Synthesis and characterization of p2DP
The p2DP film was synthesized via the polymerization of porphyrin monomer on the water surface with the assistance of a surfactant monolayer and microwave irradiation (Fig. 1a). Sodium octadecyl sulfate monolayer was firstly prepared on the water surface in a glass well (50 mL) with a diameter of 5.5 cm, followed by the slow addition of tetrakis(4-carboxyphenyl)porphyrin (H2TCPP) and pyridine in the subphase. After 20 mins, when the monomers were diffused and adsorbed underneath the surfactant monolayer, the aqueous solution of Cu2+ was added into the subphase and the polymerization was performed under microwave irradiation (Fig. 1b). After just about 3 mins reaction, a transparent film was obtained on the water surface. The resultant film was transferred to -OH functionalized SiO2/Si substrate for characterization. Atomic force microscopy (AFM) image revealed the morphology of crystal domain size up to 4.0 µm (Supplementary Fig. 1). For further morphological and structural characterization, a continuous film with a diameter of ~ 6 cm was obtained (~ 23 cm2 in area, Fig. 1g) by improving reaction time under ambient condition. As shown in Fig. 1d, we observed plentiful p2DP crystals in the film, and the monomodal size distribution was obtained with an average domain size of 14.32 ± 2.80 µm2 (Fig. 1e-f). The AFM measurements revealed a thickness of ~ 12 nm and clear domain edge for the obtained film after 24h polymerization (Fig. 1d). Interestingly, we noted that the average crystal domain size is unchanged while the thickness of film is increasing with the reaction time (Fig. 1i). Besides, the image of scanning electron microscope (SEM) and optical microscope demonstrated the homogeneous morphology of the film, which is similar to the AFM results (Supplementary Fig. 2). To investigate the role of the surfactant monolayer, the interfacial reaction was carried out by the identical condition but without using surfactant monolayer. The results suggested that only amorphous p2DP was formed without surfactant, demonstrating the critical role in synthesizing crystalline p2DP film (Supplementary Fig. 3). Moreover, the introduction of energy using microwave irradiation is well established in organic synthetic chemistry, but little used in the interface of aqueous solution for the synthesis of the polymer film40–42. Under the above synthesis strategy, we demonstrated the high-throughput synthesis of crystalline p2DP film on the water surface accelerated by microwave irradiation for the first time.
After being transferred to target substrates (e.g., SiO2/Si, quartz), spectroscopic characterizations were carried out to gain insight into the chemical compositions of p2DP. As shown in Fig. 2a, the Raman spectra of p2DP film contain all the characteristic peaks of the porphyrin core in the range of 950–1700 cm− 1 43. Furthermore, the Fourier transform infrared spectroscopy (FT-IR) spectra were measured to reveal the coordination mode between metal ion and carboxyl groups of p2DP. Compared with the FT-IR spectra of the H2TCPP monomer, the resultant p2DP film showed two dominant peaks located at 1610 and 1405 cm− 1 (νas - νs = 205 cm− 1), indicating that the carboxyl ligands mainly adopt a bridging bidentate coordination mode (Fig. 2b)42,44. From ultraviolet-visible-near-infrared (UV-Vis-NIR) spectra, the p2DP present two characteristic absorbances at 420 nm (Soret band) and 545 nm (Q band), associated with the π-π* transition and the charge transitions from metal dπ (dxy and dyz) to porphyrin π*, respectively45,46. The absorbance intensity of the p2DP increased with the number of films transferred to the substrate and the maximum absorbance showed a good linear relationship34 (Fig. 2c and Supplementary Fig. 4). X-ray photoelectron spectroscopy (XPS) analysis revealed the presence of C, O, N, and Cu elements in p2DP, as well as Si and O from the SiO2 substrate (Supplementary Fig. 5a). The peaks at 288.65 eV and 285.75 eV were assigned to the characteristic carboxyl while the peaks at 398.65 eV and 399.55 eV were assigned to C-NH and C = N bonds in the porphyrin ring, respectively (Supplementary Fig. 6b)36,47. Meanwhile, the Cu 2p XPS spectrum showed that two main peaks at ≈ 955 and 935 eV are attributed to Cu 2p1/2 and Cu 2p3/2, respectively48. The satellite peak located between 940 and 944.5 eV corresponded to the electron-correlation effects, which was consistent with a copper oxidation state of + 2 (Supplementary Fig. 7a)38. Moreover, the elemental mapping also indicated that C, N, O, and Cu elements were homogeneously distributed in the p2DP film (Supplementary Fig. 8).
