Mimicking Synaptic Plasticity and Neural Network Using Memtranstors

Abstract

Artificial synaptic devices that mimic the functions of biological synapses have drawn enormous interest because of their potential in developing brain-inspired computing. Current studies are focusing on memristive devices in which the change of the conductance state is used to emulate synaptic behaviors. Here, a new type of artificial synaptic devices based on the memtranstor is demonstrated, which is a fundamental circuit memelement in addition to the memristor, memcapacitor, and meminductor. The state of transtance (presented by the magnetoelectric voltage) in memtranstors acting as the synaptic weight can be tuned continuously with a large number of nonvolatile levels by engineering the applied voltage pulses. Synaptic behaviors including the long-term potentiation, long-term depression, and spiking-time-dependent plasticity are implemented in memtranstors made of Ni/0.7Pb(Mg Nb )O -0.3PbTiO /Ni multiferroic heterostructures. Simulations reveal the capability of pattern learning in a memtranstor network. The work elucidates the promise of memtranstors as artificial synaptic devices with low energy consumption.

Tailoring Multilevel‐Stable Remanence States in Exchange‐Biased System through Spin‐Orbit Torque

“…Under an external stimulus, spikes or action potentials from the preneuron are transmitted through synapse to the postneuron and generate excitatory postsynaptic potentials (EPSP) or inhibitory postsynaptic potentials (IPSP), together with the synaptic weight updates. [ 42,43 ] The information storage and learning of human brains are exactly a consequence of changes in the synaptic weight. The evolution of stable multilevel remanent Hall resistance with the consecutive current pulses in the Pt/Co/IrMn stacks can be viewed as the plasticity of a synapse, which is illustrated in Figure .…”

“…So the EPSP and IPSP can be classed as long‐term plasticity. [ 5,14,43 ] It should be mentioned that the incomplete switching of magnetization in HM/FM bilayers by SOT can also be used for multilevel memory and neuromorphic computing. [ 14,15 ] However, the magnetic state could be easily changed by external magnetic field.…”

A MoS2/PTCDA Hybrid Heterojunction Synapse with Efficient Photoelectric Dual Modulation and Versatility

“…Nowadays, organic bioelectronics are attracting intense interest in neural applications due to their low‐cost processing, mechanical flexibility and tenability. [] More recently, two‐dimensional (2D) organic materials, such as perylene‐3,4,9,10‐tetracarboxylic dianhydride (PTCDA), have received increasing attention, not only due to their unique photoelectric properties but also because of their superior compatibility with most inorganic 2D materials . Beyond that, because of their unique internal and interfacial structure, and electrical and optical properties, 2D transition metal dichalcogenides (TMDCs) such as molybdenum disulfide (MoS 2 ) and tungsten selenide (WSe 2 ) have been reported as promising candidates for complex neuromorphic applications .…”

Neuromorphic Engineering for Hardware Computational Acceleration and Biomimetic Perception Motion Integration

“…Neuromorphic engineering, usually also referred to as neuromorphic computing, was proposed by Carver Mead [21] and initially described the use of VLSI analog circuits to mimic the architecture of biological nervous systems. [49] Nowadays, neuromorphic engineering has been extended to use analog, digital, and digital-analog hybrid VLSI [22] or emerging memristive devices (such as metal oxides, [24,25] phase change, [20,28] ferroelectrics, [30,50] spintronics [31,32] and vdW [51,52] memristors as well as electrolytic transistors, [53][54][55] memtransistors, [34,56,57] etc.) to implement data-centric artificial neural networks (ANN) and multiple sensory integration and motion control bionic peripheral neural systems through software and hardware strategies.…”