Figure 1a illustrates the physical process of implementing wavelength-selective imaging with a Ge-based floating gate phototransistor. A hyperspectral remote sensing image of Shaanxi, China, is shown. The left is the hyperspectral image, and the right is the imaging effect of VIS light imaging mode, NIR light imaging mode, and wide spectrum imaging mode. In VIS light imaging mode, the characteristics of the mountains are more obvious. In NIR light imaging mode, the characteristics of the river are more obvious. In wide-spectrum imaging mode, the enhancement effects for each object are consistent, and there is no obvious prominent effect. Through the wavelength-selective imaging function, the visible and infrared regions of an object can be effectively extracted, thereby improving the recognition rate of the object. Therefore, studying photodetectors with wavelength-selective imaging characteristics is highly important for improving the object recognition rate. This article uses an ingenious device structure and adjusts the Vpgm of the device to achieve three modes: single visible light imaging, visible–infrared dual-band imaging, and single infrared band imaging. A wide-spectrum image is obtained in the dual-band imaging state, and the single-visible-band imaging effect and the single-infrared-band imaging effect are obtained by further modulating the device state (Fig. 1d).
Figures 1b and 1c show schematic diagram of the device, which has a three-layer vertical structure, namely, a visible light absorber/floating gate storage layer/near-infrared light gating layer configuration. A floating-gate phototransistor (FG-PT) was fabricated on a SiO2 (300 nm)/p-Ge substrate by using a conventional dry transfer method, and the details of the fabrication processes can be found in the experimental section. MoS2/hBN/Gr were successively stacked vertically on the substrate, which act as a channel, a tunnel barrier, and a floating gate. Ge acts as the back gate material. Multilayer graphene was selected to reduce the contact barrier between MoS2 and Au. The cross-sectional information of the different regions is obtained with aberration-corrected scanning transmission electron microscopy, and the results are shown in Fig. 1e and Fig. 1f. The boundary of bottom graphene, bottom hBN and channel MoS2 can be resolved. An optical image of the FG-PT is shown in Figure S1a. The atomic force microscopy (AFM) images in Figure S1b indicate that the thicknesses of Gr, hBN and MoS2 are approximately 10.2 nm, 13.2 nm and 25.3 nm, respectively. As shown in Figure S1c, distinct Raman peaks corresponding to MoS2 and Ge can be observed. The Ge peak is located at ~ 300 cm− 1. The other two distinct Raman peaks located at ~ 382 and ~ 407 cm− 1 correspond to the\({\text{E}}_{\text{2g}}^{\text{1}}\)and \({\text{A}}_{\text{1g}}\)modes for multilayer MoS2.
The typical electrical characteristics of the FG-PT device were investigated, as shown in Fig. 2. We obtained the basic transfer curve of our devices by sweeping the gate voltage over different ranges under a constant 50 mV drain-source voltage (Vd). A sizable clockwise hysteresis, modulated by the swept amplitude of the control gate voltage, is shown in Fig. 2a. The total shift in the memory threshold voltage (Vth) gives an estimated memory window of 25.7 V between − 20 and 20 V, taken for a 1 nA constant current. Figures 2b and 2c show the voltage pulse modulations of the FG-PT conduction states. To fully switch the FG-PT to high (program) and low (erase) resistance states, − 10/10 V voltage pulses of 1 ms were applied. A high on-off current ratio of ≈ 106 can be realized with this voltage pulse programming scheme. As shown in Fig. 2b, the program and erase states exhibit almost negligible loss within 1000 s, which indicates superior nonvolatile memory capability. The multilevel behavior is illustrated in Fig. 2c, in which we show the ability to set the channel conductance with the programming voltage (Vpgm). When the device is in the initial state, the device can be switched to the program state (state 1) by applying a voltage of 10 V for 1 ms. At the same time, switching between state 1 and state 5 can also be achieved by applying a gate voltage of + 10 V (-10 V). In addition, when the pulse voltage is reduced, the device can be made to work in an intermediate state. When a 7.5 V (1 ms) pulse is applied in the initial state, the device can be adjusted to state 4; when a forward voltage pulse is applied again in state 4, the device current can continue to decrease. When the device is in the program state, the current in the device can be increased by applying a negative voltage. Applying a voltage of -7.5 (1 ms) in state 1 can increase the current to state 2, and when the same voltage is applied again in state 2, the current can continue to increase to state 3. In other words, through clever adjustment of the voltage direction and magnitude, the current of the device can be adjusted at will, laying the foundation for subsequent switching of imaging modes. In addition to its excellent on-off conductance switching performance, the FG-PT is also robust in terms of its retention and endurance properties. An endurance test of the FG-PT was further carried out by repeatedly switching the device between the program and erase states using alternating 10 and − 10 V voltage pulses, as shown in Fig. 2d. Clearly, no sizable change can be observed for either the program or erase states of our device, highlighting the minimum degradation of the FG-PT during the endurance test. In addition to better storage characteristics, the device can be switched between three states—on state, on-off state and off state—by adjusting the storage state, resulting in good electrical reconfigurability characteristics (Figure S2), which has great development potential in logic circuits.