To probe the crystallinity and lattice structure of the p2DP, we performed grazing incidence wide-angle X-ray scattering (GIWAXS) and transmission electron microscopy (TEM) measurements on the resultant films. The 2D-GIWAXS pattern obtained for a p2DP film on Si wafer displays sharp and discrete Bragg spots near Qz = 0 Å−1, indicating a high in-plane structural order and crystallinity of the p2DP film (Fig. 2d). The in-plane peaks at Qxy = 0.37 Å−1 and Qxy = 0.53 Å−1 correspond to the 100 and 110 Bragg reflections of a square lattice with a = b = 16.79 Å (Fig. 2f). The complete list of the measured and indexed peaks can be found in Supplementary Table 1. The out-of-plane peak at Qxy = 0 Å−1 and Qz = 1.31 Å−1 corresponds to the 001 Bragg reflection, suggesting a π-π stacking along the c direction with an interlayer distance of 4.8 Å (Fig. 2e). Furthermore, the selected area electron diffraction (SAED) pattern displays a square diffraction pattern along the [001] axis with the main reflection at (110) at 0.79 nm− 1, which confirms the high crystallinity of the p2DP film (Fig. 2g). The molecular structure of the p2DP film was further visualized by high-resolution transmission electron microscopy (HRTEM) imaging, which revealed the crystallographic orientation of the domain (Fig. 2h and Supplementary Fig. 9). The corresponding fast Fourier transform (FFT) image of p2DP displays a fourfold symmetry (inset in Fig. 2h). As shown in Fig. 2i, the lattice fringes indicate that the lattice constants are a = b = 1.67 nm, which is in agreement with the p2DP structure35.
Photoadaptation characteristics
The visual adaptation function of human retina mainly rely on the cone and rod cells (Fig. 3a), in which the two types of cells having intrinsic photoadaptation characteristics, i.e. scotopic adaptation and photopic adaptation, can dynamically strengthen/weaken the responsivity to dim/strong light, enabling the accurate sensing and capture of light with varying intensity (Fig. 3b-c)49–51. The porphyrin-based 2DPs film offers the features with atomic thinness and adjustable charge transport properties contributed by highly ordered π-electronic system52, enabling the ability to emulate the visual adaptation function by modulating charge trapping process53,54. Therefore, we fabricated heterostructure consisting of p2DP channel (Supplementary Fig. 10) and ZnO nanoparticles (ZnONP) layer, where the ZnONP can trap carriers to modify the source-drain behaviour55,56, owing to its electron accepting character (n-type)57 and energy band mismatch with p2DP (p-type)38. Figure 3d and Supplementary Figs. 11–14 show the configuration of the heterostructure device.
We firstly measured the light intensity-dependent current (IDS) response (ranging from 2.5 to 80 mW cm− 2) under a read voltage (VDS) of 5V, which was applied to amplify the photo-responsive signal (Supplementary Fig. 15). At the low light intensity, e.g., 2.5 mW cm− 2, the sensor shows a low photocurrent of 0.027 (normalized photocurrent) at initial state. Interestingly, the current progressively increases over time instead of being steady, finally reaching an equilibrium state of 0.4 within 5 s, with the photocurrent enhanced by 15 folds, showing a current excitation characteristic. On the other hand, applying light with a high intensity ~ 57 mW cm− 2 resulted in rapidly increased current (IDS) and substantially decreased current within the following few seconds, exhibiting a current inhibition behavior (Fig. 3e and Supplementary Fig. 16). Therefore, when exposed to light illumination with weak and strong intensity, the device can autonomously enhance and weaken the photocurrent and tune the current intensity to a moderate level at the equilibrium state, biorealistically replicating the photoadaptation.
