Brain-Inspired Reservoir Computing Using Memristors with Tunable Dynamics and Short-Term Plasticity.

Recent advancements in reservoir computing (RC) research have created a demand for analogue devices with dynamics that can facilitate the physical implementation of reservoirs, promising faster information processing while consuming less energy and occupying a smaller area footprint. Studies have demonstrated that dynamic memristors, with nonlinear and short-term memory dynamics, are excellent candidates as information-processing devices or reservoirs for temporal classification and prediction tasks. Previous implementations relied on nominally identical memristors that applied the same nonlinear transformation to the input data, which is not enough to achieve a rich state space. To address this limitation, researchers either diversified the data encoding across multiple memristors or harnessed the stochastic device-to-device variability among the memristors. However, this approach requires additional preprocessing steps and leads to synchronization issues. Instead, it is preferable to encode the data once and pass them through a reservoir layer consisting of memristors with distinct dynamics. Here, we demonstrate that ion-channel-based memristors with voltage-dependent dynamics can be controllably and predictively tuned through the voltage or adjustment of the ion channel concentration to exhibit diverse dynamic properties. We show, through experiments and simulations, that reservoir layers constructed with a small number of distinct memristors exhibit significantly higher predictive and classification accuracies with a single data encoding. We found that for a second-order nonlinear dynamical system prediction task, the varied memristor reservoir experimentally achieved an impressive normalized mean square error of 1.5 × 10-3, using only five distinct memristors. Moreover, in a neural activity classification task, a reservoir of just three distinct memristors experimentally attained an accuracy of 96.5%. This work lays the foundation for next-generation physical RC systems that can exploit the complex dynamics of their diverse building blocks to achieve increased signal processing capabilities.

Author Correction: Artificial water channels enable fast and selective water permeation through water-wire networks.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

Artificial water channels enable fast and selective water permeation through water-wire networks.

Artificial water channels are synthetic molecules that aim to mimic the structural and functional features of biological water channels (aquaporins). Here we report on a cluster-forming organic nanoarchitecture, peptide-appended hybrid[4]arene (PAH[4]), as a new class of artificial water channels. Fluorescence experiments and simulations demonstrated that PAH[4]s can form, through lateral diffusion, clusters in lipid membranes that provide synergistic membrane-spanning paths for a rapid and selective water permeation through water-wire networks. Quantitative transport studies revealed that PAH[4]s can transport >109 water molecules per second per molecule, which is comparable to aquaporin water channels. The performance of these channels exceeds the upper bound limit of current desalination membranes by a factor of ~104, as illustrated by the water/NaCl permeability-selectivity trade-off curve. PAH[4]'s unique properties of a high water/solute permselectivity via cooperative water-wire formation could usher in an alternative design paradigm for permeable membrane materials in separations, energy production and barrier applications.

Memristive plasticity in artificial electrical synapses via geometrically reconfigurable, gramicidin-doped biomembranes.

It is now known that mammalian brains leverage plasticity of both chemical and electrical synapses (ES) for collocating memory and processing. Unlike chemical synapses, ES join neurons via gap junction ion channels that permit fast, threshold-independent, and bidirectional ion transport. Like chemical synapses, ES exhibit activity-dependent plasticity, which modulates the ionic conductance between neurons and, thereby, enables adaptive synchronization of action potentials. Many types of adaptive computing devices that display discrete, threshold-dependent changes in conductance have been developed, yet far less effort has been devoted to emulating the continuously variable conductance and activity-dependent plasticity of ES. Here, we describe an artificial electrical synapse (AES) that exhibits voltage-dependent, analog changes in ionic conductance at biologically relevant voltages. AES plasticity is achieved at the nanoscale by linking dynamical geometrical changes of a host lipid bilayer to ion transport via gramicidin transmembrane ion channels. As a result, the AES uniquely mimics the composition, biophysical properties, bidirectional and threshold-independent ion transport, and plasticity of ES. Through experiments and modeling, we classify our AES as a volatile memristor, where the voltage-controlled conductance is governed by reversible changes in membrane geometry and gramicidin channel density. Simulations show that AES plasticity can adaptively synchronize Hodgkin-Huxley neurons. Finally, by modulating the molecular constituents of the AES, we show that the amplitude, direction, and speed of conductance changes can be tuned. This work motivates the development and integration of ES-inspired computing devices for achieving more capable neuromorphic hardware.

Author Correction: Dynamical nonlinear memory capacitance in biomimetic membranes.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

Dynamical nonlinear memory capacitance in biomimetic membranes.

