Piezotronic neuromorphic devices: principle, manufacture, and applications

In nanoscience, the bottom-up method means an extremely small manufacturing process which can grow the desired nanomaterials and structures on a substrate starting from atoms/molecules in a self-assembly fashion. The commonly used bottom-up methods include wet chemical synthesis and chemical/physical vapor depositions.

Generally, the wet chemical methods refer to the synthetic processes to prepare nanomaterials with liquid phase participating in the chemical reaction, including chemical liquid deposition, electrochemical deposition, sol-gel processing, hydrothermal synthesis, etc. The primary principle of wet chemical process is to sophisticatedly select one (or several) required soluble metal salts (or oxides) to prepare the reaction solution according to the composition of target synthetic materials [40–42]. In the reaction solution, the target elements in an ionic or molecular state uniformly precipitate or crystallize to form the required nanomaterials by applying (or crystallizing) suitable precipitants induced by evaporation, sublimation, or hydrolysis.

For the vapor deposition method, high-quality nanomaterials can be deposited on a substrate by using physical or chemical processes, which generally occur in the gas phase. According to the film growth mechanism of vapor deposition technique, it can be classified into chemical vapor deposition (CVD), physical vapor deposition, etc. Commonly, the vapor deposition techniques are accepted by researchers with high reactive temperature environment, expensive instruments, and comprehensive processing because these methods are preferential to fabricate various types of materials that cannot be synthesized by wet chemical synthesis, such as GaN. For instance, high quality NWs, 2D materials, etc, can be readily grown by the CVD method. During the CVD process, target oxide powder is heated in a tubular furnace and evaporated to the upper surface of the substrate by flowing Ar gas and then forming the desired material under specific temperature, pressure, atmosphere, etc. Pulsed laser deposition (PLD) is another common method for growing epitaxial thin films [43]. During the NWs growth process, the ceramic target is bombarded with a high-energy laser, transformed into a plasma, and then deposited on the top surface of the single-crystal substrate. Accordingly, a buffer layer pre-deposited on the substrate is critical to the morphology of the required target nanomaterials. For instance, a textured buffer layer prepared on the single crystal substrate is the key to the fabrication of vertically oriented NWs [44–47].

Although the bottom-up fabrication is qualified to grow large-scale nanomaterials, it also has some shortcomings, e.g. the nonuniform height, diameter, and composition of the target materials. In contrast to bottom-up method, another fabrication strategy (i.e. top-down method) is well used in the semiconductor industry. The top-down fabrication process refers to the direct preparation of desired nanostructures on the substrate (or bulk material) through a series of thin film deposition, photolithography, and etching techniques. By employing top-down patterning and inductively coupled plasma techniques, highly uniform and well-oriented NWs can be fabricated on a high-quality epitaxial single crystal film [48–51]. The top-down method has also been widely applied to the preparation of devices such as 2D materials-based transistors. For instance, by applying external shear force on the material bulk, the distance between the layers is increased and the van der Waals forces are weakened to realize the exfoliation of 2D crystals.

These two types of synthesis and fabrication methods are highly promising for preparing the active nanomaterials for piezotronic neuromorphic devices. The bottom-up fabrication has shown good compatibility and a wide range of applicability, while the wet chemical synthesis is the least expensive method in terms of both equipment and materials. For vapor deposition, the fast growth rate and large growth area make it suitable for continuous fabrication in laboratory (or industry). Top-down fabrication process is relatively expensive but it is compatible with state-of-the-art semiconductor technology and available for large-scale fabrication, which generally requires large-scale equipment paired with complicated techniques. The elaborate combination of top-down and bottom-up methods is highly required to prepare more sophisticated neuromorphic piezotronic devices in future.

Tactile sensation is an important form of perception in which people acquire information from the surrounding environment. Simulated tactile perception sensors have been developed and widely used in various scenarios. However, the integration and resolution of tactile sensors using piezoresistive or capacitive types are low, which are not suitable for high precision and tiny tactile sensing. By replacing the gate with strain to regulate the channel carriers, Wang's group proposes a two-terminal strain-gated vertical piezotronic transistor (SGVPT), as shown in figure 4(a) [20]. By integrating ZnO NW-based SGVPT into a 92 × 92 addressable pressure sensing array, active and adaptive tactile imaging is achieved. As shown in figure 4(b), the array topology diagram shows the structure and arrangement of the devices. The top and bottom electrodes are gold/indium tin oxide (Au/ITO) constructed in an orthogonal configuration for addressing external pressure stimuli. Gold electrodes and ZnO form Schottky contacts, which act as a piezoelectric gate. A mold in the shape of a letter A is pressed vertically onto the surface of the device with a pressure of 25 kPa. Figure 4(c) shows the corresponding current response contour plot. The device has a high spatial resolution of 234 dpi and a sensitivity of about 2 μS kPa⁻¹, which demonstrates its excellent pressure sensing capability. This interesting structural design based on ZnO NWs provides a new way to construct tactile imaging devices with high spatial resolution. In addition, the spatial resolution of the pressure sensor can be increased to 12 700 dpi based on an ordered array of hexagonal ZnO nanosheets [23]. By integrating a WO₃ electrochromic device film and a ZnO-NW matrix pressure sensor, Han et al simultaneously realize the visualization and recording of spatial pressure [52].

