Multi-terminal memtransistors from polycrystalline monolayer molybdenum disulfide

• Published in: Nature (London) Vol. 554; no. 7693; pp. 500 - 504
• Main Authors: Sangwan, Vinod K., Lee, Hong-Sub, Bergeron, Hadallia, Balla, Itamar, Beck, Megan E., Chen, Kan-Sheng, Hersam, Mark C.
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 22.02.2018; Nature Publishing Group
• Subjects: 639/166/987; 639/925/357/1018; 639/925/927/1007; Brain; Charge transport; Chemical vapor deposition; Computer memory; Data storage; Fabrication; Field effect transistors; Humanities and Social Sciences; Learning; letter; Long-term potentiation; Memory; Memristors; Microscopy; Molybdenum; Molybdenum compounds; Molybdenum disulfide; Monolayers; multidisciplinary; Neurons; Polycrystals; Production processes; Resistance; Scanning probe microscopy; Science; Semiconductor devices; Switching; Synapses; Transistors; Transport buildings, stations and terminals; Voltage pulses
• ISSN: 0028-0836; 1476-4687
• DOI: 10.1038/nature25747

Polycrystalline monolayer molybdenum disulfide is used to fabricate a multi-terminal device combining a memristor and a transistor, which can mimic biological neurons with multiple synapses for neuromorphic computing applications. Memtransistor mimics multiple synapses Memristors are two-terminal devices whose resistance exhibits a memory effect that depends on the current or voltage history. This memory enables such devices to mimic the behaviour of a neural synapse, making them of great interest for creating brain-inspired neuromorphic computing architectures. Basic neural functions have been demonstrated with two-terminal devices, but more complex functions, such as heterosynaptic plasticity, will probably require devices with multiple terminals. Mark Hersam and colleagues combine the restive switching behaviour of a memristor with the gate-tunability of a transistor into one multi-terminal device called a memtransistor. Based on two-dimensional layers of molybdenum disulfide, such memtransistors not only exhibit conventional neural learning behaviour but also heterosynaptic functionality, providing a platform for mimicking biological neurons with multiple synapses. Memristors are two-terminal passive circuit elements that have been developed for use in non-volatile resistive random-access memory and may also be useful in neuromorphic computing 1 , 2 , 3 , 4 , 5 , 6 . Memristors have higher endurance and faster read/write times than flash memory 4 , 7 , 8 and can provide multi-bit data storage. However, although two-terminal memristors have demonstrated capacity for basic neural functions, synapses in the human brain outnumber neurons by more than a thousandfold, which implies that multi-terminal memristors are needed to perform complex functions such as heterosynaptic plasticity 3 , 9 , 10 , 11 , 12 , 13 . Previous attempts to move beyond two-terminal memristors, such as the three-terminal Widrow–Hoff memristor 14 and field-effect transistors with nanoionic gates 15 or floating gates 16 , did not achieve memristive switching in the transistor 17 . Here we report the experimental realization of a multi-terminal hybrid memristor and transistor (that is, a memtransistor) using polycrystalline monolayer molybdenum disulfide (MoS 2 ) in a scalable fabrication process. The two-dimensional MoS 2 memtransistors show gate tunability in individual resistance states by four orders of magnitude, as well as large switching ratios, high cycling endurance and long-term retention of states. In addition to conventional neural learning behaviour of long-term potentiation/depression, six-terminal MoS 2 memtransistors have gate-tunable heterosynaptic functionality, which is not achievable using two-terminal memristors. For example, the conductance between a pair of floating electrodes (pre- and post-synaptic neurons) is varied by a factor of about ten by applying voltage pulses to modulatory terminals. In situ scanning probe microscopy, cryogenic charge transport measurements and device modelling reveal that the bias-induced motion of MoS 2 defects drives resistive switching by dynamically varying Schottky barrier heights. Overall, the seamless integration of a memristor and transistor into one multi-terminal device could enable complex neuromorphic learning and the study of the physics of defect kinetics in two-dimensional materials 18 , 19 , 20 , 21 , 22 .