In the emerging era of artificial intelligence, deep learning, and Big-data, the energy and hardware investments required for conventional high-precision digital computing are becoming increasingly unsustainable. As a result, there is a growing need for a new paradigm that prioritizes energy and resource efficiency over precision for many computing applications. Stochastic computing (SC) is a promising alternative because it can perform basic arithmetic operations using simple logic gates, unlike digital computers that require many logic gates and a high transistor volume. However, the hardware investment necessary to generate stochastic bits (s-bit), the fundamental computing primitive for SC, has hindered its widespread adoption. While traditional silicon complementary metal oxide semiconductor (CMOS) technology can accelerate SC, it still requires extensive hardware investment. Memristor and spin-based devices offer natural randomness but rely on hybrid designs involving CMOS peripherals, which increase the area and energy burden.

To overcome these limitations, we have developed a standalone SC architecture embedded in memory based on two-dimensional (2D) memtransistors. This monolithic and non-von Neumann SC architecture requires only a tiny amount of energy (< 1 nano Joules) for s-bit generation and to perform arithmetic operations, and occupies a small hardware footprint, highlighting the benefits of SC. Additionally, the researchers demonstrate the acceleration of Bayesian inference using their SC platform.

Despite Copper being the current material of choice for interconnect technology, it suffers from increased resistivity at scaled dimensions and the necessity for a barrier-liner to prevent diffusion. This has motivated the exploration of alternate materials that can overcome these limitations for scaled CMOS technology nodes. Metallic 2D materials can offer a promising option. In this work, we evaluate the properties of 2D semi-metallic material 2H Niobium diselenide (2H-NbSe2) as a candidate for realizing highly scalable interconnect technology without the need for barrier-liner. We performed detailed electrical characterization evaluating the dimensional scaling, dependence of the resistivity on temperature, device lifetime, effect of encapsulation layer etc. on ribbon devices fabricated with 2H-NbSe₂ material. 2H-NbSe₂ showed negligible change in resistivity compared to the bulk value when scaled down to thicknesses less than 15 nm. High current density transport measurements are performed on 2H-NbSe₂ ribbon devices with varied width (0.1-1um) at elevated temperature (>= 100^oC) to assess the reliability and failure characteristics. Lifetime of the NbSe₂ ribbons improved when an Al₂O₃ encapsulation layer was used. Our work provides critical insights into the potential of NbSe₂ for realizing scalable interconnects.

This talk will present my (biased!) perspective of what two-dimensional (2D) materials could be good for. For example, they could be good for applications where their ultrathin nature gives them distinct advantages, such as flexible electronics [1] or light-weight solar cells [2]. They may not be good where conventional materials work sufficiently well, like transistors thicker than a few nanometers. I will focus on 2D materials for 3D heterogeneous integration of electronics, which presents major advantages for energy-efficient computing [3]. Here, 2D materials could be monolayer transistors with ultralow leakage [4] (due to larger band gaps than silicon), used to access high-density memory [5]. Recent results from our group [6,7] and others [8] have shown monolayer transistors with good performance, which cannot be achieved with sub-nanometer thin conventional semiconductors, and the 2D performance could be further boosted by strain [9]. I will also describe some unconventional applications, using 2D materials as thermal insulators [10], heat spreaders [11], and thermal transistors [12]. These could enable control of heat in “thermal circuits” analogous with electrical circuits. Combined, these studies reveal fundamental limits and some unusual applications of 2D materials, which take advantage of their unique properties.

Refs: [1]A. Daus et al., Nat. Elec. 4, 495 (2021).[2] K. Nassiri Nazif, et al., Nat. Comm. 12, 7034 (2021). [3] M. Aly et al., Computer 48, 24 (2015). [4] C. Bailey et al., EMC (2019). [5] A. Khan et al. Science 373, 1243 (2021). [6] C. English et al., IEDM, Dec 2016. [7] C. McClellan et al. ACS Nano 15, 1587 (2021). [8] S. Das et al., Nat. Elec. 4, 786 (2021). [9] I Datye et al., Nano Lett. 22, 8052 (2022). [10] S. Vaziri et al., Science Adv. 5, eaax1325 (2019). [11] C. Koroglu & E. Pop, IEEE Elec. Dev. Lett. 44, 496 (2023). [12] M. Chen et al., 2D Mater. 8, 035055 (2021).

Memristor-based neuromorphic computing is promising for artificial intelligence. However, most of the reported memristors have limited linear computing states and consume large operation energy which hinder their applications. Herein, we report a memristor based on ionic two-dimensional CuInP₂S₆ (2D CIPS), in which up to 1350 linear conductance states are achieved by controlling the migration of internal Cu ions in CIPS. In addition, the device shows a low operation current of ~100 pA. Cu ions are proven to move along the electric field by in-situ scanning electron microscopy and energy dispersive spectroscopy measurements. Furthermore, complex signal transport among multiple neurons in the brain is imitated by 2D CIPS-based memristor arrays. Our results offer a new platform to fabricate high-performance memristors based on ion transport in 2D materials for neuromorphic computing.