Thermal-driven neuromorphic computing devices

Neuromorphic computing devices which emulate biological neural networks are vital in realizing artificial intelligence for information processing and decision-making. To date, various types of neuromorphic computing devices with multi-level resistance states have been developed, including the memristor-based devices caused by ion diffusion, the phase transition-based devices caused by threshold switching and progressive crystallization/amorphization, the spintronics-based devices caused by magnetic domain switching, etc. However, these devices may face critical challenges such as instability in integrated circuits and non-linearity in synaptic weight change. To overcome these difficulties, we present a thermal-driven multi-layer-multi-terminal neuromorphic computing device with a performance of high endurance, high reliability, and high linearity. We use a pulse current sequence to variate resistance by the asymmetric Joule heating in this device and demonstrate a wide range of synaptic functions, including potentiation, depression, and both anti-symmetric and symmetric spike-timing-dependent plasticity. Furthermore, our thermal-driven asymmetric Joule heating devices with multiple functionalities open an energy-efficient platform for future neuromorphic computing devices and artificial intelligence.