Neuromorphic applications have shown great promise not only for efficient parallel computing mode to hold certain computational tasks, such as perception and recognition, but also as key biomimetic elements for the intelligent sensory system of next-generation robotics. However, achieving such a biomimetic nociceptor that can adaptively switch operation mode with a stimulation threshold remains a challenge. Through rational design of material properties and device structures, we realized an easily-fabricated, low-energy, and reconfigurable nociceptor. It is capable of threshold-triggered adaptive bi-mode jump that resembles the biological alarm system. With a tunnel silicon nitride () we mimicked the intensity- and rehearsal-triggered jump by means of the tunneling mode transition of dielectric. Under threshold signals the device can also express some common synaptic functions with an extremely low energy density of . In addition, through the modulation of thickness it is relatively easy to fabricate the device with differing pain degree. Our nociceptor analog based on a tunneling layer provides an opportunity for the analog pain alarm system and opens up a new path toward threshold-related novel applications.
【AIGC One Sentence Reading】:A biomimetic nociceptor using tunnel silicon nitride adapts its operation mode based on stimulation threshold, resembling the biological pain alarm system, offering low-energy synaptic functions and tunable pain response.
【AIGC Short Abstract】:In this study, a biomimetic nociceptor was developed using tunnel silicon nitride, exhibiting a threshold-triggered adaptive bi-mode jump akin to biological alarm systems. This easily fabricated, low-energy device offers reconfigurability and synaptic functionality, enabling modulation of "pain" degrees through varying Si3N4 thickness. This nociceptor analog paves the way for analog pain alarm systems and novel threshold-related applications.
Note: This section is automatically generated by AI . The website and platform operators shall not be liable for any commercial or legal consequences arising from your use of AI generated content on this website. Please be aware of this.
1. INTRODUCTION
In a biological system, the skin, the first line of defense against outside potential threatening or damaging environmental inputs, provides various nerve endings whose specific obligation is to manage stimuli and then to set an adaptive alarm system. In this system, when the intensity and quality of stimuli are considered as noxious degree, which means potential injury if it continues to act, the nociceptor begins to work to cause subsequent nociceptive reactions—such as reflex withdrawal, autonomic responses, and pain—aimed at maintenance of the body’s integrity [1]. As such, the nociceptor analogs implemented by solid-state electronic devices should be an indispensable element for development of e-skin and next-generation robotic applications. Recently, neuristors (i.e., synaptic devices) as an emerging technology in an artificial intelligence system, can provide not only efficient neuromorphic computing based on parallel compute-in-memory (CIM) architecture but also a promising path to physically emulate the biological nervous behaviors [2]. Using synaptic devices currently developed, one could conceptually make multi-mode perception emulations of pressure sensing [3], visual recognition [4], and sound localization [5], etc. Unfortunately, despite the great progress in device-level neuromorphic hardware implementations, their applications in simulation of pain-generated tasks are yet limited.
To not hinder the harmless interaction with the environment, the nociceptor mediation reasonably occurs at a higher threshold compared to normal perception. This threshold property is a key for emulating a nociceptor of only sufficient stimuli to generate pain. In this concept, nociceptor analogs will require the ability of bi-mode synaptic operation upon a single device to embed two-channel perception, to mimic the no-injured and injured states (i.e., low- and high-threshold). Undoubtedly, for achieving this mode integration the high requirements in materials and device design are apparent. The electrolyte-gated indium tin oxide (ITO) vertical transistor with its own ion gate where ions respond to external bias slowly, has been experimentally described as a biomimetic nociceptor device [6], whereas its single-mode operation makes threshold-saltation characteristics unclear. As a result, the definition of behavior threshold does not depend on the device’s own electrical coupling mechanisms but is estimated in a continuous trend, which is an insufficiently accurate expression of the nature of pain. Chen et al. coupled the electrolyte-gated synaptic transistors with the electrochemical doping [7], where the bi-mode operation was achieved by tuning the input pulses to control ion doping. Unfortunately, analog pain signals only triggered by pulse bias apparently lead to the limitation of flexible programmable capability, for example, continuous stimulation cannot provide the triggered possibility. Other nociceptor analogs are mainly filament-based two-terminal memristors [8–11], but their issues—such as unstable switching and excessive write noise—need to be addressed. Therefore, achieving such a nociceptor analog that integrates bi-mode synaptic operation and presents its self-adaptive transition remains a challenge. In addition, pain neuristors also require flexible control of the device’s own parameters, to express its diversity in various intended purposes/uses.
