A Programmable Multi-Dimensional Analog Radial-Basis- Function-Based Classifier

A compact analog programmable multidimensional radialbasis-function (RBF)-based classifier is demonstrated in this chapter. The probability distribution of each feature in the templates is modeled by a Gaussian function that is approximately realized by the bell-shaped transfer characteristics of a proposed floating-gate bump circuit. The maximum likelihood, the mean, and the variance of the distribution are stored in floating-gate transistors and are independently programmable. By cascading these floating-gate bump circuits, the overall transfer characteristics approximate a multivariate Gaussian function with a diagonal covariance matrix. An array of these circuits constitute a compact multidimensional RBF-based classifier that can easily implement a Gaussian mixture model. When followed by a winner-take-all circuit, the RBFbased classifier forms an analog vector quantizer. Receiver operating characteristic curves and equal error rate are used to evaluate the performance of the RBF-based classifier as well as a resultant analog vector quantizer. It is shown that the classifier performance is comparable to that of digital counterparts. The proposed approach can be at least two orders of magnitude more power efficient than the digital microprocessors at the same task. 1 Motivations for Analog RBF Classifier The aggressive scaling of silicon technologies has led to transistors and many sensors becoming faster and smaller. The trend toward integrating sensors, interface circuits, and microprocessors into a single package or into a single chip is more and more prevalent. Fig. 1A illustrates the block diagram of a typical microsystem, which receives analog inputs via sensors and performs classification, decision-making, or, in a more general term, information-refinement tasks in the digital domain. Although fabrication and packaging technologies enable an unprecedented number of components to be packed into a small volume, the accompanying power density can be higher than ever, which has become one of the bottle-neck factors in the microsystem development. If the informationrefinement tasks can be performed in the analog domain with less power consumption, the specifications for the analog-to-digital-converters, which are usually power-hungry, can be relaxed. In some cases, analog-to-digital conversion can 2 S.-Y. Peng, P. E. Hasler, D. V. Anderson Analog Cepstrum Generator Analog RBF Classifier