The detailed mechanism for the electrical programming/erasing process of the proposed device is shown in Figs. 2e and 2f. The applied control gate voltage enables electrons to tunnel between the Gr floating gate and the MoS2 channel across the thin hBN barrier. In the programming process, electrons move from MoS2 to Gr due to the positive gate voltage and are permanently stored in the Gr floating gate due to the small barrier35,36. The negative charge induced in Gr is equivalent to a partial negative gate voltage, effectively depleting the MoS2 channel, leading to a low conductance (program state) in the MoS2 channel at Vg = 0.05 V. In contrast, electrons can be pushed back to the MoS2 layer from the Gr layer with the help of a negative gate voltage, that is, the erasing process, resulting in a high conductance (erase state) in the MoS2 channel35. Thus, the nonvolatile conductance switching behavior is determined by the charge trapping/release capability of the Gr layer. Moreover, depending on the amplitude of the gate voltage, charge trapping/release processes with different strengths can be obtained for the Gr layer, which can account for the multilevel conductance states observed in Fig. 2c. The channel electron distribution of the device in different floating gate states was simulated through TCAD. When the amount of charge in the floating gate is large, the channel can be effectively depleted (Figure S3).
Two-dimensional transition metal dichalcogenides (TMDCs), such as MoS2, have been revealed as promising optoelectronic materials37–39; owing to their unique physical properties and versatile device configurations, a wide spectrum of functional devices with unprecedented and enhanced optoelectronic characteristics have been proposed40–42. Here, the response characteristics of the device in different states were obtained using a three-layer device structure with a visible light absorption layer (Fig. 3a), a floating gate memory layer (Fig. 3b) and a near-infrared photogating layer (Fig. 3c), indicating that the device has a wide spectral response in different states. Figures 3d-3f are the simulation results of the device channel electron current density obtained through TCAD (Technology Computer Aided Design). The simulation results respectively correspond to the channel current state of the device under visible light (Fig. 3d), dark (Fig. 3e) conditions, and infrared light (Fig. 3f). The programming state of the device is an intermediate state. Under visible illumination, the channel current density of the device significantly increases through the photoconductivity effect, while under infrared illumination, the channel of the device is effectively depleted through the photogating effect, and in some areas in the channel, the current density decreases. Figures 3g-3o show the visible and infrared response characteristics of the device under different program states. Figure 3g and 3h shows the output curves and transfer curves of the device in intermediate state. Positive photoresponse under visible light and negative photoresponse under infrared light can be clearly observed. As shown in Figs. 3j and 3k, the photodetection performance of the FG-PT was investigated under 532 nm incident light at an intensity of 20.4 nW. The device shows an obvious positive response in different program states. This response mainly comes from the photoconductive effect of the intrinsic absorption of 532 nm light by the MoS2 material (Fig. 3a). Figure 3l depicts the photocurrent (Ilight-Idark) of the device in different states, which increases with increasing device dark current, which is mainly attributed to the high carrier collection efficiency of the device in the on state.
