The standard multivariate techniques of hypothesis testing and discrimination analysis have been applied to detect and classify faults in a variety of linear IC designs. These techniques are potentially useful for tracking IC failures during processing or for assessing failure mechanisms of ICs once the circuits are in field use. The results indicate that the statistical methods investigated potentially yield low detection and classification error rates.>

The authors report a methodology that enhances the compression of Landsat thematic mapper (TM) multispectral imagery, while reducing the image information loss. The method first removes interband correlation of the image data by use of the Karhunen-Loeve transform (KLT) to produce the image principal components. Each principal component is spatially decorrelated using a discrete wavelet transform. The resulting coefficients are then quantized and losslessly encoded. Image compressions of typically 80:1 demonstrate that the method should be quite suitable for rapid browsing applications where small amounts of image loss are tolerable.>

Classical discrimination analysis and neural network techniques are used to detect and classify possible faults in linear microcircuits. The success rates of simulated fault detection and classification are described for various types of analog and mixed-mode circuits.>