Hisanao Akima

Analogue spin–orbit torque device for artificial-neural-network-based associative memory operation

We demonstrate associative memory operations reminiscent of the brain using nonvolatile spintronics devices. Antiferromagnet–ferromagnet bilayer-based Hall devices, which show analogue-like spin–orbit torque switching under zero magnetic fields and behave as artificial synapses, are used. An artificial neural network is used to associate memorized patterns from their noisy versions. We develop a network consisting of a field-programmable gate array and 36 spin–orbit torque devices. An effect of learning on associative memory operations is successfully confirmed for several 3 × 3-block patterns. A discussion on the present approach for realizing spintronics-based artificial intelligence is given.

Size-dependent regulation of synchronized activity in living neuronal networks.

We study the effect of network size on synchronized activity in living neuronal networks. Dissociated cortical neurons form synaptic connections in culture and generate synchronized spontaneous activity within 10 days in vitro. Using micropatterned surfaces to extrinsically control the size of neuronal networks, we show that synchronized activity can emerge in a network as small as 12 cells. Furthermore, a detailed comparison of small (∼20 cells), medium (∼100 cells), and large (∼400 cells) networks reveal that synchronized activity becomes destabilized in the small networks. A computational modeling of neural activity is then employed to explore the underlying mechanism responsible for the size effect. We find that the generation and maintenance of the synchronized activity can be minimally described by: (1) the stochastic firing of each neuron in the network, (2) enhancement in the network activity in a positive feedback loop of excitatory synapses, and (3) Ca-dependent suppression of bursting activity. The model further shows that the decrease in total synaptic input to a neuron that drives the positive feedback amplification of correlated activity is a key factor underlying the destabilization of synchrony in smaller networks. Spontaneous neural activity plays a critical role in cortical information processing, and our work constructively clarifies an aspect of the structural basis behind this.

Majority neuron circuit having large fan-in with non-volatile synaptic weight

We present a design of a majority neuron circuit with non-volatile synaptic weights. It is based on an analog majority circuit composed of controlled current inverters (CCIs). The proposed circuit is immune to device parameter fluctuations, and its fan-in is estimated about 1000. Synaptic weights are realized on the neuron circuit by adding variable resistors. We consider a design of a non-volatile synaptic weight by using a three-terminal magnetic domain-wall motion (DWM) device. The operation of a fully connected recurrent neural network composed of the proposed circuits has been confirmed by SPICE simulation.

Design of Single Electron Circuitry for a Stochastic Logic Neural Network

Single electron devices are ultra low power and extremely small devices, and suitable for implementation of large scale integrated circuits. An artificial neural network (ANN) is one of the possible applications of single electron devices. We apply stochastic logic in which various complex operations can be done with basic logic gates. We design basic subcircuits of a single electron stochastic neural network, and confirm that backgate bias control and a redundant configuration are necessary for a feedback loop configuration by computer simulation based on Monte Carlo method. The proposed single electron circuit is well-suited for hardware implementation of a stochastic logic neural network.

Neuro-inspired quantum associative memory using adiabatic hamiltonian evolution

It is widely believed that the real parallel computation achieved by quantum computers has an enormous computing potential. In order to expand its applicable field, we have investigated the fusion of quantum and neural computations. As a first step of implementing learning function on quantum computers, we have proposed a novel quantum associative memory (QuAM) by considering an analogy between neural associative network and qubit network. The memorizing procedure of the QuAM is realized with a Hamiltonian derived from qubit-qubit interactions, and the retrieving procedure is based on the adiabatic Hamiltonian evolution. The memory capacity of the QuAM has been nominally estimated as 2N−1 where N is a number of qubits, but its retrieve property has not been discussed in our previous study. This paper proposes a retrieving process for the QuAM and evaluates its performance in detail. The results indicate that the average of the retrieving probability is over 50% even when the qubit network memorizes 2N−1 patterns and thus the QuAM is successfully implemented.

Modularity-dependent modulation of synchronized bursting activity in cultured neuronal network models

In a dissociated culture, neuronal networks spontaneously generate highly stereotypical activity characterized by synchronous bursting. With recent advancements in microfabrication technology, the topologies of cultured neuronal networks can now be engineered to have, e.g., the modular connectivity that is often found in vivo. In this paper, we construct networks of leaky integrate-and-fire neurons to theoretically investigate the effect of modular connectivity on the synchronous bursting activity of cultured neuronal networks. Modular network models are created by defining the number of modules and changing the connection formation probability within a given module, while maintaining a constant connection density. We find that the synchronized bursting frequencies in networks with the same numbers of neurons and connections are solely dependent on their modularity. We also investigate the mechanism behind the network-to-network variation of the activity in random networks, finding that local measures, such as the neuron in-degree and the self-connection, are the important factors. Out results indicate an economic advantage for networks bearing a modular structure and provides a graph-theoretical description of this mechanism.

Complexity Reduction of Neural Network Model for Local Motion Detection in Motion Stereo Vision

Spatial perception, in which objects’ motion and positional relationship are recognized, is necessary for applications such as a walking robot and an autonomous car. One of the demanding features of spatial perception in real world applications is robustness. Neural network-based approaches, in which perception results are obtained by voting among a large number of neuronal activities, seem to be promising. We focused on a neural network model for motion stereo vision proposed by Kawakami et al. In this model, local motion in each small region of the visual field, which comprises optical flow, is detected by hierarchical neural network. Implementation of this model into a VLSI is required for real-time operation with low power consumption. In this study, we reduced the computational complexity of this model and showed cell responses of the reduced model by numerical simulation.

An artificial neural network with an analogue spin-orbit torque device

Since collective spin systems store digital information as their magnetization direction, development of nonvolatile memories for computers with the von Neumann architecture is one of the mainstream outlets of spintronics research pursued in the last several decades.

Carrier properties of B atomic-layer-doped Si films grown by ECR Ar plasma-enhanced CVD without substrate heating

Abstract The atomic-layer (AL) doping technique in epitaxy has attracted attention as a low-resistive ultrathin semiconductor film as well as a two-dimensional (2-D) carrier transport system. In this paper, we report carrier properties for B AL-doped Si films with suppressed thermal diffusion. B AL-doped Si films were formed on Si(100) by B AL formation followed by Si cap layer deposition in low-energy Ar plasma-enhanced chemical-vapor deposition without substrate heating. After fabrication of Hall-effect devices with the B AL-doped Si films on unstrained and 0.8%-tensile-strained Si(100)-on-insulator substrates (maximum process temperature 350°C), carrier properties were electrically measured at room temperature. Typically for the initial B amount of 2 × 1014 cm−2 and 7 × 1014 cm−2, B concentration depth profiles showed a clear decay slope as steep as 1.3 nm/decade. Dominant carrier was a hole and the maximum sheet carrier densities as high as 4 × 1013 cm−2 and 2 × 1013 cm−2 (electrical activity ratio of about 7% and 3.5%) were measured respectively for the unstrained and 0.8%-tensile-strained Si with Hall mobility around 10–13 cm2 V−1 s−1. Moreover, mobility degradation was not observed even when sheet carrier density was increased by heat treatment at 500–700 °C. There is a possibility that the local carrier (ionized B atom) concentration around the B AL in Si reaches around 1021 cm−3 and 2-D impurity-band formation with strong Coulomb interaction is expected. The behavior of carrier properties for heat treatment at 500–700 °C implies that thermal diffusion causes broadening of the B AL in Si and decrease of local B concentration.