Metal-Organic Framework for Transparent Electronics.

Electronics allowing for visible light to pass through are attractive, where a key challenge is to make the core functional units transparent. Here, it is shown that transparent electronics can be constructed by epitaxial growth of metal-organic frameworks (MOFs) on single-layer graphene (SLG) to give a desirable transparency of 95.7% to 550 nm visible light and an electrical conductivity of 4.0 × 104 S m-1. Through lattice and symmetry match, collective alignment of MOF pores and dense packing of MOFs vertically on SLG are achieved, as directly visualized by electron microscopy. These MOF-on-SLG constructs are capable of room-temperature recognition of gas molecules at the ppb level with a linear range from 10 to 108 ppb, providing real-time gas monitoring function in transparent electronics. The corresponding devices can be fabricated on flexible substrates with large size, 3 × 5 cm, and afford continuous folding for more than 200 times without losing conductivity or transparency.

S4 immersion in 0.1 M (NH 4 ) 2 S 2 O 8 as etchant, the Cu foil was etched away, leaving SLG on top of the SAM-Si or SAM-quartz substrate. This SLG film was rinsed by deionized water and dried with nitrogen before further application.

Synthesis of Ni-CAT-1-on-SLG
Ni-CAT-1-on-SLG constructs with different thicknesses were synthesized under the same condition (main text), except for the addition of stock solutions with different concentrations. Details were listed in Table S1.

Synthesis of PPF-1 nanosheets on SLG
Zinc acetate and TCPP (tetrakis (4-carboxyphenyl) porphyrin) were dissolved by sonication in ethanol at the concentration of 0.05 mM and 0.01mM, respectively. Then 3 mL of each stock solution was mixed in a glass vial, with a SLG on SAM-Si substrate sitting at the bottom. The vial was caped and heated in an isothermal oven at 85 °C for 5 hours. After cooling down to room temperature, the as-grown PPF-1 nanosheets on SLG on silicon was washed with ethanol for 3 times and dried by nitrogen flow.

S5
Synthesis of randomly oriented Ni-CAT-1 HHTP (7 mg) and Ni (CH 3 COO) 2 ·4H 2 O (10 mg) were dissolved in 5 mL deionized water in a 20 mL glass vial. The reaction mixture was heated in an isothermal oven at 85 °C for 12 h, yielding small dark blue crystals. The reaction mixture was cooled to room temperature and the crystals were washed with deionized water (3 × 5 mL), followed by acetone (6 × 5 mL).

Atomic Force Microscopy (AFM)
AFM tapping mode was used for the morphology and height analysis of SLG at room temperature. The measurements were performed on a Bruker Edge instrument with scanning rate at 0.7 Hz.

Single layer graphene
From the AFM image of SLG, we could see that the surface was clean and continuous. Height of the graphene film was ~1 nm (Fig. S2). Corresponding height profile.

Raman Spectroscopy
Raman spectrums were used to check the characteristic vibration of SLG and Ni-CAT-1-on-SLG constructs. The measurements were performed on Horiba HR800 system using excitation laser with wavelength of 514 nm.

Raman spectrum of SLG
Optical image showed that SLG film was clean, and no obvious cracks were observed on its surface (Fig S3a). The Raman spectrum (Fig S3b) showed three typical graphene bands, the D (~1350 cm -1 ), G (~1580 cm -1 ) and 2D (~2680 cm -1 ) bands. The tiny D band reflected the high quality of single layer graphene. The sharp and symmetric 2D bands and the 2D/G intensity ratio were consistent with the Raman signatures of single layer graphene. [3]  Raman spectrum of single layer graphene.

Raman spectrum of Ni-CAT-1-on-SLG construct
As for the Raman spectrum of Ni-CAT-1-on-SLG constructs, apart from characteristic peaks assigned to SLG, we also observed peaks appeared at Raman shift ranging from 1100 to 1580 cm -1 , which were ascribed to the vibrations of C-C bond of HHTP linkers in Ni-CAT-1, [4] indicating the successful formation of Ni-CAT-1-on-SLG construct ( Fig S4).

