Neuromorphic learning, working memory, and metaplasticity in nanowire networks

Intelligent soft matter: towards embodied intelligence

Intelligent soft matter lies at the intersection of materials science, physics, and cognitive science, promising to change how we design and interact with materials. This transformative field aims to create materials with life-like capabilities, such as perception, learning, memory, and adaptive behavior. Unlike traditional materials, which typically perform static or predefined functions, intelligent soft matter can dynamically interact with its environment, integrating multiple sensory inputs, retaining past experiences, and making decisions to optimize its responses. Inspired by biological systems, these materials leverage the inherent properties of soft matter such as flexibility, adaptability, and responsiveness to perform functions that mimic cognitive processes. By synthesizing current research trends and projecting their evolution, we present a forward-looking perspective on how intelligent soft matter could be constructed, with the aim of inspiring innovations in areas such as biomedical devices, adaptive robotics, and beyond. We highlight new pathways for integrating sensing, memory and actuation with low-power internal operations, and we discuss key challenges in realizing materials that exhibit truly "intelligent behavior". These approaches outline a path toward more robust, versatile, and scalable materials that can potentially act, compute, and "think" through their inherent intrinsic material properties-moving beyond traditional smart technologies that rely on external control.

Self-organizing neuromorphic nanowire networks as stochastic dynamical systems

Neuromorphic computing aims to develop hardware platforms that emulate the effectiveness of our brain. In this context, brain-inspired self-organizing memristive networks have been demonstrated as promising physical substrates for in materia computing. However, understanding the connection between network dynamics and information processing capabilities in these systems still represents a challenge. In this work, we show that neuromorphic nanowire network behavior can be modeled as an Ornstein-Uhlenbeck process which holistically combines stimuli-dependent deterministic trajectories and stochastic effects. This unified modeling framework, able to describe main features of network dynamics including noise and jumps, enables the investigation and quantification of the roles played by deterministic and stochastic dynamics on computing capabilities of the system in the context of physical reservoir computing. These results pave the way for the development of physical computing paradigms exploiting deterministic and stochastic dynamics in the same hardware platform in a similar way to what our brain does.

Ephaptic Coupling in Ultralow-Power Ion-Gel Nanofiber Artificial Synapses for Enhanced Working Memory

Neuromorphic devices are designed to replicate the energy-efficient information processing advantages found in biological neural networks by emulating the working mechanisms of neurons and synapses. However, most existing neuromorphic devices focus primarily on functionally mimicking biological synapses, with insufficient emphasis on ion transport mechanisms. This limitation makes it challenging to achieve the complexity and connectivity inherent in biological systems, such as ephaptic coupling. Here, an ionic biomimetic synaptic device based on a flexible ion-gel nanofiber network is proposed, which transmits information and enables ephaptic coupling through capacitance formation by ion transport with an extremely low energy consumption of just 6 femtojoules. The hysteretic ion transport behavior endows the device with synaptic-like memory effects, significantly enhancing the performance of the reservoir computing system for classifying the MNIST handwritten digit dataset and demonstrating high efficiency in edge learning. More importantly, the devices in an array establish communication connections, exhibiting global oscillatory behaviors similar to ephaptic coupling in biological neural networks. This connectivity enables the array to perform working memory tasks, paving the way for developing brain-like systems characterized by high complexity and vast connectivity.

Mimicking Axon Growth and Pruning by Photocatalytic Growth and Chemical Dissolution of Gold on Titanium Dioxide Patterns

Biological neural circuits are based on the interplay of excitatory and inhibitory events to achieve functionality. Axons form long-range information highways in neural circuits. Axon pruning, i.e., the removal of exuberant axonal connections, is essential in network remodeling. We propose the photocatalytic growth and chemical dissolution of gold lines as a building block for neuromorphic computing mimicking axon growth and pruning. We predefine photocatalytic growth areas on a surface by structuring titanium dioxide (TiO2) patterns. Placing the samples in a gold chloride (HAuCl4) precursor solution, we achieve the controlled growth of gold microstructures along the edges of the indium tin oxide (ITO)/TiO2 patterns under ultraviolet (UV) illumination. A potassium iodide (KI) solution is employed to dissolve the gold microstructures. We introduce a real-time monitoring setup based on an optical transmission microscope. We successfully observe both the growth and dissolution processes. Additionally, scanning electron microscopy (SEM) analysis confirms the morphological changes before and after dissolution, with dissolution rates closely aligned to the growth rates. These findings demonstrate the potential of this approach to emulate dynamic biological processes, paving the way for future applications in adaptive neuromorphic systems.

