Vertically Integrated Dual-memtransistor Enabled Reconfigurable Heterosynaptic Sensorimotor Networks and In-memory Neuromorphic Computing

Srilagna Sahoo, Abin Varghese, Aniket Sadashiva, Mayank Goyal, Jayatika Sakhuja, Debanjan Bhowmik, Saurabh Lodha · 2024

Read Vertically Integrated Dual-memtransistor Enabled Reconfigurable Heterosynaptic Sensorimotor Networks and In-memory Neuromorphic Computing by Srilagna Sahoo, Abin Varghese, Aniket Sadashiva, Mayank Goyal, Jayatika Sakhuja, Debanjan Bhowmik, Saurabh Lodha (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-10757.txt.

Introduction

Srilagna Sahoo and co-authors stack two atom-thin transistors into one device that mimics brain-like learning, in this 2024 paper. Brain-inspired computing needs compact parts that store and process in one place. Their device layers a memory transistor over a ferroelectric one, and the two talk through electric coupling.

The top channel shows synapse-like behavior and learns pattern tasks with high accuracy, in both plain and spiking network modes. Reading both channels at once reproduces cooperation and rivalry between synapses. The team even copied a sea slug's reflex learning, with tiny power use. The device also flips into logic gates on demand.

Catalog record

CREATORSrilagna Sahoo, Abin Varghese, Aniket Sadashiva, Mayank Goyal, Jayatika Sakhuja, Debanjan Bhowmik, Saurabh Lodha
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE51 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-10757