X-ray Photoelectron Spectrometer (XPS)
XPS was used to analysis the chemical environment of Ni in Ni-CAT-1-on-SLG construct and Ni-CAT-1 powder. Data were collected on an ESCALAB 250Xi X-ray Photoelectron Spectrometer. From the XPS spectrum, we found that no obvious difference could be observed in the Ni 2p spectra of Ni-CAT-1-on-SLG and Ni-CAT-1 powder, indicating the chemical environment of Ni was identical in these two samples.

Scanning Electron Microscopy (SEM)
SEM was used to examine the morphology of Ni-CAT-1-on-SLG constructs and Ni-CAT-1 on SAM-Si substrate. All of the samples were directly subjected to electron beam without any coating. These samples were analyzed using a FEI Verios 460 with both TLD and MD detectors, under an accelerating voltage of 0.35 kV.

SEM images of Ni-CAT-1 on SAM-Si substrate
In order to illustrate the critical role of graphene for the growth of vertically aligned MOF, Ni-CAT-1 on SAM-Si sample was prepared by using SAM-Si without graphene as substrate. Other synthetic conditions duplicated with the preparation of Ni-CAT-1-on-SLG-25nm sample. As showed in the SEM images, Ni-CAT-1 crystals dispersed randomly on the SAM-Si substrate surface.

SEM images of randomly distributed Ni-CAT-1 powder
The morphology of Ni-CAT-1 powder was characterized by SEM. Ni-CAT-1 powder was placed on a clean copper tape and then attached to the SEM sample holder. The following SEM images of Ni-CAT-1 powder were collected in Zeiss Merlin Compact SEM, at an accelerating voltage of 5 kV. Rod shaped crystals with uniform sizes (~60 nm in diameter) can be obtained from the Ni-CAT-1 powder samples. Those crystals orientated in different directions ( Fig S7).

Transmission Electron Microscopy (TEM)
Crystal structure of Ni-CAT-1-on-SLG construct was confirmed by TEM (Fig 2A,   S9). TEM data were collected on a JEOL JEM-2100Plus instrument operated at 200 kV.

Transfer process of Ni-CAT-1-on-SLG construct to TEM grid
Ni-CAT-1-on-SLG construct was transferred to TEM grid through a polymer assistant method (Fig S8). The transfer process was described as follows. First, polyvinyl alcohol (PVA) aqueous solution was spin-coated on Ni-CAT-1-on-SLG construct at 1500 rpm for 30s, then baked to evaporate the solvent. After that, the PVA-Ni-CAT-1-on-SLG construct was peeled off from SAM-Si wafer carefully. Then the PVA-Ni-CAT-1-on-SLG film was transferred to a TEM grid, with the PVA side facing up. Deionized water was dropped on the surface of PVA-Ni-CAT-1-on-SLG gently to dissolve PVA. After PVA was removed with deionized water completely, we could get the Ni-CAT-1-on-SLG construct successfully on TEM grid ( Fig S8).

Scanning TEM (STEM) image of Ni-CAT-1-on-SLG construct
STEM were used to confirm the atoms arrangement in Ni-CAT-1-on-SLG construct. The STEM images in Fig 2B and 2G in the main text, Supplementary Fig.   S10 and S11 were collected with a FEI Titan Themis instrument using the iDPC (integrated differential phase contrast) and HAADF (high-angle annular dark field) modes at an accelerating voltage of 300 kV.
The iDPC images were obtained by combining signals from quadrant sections of the circular detector. [5] Similar to HAADF images, white spots in the image represent the presence of atoms, while the dark ones represent void. The high resolution images in both HAADF and iDPC images allowed us to clearly disclose the Ni SBU and HHTP linker in the framework, as well as the position of pores without further image processing ( Fig S10 and S11).