Zoom In Zoom Out Reset image size

Tactile perception based on piezotronic effect can also be applied in smart power devices. Inspired by the human reflex process, Zhang et al propose a high electron mobility transistor with a cantilever structure based on the AlGaN/AlN/GaN heterojunction (figure 4(d)) [25], which can serve as a strain-controlled power device (SPD) to achieve fast modulation of the output power by external mechanical stimulation. As shown in the inset of figure 4(b-ii), the scanning electron microscope (SEM) image shows the cantilever structure of the SPD. Figure 4(e) shows the output characteristics of the SPD under different strains, in which the gate voltage V_G is 1 V. The output current density of the cantilever increases under external stress, with a 17.5% increase in output current at 16 mN stress compared to no stress. In addition, V_G can also effectively modulate the output current of the cantilever under strain. Based on the SPD, acceleration feedback control (one of the important functions of AI applications) is successfully implemented on a microcontroller (inset of figure 4(b-iii)). The output power density of SPD versus acceleration is shown in figure 4(f). As the acceleration increases (from 1 G to 5 G), the output power increases with self-regulation capability. The proposed SPD has promising applications in AI fields, such as autopilots and intelligent robots.

In tactile perception systems, the perception and detection of pressure in different directions are also necessary. Shadi A. Oh et al prepare a flexible tactile sensing array based on dual-gate piezoelectric ZnO thin film transistors (TFTs) for sensing shear forces and autonomously adjusting closed-loop grasps [53]. For a single ZnO TFT, piezoelectric charges accumulate on the channel surface under the action of vertical forces and lead to a current variation for the output, which enables the perception of tactile sensation (figure 4(g)). Based on the reliable current switching capability of TFTs with dual-gate structure, two 8 × 16 multiplexed tactile sensor arrays are prepared on wafers for pressure mapping. As shown in figure 4(h), the average force response of the 10 different areas in the array is proportional to the magnitude of the applied force. The calculated sensitivity of (0.063 5 ± 0.014 8)%·mN⁻¹ with a linearity error of 4.35% demonstrates the excellent normal stress sensing capability of the TFT. Besides, by using 3D polydimethylsiloxane (PDMS) pillars on multiple force sensors, the lateral shear forces are converted into vertical forces for measurement, enabling the sensing of different force directions and magnitudes (figure 4(i)) [54, 55]. This high spatiotemporal resolution multiplexed tactile sensing array provides a novel approach to constructing normal and shear force sensory devices.

Vision is the sense that provides the most amount of information, so it is important to simulate visual perception for neuromorphic optoelectronics. The piezo-phototronic effect, first proposed in 2010 [11, 56, 57], reflects the coupled response of optoelectronic devices to stress and light, providing a new idea for the integration of multiple senses, such as tactile and visual sensing. Wang et al demonstrate a hybridized photodetector based on p-Si/n-ZnO heterojunction [58], as shown in figure 5(a). The piezoelectric polarization charge induced on the n-ZnO side under strain can effectively modulate the photoelectric effect on the p-Si side, thus improving the optoelectronic performance of the device. The SEM image of the device is shown in the inset of figure 5(a). First, micro-pyramid structures are etched on the p-type silicon wafer to increase the surface area and help to improve the light absorption. Then a ZnO NW layer is covered on the sample surface to form p-Si/n-ZnO heterojunction. Finally, Ag NWs are dispersed on the ZnO array with a sputtered ITO layer as the top electrode. The output current of the p–n junction at different optical power densities and strains is measured and extracted with a fixed bias voltage of 2 V (figure 5(b)). The output current increases significantly with the increase of optical power density and external strain. The performance optimization of silicon-based optoelectronic devices is achieved through a smart integration of strain and illumination, and this general strategy can be applied to various fields.

