Performance Estimate of Inverse Rashba–Edelstein Magnetoelectric Devices for Neuromorphic Computing

Abstract

We propose a new design for a cellular neural network with spintronic neurons and CMOS-based synapses. Harnessing the magnetoelectric and inverse Rashba-Edelstein effects allows natural emulation of the behavior of an ideal cellular network. This combination of effects offers an increase in speed and efficiency over other spintronic neural networks. A rigorous performance analysis via simulation is provided.INDEX TERMS Cellular neural network (CNN), CMOS, energy efficiency, magnetoelectric (ME), Rashba-Edelstein, spintronics.

Nonvolatile Spintronic Memory Cells for Neural Networks

“…The inverse Rashba-Edelstein magnetoelectric neuron (IRMEN) has been thoroughly described, including its relationship to standard CeNN cells, in [10]. A brief summary will be given here.…”

“…However, the capacitor represents an input mechanism rather than the true state. A magnetoelectric material within the capacitor is coupled to the ferromagnet (FM) that makes up one of its electrodes, thereby allowing control of the FM via electric charge on the capacitor [10], [12]- [14]. Readout is accomplished by driving a charge current first through the FM to spin-polarize the current and then through an inverse spin-orbit stack that transduces from spin current to charge potential along an axis orthogonal to both current flow and spin orientation [15]- [17].…”

“…Readout is accomplished by driving a charge current first through the FM to spin-polarize the current and then through an inverse spin-orbit stack that transduces from spin current to charge potential along an axis orthogonal to both current flow and spin orientation [15]- [17]. This is modeled as a voltage source with the inverse Rashba potential V IR and internal resistance R IR [10]. The IRMEN cells natively compute a nonlinear transfer function on their inputs by virtue of the M -H curve dictated by the anisotropy of the nanomagnet.…”

“…Another platform that uses inverse spin-orbit magnetoelectric stacks instead of MTJs has also been proposed. The latter, though more exploratory compared to MTJ-based devices, was predicted to perform basic CeNN functions both more quickly and more efficiently than the others [10]. In this article, we propose to utilize this platform as a hybrid memory/CeNN cell with a high energy efficiency that can be used as analog memory with a built-in activation function.…”

Energy-Efficient Convolutional Neural Network Based on Cellular Neural Network Using Beyond-CMOS Technologies

“…Recently, multiple efforts have investigated the cellular neural network (CeNN) in the context of emerging information processing technologies. As representative examples, CeNNs based on graphene devices, spintronic devices, and tunnel FETs (TFETs) [16]- [22] have been studied. A variety of applications, such as noise filtering, associative memory, pattern recognition, tactile sensing, and image processing application engines have been considered [17]- [20].…”