S14
HAADF image (Fig S11a), which was obtained simultaneously with the iDPC image in the main text ( Figure 2B), showed that the Ni-CAT-1-on-SLG crystals were densely packed. The pore structure of Ni-CAT-1 crystals was clearly revealed by zoomed-in HAADF STEM image of a single crystal domain ( Figure S11b), where the Ni SBU and HHTP linker were clearly identified.

Crystal Structure of SLG, CAT-1 and PPF-1
Crystal structure of graphene was resolved in hexagonal system with carbon atoms arranged on a 2D honeycomb lattice. The distances between each atom are 0.142 nm (Fig S12a).
Lattice match was a critical parameter for the epitaxial growth. In a controlled experiment, we tried to synthesis PPF-1 crystals on SLG. PPF-1 is a typical 2D MOF constructed by zinc paddle wheel clusters and TCPP linkers. The crystal structure of PPF-1 is tetragonal with I4/mmm. Porphyrin layer forms square channels along the  (c) The crystal structure of PPF-1.

AFM images of PPF-1-on-SLG construct
The morphology of PPF-1-on-SLG construct was characterized by AFM. From AFM image in a large scope, we found that PPF-1 nanosheets were randomly distributed on the SLG surface. PPF-1, quadrate crystals with layered structure ( Fig   S13b-d), were distinguishable from the Zn(CH 3 COOH) 2 crystals (Fig S13a), which were resolved in monoclinic space. The layer distance of PPF-1 crystals was measured to be ~2 nm which was in accordance with the unit cell parameters of simulated structure of PPF-1 in c axis (c=1.74 nm). Thickness of those tiny PPF-1 crystals varied from 2 nm (2 layers) to 20 nm (20 layers).

AFM images of Ni-CAT-1-on-SLG constructs
CAT-1-on-SLG constructs with different thickness, including 10 nm, 25 nm. 50 nm, 80 nm and 170 nm, were characterized by AFM (Fig 2H-2J, S14). From the AFM images, we could see that all of the CAT-1-on-SLG constructs were uniform and MOFs packed densely on SLG surface. Corresponding height profiles confirmed the height of these constructs (Table S1).

SEM images of CAT-1-on-SLG constructs with different thickness
CAT-1-on-SLG constructs with different thickness, including 10 nm, 25 nm, 50 nm, 80 nm and 170 nm, were characterized by SEM (Fig S15-16). SEM images in large scale showed that Ni-CAT-1 crystals was highly orientated with uniform height on SLG surface ( Figure S17), which was consistent with the AFM results ( Fig S14).
The zoomed-in SEM images of these constructs indicated that the Ni-CAT-1 crystals were densely packed on SLG surface, when thickness of the entire constructs was over 25 nm ( Fig S18).

S19
Domain size of the Ni-CAT-1 crystals increased from 11 to 160 nm, along with their thickness from 10 to 170 nm ( Fig S16). SEM images at the cross section ( Fig   S17) indicated the thickness of the construct which was consistent with the AFM results.

X-ray diffraction (XRD)
XRD was used to characterize the bulk purity of Ni-CAT-1-on-SLG constructs at different thickness. Two modes of XRD were performed for the characterization.
Specifically, Ni-CAT-1-on-SLG constructs with the thickness over 80 nm were collected through Powder X-ray diffraction (PXRD) mode, while the thickness less than 80 nm were collected through 2D Grazing incidence X-ray diffraction (2D-GIXRD) mode.   S20). The wavelength of beam was 1.2398 Å and incidence angle S22 (θ i ) between the beam and sample surface was 0.3°. LaB 6 was used as standard sample for the distance calibration between samples and detector. [7] Axes labels (q xy and q z ) are defined using the GIXRD convention equation q xy = 4πsinα/λ and q z = 4πsinβ/λ, where λ is the x-ray wavelength, and α and β are the horizontal and vertical diffraction angles, respectively. [8]

Mosaicity Estimate from 2D-GIXRD
In order to further estimate the angular distribution of the crystals, the azimuthal integration of [004] Bragg peaks intensity of the samples were performed (Fig S22), showing nearly identical distributions. [8] The peak width at half-height (FWHM) of these distributions was ~13.0° in Fig S22a, ~15.0° in Fig S22b and ~15.4° in Fig   S22c.

