Testing of programmable analog neural network chips

Artificial neural network chips can achieve high-speed performance in solving complex computational problems for signal and information processing applications. These chips contain regular circuit units such as synapse matrices that interconnect linear arrays of input and output neurons. The neurons and synapses may be implemented in an analog or digital design style. Although the neural processing has some degree of fault tolerance, a significant percentage of processing defects can result in catastrophic failure of the neural network processors. Systematic testing of these arrays of circuitry is of great importance in order to assure the quality and reliability of VLSI neural network processor chips. The proposed testing method consists of parametric test and behavioral test. Two programmable analog neural chips have been designed and fabricated. The systematic approach used to test the chips is described, and measurement results on parametric test are presented.This research was partially supported by DARPA under Contract MDA 972-90-C-0037 and by National Science Foundation under Grant MIP-8904172.     

Effect of biochemical constituents on macaroni quality. I.—Differences between hard red spring and durum wheats

Durum semolina and farina from hard red spring wheat were fractionated into starch, gluten, watersoluble fraction and sludge. By systematic interchange of the various fractions, a series of reconstituted ‘flours’ was formed. Macaroni was made from these reconstituted materials and the colour and cooking characteristics of these macaroni samples were evaluated. In this way, differences in the effects of the four fractions from the two wheat types on macaroni quality were studied. Macaroni made from durum wheat had superior colour, higher cooked weight, greater residue and lower firmness score than macaroni made from hard red spring wheat. Interchange of gluten and water-soluble fractions had the most pronounced effect on macaroni colour and on macaroni cooking characteristics. Interchange of starch and sludge fractions had no effect on colour and only a small effect on cooking quality.     

On theories and applications of mixed finite element methods for linear boundary-value problems

This paper is devoted to a study of mathematical properties of certain mixed finite element approximations of linear boundary-value problems, and the application of such methods to simple representative problems which are designed to test the validity of the theory. In particular, the Oden—Reddy theory[1,2] is studied in some depth. An alternate approach to convergence questions, suggested by certain theorems of Babu ka[3], is also devised, and predictions of the two theories are briefly compared. As a result of this investigation, a number of criteria for using mixed methods in practical problems are identified, and it is shown that these criteria are supported by both theoretical arguments and by numerical experiments.     

Representation of 3D surfaces by two-variable Fourier descriptors

A method for generating two-variable 3D FDs directly from a striped lighting system is developed. An iterative algorithm is proposed to compute the two-variable 3D FDs for both axisymmetric and nonaxisymmetric objects and a formula for convergence test is derived. Experiments conducted for a set of 3D objects show that the iterative algorithm converges very quickly and the two-variable 3D FD representations are attained accurately     

Surface properties and interfacial adhesion studies of aramid fibres modified by gas plasmas

The surface of aramid (Kevlar 49) fibre was modified by NH₃, O₂ or H₂O plasmas in order to improve the adhesion to epoxy resin. Electron spectroscopy for chemical analysis (ESCA) and static secondary ion mass spectroscopy (SSIMS) were used to identify the chemical compositions and functional groups of the plasma-treated fibre surface, respectively. The surface topography and mechanical properties of plasma-treated fibres were also investigated. After plasma treatments, the interfacial shear strength (IFSS) of aramid-fibre/epoxy-resin composites, as measured by the microbond pull-out technique, was markedly improved (43–83% improvement) and the fibre strength was only little affected (less than 10% loss). Significant improvement in the IFSS principally resulted from forming the covalent bonds between the newly reactive functionalities at the modified fibre surfaces and the epoxides of the resin.     

