All-in-one metal-oxide heterojunction artificial synapses for visual sensory and neuromorphic computing systems

IntroductionRecent meta-analytic findings show impaired visual perceptive performance for VLBW children. Little is known about relationships between visual sensory and visual perceptive processes in VLBW children.MethodsVLBW children (n=121) and age...

Mutations in the synaptic gene SHANK3 lead to a neurodevelopmental disorder known as Phelan-McDermid syndrome (PMS). PMS is a relatively common monogenic and highly penetrant cause of autism spectrum disorder (ASD) and intellectual disability (ID), and frequently presents with attention deficits. The underlying neurobiology of PMS is not fully known and pharmacological treatments for core symptoms do not exist. Here, we report the production and characterization of a Shank3-deficient rat model of PMS, with a genetic alteration similar to a human SHANK3 mutation. We show that Shank3-deficient rats exhibit impaired long-term social recognition memory and attention, and reduced synaptic plasticity in the hippocampal-medial prefrontal cortex pathway. These deficits were attenuated with oxytocin treatment. The effect of oxytocin on reversing non-social attention deficits is a particularly novel finding, and the results implicate an oxytocinergic contribution in this genetically defined subtype of ASD and ID, suggesting an individualized therapeutic approach for PMS.DOI: http://dx.doi.org/10.7554/eLife.18904.001

Since the introduction of a concept of memristor by Leon Chua in 1971 as the fourth fundamental passive component in electronic circuits, and its functional prototype introduced by HP Lab in 2008, research has increasingly concentrated on exploring its application ability in electronic circuits. First, the memristor principles have been exploited in various memory devices within a classical von Neumann computer architecture. The properties of memristor were found particularly suitable for resistive random-access memories (ReRAM).1 Contrary to the classical von Neumann architecture based on the binary logic, advanced methods like neuromorphic computing require an electronic element with continuously varying states depending on previous input signals. The input signal consists of a sequence of voltage spikes that stimulate the device, similarly as a neural synapse does in biological systems. These trains of spikes change the output state of the device continuously to a lower resistance state, mimicking a learning process. Having these functionalities, memristors are considered to be possible building blocks of brain-inspired neuromorphic computing and artificial neural networks.2 Memristors with neurosynaptic functionality take a continuity of resistance values with synaptic weights modulated by the number and frequency of homogeneous spikes. In order to mimic the neural synapse, memristors must exhibit analog properties including non-abrupt switching transitions, memory loss, continuously variable resistance states, and predictable response. We present such functionality on thin films of two kinds of organic materials:(i) A newly synthesized poly(methacrylamide) derivative with carbazole charge transporting group separated from the polymer backbone by an alkyl chain, was synthesized recently and found to behave as a bistable resistive memory.3 Thin films of the polymer, sandwiched between Al and ITO electrodes, exhibit rewritable flash memory behavior with bistable conductivity, with the set switching voltage ranging from 2 to 4.5 V, and the current ON/OFF ratio exceeding 100. The device demonstrates a remarkable lifetime and remains persistent for more than 104 seconds under the static voltage of 0.5 V. The main physical mechanisms driving the resistive switching have been attributed to the electric field-induced reorientation of heterocycles, which modulates charge transport, and trapping/detrapping of charges in localized states within the bulk of the polymer. Memory persistence is strengthened by the physical crosslinking caused by hydrogen bonds between amide and carbonyl groups in the side chains. Depending on the layer thickness, electrode material and applied voltage range the electrical characteristics can change from bistable to analog behavior showing memristive properties.(ii) Ditopic π-conjugated bis-terpyridine (tpy-) ligands form supramolecular structures in which the ligands are interconnected by complexation with metal ions. These molecules offer a low energy operation with improved reproducibility of memristive characteristics. They possess low-lying molecular energy levels and relatively narrow bandgap enabling the low-power consuming resistive switching and longer stability. We used a newly synthesized bis-terpyridine ligands based on 9-phenyl carbazole, Cbz, and its complex with cobalt Co2+ as an active layer of a memristor device. A two-terminal memristor architecture was employed, consisting of the ligand or its cobalt complex deposited on an ITO substrate, and aluminum, gold, or gallium as a top electrode. We show that Cbz as a memristor active layer exhibits an excellent nonvolatile bistable memory behavior, while Co-Cbz exhibits both electronic memory and synaptic plasticity. The ligand exhibited bi-stable conduction and non-volatile memory effect with persistence over 18 hours, while its cobalt complex displayed a synaptic effect characterized by subsequent potentiation and depression cycles. Voltage-induced modulation of synaptic weight was observed with a low applied voltage below 500 mV and short pulse duration below 20 ms. Pair pulse facilitation and pair pulse depression demonstrated significant synaptic weight changes, showing the cobalt complex ability to emulate biological neurons. The operational mechanism was attributed to a combination of the redox effect in the ligand and complex, and to the voltage-induced migration of perchlorate counter ions and subsequent redox processes, altering the metal-to-ligand charge transfer band and changing the conducting state of the active layer.Acknowledgments: The work was financially supported by the Czech Science Foundation, project 24-10384S, and MEYS of the Czech Republic, project LUAUS24032. Strukov, D. B., Snider, G. S., Stewart, D. R. & Williams, R. S. The missing memristor found. Nature 453, 80–83 (2008).Chang, T., Jo, S. H. & Lu, W. Short-term memory to long-term memory transition in a nanoscale memristor. ACS Nano 5, 7669–7676 (2011).Panthi, Y.R, Pfleger, J., Vyprachticky, D., Pandey, A., Thottappali, M.A., Sedenkova, I., Konefal, M. & Foulger, S.H. Rewritable resistive memory effect in poly[N-(3-(9H-carbazol-9-yl)propyl)-methacrylamide] memristor. Mater. Chem. C 11, 17093 (2023).