PXRD of Ni(OAc) 2 -on-SLG and HHTP-on-SLG
In a controlled experiment, PXRD patterns of the Ni(OAc) 2 -on-SLG and HHTP-on-SLG samples were collected. The prepared process was described as follows. Ni(OAc) 2

Transmittance of Ni-CAT-1-on-SLG constructs.
The transmittance measurements of the CAT-1-on-SLG constructs on quartz were carried out on Lambda 650S Ultraviolet-visible (UV-VIS) spectrometer.
Transmittance of single layer graphene was measured to be 97.6%, which was consistent with the previous report. [10] Transmittance of Ni-CAT-1-on-SLG constructs at 550 nm visible light was decreased as the thickness of the constructs increased ( Figure S24).

S26
In a controlled experiment, Ni-CAT-1 crystals were dispersed on SLG surface randomly through drop-casting method. The transmittance was decreased dramatically compared with Ni-CAT-1-on-SLG construct (Fig S25), demonstrating the power of epitaxial growth of MOF on SLG.

Electrical resistance of Ni-CAT-1-on-SLG
The devices based on Ni-CAT-1-on-SLG constructs were fabricated by placing a pair of indium electrodes on the surface of samples. The electronic resistance of device was tested with a multimeter analyzer at room temperature. We found that when the thickness of Ni-CAT-1-on-SLG constructs were above 80 nm, it was hard to detect the resistance. While the resistance of Ni-CAT-1-on-SLG constructs under 50 nm were measured to be 1 to 3 kΩ, in accordance with the resistance of SLG.

Gas adsorption kinetic measurements
A home-made gas sensing system was used to evaluate the sensor performance of Ni-CAT-1-on-SLG constructs. The fabricated device was placed in the quartz chamber and the two electrodes were connected to a source meter (Keithley 2400) (Fig. S27).
The whole experiment was performed at room temperature. For a typical trial, a constant potential of 100 mV was applied, and the resistance was monitored in situ by the source meter. Concentration of the gas molecules were precisely controlled by mass flow controller (MFC). Before each test, the device was activated at 100 °C for 12 h in vacuum to evaporate contaminate on the device.

S31
Pure N 2 was used to calibrate the relationship between the flow rate and pressure, which is linear with regression coefficient above 0.999 (Fig S28).

Isotherm analysis
To evaluate the electrical response of device to different gas molecules directly, the response value was normalized as R%. The R% of the device is defined as: [11] R% = ( − ) 100% Where R g is the resistance of the device after exposing to gas molecules, and R i is the initial resistance.
The curves showed characteristic time-dependent response, so reliable kinetics simulation could be fitted for the adsorption behavior. A two-compartment reaction model was used to analyze the adsorption kinetic data. The instantaneous rate values are the derivative of response equation:

Where [A] is the concentration of gas molecules which is considered as inviable during the adsorption process, [B] represents the active sites concentration of the
Ni-CAT-1. This can be determined by fitting the temporal curve with the kinetic model. Specifically, the response curves of Ni-CAT-1-on-SLG construct to NH 3 with concentration varied from 10 ppb to 10 8 ppb was carefully analyzed by first order kinetics model (Fig S29). All of the curves could be fitted well by first order kinetics model with regression coefficient above 0.98. Fig S29. The response curves and the corresponding simulated curves using first order kinetic reaction model for the ammonia molecules from 10 to 10 8 ppb.