Improved Light Extraction Efficiency in AlGaInP Light-Emitting Diodes by Applying a Periodic Texture on the Surface

In this letter, AlGaInP-GaP-based light-emitting diodes (LEDs) were fabricated with an Si substrate and an SiO₂-ITO-Ag omni-directional reflector using a metal-to-metal bonding technique. To enhance light extraction efficiency, a periodic texture was applied to the (Al_0.5Ga_0.5)_0.5In_0.5P surface layer of AlGaInP-Si LEDs by photolithography and a wet etching process. The exterior of the etched texture consists of a series of bowl-shaped recesses. With a 350-mA current injection, the typical output power of the AlGaInP-Si LEDs with and without the textured surface (LED-I and LED-II, respectively) were measured at approximately 118 and 81 mW, respectively, when the LED chips were bonded on the TO 46 without resin encapsulation. The enhancement of output power in LED-I can be attributed to a multitude of bowl-shaped notches on the surface, which resulted in a reduction of the reabsorption probability of the photons due to the fact that the photon path length in LED-I is shorter than in LED-II before the photons escape into the free space.     

Phosphor-Free GaN-Based Transverse Junction White-Light Light-Emitting Diodes With Regrown n-Type Regions

In this study, we demonstrate a GaN-based phosphor-free white-light light-emitting diode (LED), which is composed of GaN-based dual-wavelength (blue and yellow-green) multiple-quantum-wells (MQWs) and a transverse p-n junction. The device was realized by the regrowth of n-type GaN layers on the sidewall of p-type GaN and undoped MQWs. The problems related to the bias-dependent shape of the electroluminescence spectra that occur in traditional phosphor-free white-light LEDs (with vertical p-n junctions) are greatly minimized. The current-voltage performance of our device is comparable to that of the commercially available phosphor white-light LEDs. In addition, the dynamic measurement results indicate that we can attain a much higher modulation bandwidth (22 versus 3 MHz) with this device than with the currently available commercial ones.     

A Distributed Localization Scheme for Wireless Sensor Networks with Improved Grid-Scan and Vector-Based Refinement

Localization is a fundamental and essential issue for wireless sensor networks (WSNs). Existing localization algorithms can be categorized as either range-based or range-free schemes. Range-based schemes are not suitable for WSNs because of their irregularity of radio propagation and their cost of additional devices. In contrast, range-free schemes do not need to use received signal strength to estimate distances and only need simple and cheap hardware, and are thus more suitable for WSNs. However, existing range-free schemes are too costly and not accurate enough or are not scalable. To improve previous work, we present a fully distributed range-free localization scheme for WSNs. We assume that only a few sensor nodes, called anchors, know their locations, and the remaining (normal) nodes need to estimate their own locations by gathering nearby neighboring information. We propose an improved grid-scan algorithm to find the estimated locations of the normal nodes. Furthermore, we derive a vector-based refinement scheme to improve the accuracy of the estimated locations. Analysis, simulation, and experiment results show that our scheme outperforms the other range-free schemes even when the communication radius is irregular.     

Learning Association Rules and Tracking the Changing Concepts on Webpages: An Effective Pornographic Websites Filtering Approach

We applied the decision tree algorithm to learn association rules between webpage’s category (pornographic or normal) and the critical features. Based on these rules, we proposed an efficient method of filtering pornographic webpages with the following major advantages: 1) a weighted window-based technique was proposed to estimate for the condition of concept drift for the keywords found recently in pornographic webpages; 2) checking only contexts of webpages without scanning pictures; 3) an incremental learning mechanism was designed to incrementally update the pornographic keyword database.     

Strengths and weaknesses of finite element modeling deep hole drilling as compared with beam and column equations

Deep hole drilling has been studied mainly experimentally in the past. Recently, some theories using beam or column equations have been proposed, which involved complicated mathematical efforts. This work analyzed deep hole drilling by a finite element model (FEM). Results of modal analysis on the established FEM were compared with results from Euler beam equations. Further analysis showed that the FEM could also predict straightness deviation as did the column equation. In addition, FEM could analyze the effects of variable support distance which neither beam nor column equation could. Other analysis results are also presented. The results in this study showed the strength and weakness of the FEM.