-doped CQD-decorated graphene FET-based memory is shown in Figure 4a, where a bottom-gate-top-contact FET structure was applied. The Raman spectrum of the graphene transferred onto the Si/SiO 2 substrate suggested the existence of a single-layer graphene with a similar quality as previously reported CVD-grown graphene ( Figure  S4, Supporting Information). [63] After depositing N-doped CQDs on the surface of graphene, AFM characterizations were performed inside the FET channel (Figure 4b). While the present AFM experiment do not have the resolution to resolve an individual CQD, it confirms well their existence on the top surface of the graphene, likely as isolated few-nanoparticle clusters. Figure 4c exhibits a picture of a typical sample under test: Six graphene FETs were fabricated onto the same Si/SiO 2 substrate while the channels of two of them were further decorated by N-CQDs (which exhibits blue fluorescence under UV illumination). Comparing the transfer characteristics of the graphene FET measured in dark before and after depositing N-doped CQDs, a shift of the Dirac point toward negative gate voltages was observed triggered by the deposition of N-doped CQDs ( Figure S5, Supporting Information). Such a n-type doping behavior is likely due to the interactions between the surface groups and/or defects of the CQDs and the free charges inside the graphene FET channel. Importantly, the decoration of the N-doped CQDs on the graphene FET enables the device a clear photoresponse under UV illumination: A series of UV illuminations (at ex = 365 nm) of different power densities (E a ) was shone from above onto the N-doped-CQD-decorated device. As shown in Figure 4d, the UV illumination induced a further shift of the Dirac point of the transfer characteristics toward more negative gate voltages. The magnitude of this shift in Dirac point voltage increases as a larger E a was applied and the direction of this shift also indicates a further n-type doping behavior on the graphene due to the photoexcited N-doped CQDs. This observed UV-sensitivity is clearly different from that observed on graphene FETs without any CQDs, as the same device before depositing N-doped CQDs exhibited negligible shift in terms of Dirac point voltage even under the strongest UV illumination of the present study ( Figure S5, Supporting Information). This thus confirms that the observed UV photoresponse should be attributed to the existence of the N-doped CQDs and the interactions between the photoexcitation QDs and the graphene.
The Dirac point voltage shift (ΔV Dirac ) measured in the transfer characteristics of the N-doped-CQD/graphene device under different UV illumination power densities (E a ) with reference to the Dirac point measured in dark was plotted in Figure 4e. According to previous works on phototransistors and phototransistor-based memories, [19,[63][64][65] such a ΔV Dirac -E a relationship can be modeled by power law relationship with ΔV Dirac = aE a , where and are both constants with typically of values between 0 and 1. A value much smaller than 1 can, for example, originate from a saturation of the charge transfer processes between the photoabsorber and the graphene as the illumination irradiance increases due to an increased extent of charge recombination offsetting the increased charge generation. Here, an exceptional nearlinear relationship between ΔV Dirac and E a was observed with fitted as 0.99 (Figure 4e), indicating a non-saturated photoresponse under the different illumination irradiances of this study. Taking into account of the HOMO and LUMO values measured previously [40,66,67] on similar N-doped CQDs and the observed shift of ΔV Dirac over E a , we attribute the physical mechanism behind the current N-doped CQD/graphene device as the photogating effect, equivalent to a change in terms of the effective gate voltage due to illumination (Figure 5). Such a photo-gating effect has been observed in other graphene-based FETs where the graphene was decorated or coupled with another photo-absorber such as PbS QDs, [68] a perovskite halide layer, [69] black phosphorus nanosheets, [70] and TiO 2 nanocrystals. [19] Here, upon UV absorption, while both types of photoexcited carriers in the Ndoped CQDs can be transferred to graphene due to the favorable energy alignments, the observed n-type doping behavior upon illumination suggests a more significant electron transfer than hole transfer to the graphene (Figure 5a). This is possible if the photo-generated holes are transferred to the graphene either less efficiently or hindered by traps. [65] As N dopants were found previously to form mid-gap deep hole traps for other types of nanoparticles, [71][72][73] we thus conjecture that the N-dopants in the current case should serve equally as hole traps. This will be further discussed below when we compare the device behavior with non-N-doped CQDs. As a consequence, the net remaining holes left behind the N-doped CQDs induce more electrons in the graphene channel due to Coulomb interactions, leading to a more negative gate voltage to reach the Dirac point (Figure 5b). The fact that a large and near-unity was found above in the ΔV Dirac − E a relationship suggests the ample availability of hole traps in the N-doped CQDs so that the charge transfer processes are less limited by charge recombinations under the irradiances applied.