The lateral back-to-back Schottky junctions (B-B SJs) based on spontaneous trappings localized at interfacial have emerged as an easily-fabricated device platform capable of hosting completely programmable neuromorphic functions, through trapping-modulated Schottky barriers [12,13]. In our work, we project to use the B-B SJ to design a nociceptor concept where there is a key technical challenge how the high-threshold synaptic operation can adaptively be embedded into a low-threshold synaptic expression. A major strategy was used at the device level; reasonably, by means of an tunneling layer of inserted into the interface it can control the electron passing mode, the low- and high-pass modes, therefore enabling two-stage memristive manipulation. The resulting device can not only operate the common synaptic functions under no-injured conditions, such as EPSC, facilitation, dynamic pair-pulse facilitation (PPF), and consolidation, with an extremely low energy density of 33.5 fJ/, but also present both intensity- and rehearsal-triggered adaptive mode jump that resembles the biological alarm system. In addition, we used the individual capacitor based on to demonstrate the mechanism of reversible transition of tunneling modes that is thought to be a physical fundamental to hold the reconfigurable bi-mode synaptic functions. Furthermore, the , as a tunneling layer, can modulate tunneling properties by its own thickness, so that it is relatively easy to fabricate the device with differing pain degree. Therefore, the tunneling mode transition in a tunneling layer can provide an efficient and low-cost technique for building an analog pain alarm network.
Sign up for Photonics Research TOC Get the latest issue of Advanced Photonics delivered right to you!Sign up now
2. RESULTS AND DISCUSSION
A. B-B SJ Structure, Mechanisms, and Fundamental Photoelectrical Properties
The Schottky junction (SJ) devices [14,15], benefiting from fast switching speed inherent to low charge storage effect, high sensitivity, and easy fabrication, are used as an effective solution for switching in the field of photoelectronics in general. In previous work, we have developed an optical memristive application based on a planar B-B SJ that naturally provides substantial advantages in sensitive areas compared to vertical structures [13]. In this device, fast resistive switching and its retention after switching rely on synergistical optical excitation and oxide trapping modulation of the Schottky barrier. The B-B SJ structure and working mechanism are shown in Fig. 1(a), where the device structure is surprisingly simple, with only one semiconductor situated above the as a functional component. To enhance the trapping-enabled memristive mechanism, high-mobility bilayer 2D molecule crystals [dioctylbenzothienobenzothiophene ()] with a thickness of only 5.9 nm served as the active channel and excited layer for separated electrons to easily be localized at the interface (see Section 4). These trapping events result in the formation of long-lifetime free holes that will diffuse to the electrode side under the built-in electric field () of the Schottky depletion zone, which determines the barrier height and hence resistive state of the device.
Figure 1.Structure, working mechanism, photoelectrical properties, and behaviors of the B-B SJ-based synapse. (a) Schematic of the device structure and physical mechanism. (b) The device’s photoelectrical properties. The left is optical resistance switching in scanning, and the right is the changing of Schottky barrier in dark and under illumination. (c) The programmable weight states triggered by relearning.
Our B-B SJ device shows a transition in operation from a high-resistance state (HRS) in dark with a rectifying Schottky contact, to a low-resistance state (LRS) under illumination due to the decrease of barrier, as shown by the dark current and photocurrent in Fig. 1(b) (left panel). The combination of the low dark current and large gain of photocurrent is expected to modulate a low-consumption device. Note that the photocurrent traces after illumination feature a slowly fading course rather than direct recovery to the initial dark current state, which demonstrates the non-volatile switching nature. Furthermore, optical modulation in the Schottky barrier () has been investigated by the temperature-dependent measurements that provide the temperature-dependent data to plot Arrhenius curves [13]. The intrinsic Schottky barrier (), a at 0 bias, can then be extrapolated in the linear function of and the applied bias. Resulting under the illumination is -fold less than that in dark [see Fig. 1(b), right panel], which demonstrates the switching mechanism of modulation of contact barrier heights. Based on this device the mostly useful plasticity of short-to-long-term plasticity (STP-to-LTP) has been demonstrated where the relearning can not only enhance the weight connection but also prolong the modulation of conductance (forgetting), as shown in Fig. 1(c). Therefore, this B-B SJ can be reasonably viewed as a superior device-level platform for integrating specific neural functions.