In order to quantitatively evaluate the photoresponse performance of the device, we defined the dependence of photosensitivity change (Pt = ΔIt/Idark, where Idark is the dark current and It is the current after light illumination) after exposing to light illumination for different intensities17 (Fig. 3f). Under the dim bright stimuli, e.g., below 25 mW cm− 2, the photosensitivity increases with time, demonstrating the scotopic adaptation. An enhanced light intensity (~ 25–80 mW cm− 2) contributes to a decrease of photosensitivity, showing the adaptation behavior changes to photopic adaptation. Following light illumination, the device exhibits a light intensity-dependent photoadaptation (i.e., scotopic and photopic adaptation)8,17. Moreover, the other characteristic of biological adaptation (e.g., habituation behavior) are successfully emulated in the sensor58,59. Supplementary Figs. 19–21 show the transient IDS responses with repeated stimuli (e.g., interval time, conditioning time) and the photoresponse under different read voltage (Supplementary Notes. 1 and 2, respectively).
To interpret the operation mechanism of the p2DP-based vision sensor, the energy band diagrams of device are depicted in Fig. 3g and Supplementary Fig. 17. The heterostructure exhibits a large energy offset (1.8 eV) that could facilitate the spatial separation of electron-hole pairs at the interface60. Under light illumination, the electron-hole pairs were created in p2DP channel and separated at the heterostructure interface, induced by an internal electric field that leads to the band bending at the interface14. As the subsequent electron trapping in the ZnONP (charge trapping effect)54, the negatively charged nanoparticles layer induces positive carriers in the p2DP channel through capacitive coupling to adjust the source-drain behavior17,61,62(Fig. 3h). For dim-light illumination, the low-density electron-hole pairs separates at the interface, whereas it should take some time to accumulate enough trapped electrons to reach the dynamic balance between the electron trapping and de-trapping process, resulting in a relatively slow photoresponse9. For bright-light illumination, the sensor displays a steep photocurrent reis and decay, owing to first spontaneous displacement of high-density free electrons and holes63 and increasing number of carriers recombination taking place at the interface64, respectively, leading to photopic adaptation behavior. Thus, the charge trapping-dominated dynamic adaptation can be controlled by the light intensity, which is similar to the switchover between rod and cone cells65, suggesting the successful realization of the bioinspired adaptation functionality in single devices.
Visual adaptation function of sensor array
Based on the light intensity-dependent photoadaptation of devices, we constructed a 4×8-sensors array to emulate the visual adaptation of the human retina (including scotopic and photopic adaptation) for image processing (Fig. 4a). The sensor array was employed to display the pattern perception of ‘F’ under different light intensities (6 and 80 mW cm− 2, respectively). For the visual adaptation test, the changes of channel current (IDS) were normalized to 0–1 for the initial photoresponse (VDS = 5V). For the scotopic adaptation test, the devices corresponding to the pattern ‘F’ are under a dim light of 6 mW cm− 2. Under the scotopic adaptation process, the pattern of ‘F’ was successfully recognized as the image contrast increases after 10s (Fig. 4b and Supplementary Fig. 18a). As for the photopic adaptation of the sensor array, the pattern ‘F’ gradually changes to a comfortable stage under a light intensity of 80 mW cm− 2, which is similar to the photopic adaptation of a retina (Fig. 4c and Supplementary Fig. 18b).
Since the proposed p2DP-based sensor shows a light-intensity dynamic visual adaptation function, it can be used to evaluate the visual adaptation on enhanced image recognition accuracy. A vision system with an image recognition function (photopic adaptation and scotopic adaptation) was modeled by a computer simulation. The preprocessed images are fed to a convolutional neural network (CNN) to complete the image recognition, as shown in Fig. 4d66. To demonstrate the image recognition function under bright- or dim-light conditions, the standard Olivetti Faces (OF) dataset as the training set were used. Figure 4e shows the adaptation process of the OF image under scotopic adaptation. The image contrast is also enhanced from 57.5–97.5% after 10 s scotopic adaptation (Fig. 4g). Figure 4f shows the evolution of the OF image during photopic adaptation. The contrast of the image increases from 20 to 30 s and the recognition accuracy shows obvious improvement from 5–90% after 10s photopic adaption (Fig. 4h). These results show that this biomimetic vision sensor can capture and perceive patterns under different illumination conditions, which can simplify the complexity for hardware and algorithms to realize the function of image recognition at the sensor terminals.