Scalable VLSI implementations for neural networks

This paper discusses research on scalable VLSI implementations of feed-forward and recurrent neural networks. These two families of networks are useful in a wide variety of important applications—classification tasks for feed-forward nets and optimization problems for recurrent nets—but their differences affect the way they should be built. We find that analog computation with digitally programmable weights works best for feed-forward networks, while stochastic processing takes advantage of the integrative nature of recurrent networks. We have shown early prototypes of these networks which compute at rates of 1–2 billion connections per second. These general-purpose neural building blocks can be coupled with an overall data transmission framework that is electronically reconfigured in a local manner to produce arbitrarily large, fault-tolerant networks.     

On the scalability of ad hoc networks

We investigate the inherent scalability problem of ad hoc networks originated from the nature of multihop networks. First, the expected packet traffic at the center of a network is analyzed. The result shows that the expected packet traffic at the center of a network is linearly related with the network size, that is, the expected packet traffic at the center of a network is O(k), where k is the radius of a network. From the result, the upper bound of the diameter of a network D=2k, that guarantees the network is scalable, is obtained. The upper bound is given by C/r-1, where C is the channel capacity available to each node and r is the packet arrival rate at each node.     

The isolation and nucleation of misfit dislocations in strained epitaxial layers grown on patterned,ion-damaged GaAs

As shown previously, the misfit dislocation density of strained epitaxial III–V layers can be significantly reduced by isolating sections (via patterned etching) of a GaAs substrate before epitaxial growth. A disadvantage of this technique is that the wafer surface is no longer planar, which can complicate subsequent device fabrication. As an alternative, we have investigated growth of 350 nm of In_0.5Ga{0.95}As by molecular beam epitaxy at two temperatures on substrates which were patterned and selectively damaged by Xe ion implantation (300 keV, 10¹⁵ cm²). Selectively etched substrates were prepared as reference samples as well. The propagation of the misfit dislocations was stopped by the ion-implanted regions of the low growth temperature (400° C) material, but the damaged portions also acted as copious nucleation sources. The resulting dislocation structure was highly anisotropic, with dislocation lines occurring in virtually only one direction. At the higher growth temperature (500° C) the defect density fell, but the ion damaged sections no longer blocked dislocation glide. Images from cathodoluminescence and transmission electron microscopy show thatthe low growth temperature material has a dislocation density of 70,000 cm^-1 in the 110 direction and less than 10,000 cm^-1 in the 110 direction. Ion channeling and x-ray diffraction show that strain is relieved in only one direction. The strain relief is consistent with the relief derived from TEM dislocation counts and Burgers vector determination. However, even this high dislocation count is not sufficient to reach the expected equilibrium strain. Reasons for the anisotropy are discussed.     

Underwater Target Tracking Using Bearing-Only Measurements

The purpose of this work is to estimate target coordinates and velocity. Target tracking is performed by an autonomous underwater vehicle (AUV). The observer is equipped with the direction finder that measures the azimuth and elevation angles of the target and the target range. Such problems are normally solved with the aid of non-linear filters due to nonlinearity of measurements. However, most filters produce estimations with unknown bias in the absence of estimations of accuracy. An important problem is related to unobservability of the target. A pseudomeasurement method is proposed to transform the problem of estimation into a linear problem. Such an approach makes possible to estimate coordinates and velocity of a moving target and solve the problem of unobservability. Results of computer simulation prove efficiency of the proposed algorithm in the problem of passive target tracking when only angles are measured.     

Sensitivity of the Human Eye to Power Frequency Electric Fields

Reports spanning more than a century have shown that tissues associated with the eye are able to respond to local electric fields with impressions of light called electrophosphenes. A finite element analysis of the field distributions in the head near the receptors shows that the 60 Hz threshold for perception of electrophosphenes is approximately 1 V/m. Thus, tissues associated with the eye appear to be among the most sensitive to power frequency electric fields in the human body.     

Automatic-repeat-request error-control schemes

ERROR DETECTION incorporated with automatic-repeatrequest (ARQ) is widely used for error control in data communications systems. This method of error control is simple and provides high system reliability. If a properly chosen code is used for error detection, virtually error-free data transmission can be attained. This paper surveys various types of ARQ and hybrid ARQ schemes, and error detection using linear block codes.     

Coherent imaging with two-dimensional focal-plane arrays: design and applications

Scanned, single-channel optical heterodyne detection has been used in a variety of lidar applications from ranging and velocity measurements to differential absorption spectroscopy. We describe the design of a coherent camera system that is based on a two-dimensional staring array of heterodyne receivers for coherent imaging applications. Experimental results with a single HgCdTe detector translated in the image plane to form a synthetic two-dimensional array demonstrate the ability to obtain passive heterodyne images of chemical vapor plumes that are invisible to normal video infrared cameras. We describe active heterodyne imaging experiments with use of focal-plane arrays that yield hard-body Doppler lidar images and also demonstrate spatial averaging to reduce speckle effects in static coherent images.     

Dual-function detector-modulator smart-pixel module

We describe a smart-pixel circuit that permits the use of a GaAs/AlGaAs multiple quantum well diode to be used both as a detector for data input and a modulator for data output. The module provides the ability to double the number of inputs or outputs to the array and is well suited to cascaded optoelectronic system architectures that require bidirectional communition.     

Microbiological and organoleptic characteristics of beef trim and ground beef treated with acetic acid, lactic acid, acidified sodium chlorite, or sterile water in a simulated commercial processing environment to reduce Escherichia coli O157:H7 and Salmonella

The objective of this study was to validate the effectiveness of acetic and lactic acids (2% and 5%), acidified sodium chlorite (1000 ppm), and sterile water in reducing Escherichia coli O157:H7 and Salmonella Typhimurium in inoculated beef trim in a simulated processing environment. Samples were collected to assess microbial characteristics at three processing points. Results from this study indicate that all treatments, including sterile water, reduced pathogen concentrations (P < 0.05) of both E. coli O157:H7 and Salmonella Typhimurium in ground beef up to 0.5 and 0.6 log by 24 h, respectively. In some cases, there were no significant differences between the antimicrobial treatments and the sterile water using this application method. Triangle sensory test results of non-inoculated beef indicated there were no differences (P < 0.05) in the means of correct responses between controls or antimicrobial treatments at 6 or 24 h. While interventions are important for beef trim, use of the interventions must be validated under industry conditions to ensure proper effectiveness.