As shown in Fig. 2c, the channel current state can be cleverly adjusted by the floating gate layer structure through the voltage pulse, achieving switching between different program states. When the channel is modulated into the erase state through a voltage pulse (Vg< 0), there are a small number of electrons in the floating gate, and the device is in the on state (state 4 and state 5). After a forward gate voltage pulse is applied, the channel current is in the program state (state 2 and state 3), and the channel current decreases. Through effective adjustment of the dark current state of the device, the photocurrent realizes detection of a large dynamic range spanning 3 orders of magnitude. Under 1550 nm illumination (11.05 µW), the Ge material absorbs infrared light and effectively modulates the channel based on the photogating effect. Electrons accumulate due to the band trap at the interface of Ge and silicon oxide, which is equivalent to a partial negative gate voltage and effectively exhausts the channel current, thereby achieving a negative response (Fig. 3c). The details of the photoresponse mechanism and carrier transport model are shown in Figures S4 and S5. The photocurrent-time (I-t) curves under 1550 nm infrared light in different program states are shown in Figs. 3m and 3n. The absolute values of the photocurrent under 1550 nm illumination are shown in Fig. 3o. Similar to at 532 nm, a large dynamic range photoresponse spanning 3 orders of magnitude can also be achieved at 1550 nm. The device has a negative photoresponse to infrared light regardless of the current level. The responsivity of device is shown in Fig. 3i. The highest responsivity under visible light is 12.7 A/W, and the highest responsivity under infrared light is -24 mA/W. Since the current level of the device is small in some states, the light response is also reduced accordingly. In order to ensure the imaging effect in later applications, we connected the device in series with a resistor and used the output voltage to characterize the light response of the device, effectively avoiding the problem of a small photoresponse.
Combined with the memory characteristics of the device discussed above and the bidirectional photoresponse characteristics for different wavelengths of light, wavelength-selective imaging can be achieved by connecting the device in series with a suitable resistor, as shown in Fig. 4a. The use of the FG-PT as the basic building block allows us to further modulate the channel current compared to common FETs, providing additional degrees of freedom for applications in photodetectors. The gate terminal can then be used to set the state of the device using a programming voltage (Vpgm). Using this characteristic, the channel current is effectively modulated into 3 states by applying gate voltage pulses. As presented in Fig. 4b, we limit the gate voltage during regular operation to 0 V (Vg = 0V). With this, we can preserve the preprogrammed memory state (Q). When the device is in the erase state (Q = 3), the corresponding gate voltage pulse can be applied to modulate the device into the corresponding state. When a voltage of 10 V (1 ms) is applied, the device can be modulated into the off state (Q = 1). When the device is in the Q = 1 state, applying a voltage of 7.5 V (1 ms) will modulate the device into the Q = 2 state. On the basis of Q = 2, continuing to apply a voltage of 7.5 V (1 ms) can modulate the device into the Q = 3 state. Therefore, FG-PTs with different channel resistances were obtained through gate voltage pulse modulation.