Coherent with the photo-gating and nitrogen hole-trapping mechanism proposed above, in N-doped-CQD-decorated graphene devices an exceptional charge retaining behavior was observed (Figure 6). When the mobility remain unchanged (which is the case in Figure 4a), the Dirac point shift (ΔV Dirac ) triggered by UV illumination translates into photocurrents following the relation of ΔI DS = W L C i ΔV Dirac V DS assuming the FET in the linear regime, where C i is the capacitance per unit area of the gate dielectric, W and L are, respectively, the channel width and channel length, I DS and V DS are, respectively, the source-drain current and the source-drain voltage, and the photocurrent ΔI DS is the change of I DS triggered by illumination by comparison to the I DS,dark measured in dark. The evolution of ΔI DS over time of the N-doped CQD device was monitored at a gate voltage V G = 0 V and a source-drain voltage V DS = 0.5 V while a UV light ( = 365 nm, E a = 0.7 Mw cm −2 ) was shone onto the device from the top for a duration of 120 s. Under this illumination, the ΔI DS /I DS,dark ratio of the current device reach ≈16%. As shown in Figure 6a, a rapid increase of photocurrent was first observed upon UV illumination. Subsequently, after the UV illumination was switched off, the I DS was maintained at nearly the same programmed level with a slight further increasing trend (up to ≈8%) during the first 2 h after switching off the light. Such a slight increase of I DS is likely due to the device instability and/or bias-stress experienced by the device. Beyond this time window, the I DS starts to decrease very slowly, reaching back to 96% of its initial value (i.e., the I DS value right after switching off the light) 10 h after switching off the light. Based on the observed trend of how the I DS decreases over time, extrapolation suggests ≈>37% of charge retention can be obtained 10 years after switching off the light. Both the rapid increase of I DS during illumination and the subsequent charge retention after removing the illumination are coherent with the mechanism proposed where a large amount of hole traps, likely due to nitrogen-dopants, exist in the N-doped CQDs capable to firmly immobilize photoexcited holes preventing them from recombination and resulting in a long-lasting photo-gating effect (Figure 6c). By comparison to previous literature on similar graphene FET-based memories coupled with N-doped TiO 2 nanocrystals, [19] the current charge retention capability is remarkable, maintaining ≥100% of the programmed state (I DS ) for 7 h after switching off the light. In order to probe whether nitrogen dopants are indeed the origins of the hole traps and the observed memory retention, near identical device as mentioned-above was fabricated except that undoped CQDs were applied to decorate the graphene in this case. The transfer characteristics of the undoped CQD/graphene FET under UV illumination at ex = 365 nm of different power densities are shown in Figure S6, Supporting Information. In comparison to N-doped CQD device, under an illumination of ex = 365 nm and E a = 0.7 mW cm −2 , the ΔI DS /I DS,dark ratio and the photoresponsivity of the undoped device are significantly smaller, reaching only ≈1.3%. The evolution of ΔI DS over time of the undoped device was monitored the same way as the N-doped device discussed above (Figure 6b), exhibiting a very different behavior. Notably, in the undoped CQD device, the photocurrent decreases progressively down to less than 30% of the highest level within only about a thousand seconds after switching off the light. In another word, in undoped devices, the capability of light-induced charge retention is rather poor, resulting in a more than 70% of loss of programmed state within a thousand seconds (i.e., approximately half an hour) after memory writing. Indeed, as shown in the FT-IR spectrum, the current updoped CQDs also have rich chemical groups (e. g. O-H, C-H, C=O, and C-O groups) likely mostly on the surface. Even without nitrous carbon groups, in undoped CQDs these surface groups and/or defects may also serve as traps for the photoexcited carriers leading to the observed photoresponse, which is coherent with the non-instantaneous decay of the photoresponse after switching of the UV. Nevertheless, in the absence of deep traps capable of retaining charges firmly, the photo-gating effect diminishes rather rapidly after switching off the illumination due to carrier recombination (Figure 6d). The comparison of the photocurrent (ΔI DS ) evolution over time upon and after UV illumination between the N-doped and the undoped device indirectly confirms the vital role of nitrogen dopants toward a long-lasting non-volatile memory function.
