A Carbon Nanotube Neuron with Dendritic Computations

Conference proceedings : ... Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual Conference, 2009

This paper describes a carbon nanotube synapse circuit that exhibits Spike-Timing Dependant Plasticity (STDP). These synapses are found in cortical (e.g. pyramidal) neurons. Experiments with the synapse in a neuron circuit demonstrate changes in synaptic potential with pre- and post-spiking timing variations. The circuit design is biomimetic and changes in control voltages representing neurotransmitter concentration lead to changes in synaptic strength. The experiments are demonstrated with SPICE simulations using carbon nanotube transistor models.

—Carbon nanotubes (CNTs) show excellent potential materials for many applications in electrical engineering and recently in neural engineering. CNTs are made of carbons arranged in hollow cylinder shape, which can form a single or multiple sheets of graphite. Due to their strength, high electrical conductivity and electrochemical stability, they have the potential to be implemented in neural system. Lately, it has been revealed that neurons can still grow in the presence of carbon nanotubes, and boost electrical signals. Cellot et al. showed that CNTs could improve the electrical signals of neurons by forming tight contacts with neuronal cell membranes in which nanotubes can create shortcuts between the soma and the dendrite of a nerve cell. A critical analysis is conducted on the influence of proposed electrical modeling on neuron action potential to find out if the proposed modeling satisfies the requirements of generating an action potential. Cellot et al., observed the occurrence of after depolarization potential (ADP) by neuron-CNTs in which they claimed to be caused through the backpropagation of action potentials to dendritic region causing the calcium channel at dendrite to release Ca 2+ ions which are responsible for the observed ADP. Analyzing the ADP phenomena and the experimental approach to evaluate the impact of CNTs interface on neuron performance were performed. Moreover, addressing the critical problems and challenges that could face the use of CNTs on the generation of nerve impulses and on the ion channels are important to be taken into consideration. Introducing CNTs in clinical studies is a major challenge since neuron-CNTs interface must undergo a real physiological environment, which turned out to be still requiring more research to be done for implementing CNTs in medical applications.

2007

A neural synapse circuit design is presented here. The circuit models the result of an action potential applied to a biological synapse, including neurotransmitter action, membrane potentials, and ion pumps. The output of the circuit is an Excitatory PostSynaptic Potential (EPSP). The circuit is simulated using carbon nanotube SPICE models.

Advanced Materials, 2013

The human brain is composed of neuronal networks connected by ∼ 10 15 synapses. The realization of a physical device with the functions of a biological synapse is essential to emulate neuronal networks. [ 1 , 2 ] Although the synaptic functions could be emulated by large-scale CMOS neuromorphic circuits, [ 3-5 ] the CMOS circuits consumed substantially more energy than a biological synapse, which make it hard to scale up the circuits to a size comparable with the brain. Electronic synapses such as fl oating-gate silicon transistors, [ 6 ] nanoparticle organic transistors, [ 7 , 8 ] resistive switching devices, [ 9-11 ] memristors, [ 12-14 ] ionic/electronic hybrid material, [ 15 ] phase change memory, [ 16-18 ] conductive-bridging (CBRAM), [ 19 ] and carbon nanotube (CNT) transistors [ 20 ] have been demonstrated in the pursuit of the synaptic functions in a single device with scalable and low-energy consumption. In this work, we report a CNT-based electronic device that can emulate the elementary dynamic logic, memory, and learning functions of the biological synapse. The CNT synapse can process spatiotemporally correlated spikes, facilitate memory with long-term potentiation (LTP) or depression (LTD), and learn with spike-timing-dependent plasticity (STDP). The functions of the biological neuronal network could be potentially emulated by an electronic circuit of the CNT synapses for speech recognition, pattern recognition, statistic inference, and other intelligent behaviors. CNT has been considered as an alternative material to replace silicon in future electronic circuits due to its nanoscale size and unique electronic properties. [ 21 ] We designed and fabricated the CNT synapse with a transistor-like structure in order to operate it with extremely low-energy consumption (Figure 1 a). 0.5 mg single-walled CNTs (Unidym Inc.) with lengths of 0.2-2.0 μ m and diameters of 0.8-1.2 nm were dispersed in 20 mL dichloromethane solvent. A randomly aligned CNT network was coated onto the surface of a silicon dioxide layer on a silicon chip by dipping the chip into the CNT solution 20 times. The transistor channel was made from a strip of randomly distributed single-walled CNT network with a width of 8 μ m and a www.advmat.de www.MaterialsViews.com

IEEE Transactions on Circuits and Systems I: Regular Papers, 2000

We present an original method to implement neuro-inspired supervised learning for a synaptic array based on carbon nanotube devices. The device characteristics required to implement on chip learning within a crossbar of carbon nanotube field effect transistors (CNTFETs) as synaptic arrays were experimentally demonstrated and accurately modeled through a specific electrical compact model. We performed electrical simulations of learning for an array of 24 nanotube memory devices corresponding to a 3 input 3 output neural layer that revealed successful learning of separable logic functions within very few epochs, even when a realistic variability of nanotube diameter was taken into account. Such a learning approach opens the way to the use of high-density synaptic arrays as generic logic blocks in configurable circuits.

