Simulation of neuronal membrane behavior based on memristor
Abstract
Neural networks based on memristors take advantage of the use of a recent nanoscale innovation. Memristor was initially postulated by Chua in 1971 as the fourth fundamental element of circuits and validated experimentally by an HP laboratory in 2008 (1-5). It is a device that has the peculiar characteristic of storing information in form of resistance. By changing the voltage, memristor is capable of resistive switching, a phenomenon which is essential for device functioning. Because it is a cheaper and more efficient technology in terms of occupying space, due to simplicity of its structure, the memristor may come to replace the CMOS device in the future. In addition, the entire infrastructure already used for the production of CMOS components can be adapted to obtain a memristor (6). Unlike the traditional computing paradigm, computation based on the brain model is inspired by biological mechanisms. Used in computing and neuromorphic engineering, it is composed of a large number of interconnected neural devices that work in parallel to solve a specific problem. This makes Von Neumann s bottleneck an eliminated disadvantage. Artificial neural networks have been developed by researchers based on the understanding of different brain skills, such as learning and remembering. Memristors can simulate from the memory process to the functioning of the neural membrane. Since the first neural networks were proposed, research has been divided into two areas- one aimed at simulating biological phenomena and the other directed at applications. Spiking neural networks (SNN) can be constructed using plausible biological models. In SNNs, spikes are short signals used to carry information from one neuron to others. By using these spikes, the concept of time is introduced in neural networks. Memristors with threshold switching are the ideal devices for the SNN model, which can incorporate space and time as dimensions in obtaining data (7). In this study, we intend to associate a graphene-based memristor to an RC circuit to simulate neural firing behavior, simulating a neuronal membrane. Capacitors, basically, are devices that have function of storing electric charge, generating a time-dependent charge and discharge ramp. Resistors are components that have purpose of presenting resistance to the passage of electrical current in the circuit, through its material, obeying Ohm s Law. The memristor will be the component responsible for the non-linear behavior of the circuit due to its resistive switching property, where it switches from a state of high resistance (HRS) to low resistance (LRS), or vice versa, depending on the voltage. The graphene-based memristor still has a specificity, threshold switching, in which the change of state happens abruptly. Devices with three layers will be manufactured (Figure 1)- lower and upper contacts will be obtained through deposition or evaporation of thin film coating; the insulating layer will be also deposited (dip coating, IBAD sputtering) or evaporated. Electrical characterization was done through neuronal membrane simulation. An RC circuit was built and the measurements were made using a 16-bit ADC with an input impedance of 10 MΩ, to which the three-layer memristor was associated. Two graphs were obtained- voltage as a function of time for charge and discharge ramp and current as a function of voltage for charge and discharge.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 29, 2024
- Source ID
- FA95502310305
Entities
People
- Marina Medeiros
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of São Paulo