Consistent with the nitrogen-induced hole-trapping and photo-gating mechanisms proposed above, the achieved memory in the N-doped CQD device after UV programming can be successfully erased by a positive gate voltage applied. Figure 7a exhibits five complete UV-writing and V G -erasing cycles, where the I DS of the device (measured at V G = 0 V and V DS = 0.5 V) was first increased to a high level thanks to a UV illumination ( = 365 nm, E a = 0.7 mW cm −2 ) for a duration of 30 s. Then a positive gate voltage of +50 V was applied on the device for another duration of 30 s, resulting in a drop of I DS back to a level very close to its initial value before UV programming. The characteristics of 50 complete UV-writing and V G -erasing cycles are shown in Figure S7, Supporting Information, exhibiting globally good reproducibility of the programming and erasing process. The schematic proposing the erasing process is shown in Figure 7b. Under a large positive gate voltage, strong electron accumulation is resulted in the FET channel and the Fermi level of  the graphene is raised. A large number of electrons can thus be transferred back from the graphene to the N-doped CQDs leading to charge recombination and programmed state erasing. Toward a better applicability of these memory devices, one can expect to apply a lower gate voltage necessary to perform the erasing by strategies such as the application of a thinner SiO 2 gate oxide, or a higher-k gate dielectric material, or ionic gating.
Finally, by modifying the dose of the illumination, we demonstrate that multilevel memory retention can be achieved in the current N-doped CQD devices. As shown in Figure 7c, upon receiving three consecutive UV illuminations of the same irradiance (0.7 mW cm −2 ) but a different illumination duration, the device exhibits a stepwise increase of photocurrent (ΔI DS ). Such a distinctive multilevel writing capability can be potentially translated into memory applications for multi-bit information storage. Taking into account the illumination duration, the power density, and the active channel area of the device, one can calculate the illumination dose in the unit of joule. The dose information was then plotted together with the corresponding photocurrent shown in Figure 7d. A nearly linear relationship can be obtained between the measured photocurrent and the illumination dose, which is in agreement with the observed ΔV Dirac -E a behavior shown in Figure 4e where the dose was modified by changing the power density of the illumination instead of duration.

Conclusion
FET-based non-volatile multilevel optoelectronic memory is demonstrated by coupling graphene transistors and colloidal Ndoped CQDs. The hybrid N-doped CQD/graphene memory devices are sensitive to UV photons such as those with a wavelength of 365 nm. Upon UV illumination, the photo-excited Ndoped CQDs result in a more significant transfer of electrons than holes from the CQDs to the graphene, likely due to the photo-excited holes being trapped by the nitrogen dopants of the N-doped CQDs. This leads to a prominent photo-gating effect and a n-type doping of the graphene FET. In comparison to undoped CQD/graphene devices where only poor charge retention was observed, N-doped CQD devices exhibited long-lasting lightinduced memory behavior which highlights the vital role of the nitrogen dopant-induced charge trapping in the functioning of the memory device. On N-doped CQD devices, multilevel memory functions are demonstrated triggered by UV illumination of a controlled dose under a gate bias of 0 V. After UV programming, the drain current of the N-doped CQD/graphene FET memory can be maintained at a high level (to >96% of the I DS value right after switching off the light) for 10 h. Such a strong chargeretention capability and memory function can be erased by an external positive gate bias, by which strong electron accumulation was induced in the graphene FET leading to the removal of the trapped hole-retention in the N-doped CQDs through charge recombination. Taking into account the abundant availability of carbon elements, the facile solution synthesis routes, and the demonstrated long-lasting non-volatile memory function, these fully carbon-based CQD/graphene FET devices provide interesting perspectives for the further development of transistor-based optoelectronic memories.