A self-learning magnetic Hopfield neural network with intrinsic gradient descent adaption

Physical neural networks (PNN) using physical materials and devices to mimic synapses and neurons offer an energy-efficient way to implement artificial neural networks. Yet, training PNN is difficult and heavily relies on external computing resources. An emerging concept to solve this issue is called physical self-learning that uses intrinsic physical parameters as trainable weights. Under external inputs (i.e., training data), training is achieved by the natural evolution of physical parameters that intrinsically adapt modern learning rules via an autonomous physical process, eliminating the requirements on external computation resources. Here, we demonstrate a real spintronic system that mimics Hopfield neural networks (HNN), and unsupervised learning is intrinsically performed via the evolution of the physical process. Using magnetic texture-defined conductance matrix as trainable weights, we illustrate that under external voltage inputs, the conductance matrix naturally evolves and adapts Oja's learning algorithm in a gradient descent manner. The self-learning HNN is scalable and can achieve associative memories on patterns with high similarities. The fast spin dynamics and reconfigurability of magnetic textures offer an advantageous platform toward efficient autonomous training directly in materials.

Identifying winner-takes-all emergence in random nanowire networks: an inverse problem

Random nanowire networks (NWNs) are interconnects that enable the integration of nanoscopic building blocks (the nanowires) in a disorganized fashion, enabling the study of complex emergent phenomena in nanomaterials and built-in fault-tolerant processing functionalities; the latter can lead to advances in large-scale electronic devices that can be fabricated with no particular array/grid high-precision pattern. However, when various nanowires are assembled to form an intricate network, their individual features are somehow lost in the complex NWN frame, in line with the complexity hallmark "the whole differs from the sum of the parts". Individual nanowire materials and geometrical features can only be inferred indirectly by attempting to extract information about their initial conditions from a response function measurement. In this work, we present a mathematical framework that enables inference of the intrinsic properties of highly complex/intricate systems such as random NWNs in which information about their individual parts cannot be easily accessed due to their network formation and dynamical conductance behaviour falling in the category of memristive systems. Our method, named misfit minimization, is rooted in nonlinear regression supervised learning approaches in which we find the optimum parameters that minimize a cost function defined as the square least error between conductance evolution curves taken for a target NWN system and multiple configurational NWN samples composing the training set. The optimized parameters are features referent to the target NWN system's initial conditions obtained in an inverse fashion: from the response output function, we extract information about the target system's initial conditions. Accessing the nanowire individual features in a NWN frame, as our methodology allows, enables us to predict the conduction mechanisms of the NWN subjected to a current input source; these can be via a "winner-takes-all" energy-efficient scheme using a single conduction pathway composed of multiple nanowires connected in series or via multiple parallel conduction pathways. Predicting the conduction mechanism of complex and dynamical systems such as memristive NWNs is critical for their use in next-generation memory and brain-inspired technologies since their memory capability relies on the creation of such pathways activated and consolidated by the input current signal.

Advancements in Nanowire-Based Devices for Neuromorphic Computing: A Review

Neuromorphic computing, inspired by the highly interconnected and energy-efficient way the human brain processes information, has emerged as a promising technology for post-Moore's law era. This emerging technology can emulate the structures and the functions of the human brain and is expected to overcome the fundamental limitation of the current von Neumann computing architecture. Neuromorphic devices stand out as the key components of future electronic systems, exhibiting potential in shaping the landscape of neuromorphic computing. Especially, nanowire (NW)-based neuromorphic devices, with their advantages of high integration, high-speed computing, and low power consumption, have recently emerged as candidates for neuromorphic computing technology. Here, a critical overview of the current development and relevant research in the field of NW-based neuromorphic devices are provided. Neuromorphic devices based on different NW materials are comprehensively discussed, including Ag NW-based, organic NW-based, metal oxide NW-based, and semiconductor NW-based devices. Finally, as a foresight perspective, the potentials and the challenges of these NW-based neuromorphic devices for use as future brain-like electronics are discussed.