Zoom In Zoom Out Reset image size

The piezo-phototronic effect can also influence the performance of p–n junction photodetectors. Xue et al demonstrate a p–n photodiode based on MoS₂/ZnO heterojunction, and the structure schematic is shown in figure 5(c) [59]. The p-type MoS₂ films are obtained by exfoliating from bulk crystals and treated with SF-6 plasma, and the n-type ZnO films are deposited on the wafer surface by PLD. The transparent acrylic above the heterojunction can detect the amount of applied pressure and has little effect on the incident UV light. The piezo-phototronic effect can enhance the photoresponse of p-type MoS₂ and n-type ZnO photodiodes. Figure 5(d) shows the energy band diagram of the photocurrent enhanced by the piezo-phototronic effect. Under a reverse bias and light illumination, the Fermi levels of MoS₂ and ZnO will generate an energy difference, which is beneficial to the separation of photogenerated carriers. When external pressure is applied to the device, the positive and negative piezoelectric charges are separated along the c-axis, causing the energy band on the ZnO side to bend downward. This enables the regulation of the diode photocurrent by external pressure.

The vertical array structure contributes to higher computational speed and lowers the chip size and power consumption [61]. Han et al propose a UV photodetector array based on ZnO NWs (32 × 40 pixels, figure 5(e)) [60], in which the top and bottom ITO/Au electrodes are vertically aligned, and the Au and ZnO NWs form Schottky junctions. Strain-induced piezoelectric polarization charges can enhance the optoelectronic properties of ZnO NWs arrays. The imaging of the output current of the ZnO NWs array upon UV illumination is measured under the circumstances of strain-free and 40.38 MPa pressure, respectively (the UV illumination intensity is 1.38 mW cm⁻² and the applied bias voltage is 1 V). Figure 5(f) shows the current difference between the two imaging results, which indicates that the output current of the photodetector array increases significantly under applied pressure. With the piezo-phototronic effect, the UV photodetector array has achieved a 700% increase in photoresponsivity and a 600% increase in sensitivity, which provides an essential method for designing high-performance optoelectronic devices.

The piezo-phototronic effect is also used to enhance the light-emitting properties of optoelectronic devices. Pan et al construct a NW light-emitting diode (NW-LED) array for electroluminescent imaging with a spatial resolution of 2.7 μm (figure 5(g)) [21]. By growing n-type ZnO NWs on p-type GaN thin film substrates, an array of p–n junction light-emitting devices is formed. Under forward bias, the luminescence performance of each pixel is modulated by the strain magnitude. Figure 5(h) shows the enhancement E of the NW-LED as a function of the applied compressive strain, together with the corresponding luminescent status. The enhancement E reflects the change in the luminescence intensity of the device, which is proportional to the magnitude of the compression strain. When a 'PIEZO'-shaped stamp is pressed on the NW-LED array, figure 5(i) shows the corresponding optical image of the stamp and the luminescent image of the device under applied strain of 0 and −0.15%. This NW-LED array converts external mechanical signals into electroluminescent signals to visually reflect the distribution of applied pressure, enabling efficient data transmission and processing.

In addition to tactile and visual perception, other sensing modals can also be modulated by piezotronic effects. As one of the important sensory channels, hearing is the most direct way of receiving information during human/robot communication. As shown in figure 6(a-i), Han et al demonstrate a flexible piezoacoustic sensor (f-PAS) that implements speaker recognition based on a machine learning process [24]. Flexible PZT films with piezoelectric properties are transferred to a polymer substrate, and the concaved trapezoidal structure mimics the basement membrane in the human cochlear (figure 6(a-ii)). Seven channels of the interdigitated electrodes are defined by photolithography and etching processes. The vibrations induced by the human voice trigger the multi-channel piezoelectric current signal, which is a self-driven process. Using machine learning to extract the frequency features of speech signals and train them to achieve speaker recognition, figure 6(a-iii) shows the speaker recognition error rate statistics of f-PAS and commercial microphones for 30 mixed samples. The recognition error rate of two-averaging is only 2.5%, which demonstrates the excellent recognition capability and potential application advantages of f-PAS.

Zoom In Zoom Out Reset image size

Piezotronic neuromorphic devices have also been used to simulate taste perception. Zhao et al propose a self-powered electronic skin to achieve taste perception function by enzyme-modified ZnO NW arrays [62]. As shown in figure 6(b-ii), the device is composed of ZnO NWs and interdigitated electrodes, which can output a continuous piezoelectric signal through the piezoelectric-enzymatic reaction coupling effect. By modifying ZnO NWs with ascorbate acid oxidase (AAO), figure 6(b-iii) shows the piezoelectric output voltages and responses of ZnO and AAO@ZnO NWs after adding 5 ml of ascorbic acid (AA) solution to 20 ml of pure water. Compared with the unmodified ZnO NWs, the response of AAO@ZnO NWs to the AA solution is significantly enhanced, which demonstrates the favorable recognition of AA by the devices. In addition, the self-powered electronic skin is used to detect the AA concentration of different fruits (figure 6(b-iv)). Enzyme-modified ZnO NW arrays provide a new approach to achieving gustatory perception.