Astrocytes participate in synaptic transmission and plasticity through tightly regulated, bidirectional communication with pre- and post-synaptic neurons, as well as microglia and oligodendrocytes. A key component of astrocyte-mediated synaptic regulation is the cytokine tumor necrosis factor (TNF). TNF signals via two cognate receptors, TNFR1 and TNFR2, both expressed in astrocytes. While TNFR1 signaling in astrocytes has been long demonstrated to be necessary for physiological synaptic function, the role of astroglial TNFR2 has never been explored. Here, we demonstrate that astroglial TNFR2 is essential for maintaining hippocampal synaptic function and plasticity in physiological conditions. Indeed, Gfap creERT2 :Tnfrsf1b fl/fl mice with selective ablation of TNFR2 in astrocytes exhibited dysregulated expression of neuronal and glial proteins (e.g., SNARE complex molecules, glutamate receptor subunits, glutamate transporters) essential for hippocampal synaptic transmission and plasticity. Hippocampal astrocytes sorted from Gfap creERT2 :Tnfrsf1b fl/fl mice displayed downregulation of genes and pathways implicated in synaptic plasticity, as well as astrocyte-neuron and astrocyte-oligodendrocyte communication. These alterations were accompanied by increased glial reactivity and impaired astrocyte calcium dynamics, and ultimately translated into functional deficits, specifically impaired long-term potentiation (LTP) and cognitive functions. Notably, male Gfap creERT2 :Tnfrsf1b fl/fl mice exhibited more pronounced hippocampal synaptic and cellular alterations, suggesting sex-dependent differences in astroglial TNFR2 regulation of synaptic function. Together, these findings indicate that TNFR2 signaling in astrocytes is essential for proper astrocyte-neuron communication at the basis of synaptic function, and that this is regulated in a sex-dependent manner.

Neuromorphic computing systems convert multimodal signals to electrical responses for artificial intelligence recognition. Energy is consumed during both the response enhancement and depression, making the systems suffer from high energy consumption. This study presents a neuromorphic computing pathway based on supercapacitors. MXene Ti₃C₂Tx supercapacitors are fabricated and convert current stimuli to voltage responses. The response enhancement and depression are tunable through adjusting charging and discharging current stimuli, thus exhibiting synaptic plasticity. Typical synaptic behaviors are demonstrated, including short‐term memory, long‐term memory, paired‐pulse facilitation, and learning experience. Next, the voltage responses are used to recognize Braille numbers represented by 3 × 4 arrays. A charging/discharging current pulse train representing each Braille array is applied to the supercapacitor. The voltage responses are collected and converted to 12‐pixel greyscale images. Once the images representing Braille numbers 0–9 are input into artificial neural networks and deep diffraction neural networks, 100% accuracy can be achieved for recognizing the ten numbers. Because energy is stored during response enhancement in the supercapacitor and released once the response declines, this research demonstrates the potential applications of energy storage devices in neuromorphic computing, providing an innovative way to develop energy‐efficient brain‐like computing systems.