Neuromorphic neuromodulation: Towards the next generation of closed-loop neurostimulation

Neuromodulation techniques have emerged as promising approaches for treating a wide range of neurological disorders, precisely delivering electrical stimulation to modulate abnormal neuronal activity. While leveraging the unique capabilities of AI holds immense potential for responsive neurostimulation, it appears as an extremely challenging proposition where real-time (low-latency) processing, low-power consumption, and heat constraints are limiting factors. The use of sophisticated AI-driven models for personalized neurostimulation depends on the back-telemetry of data to external systems (e.g. cloud-based medical mesosystems and ecosystems). While this can be a solution, integrating continuous learning within implantable neuromodulation devices for several applications, such as seizure prediction in epilepsy, is an open question. We believe neuromorphic architectures hold an outstanding potential to open new avenues for sophisticated on-chip analysis of neural signals and AI-driven personalized treatments. With more than three orders of magnitude reduction in the total data required for data processing and feature extraction, the high power- and memory-efficiency of neuromorphic computing to hardware-firmware co-design can be considered as the solution-in-the-making to resource-constraint implantable neuromodulation systems. This perspective introduces the concept of Neuromorphic Neuromodulation, a new breed of closed-loop responsive feedback system. It highlights its potential to revolutionize implantable brain-machine microsystems for patient-specific treatment.

Spontaneous formation of monocrystalline nanostripes in the molecular beam epitaxy of antimony triselenide

Self-assembled, highly anisotropic nanostructures are spontaneously formed in the molecular beam epitaxy of antimony triselenide on GaAs substrates. These one-dimensional (1D) nanostripes have all the orientations parallel to the substrate surface and preserve the epitaxial relationship with the substrate. The shape of the nanostripes is directly related to the highly anisotropic stibnite structure of antimony triselenide which consists of 1D ribbons held together by weak van der Waals forces. The fabrication of well-ordered arrays of horizontal nanostripes aligned in directions defined by the orientation of the substrate may contribute significantly to the development of electronic circuits and networks composed of interconnected nanostructures leading to applications in neuromorphic devices, gas sensors and polarization-sensitive photodetectors.

High-Frequency rTMS Broadly Ameliorates Working Memory and Cognitive Symptoms in Stroke Patients: A Randomized Controlled Trial

OBJECTIVE

To explore the efficacy and tolerability of high-frequency repetitive transcranial magnetic stimulation (rTMS) in the treatment of post-stroke working memory (WM) impairment and its changes in brain function.

METHODS

In the present randomized, double-blinded, sham-controlled design, 10 Hz rTMS was administered to the left dorsolateral prefrontal cortex (DLPFC) of patients with post-stroke WM impairment for 14 days. Measures included WM (primary outcome), comprehensive neuropsychological tests, and the functional near-infrared spectroscopy test. Patients were assessed at baseline, after the intervention (week 2), and 4 weeks after treatment cessation (week 6).

RESULTS

Of 123 stroke patients, 82 finished the trial. The rTMS group showed more WM improvement at week 2 (t = 5.55, P < .001) and week 6 (t = 2.11, P = .045) than the sham group. Most of the neuropsychological test scores were markedly improved in the rTMS group. In particular, the rTMS group exhibited significantly higher oxygenated hemoglobin content and significantly stronger functional connectivity in the left DLPFC, right pre-motor cortex (PMC), and right superior parietal lobule (SPL) at weeks 2 and 6. Dropout rates were equal (18% [9/50 cases] in each group), and headaches were the most common side effect (rTMS: 36% [18/50 cases]; sham: 30% [15/50 cases]).

CONCLUSIONS

High-frequency rTMS was effective in improving post-stroke WM impairment, with good tolerability, and the efficacy lasted up to 4 weeks, which may be due to the activation of the left DLPFC, right PMC, and right SPL brain regions and their synergistic enhancement of neural remodeling.