B. Nociceptor Analogs and Their Structure and Mechanisms
In the biological nervous system, nociceptor-mediated neural signals can be managed via various receptors, gated channels, and key chemical active molecules, located at the synaptic junction. For example, in the case of tissue damage, chemicals will be released by the tissue and can specifically bind corresponding high-threshold receptors at fine nerve fibers that fire the gated channel of specific neuropeptides (such as substance P, neurokinin A, and CGRP) which have long-lasting effects [16,17]. With the influx of neuropeptides in cleft, the presynaptic membrane will be steadily depolarized as a consequence of great binding between neuropeptides and its receptors. Then and enter the pre-synapse increasing the release of neurotransmitters that can produce high enough EPSC to generate pain in the cerebral cortex. In normal perceptual activities, high-threshold receptors produce no effect and the neuropeptide channel is blocked. Only low-threshold receptors are fired and, as a result, a smaller number of neurotransmitters are released over a shorter timescale compared to the nociceptor-mediated case.
At the analog device, we use the simple B-B SJ structure inserting an electron tunneling layer of , and its structure and operation principle are presented in Fig. 2. This is essentially similar to an adaptive switch that controls the electron flux through and then is localized at the interface. When input pulses are applied with a low intensity that is viewed as non-nociceptive stimuli, excited electrons in the channel are so limited that only few electrons tunnel through to be trapped at the interface, in which most of the electrons still end in the recombination with holes. This low-pass course will produce a small photogating electrical field () that is formed by localized states at the interface, to modulate the channel. The weak can induce device results at the low-threshold stage that have further been divided into the following three aspects: (i) the conductance of the channel modulated by photogating bias is low, resulting in the low output current; (ii) inefficient dissociation of excitons further accelerates their recombination that can also result in low output current; (iii) ambiguous memristive properties inevitably lead to the low plasticity features. When programmed by sharp pulse patterns, devices can adaptively switch to a distinguishing high-threshold stage, and this is due to the high-pass mode of . To explain this, the structure stack is roughly viewed as a capacitor, with charged oxide and semiconductor corresponding to negative and positive. Once accumulated in this capacitor continuously increases until a threshold is reached, and the transition of low- and high-pass modes can occur. In this case, in channel the ratio of long-lifetime free hole will increase sharply as large numbers of electrons go through and then are trapped by , which will lead to the abrupt increase of output current and programming efficiency. In these processes, residual holes with a long lifetime in channel after trapping electrons are analogous to lasting-working neuropeptides in biological nerve actions; subsequently, these holes diffuse into the Schottky junction and decrease its barrier height, akin to the case of neuropeptide-induced depolarization of presynaptic membrane. The external charges undergo a course from injection to collection that is similar to the concept of the release and reception of neurotransmitters. Hence, our B-B SJ with an electron tunneling layer can be suggested as a similar configuration of nociceptor, which can be used not only as a non-nociceptive perception synapse but also for producing and transmitting nociceptive information.
Figure 2.Receptors, gated channels, and key chemical active molecules synergistically control the generation and transfer of pain signals in a biological nociceptor (left). The operating mechanism of our analog nociceptor based on the B-B SJ structure (right).
C. Film Characteristics and Its Electrical Properties
In our analog device, the high-quality bilayer 2D molecular crystals were prepared on a heterogeneous-insulator substrate (80 nm ) via a solution process of “floating-coffee-ring-driven assembly” [18], to act as a semiconductor. Its good hole-transport capabilities and close inherent interface coupling at 2D limits can enable device outstanding light responsivity () and specific detectivity, therefore decreasing the energy consumption. To characterize the two-dimensional features of films, the surface morphology and crystalline properties were measured using an atomic force microscope (AFM) and a polarizing microscope. In Fig. 3(a), the optical image shows the uniform large-sized bilayer with clear layer definition (at the left panel). The AFM picture not only demonstrates the layered nature of grown films but also precisely measures the surface roughness and height of each step. The results show each step height of approximately 2 nm (molecule length of ), with the atom-level roughness of 0.342 nm, which indicates the obliquely stacking molecule structure inside films. The cross-polarized optical images further confirm such orderly stacking structure through the uniformly changing brightness of films with sample angles [Fig. 3(b)]. Therefore, these results illustrate the nature of 2D molecular single crystals of grown films. The scanning was applied to demonstrate the working mode of our B-B SJ, where the photodiode operation in the dark is verified by a rectifying Schottky contact with a negative reverse bias, as shown in Fig. 3(c). When an light is applied, the photocurrent at is boosted by around 1800-fold while the photocurrent trace goes to symmetry that indicates the quasi-ohmic contact resulting from the decreasing Schottky barrier. The noise current (i.e., dark current) is as low as 8 pA at read voltage because of the blocking contact at the electrode/semiconductor interface. Hence, combination of the low noise current and large photocurrent gain is expected to yield the high optical sensitivity in our device that is a key factor for the high energy efficiency.