Gas sensors are capable of converting the composition and concentration information of the measured gas into electrical signals [65], which can be considered as simulated olfactory perception. Zhou et al propose a piezotronic effect enhanced gas sensor based on ZnO NW, which can be used to detect hydrogen (H₂) and nitrogen dioxide (NO₂) gases (figure 6(c-i)) [63]. By applying strain to the ZnO NW through the displacement of a linear motor, they have successfully modulated the sensor sensitivity and realized the gas detection at room temperature. Figure 6(c-ii) shows the current variation of the device under H₂ atmosphere and compressive strain conditions. An increase in either H₂ concentration or strain results in a higher output current, which reflects the high sensitivity and resolution of the gas sensor. The energy band diagram reflects the operating principle of the ZnO NW gas sensor (figure 6(c-iii)). Under compressive strain, the energy band of ZnO bends downward. When H₂ gas is passed into the cavity, the H₂ molecules dissociate by reacting with O²⁻, and electrons are injected into the conduction band and accumulate on the ZnO surface, which increases the response current. By contrast, when NO₂ gas is passed into the cavity, the NO₂ molecules associate with the oxygen vacancies. Electrons are discharged from the conduction band to induce a depletion layer on the surface, which reduces the output current of the device. Thus, the piezotronic effect can improve the performance of gas sensors by changing the energy band structure of the M–S interface.

In addition, piezo-phototronic effects in 2D semiconductors are also used in gas sensing applications. Guo et al propose a flexible NO₂ gas sensor based on MoS₂ FET (figure 6(c-iv)) [64], and the current variation of the device reflects the ability of gas sensing. The device is obtained by transferring single-layer MoS₂ grown by CVD to a flexible PET substrate with defined source and drain electrodes on both sides. As shown in figure 6(c–v), when the NO₂ concentration is 400 ppb, the sensitivity of the device is improved to 671% under the tensile strain of 0.67% and the 625 nm optical illumination of 4 mW cm⁻² (compared with no strain and light conditions). In addition, the response/recovery time of the sensor for NO₂ gas is reduced considerably. The combination of excellent physical properties of 2D materials and piezoelectric effects facilitates the construction of highly sensitive gas sensors.

In addition to simulating biological perception, piezotronic neuromorphic devices are required to have the function of data memory/storage after receiving and converting the external stimulus signals. For instance, the modulation of the metal-electrolyte interface by the piezopotential can impact the memory switching characteristics of a memristor [2, 66–68]. Wu and Wang construct the first resistive switching device by using ZnO NW [69], which effectively modulates the memory properties of resistive memory cells by strain (figure 7(a)). When ZnO NW is subjected to external stress, the hysteretic I–V characteristics of the resistive memory cell change accordingly. As shown in figure 7(b), when the device is under tensile strain ( = 1.17%), the hysteretic switching curve shifts in the negative voltage direction; when the device is under compressive strain ( = −0.76%), the I–V characteristic curve shifts in the positive voltage direction. V_th,S and V_th,D represent the threshold switching voltages in the I–V characteristic curves, which will increase or decrease according to the applied strain (figure 7(c)). It should be noted that the window width of the threshold voltage remains almost constant under different strain states. The demonstrated ZnO NW-based nonvolatile resistive switching memories have a large range of applications.

Zoom In Zoom Out Reset image size

In addition, Liu et al construct piezotronic memristors with multilevel switching characteristics by synthesizing bamboo-like GaN micro-wires via CVD method [70]. Figure 7(d) shows the schematic of the memristor and the SEM image of the GaN micro-wires. Typical GaN micro-wires have distinct bamboo-shaped knots, which are formed by nucleation and radial growth of GaN in the form of seed crystals during the growth process. The memory switching characteristics of piezotronic memory resistors are mainly attributed to nitrogen-deficient defects in the GaN knot region [72, 73]. By applying compressive strain to the device using a 3D mechanical stage, they explore the influence of the piezoelectric effect on the performance of GaN memristors. Figure 7(e) shows the I–V characteristic curves of the piezotronic memristor under different compressive strains, which exhibit a larger set voltage (the process of switching the device from high resistance state (HRS) to low resistance state (LRS)) at larger compressive strains. Figure 7(f) shows the set voltages at different strains extracted from figure 7(e). As the applied compression strain increases from 0 to −0.76%, the set voltage increases from 1.42 to 2.15 V. Piezotronic memristors can readily optimize the storage performance by modulating the operating voltage through piezotronic effects.