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Vertically Integrated Dual-memtransistor Enabled Reconfigurable Heterosynaptic Sensorimotor Networks and In-memory Neuromorphic Computing Srilagna Sahoo, Abin Varghese, Aniket Sadashiva, Mayank Goyal, Jayatika Sakhuja, Debanjan Bhowmik, Saurabh Lodha Abstract Neuromorphic in-memory computing requires area-efficient architecture for seamless and low latency parallel processing of large volumes of data. Here, we report a compact, vertically integrated/stratified field-effect transistor (VSFET) consisting of a 2D non-ferroelectric MoS 2 FET channel stacked on a 2D ferroelectric In 2 Se 3 FET channel. Electrostatic coupling between the ferroelectric and non-ferroelectric semiconducting channels results in hysteretic transfer and output characteristics of both FETs. The gate-controlled MoS 2 memtransistor is shown to emulate homosynaptic plasticity behavior with low nonlinearity, low epoch, and high accuracy supervised (ANN - artificial neural network) and unsupervised (SNN - spiking neural network) on-chip learning. Further, simultaneous measurements of the MoS 2 and In 2 Se 3 transistor synapses help realize complex heterosynaptic cooperation and competition behaviors. These are shown to mimic advanced sensorimotor neural network-controlled gill withdrawal reflex sensitization and habituation of a sea mollusk (Aplysia) with ultra-low power consumption. Finally, we show logic reconfigurability of the VSFET to realize Boolean gates thereby adding significant design flexibility for advanced computing technologies. † † affiliation: Department of Electrical Engineering, Indian Institute of Technology Bombay, Mumbai 400076, India † † affiliation: Department of Engineering, King’s College London, London WC2R 2LS, UK † † email: [email protected] † † affiliation: Department of Electrical Engineering, Indian Institute of Technology Bombay, Mumbai 400076, India † † affiliation: https://orcid.org/0000-0002-0690-3169 1 Keywords 2D materials, vdW, Heterostructure, Multiterminal, Ferroelectric, In 2 Se 3 , MoS 2 , Homosynaptic and Heterosynaptic plasticity, Synaptic interactions, Synaptic cooperation, Synaptic competition, Logic gates 2 Introduction In today’s data-intensive world, research in future computing paradigms is pursuing alternate paths compared to the traditional scaling of complementary metal oxide semiconductor (CMOS) transistor technology. Applications such as artificial intelligence, challenged by the von Neumann bottleneck, demand high performance as well as energy- and area-efficient computing devices with logic-in-memory functionality. This can be addressed by brain-inspired neuromorphic computing devices that offer in-memory computing with parallel processing capability along with low energy consumption and latency. Thakar et al. [2023] Their fundamental components are neurons (responsible for processing) that are connected to each other by synapses (responsible for storage) to form a neural network (NN). However, the challenges in realizing neuromorphic devices lie in material limitations that hinder the effective integration of memory and logic and, secondly, architecture that is both energy- and area-efficient. Ferroelectric field effect transistors (Fe-FETs) offer one such option of integrating logic and memory in a single, area-efficient device. Ajayan et al. [2023] Conventional Fe-FETs involve stacking the semiconducting channel layer with an out-of-plane ferroelectric dielectric or polymer as gate dielectric to demonstrate a modulated memtransistor for synapses and logic circuits. Lipatov et al. [2015] ; Tian et al. [2019] ; Chen et al. [2020] ; Ram et al. [2023] However, conventional bulk ferroelectric dielectrics (BaTiO 3 , Pb(Zr, Ti)O 3 , BiFeO 3 ) suffer from degradation of ferroelectric properties with thickness scaling, charge trapping, and poor interfaces. Wang et al. [2023] In recent years, two-dimensional (2D) van der Waals (vdW) ferroelectrics have drawn attention because of their room temperature atomic level ferroelectricity, a dangling bond-free surface and high-quality heterointerfaces due to the absence of lattice mismatch with other materials. Baek et al. [2022] ; Liu et al. [2016] ; Wang et al. [2023] ; Ghosh et al. [2022] ; Varghese et al. [2020] CuInP 2 S 6 (CIPS) is one such 2D vdW ferroelectric dielectric that shows out-of-plane (OOP) ferroelectricity in ultra-thin flakes. Belianinov et al. [2015] ; Liu et al. [2016] ; Baek et al. [2022] Unlike conventional insulating ferroelectrics, 2D vdW ferroelectrics can also be semiconducting in nature. This offers an additional degree of freedom in the design of ferroelectric semiconductor (FeS) FETs (FeS-FETs) for logic-in-memory functionality. Si et al. [2019] ; Liao et al. [2023] Ferroelectric 2D vdW semiconductors such as SnSe, Chang et al. [2020] and SnS Higashitarumizu et al. [2020] ; Bao et al. [2019] show in-plane (IP) ferroelectric polarization. Wang et al. [2023] In 2 Se 3 on the other hand shows both IP and OOP ferroelectricity, which makes it suitable for multi-directional conductivity control device architectures. Xiao et al. [2018] ; Cui et al. [2018] Several memory device and neural network designs using In 2 Se 3 as a semiconductor channel or a ferroelectric gate have been reported in conjunction with CIPS Baek et al. [2022] , MoS 2 Park et al. [2023] , and hBN Wang et al. [2021] ; Kang et al. [2024] . However, reported structures are either area-inefficient (lateral or multi-device architecture), and/or they do not simultaneously exploit their semiconducting and correlated IP and OOP ferroelectric polarization properties. In this work, we demonstrate a compact, vertically stacked heterostructure FET where the 2D vdW semiconductor (S) MoS 2 is stacked upon ferroelectric semiconductor (FeS) In 2 Se 3 on gate insulator (I) hBN aligned over the metal (M) gate to form a MIFeSS structure in a top-down approach. Lateral electrodes are placed on both semiconductors such that they can be readily accessed simultaneously. The FeS In 2 Se 3 is stacked on top of the gate electrode and beneath the MoS 2 to gain maximum control on its polarization and mobile charge, while its OOP polarization can be leveraged to tune MoS 2 conduction. This densely packed vertical stratified field-effect transistor (VSFET) architecture utilizing a non-FeS/FeS heterostructure harnesses not only the semiconducting properties of each layer but also the interface FeS-S coupling between them. This work consists of three parts. The first part utilizes the electrostatic coupling between MoS 2 (S) and In 2 Se 3 (FeS) to demonstrate hysteretic behavior in transfer (drain current vs. gate voltage: I_{D} - V_{G} ) and output (drain current vs. drain voltage: I_{D} - V_{D} ) characteristics of the MoS 2 FET. The gate-modulated MoS 2 memtransistor is shown to emulate biological input-specific Hebbian homosynaptic plasticity behavior with low nonlinearity, and high accuracy supervised (ANN - artificial neural network) and unsupervised (SNN - spiking neural network) on-chip learning. In the second part, we demonstrate advanced biomimetic heterosynaptic cooperation and competition between the two simultaneously measured MoS 2 and In 2 Se 3 memtransistors representing two distinct but coupled synapses. This heterosynaptic behavior is shown to replicate the highly developed sensorimotor NN governing the gill reflexes of a sea mollusk. Finally, the realization of two-input Boolean logic gates comprehensively shows the area-efficient, logic, and multi-memory reconfigurability of the VSFET architecture. 3 Results Device Architecture: A three-dimensional (3D) schematic of the VSFET stack consisting of Au (bottom gate)/hBN (dielectric)/n-type In 2 Se 3 (FeS)/n-type MoS 2 (top S) is shown in Figure 1 a. 2D cross-section of the MoS 2 /In 2 Se 3 VSFET that shows heterostructure stacking of two individual FET channels (top MoS 2 : HT-MoS 2 and bottom In 2 Se 3 : HB-In 2 Se 3 ) is depicted in Figure 1 b, enabling 2D vdW FeS-S coupling that can modulate the HT-MoS 2 electron conduction (Figure 1 c). The bottom metal gate was patterned on a 285 nm SiO 2 /Si substrate using electron beam (e-beam) lithography

Read the complete text