Title :
Cort-X II: The Low-Power Element Design for a Dynamic Neural Network
Author :
Yuan, Jie ; Song, Ning ; Farhat, Nabil ; Van der Spiegel, Jan
Author_Institution :
Hong Kong Univ. of Sci. & Technol., Hong Kong
Abstract :
The parametrically coupled logistic map network (PCLMN) can serve as the front-end dynamic neural network (DNNs) for clustering and generation of spatio-temporal patterns. In this brief, the element of the PCLMN is designed in a 0.25-mum 2.5-V CMOS process for low power consumption. The analog design employs self-calibration techniques to improve the accuracy of the low-power element. After calibration, the fabricated element is able to generate a 1-D map and the nonlinear interconnection in 4-bit resolution for driving signals up to 1 MHz, with a power consumption of 12 mW.
Keywords :
CMOS analogue integrated circuits; low-power electronics; neural chips; pattern clustering; CMOS process; DNN; PCLMN; dynamic neural network; front-end dynamic neural network; low-power element design; parametrically coupled logistic map network; power 12 mW; size 0.25 micron; voltage 2.5 V; Analog artificial neural networks (ANNs); Cort-X; low-power nonlinear circuits; parametrically coupled logistic map network (PCLMN); spatio-temporal (ST) pattern clustering;
Journal_Title :
Circuits and Systems II: Express Briefs, IEEE Transactions on
DOI :
10.1109/TCSII.2007.905882