Mechano-Driven Logic-in-Memory with Neuromorphic Triboelectric Charge-Trapping Transistor

One of the trends in the era of post-Moore's Law is to develop sophisticated electronic devices that integrate the function of sensation/data-storage/computation for diverse applications. Two-dimensional semiconductor transistor based on charge storage mechanism is a promising alternative for future information devices. Here, we propose a neuromorphic triboelectric charge-trapping MoTe2 transistor with stacked high-k dielectric structure, aiming at mechano-driven logic-in-memory for neuromorphic computation. Gating by triboelectric potential, the neuromorphic device exhibits excellent electrical performance, including a high switching ratio (>105), low off-state current (~0.6 pA), and robust cyclic stability. By controlling the trapped charges in the stack gate structure via tribopotential modulation, the conductivity state of the MoTe2 channel can be readily set, realizing an excellent mechano-driven nonvolatile memory with a retention time up to 104 seconds, stable switching behavior over 100 cycles, and 8 levels of multi-level data storage capability. A mechano-driven programmable inverter can also be realized by serially connecting a load resistor. Besides, the triboelectric charge-trapping transistor is ready to emulate typical synaptic characteristics in low energy levels (~147 fJ). Based on the finely tunable conductivity through tribopotential, a mechano-assisted artificial neural network is demonstrated to recognize handwritten digits with an accuracy of approximately 88.59%. These results highlight the great significance of the triboelectric charge-trapping transistor in mechanical-assisted real-time interaction, low-energy data storage, and neuromorphic computation.