Figure 3.Film characteristics and device properties. (a) The optical microscope (left panel) and AFM (right panel) images. (b) The cross-polarized optical images at different sample angles. The viewing area shows the uniformly changing brightness. (c) The curve of the analog device and the noise curve of the working device.
D. Non-nociceptive and Nociceptive Properties and Features
To demonstrate the adaptive jumping between the low- and high-threshold stages, a set of single pulses with differing light power () was applied, as shown in Fig. 4(a). At a power section of a marked jumping is observed as expected, which can be explained by an adaptive transition of tunneling modes of (discussed below). Reasonably, the EPSC with this condition is thought to have a specific threshold to manage stimuli and when this is excessed, the device presents a nociceptor-mediated high-threshold stage; otherwise, it is defined as a non-nociceptive low-threshold stage. Therefore, at the device level the behaviors are indeed consistent with the concept of a nociceptor. Subsequently, a series of low-threshold synaptic functions was operated with an extremely low non-nociceptive stimulation of , including EPSC, PPF, and dynamic PPF with intervals. In Fig. 4(b), it is observed that the single-pulse EPSC out features a normal spike characteristic corresponding to both switch from HRS to LRS and decay from LRS to HRS. Such a spike pattern indicates a CIM mode where the stored history in real time modulates outputs. The device’s energy consumption can commonly be evaluated by a single-pulse event, using the following formula: where , , , , and () are the response current, spike duration time, read voltage, light power, and effective device area, respectively. The resulting consumption is about 13.4 pJ with an extremely low energy density of 33.5 fJ/. In a biological system, at some synapses with repeated train, the facilitatory process dominates in which the second EPSC is higher than that of the first and this is highly dependent on the details of the timing of synaptic activation [19]. The amplitude of facilitation always decreases as the interpulse interval () is increased. Mathematically, the PPF ratio is defined as , where and correspond to the second and first spike currents, which can be approximated by a double exponential decay of the form where and correspond to the characteristic timescales of rapid phase (F1) and slow phase (F2). In our device, the facilitation was also easy to establish because of the inherent manipulation of the memristive mechanism, and its values under differing intervals are pointed out in Fig. 4(c). Taking the double exponential fit for interval-modulated trace, we find high consistency, further obtaining two timescales of 16.7 and 840 ms that are similar to that of a biological system. In addition, the short-to-long-term plasticity that features the prolonging state-retention time, was observed in a state-enhanced course that is triggered by more repetitive stimulations, as shown in Fig. 4(d). This extension of decay through repetitive learning resembles a key feature of consolidation in a biological synapse [20]. Amazingly, when the continuous pulses were applied with a specific input pattern, the system would generate a clear mode jumping with a threshold of approximately 20.5 pA, as shown in Fig. 4(e). Such a jumping was also observed in the power-modulated spike characteristic, and we attributed it to the same intrinsic mechanism of enhancing . The enhancing rate, which is demonstrated by the ratio of weight change (), in the high-threshold stage, was faster than that in low-threshold stage, which correlates well to the pain alarm in biology. In Fig. 4(e), we also show the reconfigurable pain signals where conditional decay of post-pain current (decreases below threshold) transforms high-threshold to low-threshold stage, and then, an almost same pain process again develops with the number of input pulses. It indicates the inherently no structural damage of such over-threshold jumping, which resembles that the pain alarm is suggested as a safety protection for living organisms.
Figure 4.Related analog behaviors in our device. (a) The intensity-triggered EPSC jump, enabling the device to exhibit two operation stages of low- and high-threshold stages. (b) The single-pulse EPSC and double-pulse facilitation. The PPF ratio is defined as . (c) The dynamic PPF depended on intervals. The PPF ratio is fitted by a double-exponential form. (d) The EPSC programmed by different pulse numbers. The post-stimulation current decay is prolonged with the increasing pulse number. (e) The rehearsal-triggered EPSC jump. The weight changing is defined as . (b), (c), and (d) are low-threshold synaptic functions.