Small, 2014

amplitudes are usually less than 1 nA, which enables the spikes to be processed with extremely low power consumption in a neural network. Due to the synaptic plasticity, the amplitude of a PSC triggered via a synapse can be increased or decreased by the spikes, which establishes the foundation for learning and memory in a neural network. Silicon (Si)-based circuits have been utilized to emulate neural networks, [ 1,3-5 ] but the Si circuits consumed considerably more energy than a biological network and were unable to be integrated at a scale comparable with the biological neural network. [ 6-11 ] There have been extensive attempts to emulate the functions of synapses and neurons utilizing various electronic devices, such as fl oating gate silicon transistors, [ 12 ] nanoparticle organic transistors, [ 13 ] resistive switches, [ 14 ] memristors, [ 15,16 ] phase change memory, [ 17 ] and carbon nanotube (CNT) transistors. [ 2,18-20 ] However, the devices lack the analog memory, [ 12-14,21 ] plasticity, [ 10 ] or the function for spikes to trigger PSCs for spike signal processing. [ 6,15,17 ] Spikes can trigger PSCs in other synaptic devices, but the PSC amplitudes range between 10 −3-10 −6 A, [ 14,15,21 ] resulting in signifi cantly higher power consumptions of the devices than those of biological synapses. Due to their nanoscale diameter, high carrier mobility, and sensitivity to environmental changes, CNTs have been extensively explored as an alternative to Si in nano-electronic devices. [ 22 ] Large-scale devices

Beilstein journal of nanotechnology, 2014

The growth of cortical neurons on three dimensional structures of spatially defined (structured) randomly oriented, as well as on vertically aligned, carbon nanotubes (CNT) is studied. Cortical neurons are attracted towards both types of CNT nano-architectures. For both, neurons form clusters in close vicinity to the CNT structures whereupon the randomly oriented CNTs are more closely colonised than the CNT pillars. Neurons develop communication paths via neurites on both nanoarchitectures. These neuron cells attach preferentially on the CNT sidewalls of the vertically aligned CNT architecture instead than onto the tips of the individual CNT pillars.

ACS nano, 2018

This paper presents aligned carbon nanotube (CNT) synaptic transistors for large-scale neuromorphic computing systems. The synaptic behavior of these devices is achieved via charge-trapping effects, commonly observed in carbon-based nanoelectronics. In this work, charge trapping in the high- k dielectric layer of top-gated CNT field-effect transistors (FETs) enables the gradual analog programmability of the CNT channel conductance with a large dynamic range ( i. e., large on/off ratio). Aligned CNT synaptic devices present significant improvements over conventional memristor technologies ( e. g., RRAM), which suffer from abrupt transitions in the conductance modulation and/or a small dynamic range. Here, we demonstrate exceptional uniformity of aligned CNT FET synaptic behavior, as well as significant robustness and nonvolatility via pulsed experiments, establishing their suitability for neural network implementations. Additionally, this technology is based on a wafer-level techniqu...

Nature communication, 2018

In contrast to AI hardware, neuromorphic hardware is based on neuroscience, wherein constructing both spiking neurons and their dense and complex networks is essential to obtain intelligent abilities. However, the integration density of present neuromorphic devices is much less than that of human brains. In this report, we present molecular neuromorphic devices, composed of a dynamic and extremely dense network of single-walled carbon nanotubes (SWNTs) complexed with polyoxometalate (POM). We show experimentally that the SWNT/POM network generates spontaneous spikes and noise. We propose electron-cascading models of the network consisting of heterogeneous molecular junctions that yields results in good agreement with the experimental results. Rudimentary learning ability of the network is illustrated by introducing reservoir computing, which utilises spiking dynamics and a certain degree of network complexity. These results indicate the possibility that complex functional networks can be constructed using molecular devices, and contribute to the development of neuromorphic devices.

Modeling the dendritic arbor of a cortical neuron is important because each dendritic arbor contains thousands of synapses along with dendritic computations on the signals generated by the synapses in the arbor. Nanometer-scale CMOS circuits, important for the scale of future synthetic neural systems, also raise the inherent issue of power consumption. This paper explores the cost of implementing basic dendritic arbors in terms of area and power consumption. The dendritic computational components used are capable of linear and nonlinear summations and are simulated in SPICE to show biomimetic capability. A comparison of possible components is presented to illustrate the possibilities of implementation.