S34
We also explored other models to simulate the response curves. Apparently, zero order kinetic was not suited for this study (Fig. S30). Second order kinetic equation was also used to simulate the kinetics in the response curves. The linear form of second order equation matched with the experimental data with the regression coefficient above 0.97. However, k values extracted from the simulation were not related with the concentration, so second order model was not suitable in this study (Fig. S31).   When the concentration of NH 3 was high, the initial high responses didn't allow for the first order fitting. Therefore, the first data points were used to directly derive k values with concentration from 100 to 1000 ppm. We found that the k was also in S36 proportional with the concentration of NH 3 (Fig. S34b). It is worth noting that the k value obtained at higher NH 3 concentration didn't share the same slope with those at concentration below 2 ppm in the plot of k versus NH 3 concentration.

Langmiur fitting and Temkin fitting
In order to figure out the adsorptive type, the relationship between r 0 and c was carefully investigated. We found that the Freundlich chemical adsorption model matched well for NH 3 adsorption isotherms in our study, and other models showed poor regression coefficient in the NH 3 adsorption process. The detailed results using different models were compared as following, where the regression coefficient was 0.53 for Langmiur model and 0.93 for the Temkin model (Fig S36).

S38
Electrical response to different gas molecules at the same concentration were tested both on CAT-1-on-SLG-25nm and SLG devices. Six kinds of gases were introduced into the system separately, including N 2 , H 2 , CH 4 , NH 3 , CO, O 2 . We found that the device based on Ni-CAT-1-on-SLG-25nm constructs exhibit observable electrical response to NH 3 , CO and O 2 , and negligible response for N 2 , H 2 and CH 4 ( Figure S37). This can be attributed to the specific interaction between these gas molecules and the Ni-CAT-1, demonstrating that this MOF-on-SLG construct were selective to different gas molecules. purge was all set to be 30 min, and the data acquiring interval was 5 s. Once exposed to NH 3 , the device based on Ni-CAT-1-on-SLG-25nm construct exhibited an instant positive response within 5 s and gradually reached to an equilibrium state. In the recovery range, an instant decrement in response was observed. While for CO molecules, the sensor exhibited slowly decreased and increased response in the exposure and recovery range (Fig. S38), demonstrating that the adsorption and desorption curves were characteristic for each gas molecules. Fig S37. The response curves and recovery curves for NH 3 and CO molecules, red and blue dots, respectively, both at the concentration of 100 ppm.

S40
First order kinetic equation was also used to simulate the response curves of different gas molecules. All of the curves showed high correlation, and the regression coefficient was above 0.99 for CO and 0.87 for O 2 (Fig S39). The rate parameter k and r 0 were extracted for CO and O 2 molecules, which was critical to differentiate the gas species (Fig 4E).

S41
Furthermore, we investigated the impact of constructs thickness to the device performance. The response to NH 3 at the same concentration were examined in three devices, based on Ni-CAT-1-on-SLG-10nm, 25nm, 50nm constructs, respectively. The concentration of NH 3 was set to be 1 ppm. Among these devices, we found that the device with thickness at 25 nm exhibited the best performance (Fig. S40).

S42
The stability and cycling performance of the device based on the Ni-CAT-1-on-SLG-25nm construct were tested. We found that the device remained similar response after 10 cycles and showed high repeatability (Fig. S41). construct was used for the detection of NH 3 . The electrical resistance was measured to be 1.4410 8 Ω, 5 magnitudes higher than that of Ni-CAT-1-on-SLG-25nm construct.
Due to the low conductivity, the response to NH 3 at low concentration was unreadable and the background was too noisy (Fig. S43), thus the accuracy was limited.

S44
We also compared electrical response of CAT-1-on-SLG-25nm construct and SLG to NH 3 at low concentration (1 ppm). Device based on SLG showed response to NH 3 , where the value was too low. In contrast, the device based on Ni-CAT-1-on-SLG-25nm construct exhibited obvious response to NH 3 , which was 10 times higher than SLG (Fig. S44).