Sensory memory can hold transient stimuli signals for a short period of time, which is essential in individuals' real-time interaction with the environment [74]. As shown in figure 7(g), after receiving an external stimulus, skin receptors are able to store and retain sensory information for certain time, mainly determined by the action of sensory receptors and memory units. Inspired by the skin memory process, Hua et al propose a piezotronic sensory memory based on a single gallium nitride (GaN) microwire [71], which realizes the implementation of sensory and memory functions on a single device. Figure 7(h) shows the typical I–V characteristics of the piezotronic memory before and after applying strains. When applying −1.57% compressive strain to the device, the current of the device in the I–V curve decreases significantly, which indicates that the resistance of the device transitions from LRS to HRS. When the compressive strain is released, the I–V curve remains essentially constant and the resistance maintains at HRS. This demonstrates the good resistive memory capability of the piezotronic memory. In addition, the device's resistance can be reversibly returned to the LRS by applying a positive erase voltage. As shown in figure 7(i), the endurance characteristic of the piezotronic memory for 100 cycles is measured under writing process with a strain of −1.57% and erasing processing with the electrical voltage at 3 V. Piezotronic sensory memory enables tactile retention in a single micro/NW, contributing to the development of interactive neuromorphic devices and memories.

The transfer characteristics of the piezotronic strain-gated transistor (SGT) can be directly modulated by the applied external stress. The integration of multiple SGTs is expected to enable multiple logic operation functions for stress control. As early as 2010, Wu et al first implement piezotronic logic operations with ZnO NWs [19]. With continuous research and development on piezotronic logic devices, piezotronics is being broadened to various piezoelectric materials toward more sophisticated applications. Yu et al propose a piezotronic SGT based on GaN nanoribbons and demonstrate the universal logic operation [75]. As shown in figure 8(a), they prepare GaN nanoribbons in uniform size by strain-controlled cracking of GaN films, and then fix the two ends of GaN nanoribbons with silver paste as the source and drain electrodes of the device. The inset of figure 8(a) shows the optical image of the GaN nanoribbon piezotronic SGT. Based on the piezotronic effect of GaN, the output current of the piezotronic SGT exhibits 'off' and 'on' states under tensile and compressive strains, respectively (under a fixed bias). The piezotronic inverter is made by encapsulating two GaN nanoribbon SGTs on the same substrate. Figure 8(b) shows the variation of the output voltage of the piezotronic inverter for different strain inputs, where logical '0' and '1' are obtained for positive and negative strains, respectively. In addition, other logic operations, including AND, OR, NAND, NOR, and XOR gates, can also be implemented based on the SGT device with GaN nanoribbon. This concept of strain-gated piezotronic logic has great potential for application in intelligent electronic devices.

Zoom In Zoom Out Reset image size

The piezo-phototronic effect can provide a more flexible and convenient modulation strategy for logic computing devices. Due to the visible light response and inherent piezoelectric properties of cadmium sulfide (CdS), Yu et al prepare an light-strain-gated transistor (LSGTs) based on CdS NWs [76]. Controlling the carrier transport process by piezo-phototronic effect, figure 8(c) shows the schematic diagram of LSGT and corresponding energy band under optical and mechanical stimulation. The SBH at the interface between CdS and the two terminal electrodes is reduced under illumination, resulting in a significant enhancement of the photocurrent (figure 8(c–i)). Under compressive strain, the positive and negative piezoelectric polarization charges appear near the interface at both terminals, making the potential barrier height of the source contact decreases while the potential barrier height of the drain contact increases (figure 8(c-ii)). When the light and strain are applied simultaneously, the strain magnitude affects the transport behavior of the photogenerated carriers. A small compressive strain (e.g. −0.24%) can help to alleviate the recombination of light-induced electron-hole pairs, resulting in an increased photoconductivity of the CdS NW device (figure 8(c-iii)). At large compressive strains, the valence band edge of the CdS NW is above the Fermi level of the contact electrode and the separation of light-induced electron-hole pairs is suppressed, which thereby leads to a significant reduction in photoconductivity (figure 8(c-iv)). Due to the significant changes in the conductance of LSGTs under applied light and strain, the light/strain-controlled piezo-phototronic logic gates such as 'AND' and 'OR' can be implemented as shown in figure 8(d). The logic '0' of the light input L_in is defined to apply light to #2 and #4 LSGTs, while the logic '0' of the mechanical input _in is defined to apply strain to #1 and #2 LSGTs. The optical and mechanical logic '1' is defined as the application of light or strain to the remaining two LSGTs, and the logic output is the amplitude of the output voltage. Other universal logic units have been successfully demonstrated by designing different electrical circuits. Piezo-phototronic circuits provide a new idea for the construction of logic computing devices.