Hierarchical self-assembly of Au-nanoparticles into filaments: evolution and break

We compare the assembly of individual Au nanoparticles in a vacuum and between two Au(111) surfaces via classical molecular dynamics on a timescale of 100 ns. In a vacuum, the assembly of three nanoparticles used as seeds, initially showing decahedral, truncated octahedral and icosahedral shapes with a diameter of 1.5-1.7 nm, evolves into a spherical object with about 10-12 layers and a gyration radius ∼2.5-2.8 nm. In a vacuum, 42% show just one 5-fold symmetry axis, 33% adopt a defected icosahedral arrangement, and 25% lose all 5-fold symmetry and display a face-centred-cubic shape with several parallel stacking faults. We model a constrained version of the same assembly that takes place between two Au(111) surfaces. During the dynamics, the two Au(111) surfaces are kept fixed at distances of 55 Å, 55.5 Å, 56 Å, and 56.5 Å. The latter distance accommodates 24 Au layers with no strain, while the others correspond to nominal strains of 1.5%, 2.4%, and 3.3%, respectively. In the constrained assembly, each individual seed tends to reorganize into a layered configuration, but the filament may break. The probability of breaking the assembled nanofilament depends on the individual morphology of the seeds. It is more likely to break at the decahedron/icosahedron interface, whilst it is more likely to layer with respect to the (111) orientation when a truncated octahedron sits between the decahedron and the icosahedron. We further observe that nanofilaments between surfaces at 56 Å have a >90% probability of breaking, which decreases to 8% when the surfaces are 55 Å apart. We attribute the dramatic change in probability of breaking to the peculiar decahedron/icosahedron interface and the higher average atomic strain in the nanofilaments. This in silico experiment can shed light on the understanding and control of the formation of metallic nanowires and nanoparticle-assembled networks, which find applications in next-generation electronic devices, such as resistive random access memories and neuromorphic devices.

SiC@NiO Core-Shell Nanowire Networks-Based Optoelectronic Synapses for Neuromorphic Computing and Visual Systems at High Temperature

1D nanowire networks, sharing similarities of structure, information transfer, and computation with biological neural networks, have emerged as a promising platform for neuromorphic systems. Based on brain-like structures of 1D nanowire networks, neuromorphic synaptic devices can overcome the von Neumann bottleneck, achieving intelligent high-efficient sensing and computing function with high information processing rates and low power consumption. Here, high-temperature neuromorphic synaptic devices based on SiC@NiO core-shell nanowire networks optoelectronic memristors (NNOMs) are developed. Experimental results demonstrate that NNOMs attain synaptic short/long-term plasticity and modulation plasticity under both electrical and optical stimulation, and exhibit advanced functions such as short/long-term memory and "learning-forgetting-relearning" under optical stimulation at both room temperature and 200 °C. Based on the advanced functions under light stimulus, the constructed 5 × 3 optoelectronic synaptic array devices exhibit a stable visual memory function up to 200 °C, which can be utilized to develop artificial visual systems. Additionally, when exposed to multiple electronic or optical stimuli, the NNOMs effectively replicate the principles of Pavlovian classical conditioning, achieving visual heterologous synaptic functionality and refining neural networks. Overall, with abundant synaptic characteristics and high-temperature thermal stability, these neuromorphic synaptic devices offer a promising route for advancing neuromorphic computing and visual systems.

Interfacial-Assembly-Induced In Situ Transformation from Aligned 1D Nanowires to Quasi-2D Nanofilms