Spectroscopy studies on the interaction sites between MOF and NH 3
The interaction sites between MOF and gas molecule are important. Chemical structural drawing of the interaction sites between this MOF (Ni-CAT-1) and NH 3 was prepared as Figure S44. The interaction sites between MOF and NH 3 were the metal oxide clusters. Possibly, there are two ways for this electro-chemical response. The first way is that NH 3 molecules are likely to coordinate with the metal (Ni) to form coordination bond. The second way is that hydrogen atoms in NH 3 molecules are possibly to link with the oxygen sites to form hydrogen bond as illustrated in this scheme.

Preparation of the heterojunction between Ni-CAT-1 and SLG
The heterojunction was prepared as the following.  S45). adsorption were performed. The data were collected on a Horiba HR800 system using excitation laser with wavelength of 514 nm. The 2D peak of SLG were carefully analyzed before and after NH 3 adsorption. Statistic results indicated that the Raman shift of the 2D peak of SLG was redshift about 4 cm -1 (Fig S48c). Raman spectrum of the 2D peak for SLG before (blue) and after NH 3 adsorption (red). (c) Statistic result for the Raman shift of the 2D peak of SLG before and after NH 3 adsorption.

Ultraviolet-visible (UV-VIS) spectrum
Ultraviolet-visible (UV-VIS) spectroscopy was performed to calculate the band gap of Ni-CAT-1 powder and Ni-CAT-1-on-SLG constructs. The measurements were carried out on Shimadzu UV-2500.

UV-VIS spectrum of HHTP
In a controlled experiment, the UV-VIS spectrum of HHTP was collected. 0.1 mM HHTP was dissolved in ethanol and loaded in a quartz cell. The sharp absorption peaks at wavelength of 250-300 cm -1 were characteristic peaks for HHTP (Fig. S46). is Eg. The band gap of Ni-CAT-1-on-SLG was 1.74 eV (Fig S47).

UV-VIS spectrum of Ni-CAT-1-on-SLG construct
The UV-VIS absorption spectrums were collected using SLG on quartz as background. The bandgap of Ni-CAT-1-on-SLG constructs was calculated based on solid-state UV-VIS absorption spectrums (Fig S48-49). Ni-CAT-1-80nm construct estimated from UV-VIS reflectance spectrum.

Field effect transistor (FET) device of SLG
The electrical property of SLG was investigated by fabricating FET deceive. SLG on SAM-Si was fabricated into FET device using a micro-grid as mask.
Electron-beam evaporation was used to make electrodes with Cr (5 nm) followed by Au (25 nm) for electrical contact at both ends of the graphene FET channel ('source' and 'drain'), where p+ doped Si was used as the bottom gate electrode and SiO 2 (300 nm thermal oxide) as the gate dielectric layer. [21] Electrical measurements of the graphene-based FET samples were carried out in vacuum, and the performance was shown in Fig S50.

Ni-CAT-1-on-SLG constructs on other transparent substrates
Ni-CAT-1-on-SLG constructs were also successfully fabricated on other transparent flexible substrates, such as PET and PDMS. Transmittance spectrum of the Ni-CAT-1-on-SLG constructs on those substrates were showed in Figure S53 to S55. The experiment results agreed well with the spectrums on quartz within a reasonable error range. Transmittance reduced as the thickness of Ni-CAT-1-on-SLG constructs increased (Fig S55c).   By using PET as substrate, the electrical connection to the device, fabricated by the Ni-CAT-1-on-SLG-25nm construct, successfully light up a LED lamp ( Figure   S56a). And the light remained on even the device was bending at the radius of 3 mm ( Figure S57b). Optical image of CAT-1-on-SLG-25 nm construct on PET substrate at the bending state. Bending radius is 3 mm.