Hardware implementation of an artificial neural network using fieldprogrammable gate arrays (FPGA's)

In this paper, the authors present a hardware implementation of a fully digital multilayer perceptron artificial neural network using Xilinx Field Programmable Gate Arrays (FPGAs). Each node is implemented with two XC3042 FPGAs and a 1 K×8 EPROM. Training is done offline on a PC. The authors have tested successfully the performance of the network     

A novel scheme for protection of power semiconductor devicesagainst short circuit faults

Power semiconductor devices find wide application in modern power electronic converters. Protection of these devices against overload/short circuit conditions is of paramount importance. Present day protection topologies employing different circuits have invariably one main drawback in that the fault current reaches the set value before action is initiated to trip the system. This poses a severe stress on the device. Hence an adequate safety margin has to be necessarily provided to prevent excessive device stresses and care has to be taken to see that the device is operated well within its safe operating areas. The present paper proposes a method wherein the slope or rate of rise of the fault current is detected and once the slope exceeds the set reference, action is initiated to trip the system much before the fault current reaches dangerous levels. The method provides a fast means of detection of overload and short circuit currents and can be conveniently adopted for the protection of devices in power transistor/IGBT based inverters against short circuited load conditions or shoot through faults. The possible reduction of stresses in the power devices are also highlighted     

Modeling the I-V characteristics of fully depleted submicrometerSOI MOSFET's

In this paper, an analytic current-voltage model for submicrometer fully-depleted (FD) silicon-on-insulator (SOI) MOSFET's is presented. This model takes into account the source/drain series resistances which can be especially high in thin film SOI devices. The effect of drain induced conductivity enhancement is also included, which is important for submicrometer channels. The model is verified by comparison to measured SOI I-V characteristics. Good agreement is obtained for SOI film thicknesses ranging from 40 to 220 nm and effective channel lengths down to 0.25 μm     

Suppression of boron penetration in p+ polysilicon gateP-MOSFETs using low-temperature gate-oxide N2O anneal

It has been reported that high-temperature (~1100°C) N₂ O-annealed oxide can block boron penetration from poly-Si gates to the silicon substrate. However, this high-temperature step may be inappropriate for the low thermal budgets required of deep-submicron ULSI MOSFETs. Low-temperature (900~950°C) N₂O-annealed gate oxide is also a good barrier to boron penetration. For the first time, the change in channel doping profile due to compensation of arsenic and boron ionized impurities was resolved using MOS C-V measurement techniques. It was found that the higher the nitrogen concentration incorporated at Si/SiO₂ interface, the more effective is the suppression of boron penetration. The experimental results also suggest that, for 60~110 Å gate oxides, a certain amount of nitrogen (~2.2%) incorporated near the Si/SiO₂ interface is essential to effectively prevent boron diffusing into the underlying silicon substrate     

E-BAMA vs. RAMA [multiple access protocols]

The article presents a performance comparison of two random access protocols for wireless mobile signaling in which a single channel is dedicated to the signaling function, enhanced beacon assisted multiple access (E-BAMA) and resource auction multiple access (RAMA). Data traffic is transported separately on a set of orthogonal channels. The beacon assisted multiple access (BAMA) protocol was first presented as a method of providing mobility management functions, e.g., handover, while minimizing the processing burden placed on the mobile. In BAMA, throughout the duration of its call, an active user repeatedly and quasi-periodically broadcasts a beacon containing its ID using the Aloha protocol. Quasi-periodicity prevents a pair of users from repeatedly colliding with each other. When a base successfully receives the beacon and assigns a channel, it uses a separate downstream channel to send to the mobile an acknowledgement that contains the number of the assigned channel. The BAMA protocol includes a scheme to maintain lists of active mobiles in nearby cells and to exchange periodically these lists among the base-stations. The authors evaluate the capacity and delay performance of E-BAMA and RAMA. Then, they present a numerical comparison of the parameters. Finally, the results are summarized qualitatively. Some additional derivation is included in the appendix     

The NC-REN meets the North Carolina Information Highway

The North Carolina Research and Education Network (NC-REN), formerly known as CONCERT, is an existing video and data network, owned and operated by MCNC. NC-REN's purpose has been to provide network-based support for collaboration in the research community within the state since 1985. The first major application planned for the North Carolina Information Highway (NCIH) is an interactive video distance learning system. A secondary application proposed is the use of Switched Multimegabit Data Service (SMDS) as a means of providing data communications services over NCIH. The user community currently supported by NC-REN has come to expect high levels of service, reliability, interoperability and performance for data and video communications. MCNC is committed to see that NC-REN users receive service that is equal to or better than what is currently provided by the existing network. The prospect of migrating NC-REN-provided services to NCIH-provided transport has raised challenging technical and unique service issues     

Phase II study of vinorelbine (Navelbine) in previously treated small cell lung cancer patients. EORTC Lung Cancer Cooperative Group

26 previously treated patients with progressive recurrent small cell lung cancer (SCLC) were given vinorelbine (Navelbine), 30 mg/m2 weekly. All patients had responded to first-line chemotherapy and were off therapy for at least 3 months. Partial response was observed in 4 out of 25 eligible patients (16%; 95% confidence interval 4-36%), stable disease in 7 patients and progression in 12 patients. The limiting toxicity was a non-cumulative leucopenia (80%, 32% WHO grade 3-4). Reaction at the site of injection was observed in 5 patients, causing treatment discontinuation in 2 cases. Other non-haematological toxicities were moderate. These results suggest acceptable toxicity and some antitumour activity of vinorelbine in pretreated SCLC patients.     

Analysis of interfacial failure in a composite microbundle pull-out experiment

Modeling of the failure of polymer-matrix composites requires substantial information about the mechanisms of failure at the interface, and load redistribution around fiber breaks in the composite. Current interface experiments involving the use of ‘microcomposites’ of single embedded fibers in a matrix generally do not include all the key geometric features of the real composite; in particular, they do not include the effects of fiber volume fraction and the higher matrix shear resulting from closely neighboring fibers. A new experiment was recently devised to assess some of these effects: it is referred to as the single-fiber pull-out from microbundle (SFPOM) experiment. It consists of a hexagonal arrat of seven fibers in a matrix where the outer six fibers are restrained and the center fiber is pulled out. Recent experimental data from tests with this geometry are analyzed here using three mechanical models of the failure process, and parametric studies of the data are performed to assess the appropriateness of each model. Two of the models, based on fracture energy considerations as applied earlier to single embedded fibers in a matrix and adapted to our geometry, were found to model data from the SFPOM experiments poorly. The third model assumes the existence of three zones near a fiber break, including elastic, plastic and frictional debond zones, and was found to provide reasonable fit to the data under realistic assumptions.