Experimental Section
Synthesis of Colloidal CQDs: N-doped CQDs were synthesized by a previously reported hydrothermal route. [41] Citric acid (1.0507 g) and EDA (335 μL) was dissolved in 10 mL of deionized (DI) water under vigorous stirring for 30 min. The resultant precursor solution was poured into a poly(tetrafluoroethylene)-lined autoclave. After sealing the autoclave, it was placed in an oven at 200°C for 5 h. The autoclave was then allowed to naturally cool down to room temperature. The resulting product was a transparent brown-color solution containing CQDs. To purify these CQDs, dialysis was performed in 2K molecular-weight cut off dialysis cassettes in DI water for 48 h. For the purpose of comparison, CQDs without nitrogen doping (referred below as "undoped CQDs") were also synthesized according to a previously published procedure. [41] Their synthesis and purification procedures were identical to those stated above on N-doped CQDs except that the N-source, amine (EDA), was not added into the reaction.
Fabrication of CQDs/Graphene FET-Based Optoelectronic Memories: CVD-grown monolayer graphene on copper (Cu) foils was obtained from Graphenea Semiconductor SLU. It was then transferred by a method established previously [74] onto heavily p-doped Si substrates on which a 300nm-thick thermal SiO 2 layer was grown as the gate dielectrics. To perform the graphene transfer, poly(methyl methacrylate) (PMMA) was first dissolved in butyl-acetate with a concentration of 40 mg mL −1 . This PMMA solution was then spun onto the graphene/Cu sample at a spin speed of 2000 rpm for 120 s followed by solvent drying at 80°C for 30 min. PMMA-protected graphene/Cu foil was then cut into a desired dimension, large enough to build the FETs mentioned below. They were then carefully placed onto the surface of an etchant formed by 10 wt% of FeCl 3 aqueous solution in order to dissolve entirely the Cu foil, leaving the PMMA-protected graphene film floating on the solution's surface. The PMMA/graphene layer was then transferred (with the help of a glass slide) onto three different clean DI-water baths to remove residual etchant for a duration of 10 min in each bath. A thoroughly cleaned Si/SiO 2 (300 nm) substrate was then dipped into the last water bath to pick up the floating PMMA/graphene layer, resulting in a smooth film on the Si/SiO 2 surface. This sample was then allowed for a >12-h heat treatment at 150°C to promote graphene adhesion onto the substrate. After cooled down to room temperature, the PMMA layer of the sample was removed by acetone rinsing. After graphene transfer, gold source and drain electrodes were then thermally evaporated through a shadow mask onto the graphene layer, defining a conductive channel with a channel width (W) and length (L) dimension of W = 3 mm and L = 0.25 mm. A drop (about 36 μL) of CQD DI-water solution with a concentration of about 12 mg mL −1 was then drop-casted onto the surface of graphene FET followed by drying off the solvent in air. A step-by-step device fabrication procedure together with images is described in the Supporting Information.
Material and Device Characterization Methods: TEM characterizations were performed by a JEOL 2010 TEM microscope operated at 200 kV. UVvisible absorbance measurements were performed by a JASCO V770 UVvis spectrometer. Excitation-dependent PL and PLE spectra of N-doped CQD solutions were obtained by an Edinburgh Instruments FLS900 Fluorescence spectrometer. FT-IR spectroscopy was measured on CQDs by a PerkinElmer Spectrum Two instrument in an attenuated total reflection infrared mode. The Raman spectrum of CQDs was obtained by an Anton Paar Cora 5001 Raman spectrometer with a = 785 nm laser. AFM characterizations were performed by a Bruker Dimension Icon AFM in tapping mode. XRD spectrum was measured on CQDs deposited onto a glass substrate by a PANalytical X'Pert X-ray diffractometer (with Cu-K radiation). XPS measurements were performed by a SPECS photoelectron spectroscopy system with a mono X-Ray source Al K excitation (1486.6 eV). Graphene FET (before CQD deposition) and CQD-decorated graphene FET-base memories were placed in a probe-station in vacuum (with a transparent top window allowing illumination when necessary) where their current/voltage and current-time characteristics were measured by a computer-controlled Keithley 2634 SMU. To program the memories under test, an UV LED lamp ( = 365 nm) from Ocean Optics with a fiber-coupled output was shed onto the device through the top window. www.advancedsciencenews.com

www.advelectronicmat.de
To measure the transfer characteristics, light illumination at different irradiance was kept for 1 min before measurement.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.