As the dimensionality of materials generally affects their characteristics, thin films composed of low-dimensional nanomaterials, such as nanowires (NWs) or nanoplates, are of great importance in modern engineering. Among various bottom-up film fabrication strategies, interfacial assembly of nanoscale building blocks holds great promise in constructing large-scale aligned thin films, leading to emergent or enhanced collective properties compared to individual building blocks. As for 1D nanostructures, the interfacial self-assembly causes the morphology orientation, effectively achieving anisotropic electrical, thermal, and optical conduction. However, issues such as defects between each nanoscale building block, crystal orientation, and homogeneity constrain the application of ordered films. The precise control of transdimensional synthesis and the formation mechanism from 1D to 2D are rarely reported. To meet this gap, we introduce an interfacial-assembly-induced interfacial synthesis strategy and successfully synthesize quasi-2D nanofilms via the oriented attachment of 1D NWs on the liquid interface. Theoretical sampling and simulation show that NWs on the liquid interface maintain their lowest interaction energy for the ordered crystal plane (110) orientation and then rearrange and attach to the quasi-2D nanofilm. This quasi-2D nanofilm shows enhanced electric conductivity and unique optical properties compared with its corresponding 1D geometry materials. Uncovering these growth pathways of the 1D-to-2D transition provides opportunities for future material design and synthesis at the interface.

Li Promoting Long Afterglow Organic Light-Emitting Transistor for Memory Optocoupler Module

The artificial brain is conceived as advanced intelligence technology, capable to emulate in-memory processes occurring in the human brain by integrating synaptic devices. Within this context, improving the functionality of synaptic transistors to increase information processing density in neuromorphic chips is a major challenge in this field. In this article, Li-ion migration promoting long afterglow organic light-emitting transistors, which display exceptional postsynaptic brightness of 7000 cd m-2 under low operational voltages of 10 V is presented. The postsynaptic current of 0.1 mA operating as a built-in threshold switch is implemented as a firing point in these devices. The setting-condition-triggered long afterglow is employed to drive the photoisomerization process of photochromic molecules that mimic neurotransmitter transfer in the human brain for realizing a key memory rule, that is, the transition from long-term memory to permanent memory. The combination of setting-condition-triggered long afterglow with photodiode amplifiers is also processed to emulate the human responding action after the setting-training process. Overall, the successful integration in neuromorphic computing comprising stimulus judgment, photon emission, transition, and encoding,  to emulate the complicated decision tree of the human brain is demonstrated.

Bi-sigmoid spike-timing dependent plasticity learning rule for magnetic tunnel junction-based SNN

In this study, we explore spintronic synapses composed of several Magnetic Tunnel Junctions (MTJs), leveraging their attractive characteristics such as endurance, nonvolatility, stochasticity, and energy efficiency for hardware implementation of unsupervised neuromorphic systems. Spiking Neural Networks (SNNs) running on dedicated hardware are suitable for edge computing and IoT devices where continuous online learning and energy efficiency are important characteristics. We focus in this work on synaptic plasticity by conducting comprehensive electrical simulations to optimize the MTJ-based synapse design and find the accurate neuronal pulses that are responsible for the Spike Timing Dependent Plasticity (STDP) behavior. Most proposals in the literature are based on hardware-independent algorithms that require the network to store the spiking history to be able to update the weights accordingly. In this work, we developed a new learning rule, the Bi-Sigmoid STDP (B2STDP), which originates from the physical properties of MTJs. This rule enables immediate synaptic plasticity based on neuronal activity, leveraging in-memory computing. Finally, the integration of this learning approach within an SNN framework leads to a 91.71% accuracy in unsupervised image classification, demonstrating the potential of MTJ-based synapses for effective online learning in hardware-implemented SNNs.

Brain topology improved spiking neural network for efficient reinforcement learning of continuous control

The brain topology highly reflects the complex cognitive functions of the biological brain after million-years of evolution. Learning from these biological topologies is a smarter and easier way to achieve brain-like intelligence with features of efficiency, robustness, and flexibility. Here we proposed a brain topology-improved spiking neural network (BT-SNN) for efficient reinforcement learning. First, hundreds of biological topologies are generated and selected as subsets of the Allen mouse brain topology with the help of the Tanimoto hierarchical clustering algorithm, which has been widely used in analyzing key features of the brain connectome. Second, a few biological constraints are used to filter out three key topology candidates, including but not limited to the proportion of node functions (e.g., sensation, memory, and motor types) and network sparsity. Third, the network topology is integrated with the hybrid numerical solver-improved leaky-integrated and fire neurons. Fourth, the algorithm is then tuned with an evolutionary algorithm named adaptive random search instead of backpropagation to guide synaptic modifications without affecting raw key features of the topology. Fifth, under the test of four animal-survival-like RL tasks (i.e., dynamic controlling in Mujoco), the BT-SNN can achieve higher scores than not only counterpart SNN using random topology but also some classical ANNs (i.e., long-short-term memory and multi-layer perception). This result indicates that the research effort of incorporating biological topology and evolutionary learning rules has much in store for the future.