Two-terminal memory elements, or memelements, capable of co-locating signal processing and memory via history-dependent reconfigurability at the nanoscale are vital for next-generation computing materials striving to match the brain's efficiency and flexible cognitive capabilities. While memory resistors, or memristors, have been widely reported, other types of memelements remain underexplored or undiscovered. Here we report the first example of a volatile, voltage-controlled memcapacitor in which capacitive memory arises from reversible and hysteretic geometrical changes in a lipid bilayer that mimics the composition and structure of biomembranes. We demonstrate that the nonlinear dynamics and memory are governed by two implicitly-coupled, voltage-dependent state variables-membrane radius and thickness. Further, our system is capable of tuneable signal processing and learning via synapse-like, short-term capacitive plasticity. These findings will accelerate the development of low-energy, biomolecular neuromorphic memelements, which, in turn, could also serve as models to study capacitive memory and signal processing in neuronal membranes.

Assembly and Characterization of Biomolecular Memristors Consisting of Ion Channel-doped Lipid Membranes.

The ability to recreate synaptic functionalities in synthetic circuit elements is essential for neuromorphic computing systems that seek to emulate the cognitive powers of the brain with comparable efficiency and density. To date, silicon-based three-terminal transistors and two-terminal memristors have been widely used in neuromorphic circuits, in large part due to their ability to co-locate information processing and memory. Yet these devices cannot achieve the interconnectivity and complexity of the brain because they are power-hungry, fail to mimic key synaptic functionalities, and suffer from high noise and high switching voltages. To overcome these limitations, we have developed and characterized a biomolecular memristor that mimics the composition, structure, and switching characteristics of biological synapses. Here, we describe the process of assembling and characterizing biomolecular memristors consisting of a 5 nm-thick lipid bilayer formed between lipid-functionalized water droplets in oil and doped with voltage-activated alamethicin peptides. While similar assembly protocols have been used to investigate biophysical properties of droplet-supported lipid membranes and membrane-bound ion channels, this article focuses on key modifications of the droplet interface bilayer method essential for achieving consistent memristor performance. Specifically, we describe the liposome preparation process and the incorporation of alamethicin peptides in lipid bilayer membranes, and the appropriate concentrations of each constituent as well as their impact on the overall response of the memristors. We also detail the characterization process of biomolecular memristors, including measurement and analysis of memristive current-voltage relationships obtained via cyclic voltammetry, as well as short-term plasticity and learning in response to step-wise voltage pulse trains.

The voltage-dependence of MscL has dipolar and dielectric contributions and is governed by local intramembrane electric field.

Channels without canonical voltage sensors can be modulated by voltage acting on other domains. Here we show that besides protein dipoles, pore hydration can be affected by electric fields. In patches, both WT MscL and its V23T mutant show a decrease in the tension midpoint with hyperpolarization. The mutant exhibits a stronger parabolic dependence of transition energy on voltage, highly consistent with the favourable dielectric contribution from water filling the expanding pore. Purified V23T MscL in DPhPC droplet interface bilayers shows a similar voltage dependence. When reconstituted in an asymmetric DOPhPC/DPhPC bilayer carrying a permanent bias of ~130 mV due to a dipole potential difference between the interfaces, the channel behaved as if the local intramembrane electric field sets the tension threshold for gating rather than just the externally applied voltage. The data emphasize the roles of polarized water in the pore and interfacial lipid dipoles in channel gating thermodynamics.

Encapsulating Networks of Droplet Interface Bilayers in a Thermoreversible Organogel.

The development of membrane-based materials that exhibit the range and robustness of autonomic functions found in biological systems remains elusive. Droplet interface bilayers (DIBs) have been proposed as building blocks for such materials, owing to their simplicity, geometry, and capability for replicating cellular phenomena. Similar to how individual cells operate together to perform complex tasks and functions in tissues, networks of functionalized DIBs have been assembled in modular/scalable networks. Here we present the printing of different configurations of picoliter aqueous droplets in a bath of thermoreversible organogel consisting of hexadecane and SEBS triblock copolymers. The droplets are connected by means of lipid bilayers, creating a network of aqueous subcompartments capable of communicating and hosting various types of chemicals and biomolecules. Upon cooling, the encapsulating organogel solidifies to form self-supported liquid-in-gel, tissue-like materials that are robust and durable. To test the biomolecular networks, we functionalized the network with alamethicin peptides and alpha-hemolysin (αHL) channels. Both channels responded to external voltage inputs, indicating the assembly process does not damage the biomolecules. Moreover, we show that the membrane properties may be regulated through the deformation of the surrounding gel.

Memristive Ion Channel-Doped Biomembranes as Synaptic Mimics.