The emulation of functions and behaviors of biological synapses using electronic devices has inspired the development of artificial neural networks (ANNs) in biomedical interfaces. Despite the achievements, artificial synapses that can be selectively responsive to non‐electroactive biomolecules and directly operate in biological environments are still lacking. Herein, we report an artificial synapse based on organic electrochemical transistors and investigate the selective modulation of its synaptic plasticity by glucose. The enzymatic reaction between glucose and glucose oxidase results in long‐term modulation of the channel conductance, mimicking selective binding of biomolecules to their receptors and consequent long‐term modulation of the synaptic weight. Moreover, the device shows enhanced synaptic behaviors in the blood serum at a higher glucose concentration, which suggests its potential application in vivo as artificial neurons. This work provides a step towards the fabrication of ANNs with synaptic plasticity selectively mediated by biomolecules for neuro‐prosthetics and human‐machine interfaces.

Neuromorphic computing offers a promising approach to overcoming the limitations of von Neumann architecture by mimicking biological synapses. While optoelectronic synapses have demonstrated synaptic plasticity through optical and electrical stimuli, most studies rely on ambient light conditions, limiting their robustness and functional complexity. Here, a WSe2/h-BN/SiO2 heterostructure-based optoelectronic synapse is presented that achieves precise synaptic weight modulation through co-stimuli of electrical and optical pulses. The device exhibits enhanced paired-pulse facilitation (PPF) and long-term plasticity (LTP/LTD), demonstrating stable and linear synaptic behavior. Notably, the study systematically analyzes the effects of co-stimuli firing conditions, revealing that both the intensity of light and voltage magnitude influence synaptic weight updates. The device achieves outstanding nonlinearity, high Gmax/Gmin, and stable depression recovery, essential for high-performance neuromorphic computing. Furthermore, ANN-based cognitive simulations using MNIST digits validate their potential for inference tasks, demonstrating near-ideal accuracy. These findings underscore the potential of co-stimuli-driven synapses for multi-modal cognitive systems, paving the way for advanced neuromorphic architectures beyond single-species stimuli constraints.

Striatal Plasticity and Basal Ganglia Circuit Function

Li+ electrolyte-gated transistors (EGTs) have attracted significant attention as artificial synapses because of the fast response of Li+ ion, low operating voltage, and applicability to flexible electronics. Due to the inherent nature of Li+ ion, Li+ EGTs show, however, limitations, such as poor long-term synaptic plasticity and nonlinear/nonsymmetric conductance update, which hinder the practical applications of artificial synapses. Herein, Li+ EGTs integrated with poly(vinylidene fluoride-co-trifluoroethylene) (PVDF-TrFE) ferroelectric polymer as a channel–electrolyte interlayer are presented. Owing to the polarized domains of PVDF-TrFE, the transport of Li+ ions at the channel–electrolyte interface is accelerated, and Li+ ions effectively penetrate the channel. Moreover, the self-diffusion of Li+ ions from the channel to the electrolyte is suppressed by the downward polarized domains. Li+ EGTs, therefore, successfully demonstrate synaptic characteristics, including excitatory postsynaptic current, short-/long-term synaptic plasticity, and paired-pulse facilitation. Also, conductance update in Li+ EGTs shows a dynamic range (Gmax/Gmin) of 92.42, high linearity, and distinct stability over 100 cycles. Based on their synaptic characteristics, inference simulations using a convolution neural network for the CIFAR-10 dataset imply that Li+ EGTs are suitable as artificial synapses with an inference accuracy of 89.13%. The new methodological approach addressing modulation of ion dynamics at the interface is introduced for developing practical synaptic devices.

