Spin-Filtering Ferroelectric Tunnel Junctions as Multiferroic Synapses for Neuromorphic Computing

Abstract

As nanoelectronic synapses, memristive ferroelectric tunnel junctions (FTJs) have triggered great interest due to the potential applications in neuromorphic computing for emulating biological brains. Here, we demonstrate multiferroic FTJ synapses based on the ferroelectric modulation of spin-filtering BaTiO3/CoFe2O4 composite barriers. Continuous conductance change with an ON/OFF current ratio of ∼54 400% and long-term memory with the spike-timing-dependent plasticity (STDP) of synaptic weight for Hebbian learning are achieved by controlling the polarization switching of BaTiO3. Supervised learning simulations adopting the STDP results as database for weight training are performed on a crossbar neural network and exhibit a high accuracy rate above 97% for recognition. The polarization switching also alters the band alignment of CoFe2O4 barrier relative to the electrodes, giving rise to the change of tunneling magnetoresistance ratio by about 10 times and even the reversal of its sign depending upon the resistance states. These results, especially the electrically switchable spin polarization, provide a new approach toward multiferroic neuromorphic devices with energy-efficient electrical manipulations through potential barrier design. In addition, the availability of spinel ferrite barriers epitaxially grown with ferroelectric oxides also expends the playground of FTJ devices for a broad scope of applications.

Ferroelectric-based synapses and neurons for neuromorphic computing

“…While PCM and ReRAM technologies have been widely used in neuromorphic systems in the past years [78], ferroelectric technology has only recently been investigated for these applications [15]. The systems reported are mainly at simulation level, for both FTJs [34,35,37] and FeFETs [57,[79][80][81][82].…”

“…The closer in time the two spikes are, the higher the weight change. Simulations using FTJ-based synapses and an STDP-based learning showed network accuracy for pattern classification >96% for a custom set of letters [34], whereas when standard Modified National Institute of Standards and Technology (MNIST) database is used, accuracy between 92.8% [37] and 97.3% [35] are reported. A hardware demonstration of a 3 × 3 pattern on a crossbar was shown in [38].…”

“…In particular, FTJs have been recently considered to emulate synaptic properties and mechanisms underlying learning. From the material point of view, alongside the classic devices, also in 3D architectures [34] or using multiferroic layers [35], synaptic properties are reported in organic [36] and flexible [37] ferroelectric devices.…”

“…The learning algorithms so far demonstrated in FTJ-based synapses are mainly Hebbian, with spike timing dependent plasticity (STDP) dominating the landscape [24,25,[34][35][36][37][38]. STDP is a well known biological learning rule which determines the magnitude and direction of the weight update depending on the timing difference between two spikes fired by two neurons, namely the pre-neuron and the post-neuron, connected by a synapse [40].…”

Ferroelectric photosensor network: an advanced hardware solution to real-time machine vision

“…Notably, the polarization switching can induce not only the magnitude change but also the sign reversal of photoresponse 19,21 , enabling a single FE-PS to represent both positive and negative weights and hence reducing the hardware overhead for network construction. Moreover, the nonvolativity, high controllability, and ultrafast switching kinetics (<1 ns) of polarization as demonstrated in various ferroelectric memory and neuromorphic devices [29][30][31][32][33][34] , along with the intimate coupling between polarization and photoresponse 35 , endow the FE-PS with good reliability and high write speed. Also noteworthy are the high photosensitivity and ultrashort photoresponse time (<1 ns) of FE-PS 24,25 , allowing a high-speed readout.…”