The prevalent von Neumann architecture in today’s processors involves memory and processing units to reside in physically separate locations. With memory speeds lagging behind the processor speeds, latency in accessing data from memory has resulted in the “von Neumann bottleneck”. To alleviate this issue, several alternative non-von Neumann architectures have been explored. Neuromorphic computing is one such non-von Neumann approach, inspired by the human brain’s ability of cognitive recognition. The brain operates through a network of neurons that are connected to each other by synapses. In 2008, after the discovery of memristors, the fourth circuit element, researchers have explored the idea of mimicking synaptic behavior with a single memristive device.^1-6 Albeit significant advances, synaptic devices using phase change materials-based memristors^3,7-9 and metal oxide-based resistive switching devices^5,6,10-13 have some limitations. These devices exhibit high programming current in the range of µA to mA. Additionally, the synaptic weight update, i.e. the increase (decrease) of the synapse’s conductance with the application of a continuous stream of identical positive (negative) input voltage pulses, is non-linear. Non-linearity in weight update increases the complexity of using these devices for real-time unsupervised learning. So, it becomes necessary to employ a materials system which exhibits low programming current as well as a linear weight update.¹⁴

Recently, two-dimensional (2D) materials are being largely explored to demonstrate their viability as electronic synapses and neurons.^15-21 In this talk, we shall discuss the realization of a synaptic device using graphene/MoS₂ heterostructures. In these devices, CVD-grown monolayer graphene acts as an electrode to CVD MoS₂. These memristive devices exhibit low programming currents and a high dynamic range from 1 nA to 1 mA. In contrast with oxide-based or PCM-based synapses, these devices exhibit a gradual set and reset process when symmetric input voltage pulses are applied, resulting in a near-linear weight update. We shall also present the demonstration of an integrate-and-fire (IF) neuron using Ag/MoS₂/Au vertical structures. These devices possess the four crucial features of an IF neuron – all-or-nothing spiking, threshold-driven firing, post-firing refractory period and stimulus strength based frequency response. Realizing neurons and synapses using the same materials system allows the monolithic integration of the essential building blocks of neuromorphic hardware, and bears potential for a highly scalable spiking neural networks suitable for unsupervised learning applications.

References:

1 Chua, L. IEEE Trans. Circuit Theory 18, 507-519 (1971).

2 Jo, S. H. et al. Nano Lett. 10, 1297-1301 (2010).

3 Kuzum, D. et al. Nano Lett. 12, 2179-2186 (2011).

4 Strukov, D. B. et al. Nature 453, 80 (2008).

5 Yu, S. et al. Adv. Mater. 25, 1774-1779 (2013).

6 Yu, S. et al. IEEE Trans. Electron Devices 58, 2729-2737 (2011).

7 Jackson, B. L. et al. ACM Journal on Emerging Technologies in Computing Systems (JETC) 9, 12 (2013).

8 Li, Y. et al. Sci. Rep. 3, 1619 (2013).

9 Suri, M. et al. in Electron Devices Meeting (IEDM), 2011 IEEE International. 4.4. 1-4.4. 4 (IEEE).

10 Lee, S. R. et al. in VLSI Technology (VLSIT), 2012 Symposium on. 71-72 (IEEE).

11 Prakash, A. et al. IEEE Electron Device Lett 36, 32-34 (2015).

12 Zhang, L. et al. IEEE Electron Device Lett. 31, 966-968 (2010).

13 Chen, P.-Y. et al. in Proceedings of the IEEE/ACM International Conference on Computer-Aided Design. 194-199 (IEEE Press).

14 Yu, S. Proc. IEEE 106, 260-285 (2018).

15 Arnold, A. J. et al. ACS Nano 11, 3110-3118, (2017).

16 Jiang, J. et al. Small 13, 1700933, (2017).

17 Sangwan, V. K. et al. Nature 554, (2018).

18 Zhao, H. et al. Advanced Materials, 1703232-n/a, doi:10.1002/adma.201703232.

19 Shi, Y. et al. Nature Electronics 1, 458-465, (2018).

20 Kalita, H. et al. Scientific Reports 9, 53, (2019).

21 Kalita, H. et al. in 2018 76th Device Research Conference (DRC). 1-2.