Optofluidic nano/microsystems combine microfluidic and optical technologies to achieve great progress in the large-scale integration of optoelectronics. Purusothaman et al propose a piezopotential modulated optofluidic logic device [77]. Figure 8(e) shows the schematic of Y-channel optofluidic device (Y-OF) and the SEM image of layer-by-layer structure. The sensing layer is made by in-situ growth of zinc oxide nanorods (ZnO NR) on a polyvinylidene fluoride (PVDF) substrate, which can detect three sources of optical, fluidic, and strain stimuli. The fluid flows from two inlet channels (CH₁/CH₂) of the Y-shaped PDMS groove, converges on the PVDF/ZnO NR surface, and then flows out. Figure 8(f) shows a simplified schematic of the Y-OF implemented OR logic gate and AND logic gate. Two kinds of solvents (e.g. ethanol and decanol), can be identified as logical '0' and logical '1' according to the high and low response of the photocurrent. Only when the ethanol flows as a fluid source in both channels, the photocurrent of the Y-OF is low, corresponding to the OR gate condition '0:0 = 0'. When other combinations of fluid sources flow through the channels, the photocurrent of Y-OF will be higher, corresponding to the OR gate conditions '1:0 = 1', '0:1 = 1', and '1:1 = 1'. Therefore, the Y-OF can achieve OR gate under unstrained conditions. In contrast, the photocurrent response of Y-OF under compressive strain (−1%) will be different. For instance, the photocurrent will be high only when decanol flows as a fluid source in both channels, while it will be lower than the threshold in other combinations of fluid sources. Accordingly, the Y-OF can achieve AND gate under compressive strain. The system integration of optofluidics and piezoelectricity enables convenient switching of multiple logic functions, which contributes to the development of interactive optoelectronics.

Yang et al propose a low-power method for coupling ion-gel gating with piezoelectric effects by exploiting ZnO NW and ion-gel dielectrics [78]. As shown in figure 8(g), a tunable piezotronic logic inverter is constructed by connecting a 10 MΩ resistor in series to an ion-gel-gated ZnO NW device. Figure 8(h) shows the strain versus output voltage (V_OUT) at different gate voltages (V_G = 0, 2, and −2 V). The variation of V_OUT with strain is larger at negative gate voltage than at positive or zero gate voltage. By periodically applying and removing 0.96% tensile strain to the ZnO NW at different gate voltages, the real-time V_OUT of the logic device is shown in figure 8(i). The states with no strain and with 0.96% tensile strain applied are considered as '0' and '1' of the logic input, respectively. V_OUT decreases significantly after applying a tensile strain, enabling a logic transition from '1' state to '0' state, which corresponds to the NOT gate. The gate voltage can affect the variation of V_OUT in the NOT gate. V_OUT increases from 0.69 V to 1.48 V at a V_G of −2 V, with an increase of 97.5%. The modulation of the logic inverter is implemented by the combination of ion-gel gate control and piezotronic effect, which provides a novel approach to realizing low-power and sophisticated piezotronic devices.

Except the piezopotential interface modulation for conventional piezotronic devices, a series of general-conceptual piezotronic devices can be regarded as various electronic devices gated/driven/modulated by PENG. In this way, more abundant piezoelectric polymers with good flexibility and low-temperature processability can be utilized to fabricate multifunctional sensory devices. The coupling behavior between piezoelectric polarization and semiconductor transport properties can also be used to emulate an artificial sensory neuron (or afferent nerve) to implement parallel sensation and computation on the external mechanical stimuli. The neuromorphic devices based on PENG-gated transistors have attracted much attention in recent years due to the advantages of self-powering and directly sensing external stimuli. Sun et al propose a PENG-gated graphene transistor and construct an active strain sensor array [13]. The device consists of a polyvinylidene fluoride-trifluoroethylene [P(VDF-TrFE)]-based PENG and an ion-gel-gated graphene transistor, in which the graphene electrode on the bottom of the PENG is connected to the ion-gel layer of the transistor (figure 9(a)). The strain-induced piezopotential in the PENG is coupled to the gate of the transistor, which effectively modulates the Fermi level of the graphene channel and changes the carrier concentration. Figure 9(c) shows the output characteristics of the graphene transistor when the PENG is under different strain conditions. As the tensile strain increases, the drain current of the transistor increases. Figure 9(d) shows the dynamic current response of the PENG-gated graphene transistor under 0.06% tensile strain and −0.06% compressive strain, which demonstrates a significant and fast feedback to the piezopotential. The strain sensor exhibits excellent sensitivity at lower strains (figure 9(e)). In addition, an active strain sensor array is constructed based on PENG-gated graphene transistors. The output currents in different regions of the array under the diagonal direction of strain are shown in figure 9(b), and the magnitude of the output currents indicates the distribution of strain.

