Compute-in-sensor neuromorphic devices allow in situ weight iteration and computation within hardware, significantly reducing power consumption and latency caused by data transmission. These devices are essential building blocks for developing adaptive perception and interaction systems in the next generation of robotics. Pain, as the first line of defense against potential harm, serves as an alarm triggered by external noxious stimuli, initiating nociceptive responses to prevent further damage. However, there remain considerable challenges in stimulating nociceptor-mediated synaptic behavior. For instance, current nociceptor devices primarily simulate pain generation but cannot accurately classify pain levels, which hampers the development of adaptive pain alarm systems. Through the rational design of device structures, we have developed a multilevel nociceptor with several advantages, including ease of fabrication, reconfigurability, and clear threshold behavior. The device utilizes two tunneling-mode jumps in a 100 nm Si₃N₄ layer to modulate two synaptic mode transitions, akin to a two-level pain alarm system. These threshold-managed transitions are triggered by high-intensity and repetitive stimuli. In this paper, we characterize subthreshold normal synaptic perception behaviors and suprathreshold two-level nociceptor-mediated synaptic operations in response to varying stimulus intensities and continuous stimulation.

Our device is based on a back-to-back Schottky junction synaptic structure with an inserted 100 nm Si₃N₄ tunneling layer. The preparation process involves the deposition of a C₈-BTBT film on a Si₃N₄/SiO₂/Si substrate, followed by the transfer of two Au electrodes (each 50 μm×40 μm onto its surface. The “stamping” method is used to position the Au stripes, with channel widths and lengths of approximately 40 and 20 μm, respectively. Both Au electrodes must attach to the same crystal domain to avoid defects from step lines. The device operates through the management of three distinct tunneling modes, which control the synaptic behaviors representing normal and two-level nociceptor-mediated synaptic responses. By adjusting the stimulation patterns, we enable adaptive, threshold-managed mode transitions.

The multilevel nociceptor demonstrates distinct threshold-managed features, with subthreshold normal and suprathreshold two-level nociceptor-mediated synaptic behaviors. At lower stimulus intensities (6.2 μW/cm2), the device exhibits subthreshold synaptic behaviors, such as excitatory post-synaptic currents (EPSC) and paired-pulse facilitation (PPF). Key short-term plasticity behaviors, including interval-dependent PPF, are successfully mimicked, with characteristic times of 41 and 571 ms. As the stimulus intensity increases between 35 and 173 μW/cm2, the device enters the level-I pain mode. At higher intensities, ranging from 285 to 577 μW/cm2, the device transitions to the level-II pain mode. In addition, the device demonstrates reconfigurability and stability, which we attribute to the tunneling properties of the 100 nm Si₃N₄ layer. We build a capacitor based on this tunneling layer and study its properties by measuring conduction current under a scanning field. The results show three distinct tunneling modes at low fields, with two mode jumps, consistent with the observed two-level pain modes. Moreover, the device exhibits low energy consumption of approximately 22.5 fJ/μm2 per single spike.

We propose an optical nociceptor capable of stimulating a two-level pain perception mechanism. First, the Schottky barrier is modulated by light, enabling efficient resistance switching. Second, the Si₃N₄ tunneling layer controls electron trapping events at the SiO₂ interface, where memristive modes are modulated by tunneling behaviors. Experimentally, our nociceptor has been shown to replicate a range of normal synaptic functions under low-intensity stimulation, including EPSC, PPF, interval-dependent dynamic PPF, and simulated synaptic consolidation. More importantly, under increased stimulus intensity and duration, the device demonstrates clear two-level adaptive mode jumps, closely resembling a two-level pain alarm. This mode jumping is reversible, providing a physical foundation for a reconfigurable two-level pain alarm system. The tunneling modulation approach used in this device offers a promising and efficient method for constructing adaptive multilevel pain alarm systems.