The schematic energy band diagram of electron tunneling is shown in Fig. 5(a). The electron tunneling course is governed by the tunneling barrier () at the interface of , which undergoes the natural decrease at high field. Under low-intensity stimulation, the electron through the is very limited due to few excited electrons and high , which leads to a very low field of . With the increase of stimulation intensity, more electrons will accumulate at the interface, and this is accompanied by the increasing , which can inherently decrease the height and width of the barrier. Finally, until the decreases to a certain level the tunneling mode will be changed from the low-pass to high-pass mode. In addition, the electron tunneling features in the heterogeneous stack of that works fundamentally as a plate capacitor mentioned above, can explain the origin of the threshold-based jumping of operation modes of our synaptic device; the schematic is shown in Fig. 5(b) (at the left panel). Continuously accumulated is equivalent to an applied scanning field on . To demonstrate the tunneling nature of 80 nm we fabricated an individual capacitor based on this to repetitively measure the conduction current with the scanning field; the results are shown in Fig. 5(b). They present some clear properties, and it is easy to relate these to our device behaviors. Here, we draw the following conclusions from our experiments. (i) The tunneling (leakage) current roughly undergoes three stages before structural breakdown, including the low-pass stage at the low field, reversible transition, and the high-pass stage at the high field. Such an increase of leakage with field is due to the decrease of barrier width and height and is a common phenomenon in all dielectric materials [21]. The field-manipulated transition from low-pass to high-pass modes makes the sharply increasing electron tunneling rate inherent to result in a higher current output. Hence, in our device, the , equivalent to an external scanning field, triggers a jump from the low- to high-threshold stage as accumulating to a threshold, physically originating from the mode transition of electron tunneling. (ii) The field scan was applied six times with completely same measurement conditions and all results maintain high consistency. As such, this jumping process is reversible, meaning no damage transition that is a physical fundamental for a reconfigurable nociceptor. In Fig. 5(c), as the field persists to increase after the end of reversible transition, the structural breakdown occurs, resembling the local destruction of living organisms. In a biological system, the specific nociceptors operate to produce various degrees of pain with site of changes, and therefore require analogs flexible tunability. As mentioned above, the operating features of a nociceptor are highly related with the resistance to tunneling of where the thicker theoretically yields a higher triggering threshold on the same device. We therefore designed five sets of devices that only differ in thickness (10, 30, 50, 80, and 90 nm), to study the thickness-dependent pain triggering. There is a pain gain, defined as a ratio of the first spike after threshold () and the last spike before threshold (), to be used as the analog pain degree, which can be estimated from their rehearsal-triggered pain tests. In Fig. 5(d), under the same test conditions the pain gain presents a remarkable thickness dependence, where the thinner yields the higher gain, resembling the pain escalation. In addition, our nociceptor analogs also show a very desired advantage of air-operated robustness, and it is worth mentioning that all tests in that work are carried out in ambient air. The detailed mechanism discussion is included in our previous work [22].
Figure 5.Related behaviors in the individual capacitor. (a) The schematic energy band diagram of electron tunneling at stacked . (b) The schematic of the tunneling mechanism and the conduction current density under the field scanning. (c) The breakdown curve of the capacitor. (d) The tunable pain gain in differing thickness.
We have reported a new type of optical nociceptor analog application that takes advantage of the Schottky barrier to enable its efficient resistance switching, and that uses a tunnel silicon nitride for bi-mode synaptic functions (i.e., low- and high-threshold synapses). At the low-threshold stage, we determined experimentally that our nociceptor mimics some common functions, including EPSC, facilitation, PPF, and consolidation, with an extremely low energy density of 33.5 fJ/. Most importantly, our device shows both intensity- and rehearsal-triggered adaptive mode jump that resembles the biological alarm system. Through the field scanning tests two tunneling modes of low- and high-pass were demonstrated in individual capacitors. With the increasing field the reversible transition from low-pass to high-pass naturally occurred, which is thought to be a physical fundamental to hold the reconfigurable bi-mode synaptic functions. The tunneling mode transition in the tunneling layer opens up a highlighted path to threshold-related novel applications, such as nociceptors, threshold-switching memristors, and multi-level alarm systems.
4. EXPERIMENTAL SECTION/METHODS
Growth of ultrathin molecular crystals of for synaptic device. A piece of Si substrate (roughly ) successively grown 100-nm-thick layer of and 80-nm-thick layer of was cleaned by a standard cleaning process. Then, molecular layer-defined molecular crystals were easily prepared on this substrate via the floating-coffee-ring-driven assembly. The detailed growth procedure involved the following steps. The growing solvent was first allocated by mixing anisole and p-anisaldehyde (0.5%, mass fraction). The was then dissolved in this mixed solvent to obtain a growing solution. Subsequently, 3 μL solution was dropped onto the pre-cleaned substrate and then a pump was used to vent the air. The resulting high-quality molecular crystals were grown along the pulling track of the liquid drop.
Device fabrication. The film was prepared on a substrate; subsequently, two Au electrodes with an area of were transferred onto its surface by “stamping” Au stripes with the channel width and length of and 10 μm, respectively. Notably, the two Au electrodes should be attached onto the same crystal domain, ensuring the absence of defects from step lines.