Exploiting Nanoscale Low-Power Flexible Memristor Based on Carbon Quantum Dots for Neuromorphic Computing and Pattern Recognition

Wearable non-volatile memory based on flexible materials has shown an unstoppable development trend in the field of portable artificial intelligence technology. In this work, nanoscale flexible memristors based on carbon quantum dots (CQDs) prepared by an electrochemical ablation process are exploited. Importantly, benefit from the excellent flexibility and stability of the polyethylene terephthalate (PET) substrate, the device can still exhibit resistance switching characteristics similar to the initial state after being subjected to multiple destructive folding, such as current-voltage curve (I-V), long-term potentiation/long-term depression (LTP/LTD). The threshold powers of the devices are as low as 10-7 W, demonstrating the potential of low-power devices. In addition, the classical synaptic behaviors such as spike-timing-dependent plasticity (STDP), paired-pulse facilitation (PPF), and transition from short-term memory (STM) to long-term memory (LTM) can be mimiced by devices, which operating much faster (ms level) than the human brain. Finally, the artificial neural network model based on the Crosimm platform is used to iteratively train and recognize MNIST handwriting, and the results are all between 94.2% and 94.9%, which further verifies the reliability of Ag/CQDs/ITO/PET based devices. This work provides a new solution for the development of a new generation of flexible memory devices.