Zoom In Zoom Out Reset image size

PENG-gated FETs can also be used in mechanical information storage. A piezopotential-controlled nonvolatile memory array and the output characteristics under strain are shown in figure 9(f) [14]. The channel material of the FET is indium gallium zinc oxide (IGZO), and the piezoelectric layer of the PENG is P(VDF-TrFE) material. The piezopotential generated by the PENG under strain applies programming and erasing operations to the memory, resulting in lower power consumption of the device. The output characteristic curves under different strains demonstrate the favorable strain sensitivity of the device. Figure 9(g) shows the multilevel storing/erasing steps of the piezopotential-controlled memory by coupling four PENGs to the same ion-gel gate dielectric of the FET. The programming operation can be applied sequentially to each PENG by applying strain and disconnecting the ground terminal of the electrode. And the erasing operation is performed by connecting the electrodes of each PENG to the ground sequentially. The retention of drain current at different levels demonstrates the excellent multilevel data storage performance of the device.

By using the similar strategy, the piezopotential of the PENG can also be used to modulate the Schottky barrier in a graphene barristor. The device includes the PENG with a P(VDF-TrFE) piezoelectric layer and a vertical Schottky barrier transistor with a graphene/IGZO/ITO structure (figure 9(h)) [79]. Based on the capacitive coupling of P(VDF-TrFE) and graphene by ion-gel dielectrics (resulting in electric double layers (EDLs) at the ion-gel interfaces), the piezopotential induced by the strain of PENG can effectively modulate the work function of graphene. The multilevel modulation of the Schottky barrier and the current density of the channel are achieved by coupling two PENGs to the graphene electrode of the same device (figure 9(i)). When a PENG generates a positive piezopotential under tensile strain (0.08%), the Fermi level of graphene moves upward and the current density of the device increases. Meanwhile, the same tensile strain is applied to another PENG, and the Fermi level of graphene moves further upward, and the current density of the device exhibits a stepwise rise. When the strains of the two PENGs are released sequentially, the Fermi level of graphene moves downward to achieve reversible modulation on the current density.

The realization of high-performance artificial synaptic electronics is crucial for neuromorphic computing [80]. In the human sensory system, tactile receptors receive external mechanical stimuli and cause the variations of action potential, which are subsequently transmitted to the central nervous system via neurons and synapses (figure 10(a)) [17]. Inspired by the biological sensory system, Chen et al propose a piezotronic graphene artificial sensory synapse as shown in figure 10(b). The system consists of two parts, in which the PENG is used for energy supply and pressure sensing, while the ion-gel-gated transistor is used to simulate synaptic behavior. The piezopotential triggered by external mechanical stimuli causes the formation of EDLs in the ionic gel, which modulates the carrier concentration in the channel and induces changes in the postsynaptic current (PSC). The synaptic characteristics exhibited by ion-gel-gated transistors can be affected by different strain conditions. Figure 10(c) shows the excitatory postsynaptic currents (EPSCs) of the artificial sensory synapse at different magnitudes of stretching strain (0.05%–0.8%). As the applied tensile strain gradually increases, the peak value of the EPSC increases. Temporal and spatial information such as externally applied strain amplitude and duration can be deduced and identified by the EPSC signal. This flexible and self-powered artificial sensory synapse enables direct perception of external stimuli and efficient modulation by piezoelectric effects.

Zoom In Zoom Out Reset image size

In addition to replacing the gate voltage in the transistor, the piezopotential generated by the PENG can directly supply energy to the two-terminal device to realize self-driven neuromorphic devices. Kim et al propose an artificial synapse for self-powered biomedical devices [16]. The amorphous (Na_0.5K_0.5)NbO₃ (NKN) thin films possess favorable piezoelectric properties and biocompatibility. The device consists of a Pt/NKN/TiN structured memristor and an NKN-based PENG, which are connected via a bridge rectifier circuit (figure 10(d)). Under the pressure of a finger, the NKN PENG generates an output voltage of more than −2 V to drive the NKN memristor. The NKN memristor exhibits typical synaptic behavior, including short-term plasticity and long-term plasticity (LTP), which is attributed to the movement of oxygen vacancy and changes in the shape of oxygen vacancy filaments in the NKN. Metaplasticity refers to the application of priming stimuli to modify synaptic plasticity before the application of a primary spike to induce synaptic behavior. The priming spike modulation on the metaplasticity is reflected in the spike time-dependent plasticity as shown in figure 10(e). The combination of NKN-based PENG and memristor provides a feasible strategy for applications in biomedical and human-machine interaction fields.

