Neural-MOS Threshold Gate as a Way to Design On-Chip Learning Neuron Structures
نویسنده
چکیده
Hardware implementation of artificial neuron networks (ANN) based on MOS-transistors with floating gates (Neuron MOS or νMOS) is discussed. Comparison of two type on-chip learning neurons with digital and analog input weight storing is provided. The main problem in design the neuron with analog input weight memory is tolerance to deviations of circuit elements parameters and supplied voltage deviation. New neuron circuit that can compensate all kind of deviations is proposed and investigated. Design methodology of such a circuit and result of simulation are shown. Key-words: Neural-MOS structures, νMOS transistors, artificial neural networks, analog voltage weight representation, deviation tolerance.
منابع مشابه
Delay Analysis of neuron-MOS and Capacitive Threshold- Logic
A model for the delay of neuron-MOS (neuMOS) and Capacitive Threshold-Logic (CTL) based logic circuits is presented for the first time. It is based on the analysis of the basic neuron-MOS [l] and CTL gate structures [a]. A closed form analytic expression for the delay of the threshold gate is derived. A relation for the delay in terms of an ordinary CMOS inverter delay expressed as a function o...
متن کاملA Proposition on Floating Gate Neuron MOS Macromodeling for Device Fabrications
A neuron MOS transistor has a floating gate and multiple input gates which are capacitively coupling with the floating gate. Dramatic reduction in the number of transistors and interconnections was achieved by employing the neuron MOS in circuit designs. Since the neuron MOS gate electrode is electrically floating, it is not necessarily easy to calculate the floating gate potential using circui...
متن کاملνMOS Enhanced Differential Current-Switch Threshold Logic Gates
This paper presents a way to enhance the Differential Current-Switch Threshold Logic gate (DCSTL) in order to allow the gate to perform more complex functions. This enhancement is achieved by replacing the MOS-transistors at the dataand threshold mapping bank of the DCSTL gate with neuronMOS transisors. First, we introduce the neuronMOS-enhanced DCSTL gate. Then, the results of HSPICE simulatio...
متن کاملAn FPGA-Based General Purpose Neural Network Chip With On-Chip Learning
In this paper, a description of a general purpose neural network chip with on-chip learning is given. The design is implemented using Xilinx Vertex II XCV 1000 Field Programmable Gate Array (FPGA). An XOR gate simulation was used as a testing application. Results and comparison of both software and hardware implementations are listed. A second testing application in noise cancellation and voice...
متن کاملFast Neural
This article is focused on implementation of neural networks in hardware. We give an overview of so-called shift-add neural arithmetics, which provides a complete set of functions suitable for fast perceptron and RBF neuron implementation. The functions use linear approximation to reach suucient simplicity. Since the linear approximation produces imprecise results, precision analysis and tests ...
متن کامل