A new type artificial synapse based on the organic copolymer memcapacitor

The human brain, with high energy‐efficient and parallel processing ability, inspires to mitigate power issues perplexing von Neumann architecture. As one of the essential components constructing the human brain, the emulation of biological synapses exploiting electronic devices consuming power at a biological level lays the foundation for the implementation of energy‐efficient neuromorphic computing. Besides, signal matching between biologically‐related stimuli and the driving voltage of artificial synapses helps to realize intelligent neuromorphic interfaces and sustainable energy. Here, ultra‐sensitive artificial synapse stimulated at 1 mV with energy consumption of 132 attojoule/synaptic event is demonstrated. Biological signal matching and low power application are realized simultaneously based on sodium acetate (NaAc) doped polyvinyl alcohol (PVA) electrolyte. The biphasic current, which comprises the electrical‐ and ion‐mediation current component, contributes to enrich synaptic functions compared to monophasic synaptic behavior. Moreover, freestanding NaAc‐doped PVA membrane functions as both dielectric layer and mechanical support and facilitates to achieve flexible, transferable artificial synapse, which maintains functional stability at an ultralow voltage and power even after bending tests. Thus, encompassing superior sensitivity, low energy, and multiple functionalities with flexible, self‐supported, biocompatible property, takes a step to construct energetically‐efficient, complex neuromorphic systems for wearable, implantable medicines as well as smart bio‐electronic interfaces.

Ferrimagnetic materials with antiparallel exchange coupling, exhibit spin‐orbit‐torque‐induced dynamics, offering an emerging platform for realizing neuromorphic devices, such as artificial synapses. However, the state‐of‐the‐art artificial synapses based on ferrimagnet suffer from poor analog switching linearity, which serves as a bottleneck for achieving complex tasks with high accuracy in neuromorphic computing. Here, an artificial synapse is reported with high‐weight update linearity in a compensated ferrimagnetic crossbar device. In particular, the linear weight update of the synapses is enhanced by engineering the current density distribution. Using experimentally derived device parameters, handwritten digit recognition can be achieved with an accuracy of over 95% in a three‐layer fully connected artificial neural network. The work provides a universal method to improve the synaptic linearity, which also paves the way for applying the spin‐orbit device in neuromorphic computing.

In contrast to the commonly present UV light-stimulated synaptic oxide thin-film transistors, this study demonstrates a violet light (wavelength of 405 nm) stimulated zinc–tin oxide (ZTO) photoelectric transistor for potential application in optical neuromorphic computation. Owing to the light-induced oxygen vacancy ionization and persistent photoconductivity effect in ZTO, this device well imitates prominent synaptic functions, including photonic potentiation, electric depression, and short-term memory (STM) to long-term memory (LTM) transition. A highly linear and broad dynamic range of photonic potentiation can be achieved by modulating the light power density, while electric depression is realized by gate voltage pulsing. In addition, the brain-like re-learning experience with extended forgetting time (200 s) is well mimicked by the ZTO photoelectric transistor. As a result, the ZTO photoelectric transistor provides excessive synaptic function with multi-series of synaptic weight levels (90 levels for each given light power density), which makes it prevalent in the neuromorphic computation of massive data as well as in learning-driven artificial intelligence computation.

High sensitivity artificial synapses using printed high-transmittance ITO fibers for neuromorphic computing

Neuronal avalanches represent one important way in which the network dynamics of neural systems may be propagated. Experiments utilizing multi-channel recordings of neural cultures have shown that the sizes of neuronal avalanches follow a power-law distribution, which is a key feature of a self-organized criticality [1]. Although self-organized criticality provides certain advantages for the processing of neural information in neural networks [1-3], the underlying mechanisms of these dynamics remain unclear. One hypothesis is that synaptic plasticity is the major prerequisite for self-organized criticality [4,5]. In the present study, we focused on the roles of short-term plasticity and the relationship between synaptic weights and self-organized criticality. We simulated neural networks consisting of 300 simple integrate-and-fire neurons with all-to-all connections. Each synapse had depressive short-term plasticity and random initial synaptic weights. Small external currents were injected into individual neurons that were randomly selected at each step, and the sizes of the neuronal avalanches at each step were recorded. After approximately two million steps, the synaptic weights were altered by short-term plasticity, the sizes of the neuronal avalanches followed a power-law distribution, and a pattern of converging exponents was observed. To investigate the effects of synaptic weights on self-organized criticality, neural networks were re-simulated without short-term plasticity; instead, the synaptic weights from the above simulation, which had been altered by short-term plasticity, were used. In this simulation, the sizes of the neuronal avalanches did not follow a power-law distribution, and the exponents of the power function always remained between -1.2 and -1.9. The same results were observed for randomly connected neural networks. Although the synaptic weights of the simulated neural networks adjusted when the neural networks exhibited a self-organized criticality, neuronal avalanche size did not follow a power-law distribution when short-term plasticity was turned off. These results show that the key mechanism for self-organized criticality is synaptic plasticity rather than synaptic weights. Therefore, it is important that short-term plasticity be maintained in neural networks to ensure self-organized criticality.