2, Harvard University, Cambridge, Massachusetts, United States
Neuromorphic computing systems using electronic synapses and neurons have been extensively studied in recent years due to its ability to perform analog computing for artificial intelligence. It is also expected to overcome the high energy and various limitations of today’s computing systems. Most of all analog switching device based on the two-terminal memristor, such as conductive bridging random access memory (CBRAM) and oxide based resistive random access memory (RRAM). The memristors are operated by movement of anion vacancies or forming of metal filament, which exhibit relatively good retention and endurance for neuromorphic circuits. However, since the anion vacancies or metal ions must move within the oxide by an electric field, these devices necessarily require high operating voltages of several V or more even few-nm switching oxide. The high operating voltage causes problems such as reduced overall device stability and increased device operation energy.
On the other hand, TMDC (transition metal dichacogenide), a kind of two dimensional (2D) material, has been recently developed with great interest in growth method, process technology, and so on. Especially, the TMDC growth method using the MOCVD technique provides a poly-crystalline layered film from mono-layer to few-layers with layer controllability. In addition, the transfer method using the polymer support has been intensively studied and reached to a level at which almost no defect occurs in the process.
In this presentation, we presented a memristor device using these TMDC materials and show the electrical characteristics of memristor applications. The TMDC materials effectively blocks metal ion transport due to the high bonding energy and the close packing structure of atoms within the single layer. However, the metal ions can be transported through defect sites such as grain boundaries and vacancies, thus enabling the operation of the TMDC memristor. In particular, the bi-layer TMDC, in which the defect site is effectively controlled through transfer process, has the advantage of significantly improving the device-to-device uniformity and endurance. This is because the transfer process with rotation converts most of the defect sites to point defects. In contrast, the mono-layer TMDC has high possibility of line defect such as grain boundary, so that the uniformity and endurance of memristor are not secured.
Finally, we fabricated the memristors of metal / 2L-MoS2 / metal structure using MOCVD technique and rotated transfer method. These devices show bipolar memristor behavior with a lower set/reset voltage of +0.5 V/-0.47 V than the conventional oxide based memristor. The resistance changes of the device from tens of to hundreds of ohms-um^2 is found to occur mainly in 2D materials through external voltage measurement. It also shows the possibility of analog memory whose resistance varies linearly with bias applied from 0.3 V. Also, we confirmed STDP (spike-timing-dependent plasticity) operation in a single device, and tested various applications using a cross-bar structure. This work is an important foundation for developing devices physics to understand the memristive characteristics of 2D material-based memristor and providing neuromorphic computing systems based on such memristors.