Silver Nanowire Networks with Moisture-Enhanced Learning Ability

The human brain possesses a remarkable ability to memorize information with the assistance of a specific external environment. Therefore, mimicking the human brain's environment-enhanced learning abilities in artificial electronic devices is essential but remains a considerable challenge. Here, a network of Ag nanowires with a moisture-enhanced learning ability, which can mimic long-term potentiation (LTP) synaptic plasticity at an ultralow operating voltage as low as 0.01 V, is presented. To realize a moisture-enhanced learning ability and to adjust the aggregations of Ag ions, we introduced a thin polyvinylpyrrolidone (PVP) coating layer with moisture-sensitive properties to the surfaces of the Ag nanowires of Ag ions. That Ag nanowire network was shown to exhibit, in response to the humidity of its operating environment, different learning speeds during the LTP process. In high-humidity environments, the synaptic plasticity was significantly strengthened with a higher learning speed compared with that in relatively low-humidity environments. Based on experimental and simulation results, we attribute this enhancement to the higher electric mobility of the Ag ions in the water-absorbed PVP layer. Finally, we demonstrated by simulation that the moisture-enhanced synaptic plasticity enabled the device to adjust connection weights and delivery modes based on various input patterns. The recognition rate of a handwritten data set reached 94.5% with fewer epochs in a high-humidity environment. This work shows the feasibility of building our electronic device to achieve artificial adaptive learning abilities.

Electrochemical rewiring through quantum conductance effects in single metallic memristive nanowires

Memristive devices have been demonstrated to exhibit quantum conductance effects at room temperature. In these devices, a detailed understanding of the relationship between electrochemical processes and ionic dynamic underlying the formation of atomic-sized conductive filaments and corresponding electronic transport properties in the quantum regime still represents a challenge. In this work, we report on quantum conductance effects in single memristive Ag nanowires (NWs) through a combined experimental and simulation approach that combines advanced classical molecular dynamics (MD) algorithms and quantum transport simulations (DFT). This approach provides new insights on quantum conductance effects in memristive devices by unravelling the intrinsic relationship between electronic transport and atomic dynamic reconfiguration of the nanofilment, by shedding light on deviations from integer multiples of the fundamental quantum of conductance depending on peculiar dynamic trajectories of nanofilament reconfiguration and on conductance fluctuations relying on atomic rearrangement due to thermal fluctuations.

Self-Assembled Monolayer and Nanoparticles Coenhanced Fragmented Silver Nanowire Network Memristor

Silver nanowire (AgNW) networks with self-assembled structures and synaptic connectivity have been recently reported for constructing neuromorphic memristors. However, resistive switching at the cross-point junctions of the network is unstable due to locally enhanced Joule heating and the Gibbs-Thomson effect, which poses an obstacle to the integration of threshold switching and memory function in the same AgNW memristor. Here, fragmented AgNW networks combined with Ag nanoparticles (AgNPs) and mercapto self-assembled monolayers (SAMs) are devised to construct memristors with stable threshold switching and memory behavior. In the above design, the planar gaps between NW segments are for resistive switching, the AgNPs act as metal islands in the gaps to reduce threshold voltage (Vth) and holding voltage (Vhold), and the SAMs suppress surface atom diffusion to avoid Oswald ripening of the AgNPs, which improves switching stability. The fragmented NW-NP/SAM memristors not only circumvent the side effects of conventional NW-stacked junctions to provide durable threshold switching at >Vth but also exhibit synaptic characteristics such as long-term potentiation at ultralow voltage (≪Vth). The combination of NW segments, nanoparticles, and SAMs blazes a new trail for integrating artificial neurons and synapses in AgNW network memristors.

