Available on-demand - F.NM07.08.06
2D Memtransistors and Gaussian Heterojunction Transistors as Artificial Synapses and Spiking Neurons
Vinod Sangwan1,Hong-Sub Lee1,Megan Beck1,Mark Hersam1
Northwestern University1
Show Abstract
Hardware implementation for neuromorphic computing aims to overcome the von Neumann bottleneck by co-locating memory and logic. In this context, resistive switches, phase-change devices, electrochemical systems, and complementary metal-oxide semiconductor (CMOS) transistors have been explored extensively for artificial synapses and spiking neurons in artificial neural networks. While each of these strategies has its attributes, none of them has clearly emerged as a universal approach for neuromorphic hardware, which suggests that alternatives should continue to be evaluated. For example, neuromorphic devices based on nanomaterials such as quantum dots, nanotubes, and two-dimensional (2D) monolayers have begun to show promise due to their controlled chemical composition, unique architectures, and compatibility with ultimate scaling and speed limits [1]. Furthermore, low-dimensional materials possess low dielectric screening that provides enhanced electrostatic control, thus allowing functions that better mimic a biological neuron [2]. In this presentation, recent advances in the development of artificial synapses and neurons based on 2D semiconductors and mixed-dimensional van der Waals heterojunctions will be discussed [3]. In particular, dual-gated memtransistors based on polycrystalline MoS2 show novel synaptic responses in crossbar arrays [4]. The non-volatile memory states of these devices are tuned by both biasing history and the field-effect from the two gate terminals [5]. In this manner, the dual gates allow independent addressing of nodes in the crossbar by minimizing sneak currents in addition to facilitating tunability of synaptic weight update rules that resulted in a 94% recognition rate of hand-written digits using a multi-level perceptron model [4]. Since neuromorphic architectures require both artificial synapses and neurons, this talk will also introduce a Gaussian heterojunction transistor design that realizes the functionality of an integrate-and-fire neuron [6]. Specifically, dual-gated, self-aligned, mixed-dimensional p-n heterojunctions based on carbon nanotubes and MoS2 operate as anti-ambipolar transistors where all parameters of the Gaussian transfer curves can be controlled by the applied gate potentials. These Gaussian heterojunction transistors enable a wide range of spiking behavior including constant spiking, latency, and bursting [6]. Since atomically thin dual-gated memtransistors and Gaussian heterojunction transistors are prepared using standard materials processing and fabrication schemes, they are well-poised to be integrated together and/or with traditional electronic devices for next-generation neuromorphic computing architectures.
References
[1] V. K. Sangwan and M. C. Hersam, Nature Nanotechnology, DOI: 10.1038/s41565-020-0647-z (2020).
[2] M. E. Beck and M. C. Hersam, ACS Nano, 14, 6498-6518 (2020).
[3] D. Jariwala, T. J. Marks, and M. C. Hersam, Nature Materials, 16, 170-181 (2017).
[4] H.-S. Lee, V. K. Sangwan, H. Bergeron, H. Y. Jeong, K. Su, and M. C. Hersam, submitted, 2020.
[5] V. K. Sangwan, H.-S. Lee, H. Bergeron, I. Balla, M. E. Beck, K.-S. Chen, and M. C. Hersam, Nature, 554, 500-504 (2018).
[6] M. E. Beck, A. Shylendra, V. K. Sangwan, S. Guo, W. A. Gaviria Rojas, H. Yoo, H. Bergeron, K. Su, A. R. Trivedi, and M. C. Hersam, Nature Communications, 11, 1565 (2020).