Heterosynaptic MoSe₂ memtransistor array with ultra-low operating voltage and linear plasticity for neuromorphic computing

In the context of artificial intelligence, the development of artificial neural networks that emulate the efficient information processing capabilities of the human brain is of great importance. Synaptic devices are the integral components in achieving this goal. Current synaptic devices based on memristors usually suffer from low linearity/symmetry and/or high operating voltages, which pose significant challenges to the advancement of next-generation neuromorphic computing. This study presents a heterosynaptic MoSe₂ memtransistor with ultra-low operating voltage and high linearity/symmetry. This device incorporates heterogeneous ions into the MoSe₂ channel through gold-assisted exfoliation process, resulting in a substantial reduction in both turn-on voltage to 5 mV and power consumption to 10 fW. Additionally, the addition of gate electrode as an extra modulation terminal enhances the tunability of the device weight with greatly improved linearity and symmetry. The calculated asymmetric ratio of the weight modulation is as low as 0.058, approaching the theoretical limit of 0. We finally fabricated a MoSe₂ memtransistor array consisting of 16 devices, which is employed for the Modified National Institute of Standards and Technology fashion image recognition and Sandia document types, resulting in a significant enhancement in accuracy of around 6 % as compared to the homosynaptic counterparts.