• Journal of Sensor Science and Technology
• /
• v.28 no.5
• /
• pp.284-288
• /
• 2019

A calibration-free analog neuron circuit is proposed as a viable alternative to the power hungry digital neuron in implementing a deep neural network. The conventional analog neuron requires calibrations because a voltage-mode link is used between the soma and the synapse, which results in significant uncertainty in terms of current mapping. In this work, a current-mode link is used to establish a robust link between the soma and the synapse against the variations in the process and interconnection impedances. The increased hardware owing to the adoption of the current-mode link is estimated to be manageable because the number of neurons in each layer of the neural network is typically bounded. To demonstrate the utility of the proposed analog neuron, a simple neural network with $4{\times}7{\times}3$ architecture has been designed for classifying iris flowers. The chip is now under fabrication in 0.35 mm CMOS technology. Thus, the proposed true current-mode analog neuron can be a practical option in realizing power-efficient neural networks for edge computing.

• Journal of IKEEE
• /
• v.7 no.2 s.13
• /
• pp.135-143
• /
• 2003

This paper describes the design techniques of the data converters for Multiple-Valued Logic(MVL). A 3.3V low power 4 digit CMOS analog to quaternary converter (AQC) and quaternary to analog converter (QAC) mainly designed with the neuron MOS down literal circuit block has been introduced. The neuron MOS down literal architecture allows the designed AQC and QAC to accept analog and 4 level voltage inputs, and enables the proposed circuits to have the multi-threshold properity. Low power consumption of the AQC and QAC are achieved by utilizing the proposed architecture.

• Journal of the Institute of Electronics Engineers of Korea SC
• /
• v.39 no.3
• /
• pp.201-210
• /
• 2002

This paper describes a 3.3 (V) low power 4 digit CMOS quaternary to analog converter (QAC) designed with a neuron MOS($\upsilon$MOS) down literal circuit block and cascode current mirror source block. The neuron MOS down literal architecture allows the designed QAC to accept not only 4 level voltage inputs, but also a high speed sampling rate quaternary voltage source LSB. Fast settling time and low power consumption of the QAC are achieved by utilizing the proposed architecture. The simulation results of the designed 4 digit QAC show a sampling rate of 6(MHz) and a power dissipation of 24.5 (mW) with a single power supply of 3.3 (V) for a CMOS 0.35${\mu}{\textrm}{m}$ n-well technology.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.33B no.2
• /
• pp.131-139
• /
• 1996

This paper describes the chaotic neuron model to represent the complicated states of brain and analyzes the dynamical responses of chaotic neuron such as periodic, bifurcation, and chaotic phenomena which are simulated iwth numerical analysis. Next, the chaotic neuron circuit is implemented w ith the analog electronic devices. The transfer function of chaotic neuron is given by summed the linear and nonlinear property. The output function of chaojtic neuron is designed iwth the two cMOS inverters and a feedback resistor. By adjusting the external voltage, the various dynamical properties are demonstrated. In addition, we construt the chaotic neural networks which are composed of the interconnection of chaotic neuroncircuit such as serial, paralle, and layer connection. On the board experiment, we proved the dynamci and chaotic responses which exist in the human brain.

• JSTS:Journal of Semiconductor Technology and Science
• /
• v.17 no.2
• /
• pp.174-179
• /
• 2017

We designed the CMOS analog integrate and fire (I&F) neuron circuit for driving memristor based on resistive-switching random access memory (RRAM). And we fabricated the RRAM device that have $HfO_2$ switching layer using atomic layer deposition (ALD). The RRAM device has gradual set and reset characteristics. By spice modeling of the synaptic device, we performed circuit simulation of synaptic device and CMOS neuron circuit. The neuron circuit consists of a current mirror for spatial integration, a capacitor for temporal integration, two inverters for pulse generation, a refractory part, and finally a feedback part for learning of the RRAM. We emulated the spike-timing-dependent-plasticity (STDP) characteristic that is performed automatically by pre-synaptic pulse and feedback signal of the neuron circuit. By STDP characteristics, the synaptic weight, conductance of the RRAM, is changed without additional control circuit.

• Proceedings of the KIEE Conference
• /
• 1991.11a
• /
• pp.189-192
• /
• 1991

This paper presents a modular structure design of analog Hopfield neural network. Each multiplier consists of four MOS transistors which are connected to an op-amp at the front end of a neuron. A pair of MOS transistor is used in order to maintain linear operation of the synapse and can produce positive or negative synaptic weight. This architecture can be expandable to any size neural network by forming tree structure. By altering the connections, other nework paradigms can also be implemented using this basic modules. The stength of this approach is the expandability and the general applicability. The layout design of a four-neuron fully connected feedback neural network is presented and is simulated using SPICE. The network shows correct retrival of distorted patterns.

• The Transactions of the Korean Institute of Electrical Engineers
• /
• v.38 no.10
• /
• pp.795-804
• /
• 1989

A parallel analog-to digital converter with neuron-like elements is designed and optically implemented. Its operation principle is based on the simultaneous estimation of bit values for a given analog input. The architecture of the proposed analog-to-digital converter is simpler than that of an earlier one designed by the energy minimization technique, and its digital output is independent of the initial state. Mixed binary-to-full binary converters are also designed by using out analog-to-digital converters as basic computing elements. These converters have simple structures and fast conversion times compared with earlier ones.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.31B no.4
• /
• pp.155-165
• /
• 1994

An analog neural network circuit of rhigh density integration is introduced. It's prototype chip is designed in 3 by 3 mm2 die. It uses only one MOSFET to implement a synapse. The number of synapses per neuron can be expanded by cascading several chips. The influence of nonlinearity in synapses is analyzed. A formalization of the back propagation which can be applied to this circuit is shown. Some simulation results are shown and disscussed.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.30B no.11
• /
• pp.66-73
• /
• 1993

This paper presents an analog circuit implementation and experimental resuls of the Hopfield type neural network. The proposed architecture enables the reconfiguration betwewn feedback and feedforward networks and employs new circuit designs for the weight supply and storage, analog multilier, nd current-voltage converter, in order to achieve area efficiency as well as function al versatility. The layout design of the eight-neuron neural network is tested as an associative memory to verify its applicability to real world.

• Journal of the Institute of Convergence Signal Processing
• /
• v.3 no.1
• /
• pp.80-86
• /
• 2002

A neural network structure which is able to perform the operations of analog addition and linear equation is proposed. The network employs Hopfkeld's model of a neuron with the connection elements specified on the basis of an analysis of the energy function. The analog addition network and linear equation network are designed by using Hopfield's A/D converter and linear programming respectively. Simulation using Pspice has shown convergence predominently to the correct global minima.