The resistance of the FG-PT changes in different program states, and the corresponding proportional relationship with the fixed resistance also changes. The Vout between the FG-PT and fixed resistor exactly reflects the proportional relationship between them. As presented in Fig. 4c, the charge present in the floating gate strongly depletes the channel, which remains closed (Q = 1). At this time, the channel resistance is much greater than the fixed resistance, the device voltage division is large, and Vout is approximately 0. Under infrared light irradiation, the device shows a negative response, the equivalent resistance of the device increases, the proportional relationship between the device and series resistances does not significantly change, and Vout does not significantly change (details in Figure S5a). Therefore, the infrared light signal is effectively shielded at the voltage output end. Under visible light, MoS2 absorbs visible light, the carrier concentration increases, and the equivalent resistance decreases. The device resistance begins to become equivalent to the fixed resistor resistance, the device voltage division decreases, and Vout increases and shows an obvious visible light response. Based on the above analysis, when the device is in the Q = 1 state, visible light signal extraction and infrared light filtering are achieved. When the device is in the Q = 2 state, the resistance of the device is equivalent to the resistance of the series resistor. At this time, Vout is at the middle position between 0 and Vdd. When visible light is irradiated, Vout shows an obvious positive response. When exposed to infrared light, Vout also an obvious negative response. Therefore, in the intermediate state, both the positive and negative responses of the device can be completely output. When the device is in the Q = 3 state, the device resistance decreases, the device resistance is much lower than the series resistance, and the voltage division of the series resistor becomes larger, so Vout is approximately equal to Vdd. On this basis, when visible light is applied, the device resistance decreases, the voltage division of the series resistor does not significantly change, and Vout does not significantly change. Therefore, the visible light signal is effectively shielded in this state. When exposed to infrared light, the device resistance increases, the device voltage division increases, and Vout shows an obvious negative response. Visible light shielding and infrared light signal extraction are achieved in this state. When the equivalent resistance of the device is much lower than the series resistance, the device will not exhibit a photoresponse in either the visible or infrared region (as shown in Figure S6). Based on the above characteristics, by modulating the device state, the device exhibits a visible–infrared broad-spectrum photoresponse and wavelength-selective single-wavelength imaging. The simulation results of the carrier distribution in the device channel in the Q = 1 and Q = 3 states in the dark, under 532 nm illumination and under 1550 nm illumination are shown in Figure S7.
Broad-spectrum imaging has broad application prospects in satellite remote sensing, image recognition, and feature extraction. As shown in Fig. 1a, the visible and infrared information of an object can be extracted through the wavelength extraction function, which has profound significance for image detection and image recognition. Based on this, we studied the functionality of the device in wide-spectrum imaging and wavelength extraction. Figures 5 show a schematic of the experimental setup for dual-band imaging. We constructed a wooden formwork with the term ‘XDU’ to verify the imaging capability of the device. The ‘X’ is imaged using visible light, and the ‘U’ is imaged using infrared light. The visible and infrared light directed toward the front of the sample is transmitted (transmission imaging). As shown in Figs. 5d and 5j, when the device is in the program state (Q = 1, Fig. 5a), there is only a visible light response and no infrared response, so only the visible information ‘X’ can be seen. When the device is in the intermediate state (Q = 2, Fig. 5b), the device responds to visible and infrared light, and under dual-band imaging, the term ‘X’ and ‘U’ can be observed. When the device is in the erase state (Q = 3, Fig. 5c), the device can only respond to infrared light; therefore, under the irradiation of two beams of visible and infrared light, only the letter ‘U’ can be seen, and the results demonstrate that this state effectively extracts infrared information. Therefore, the device has a wide-spectrum response characteristic and can simultaneously present a visible–infrared dual-band response effect in the intermediate state (Q = 2), while in the program state (Q = 1) and the erase state (Q = 3), information with visible and infrared characteristics is effectively extracted, achieving wavelength-selective imaging. Devices with wavelength-selective imaging can effectively distinguish visible information and infrared information of objects, which is highly important in image recognition and can further improve the recognition rate. In addition, the wavelength-selective function can also be applied to encryption scenarios, and encryption and decryption functions can be achieved by adjusting the device storage state (Figure S8).
