Design of Programmable Current-Mode Cellular Neural Networks

• Main Authors: Jow-Dong Chen, 陳有棟
• Other Authors: Bin-Da Liu
• Format: Others
• Language: en_US
• Published: 1996
• Online Access: http://ndltd.ncl.edu.tw/handle/68440301270212813262

碩士 === 國立成功大學 === 電機工程研究所 === 84 === This thesis describes an analog current-mode realization of Cellular Neural Net-works (CNNs) that are programmable with continuous valued weights and operate in continuous time. At first, a Fixed-weight current-mode cellular neural network is proposed to implement the Connected Component Detection (CCD) with low supply voltage (3.3 v) architecture. Base on this work, A programmable CNN is invented. The features of our design are described as follow: 1) The use of fully differential architecture for reducing distortion by canceling even-order harmonics and increasing the dynamic range. 2) The use of a simple four-quadrant multiplier for the control and feedback coef-ficients. 3) An efficient realization of the scaling block to half the number of the multipliers. 4) The use of full signal range model to implement an efficient cell architectures. 5) The signals transmitted in the cells are current-mode and between the cells are voltage-mode. Hence, some important parts of the CNN Universal Machine concept are implement. Our system is designed to operate in two phases to reduce the hardware com-plexity. The original and modified mathematical models are both discussed . Some implementation methods of other researchers are mentioned and compared to our design. All the circuit components are described and HSPICE simulation results are given for the 0.8 mm CMOS digital process. A layout has been designed for a chip with 8 by 8 cells on a square grid realizing a one-neighborhood with 9 feedback and 9 control coefficients.