Hierarchies in Visual Pathway: Functions and Inspired Artificial Vision

The development of artificial intelligence has posed a challenge to machine vision based on conventional complementary metal-oxide semiconductor (CMOS) circuits owing to its high latency and inefficient power consumption originating from the data shuffling between memory and computation units. Gaining more insights into the function of every part of the visual pathway for visual perception can bring the capabilities of machine vision in terms of robustness and generality. Hardware acceleration of more energy-efficient and biorealistic artificial vision highly necessitates neuromorphic devices and circuits that are able to mimic the function of each part of the visual pathway. In this paper, we review the structure and function of the entire class of visual neurons from the retina to the primate visual cortex within reach (Chapter 2) are reviewed. Based on the extraction of biological principles, the recent hardware-implemented visual neurons located in different parts of the visual pathway are discussed in detail in Chapters 3 and 4. Furthermore, valuable applications of inspired artificial vision in different scenarios (Chapter 5) are provided. The functional description of the visual pathway and its inspired neuromorphic devices/circuits are expected to provide valuable insights for the design of next-generation artificial visual perception systems.

Connectome-based reservoir computing with the conn2res toolbox

The connection patterns of neural circuits form a complex network. How signaling in these circuits manifests as complex cognition and adaptive behaviour remains the central question in neuroscience. Concomitant advances in connectomics and artificial intelligence open fundamentally new opportunities to understand how connection patterns shape computational capacity in biological brain networks. Reservoir computing is a versatile paradigm that uses high-dimensional, nonlinear dynamical systems to perform computations and approximate cognitive functions. Here we present conn2res: an open-source Python toolbox for implementing biological neural networks as artificial neural networks. conn2res is modular, allowing arbitrary network architecture and dynamics to be imposed. The toolbox allows researchers to input connectomes reconstructed using multiple techniques, from tract tracing to noninvasive diffusion imaging, and to impose multiple dynamical systems, from spiking neurons to memristive dynamics. The versatility of the conn2res toolbox allows us to ask new questions at the confluence of neuroscience and artificial intelligence. By reconceptualizing function as computation, conn2res sets the stage for a more mechanistic understanding of structure-function relationships in brain networks.

Online dynamical learning and sequence memory with neuromorphic nanowire networks

Nanowire Networks (NWNs) belong to an emerging class of neuromorphic systems that exploit the unique physical properties of nanostructured materials. In addition to their neural network-like physical structure, NWNs also exhibit resistive memory switching in response to electrical inputs due to synapse-like changes in conductance at nanowire-nanowire cross-point junctions. Previous studies have demonstrated how the neuromorphic dynamics generated by NWNs can be harnessed for temporal learning tasks. This study extends these findings further by demonstrating online learning from spatiotemporal dynamical features using image classification and sequence memory recall tasks implemented on an NWN device. Applied to the MNIST handwritten digit classification task, online dynamical learning with the NWN device achieves an overall accuracy of 93.4%. Additionally, we find a correlation between the classification accuracy of individual digit classes and mutual information. The sequence memory task reveals how memory patterns embedded in the dynamical features enable online learning and recall of a spatiotemporal sequence pattern. Overall, these results provide proof-of-concept of online learning from spatiotemporal dynamics using NWNs and further elucidate how memory can enhance learning.

Tomography of memory engrams in self-organizing nanowire connectomes

Self-organizing memristive nanowire connectomes have been exploited for physical (in materia) implementation of brain-inspired computing paradigms. Despite having been shown that the emergent behavior relies on weight plasticity at single junction/synapse level and on wiring plasticity involving topological changes, a shift to multiterminal paradigms is needed to unveil dynamics at the network level. Here, we report on tomographical evidence of memory engrams (or memory traces) in nanowire connectomes, i.e., physicochemical changes in biological neural substrates supposed to endow the representation of experience stored in the brain. An experimental/modeling approach shows that spatially correlated short-term plasticity effects can turn into long-lasting engram memory patterns inherently related to network topology inhomogeneities. The ability to exploit both encoding and consolidation of information on the same physical substrate would open radically new perspectives for in materia computing, while offering to neuroscientists an alternative platform to understand the role of memory in learning and knowledge.