Solid-state neuromorphic systems based on transistors or memristors have yet to achieve the interconnectivity, performance, and energy efficiency of the brain due to excessive noise, undesirable material properties, and nonbiological switching mechanisms. Here we demonstrate that an alamethicin-doped, synthetic biomembrane exhibits memristive behavior, emulates key synaptic functions including paired-pulse facilitation and depression, and enables learning and computing. Unlike state-of-the-art devices, our two-terminal, biomolecular memristor features similar structure (biomembrane), switching mechanism (ion channels), and ionic transport modality as biological synapses while operating at considerably lower power. The reversible and volatile voltage-driven insertion of alamethicin peptides into an insulating lipid bilayer creates conductive pathways that exhibit pinched current-voltage hysteresis at potentials above their insertion threshold. Moreover, the synapse-like dynamic properties of the biomolecular memristor allow for simplified learning circuit implementations. Low-power memristive devices based on stimuli-responsive biomolecules represent a major advance toward implementation of full synaptic functionality in neuromorphic hardware.

Ferrofluid-Based Droplet Interface Bilayer Networks.

Droplet interface bilayer (DIB) networks allow for the construction of stimuli-responsive, membrane-based materials. Traditionally used for studying cellular transport phenomena, the DIB technique has proven its practicality when creating structured droplet networks. These structures consist of aqueous compartments capable of exchanging their contents across membranous barriers in a regulated fashion via embedded biomolecules, thus approximating the activity of natural cellular systems. However, lipid bilayer networks are often static and incapable of any reconfiguration in their architecture. In this study, we investigate the incorporation of a magnetic fluid or ferrofluid within the droplet phases for the creation of magnetically responsive DIB arrays. The impact of adding ferrofluid to the aqueous phases of the DIB networks is assessed by examining the bilayers' interfacial tensions, thickness, and channel activity. Once compatibility is established, potential applications of the ferrofluid-enabled DIBs are showcased by remotely modifying membrane qualities through magnetic fields. Ferrofluids do not significantly alter the bilayers' properties or functionality and can therefore be safely embedded within the DIB platform, allowing for remote manipulation of the interfacial bilayer properties.

Multifunctional, Micropipette-based Method for Incorporation And Stimulation of Bacterial Mechanosensitive Ion Channels in Droplet Interface Bilayers.

MscL, a large conductance mechanosensitive channel (MSC), is a ubiquitous osmolyte release valve that helps bacteria survive abrupt hypo-osmotic shocks. It has been discovered and rigorously studied using the patch-clamp technique for almost three decades. Its basic role of translating tension applied to the cell membrane into permeability response makes it a strong candidate to function as a mechanoelectrical transducer in artificial membrane-based biomolecular devices. Serving as building blocks to such devices, droplet interface bilayers (DIBs) can be used as a new platform for the incorporation and stimulation of MscL channels. Here, we describe a micropipette-based method to form DIBs and measure the activity of the incorporated MscL channels. This method consists of lipid-encased aqueous droplets anchored to the tips of two opposing (coaxially positioned) borosilicate glass micropipettes. When droplets are brought into contact, a lipid bilayer interface is formed. This technique offers control over the chemical composition and the size of each droplet, as well as the dimensions of the bilayer interface. Having one of the micropipettes attached to a harmonic piezoelectric actuator provides the ability to deliver a desired oscillatory stimulus. Through analysis of the shapes of the droplets during deformation, the tension created at the interface can be estimated. Using this technique, the first activity of MscL channels in a DIB system is reported. Besides MS channels, activities of other types of channels can be studied using this method, proving the multi-functionality of this platform. The method presented here enables the measurement of fundamental membrane properties, provides a greater control over the formation of symmetric and asymmetric membranes, and is an alternative way to stimulate and study mechanosensitive channels.

Activation of bacterial channel MscL in mechanically stimulated droplet interface bilayers.

MscL, a stretch-activated channel, saves bacteria experiencing hypo-osmotic shocks from lysis. Its high conductance and controllable activation makes it a strong candidate to serve as a transducer in stimuli-responsive biomolecular materials. Droplet interface bilayers (DIBs), flexible insulating scaffolds for such materials, can be used as a new platform for incorporation and activation of MscL. Here, we report the first reconstitution and activation of the low-threshold V23T mutant of MscL in a DIB as a response to axial compressions of the droplets. Gating occurs near maximum compression of both droplets where tension in the membrane is maximal. The observed 0.1-3 nS conductance levels correspond to the V23T-MscL sub-conductive and fully open states recorded in native bacterial membranes or liposomes. Geometrical analysis of droplets during compression indicates that both contact angle and total area of the water-oil interfaces contribute to the generation of tension in the bilayer. The measured expansion of the interfaces by 2.5% is predicted to generate a 4-6 mN/m tension in the bilayer, just sufficient for gating. This work clarifies the principles of interconversion between bulk and surface forces in the DIB, facilitates the measurements of fundamental membrane properties, and improves our understanding of MscL response to membrane tension.