Zhang et al propose an artificial spiking afferent nerve (ASAN) based on NbO_x Mott memristors, and build an artificial spiking mechanoreceptor system (ASMS) with a piezoelectric device as the tactile sensor (figure 10(f)) [81]. Pressing and lifting of the finger can generate positive and negative voltage signals in the piezoelectric device. And the produced sinusoidal voltage signal is fed into the ASAN and converted to a peak signal. The input voltage of the ASAN as a function of response frequency is shown in figure 10(g), exhibiting excitability under low input voltages and protective inhibition under high input voltages. Figure 10(h) shows the frequency response of the ASMS under different pressure intensities. The high peak voltage causes the ASAN to stop peaking (protective behavior) under applied high pressures. By combining piezoelectric devices with artificial afferent nerves, external sensory signals are successfully converted into dynamic spike frequencies.

Due to the inherent piezotronic and piezo-phototronic effects of the piezoelectric semiconductor materials in artificial synaptic devices, the synaptic plasticity can be easily modulated by external mechanical stimuli and illumination. Hua et al propose a piezotronic synapse based on a single GaN microwire for strain sensing and synaptic functions [82]. The structure of the artificial synapse and the principle of piezotronic modulation are shown in figure 11(a). The hexagonal GaN NW can generate piezopotential under strain to achieve tactile sensing. More importantly, the piezotronic effect can effectively modulate the synaptic behavior of GaN microwire, which is mainly because the nitrogen vacancies at the junction region are redistributed on the positive charge side under the effect of stress-induced piezopotential. Figure 11(b) shows the current response of the artificial synaptic device to a voltage pulse (2.5 V, 1 ms). When the pulse voltage is reduced to 0.1 V, the output current gradually decays over a period of time, which is mainly attributed to the process of re-releasing the trapped electrons after the withdrawal of the electric field. The decay process of the current can be fitted accurately with an exponential function (the inset of figure 11(b)). The synaptic weights versus the voltage pulse number (2.7 V, 1 ms, Δt = 3 ms) under different strains are shown in figure 11(c). When the strain is increased from 0 to −0.36%, the synaptic weight of the piezotronic synaptic device increases significantly by 330%. In addition, artificial synaptic devices based on ZnO NW can also be modulated by both piezotronic and piezo-phototronic effects. By coupling the piezotronic effect with the optoelectronic properties, Hu et al propose a flexible photonic synapse constructed by a single ZnO micro/NW (figure 11(d)) [83]. The conductance changes of the synaptic device under the impact of light pulses match the behaviors of biological synapse, and the strain-induced piezopotential can effectively modulate the synaptic plasticity. The I–V characteristic curves of the device are modulated by the applied strain and 365 nm UV light (figure 11(e)), exhibiting a significant increase in photocurrent at −0.18% compressive strain and illumination. The contact properties between ZnO and silver paste electrodes lead to a nonlinear fluctuation of the curve [32, 57, 63]. As shown in figure 11(f), the conductance exhibits a gradual increase of enhanced synaptic plasticity under the action of 10 consecutive light pulses (0.227 mW cm⁻², 1 s, Δt = 1 s), and the applied compressive strain can promote the increase of conductance.

Zoom In Zoom Out Reset image size

Kumar et al propose an artificial sensory synapse based on NiO/ZnO p–n junction for neuromorphic tactile perception [84]. The structure and energy bands of the peroxide-based artificial synapse under high strain are shown in figure 11(g). The device has a vertically aligned structure and bends in response to external mechanical stimuli. As the applied strain increases, more charges are trapped at the interface of the NiO/ZnO p–n junction, thus realizing the modulation of synaptic behavior with strain. Figure 11(h) shows the modulation of multipulse synaptic plasticity by the applied strain. When 50 consecutive positive voltage pulses (+2 V, 2 ms, Δt = 2 ms) and 50 negative voltage pulses (−2 V, 2 ms, Δt = 2 ms) are applied, the device exhibits typical LTP and long-term depression behaviors. In addition, the PSC increases significantly as the strain increases from 0 to 0.6%. Similarly, the paired-pulse facilitation (PPF), a major important characteristic in biological synapses, can be well modulated by strain. The PPF index increases as the pulse interval time decreases and it is enhanced by the applied strain (figure 11(i)). As previously mentioned, NiO/ZnO-based artificial synapses can be efficiently modulated by external mechanical stimuli, which contributes to the development of low-complexity and low-power neuromorphic tactile perception systems.