Artificial Synapse based on MoS2 Memtransistor for In-Memory Computing

摘要

We demonstrate a large-scale three-terminal MoS2 memtransistor crossbar array featuring self-selective function via independent gate control which effectively suppress the sneak current to be less than 0.1 nA, thus eliminating the cross talk issues of a typical 1-R array. The oriented grain boundary of ‘poly-Mo'S, along the channel enables a low switching voltage of 0.42 V and a switching energy per transition of 20 fl/bit, Such architecture is capable of performing vector matrix multiplication (VMM) operation, a core computing task for pattern classification, wherein the linear synaptic plasticity and low wire resistance lead to a high MNIST recognition accuracy of 96.87 %.