The above discussion mainly describes the rendering effect when visible or infrared light is illuminated alone or when it illuminates different areas. In real life, the illumination source is usually a broad-spectrum light source. Visible light and infrared light are integrated and illuminate the object at the same time. For this reason, we studied the effect of the device under the illumination of a visible–infrared broad-spectrum light source. In conventional imaging systems, if a single-band imaging effect needs to be obtained, then whether a wide-spectrum detector or a visible–infrared combination detector is used, a filter or grating component needs to be added to filter out excess light sources to achieve a single-band light source (Figure S9). In this system, because wavelength extraction is implemented on the device side, complex optical designs such as optical paths and optical components are eliminated, which effectively improves system integration and greatly reduces complexity. The imaging effects in mixed lighting mode are shown in Figs. 5g-i. The letter ‘D’ represents the mixed lighting mode of visible light and infrared light. When there is only visible light, an obvious ‘X’ appears in the Q = 1 and Q = 2 states, whereas there is only a blurry image in the Q = 3 state. When there is only infrared illumination, the imaging effect is very blurry in the Q = 1 state, while a very obvious 'U' is observed in both the Q = 2 and Q = 3 states. In the mixed illumination mode of visible light and infrared light, when Q = 1, the device shows an obvious positive response to visible light, and the clear part of the image is equivalent to the letter ‘X’. When Q = 3, the device shows an obvious negative response to infrared light, and similar imaging effects are obtained for the ‘D’ and ‘U’ characters. This also shows that in the Q = 1 state, the device can effectively extract visible light signals and filter infrared light signals. In the Q = 3 state, infrared light signals can be extracted while achieving visible light filtering. In the Q = 2 state, the positive response to visible light and the negative response to infrared light have a certain offset effect. Surprisingly, due to the angle of the two beams of light and the bidirectional photoresponse of the device, the boundaries of the pattern become clear, and very good three-dimensional imaging rendering is obtained, indicating that the device has great potential to play a role in the field of three-dimensional imaging. Therefore, in addition to achieving wavelength-resolvable visible positive response and infrared negative response functions, the device can also be combined with the floating gate storage function to achieve extraction of a single visible light image and extraction of a single infrared light image. Compared with traditional imaging systems, the structural complexity is greatly reduced, and the system integration is improved.
Figure 5 show the imaging effect of the device in active imaging mode, using a laser to illuminate the object for imaging. In order to obtain the imaging effect of the device in passive imaging mode, we imported the device data into the PRISM (Physically Reasonable Image Simulation Module) written on the C + + platform to simulate the imaging effect of the device. Figure 6 shows the imaging effect of a certain scene using the response information of the device in different states. First, the responsivity and detectivity of the visible and infrared of the device in three states are obtained, and then the information is input into the module. The software uses the digital model of the imaging system to perform optical modulation and spatial sampling of the radiation field to obtain the radiation intensity distribution that reaches the photodetector surface. Finally, the irradiance is photoelectrically converted according to the photodetector response model to generate a digital signal, which is stored as an 8-bit grayscale image. The specific simulation process is described in S10. Figure 6a is the original picture of the scene, show the imaging effects of the device under visible–infrared wide-spectrum illumination. Figures 6b show the imaging effects of the device in visible imaging mode corresponding to the Q = 1 state. Figure 6c show the imaging effects of the device in the infrared imaging mode corresponding to the Q = 3 state. Under the wide-spectrum photodetector, the elements in the scene are relatively complete, but there are no obvious features. On the other hand, in the separate visible light detection mode (Figs. 6b), the "mountain" information can be clearly extracted compared with the wide-spectrum rendering effect (Figs. 6a). In the separate infrared detection mode, the "river" information can be extracted (Fig. 6c). Therefore, the visible (infrared) wavelength extraction function can effectively improve the recognition rate of objects. Table 1 compares the photodetector in this article with the traditional wide-spectrum photodetector. Although a wide-spectrum photodetector composed of multiple sensors can simultaneously achieve switching between single-wavelength imaging and wide-spectrum imaging, on the one hand, modification of the external circuit is required, and different modes correspond to different detector control modes. On the other hand, the coordination between multiple detectors will inevitably cause problems such as complex structures. Although advanced broad-spectrum photodetectors have simple structures, they cannot achieve single-wavelength image extraction. The photodetector in this paper effectively solves the above problems and achieves wide-spectrum detection and single-wavelength imaging functions through a simple circuit structure, laying the foundation for the subsequent development of photoelectric detectors.
Table 1
Performance analysis and comparison of different types of devices
Device structure
|
Structural complexity
|
Working mode
|
Broad spectrum imaging
|
Single wavelength feature extraction
|
Image recognition rate
|
Filter Device + Photodetector
|
high
|
single wavelength imaging
|
low
|
Yes■ No□
|
high
|
Broad Spectrum Detector
|
low
|
Broad spectrum imaging effect
|
high
|
Yes□ No■
|
low
|
This work
|
low
|
single wavelength imaging+
Broad spectrum imaging effect
|
high
|
Yes■ No□
|
high
|