In recent years, PC cluster systems have become widely accepted in high performance computing (HPC). Such systems are constituted by connecting PCs via a commodity high-speed network system with clustering software. NEC developed an SCore-III PC cluster, consisting of 1,024 Intel Pentium -III microprocessors connected by a Myricom Myrinet -2000 gigabit network. The system achieved a 618.3GFLOPS Linpack benchmark. Its performance ranked 57th on the TOP500 supercomputers list in June, 2002. Its system software called "SCore cluster system software" builds on the Linux \\ OS. This paper shows the key features of the SCore cluster system for HPC. These key features are high scalability, high-performance communication, a parallel programming environment, job management and checkpoint.
We have developed a new concept FCBGA (Flip-Chip Ball Grid Array) based on MLTS (Multi-Layer Thin-Substrate) packaging technology in order to meet the strong demand for high-density, high-performance, and low-cost LSI packages. The most important feature of MLTS packaging is that, only a high-density and high-performance multi-layer thin-substrate remains by removing the metal plate after mounting an LSI chip. The MLTS packaging offers the advantages of 1) good registration accuracy, which makes higher-density and finer-pitch patterning possible, 2) an ideal multi-layer structure that is highly suitable for high-speed and high-frequency applications, 3) excellent flip-chip mounting reliability, which makes higher-pin-count and finer-pitch area array flip-chip interconnection possible, 4) excellent reliability, supported by use of high T-g (Glass Transition Temperature) resin, and 5) cost-effective design achieved as a result of fewer layers fabricated with fine-pitch patterning. We successfully produced a high-performance FCBGA prototype based on our MLTS packaging technology. The prototype comprises an LSI chip connected to approximately 2,500 bonding pads arranged in a 240 mum pitch area array, and 1,296 I/O pads for BGA. The prototype FCBGA's excellent long-term reliability was demonstrated through a series of tests conducted on it.
NEC has been developing vector supercomputers since 1983 to meet the growing demand for high performance computing (HPC), especially in scientific and engineering fields. SX-6 and SX-7 were developed to meet the limitless requirement for computation. A single-chip vector CPU including eight vector pipelines and a scalar unit was developed for SX-6 and SX-7. Vector pipeline and scalar unit design will be discussed. To achieve high performance and realize high sustained performance, a combination of vector type processor and high bandwidth architecture was exploited. For more performance a multi-node system was developed. A single-stage crossbar switch was developed to achieve high bandwidth and to minimize the delay because of contention. As a result, theoretical peak performance of 64GFLOPS was achieved for 8 CPU, and up to 8TFLOPS for a multimode system. To develop the low cost and high-performance supercomputer SX-6, all functions of the vector CPU are integrated on a single die. This was achieved by 0.15mum Cu interconnection CMOS technology.
Linux is one of the operating systems which can run on NEC IPF servers such as Express5800/ 1160 and the TX7/i9000 series. These large servers have characteristics such as ccNUMA (Cache Coherent Non Uniform Memory Access); large system configuration, e.g., huge number of I/O devices; reliability and serviceability features, e.g., demand for detailed analysis of a system crash; etc. As Linux is non-proprietary and is being developed as open source under the GNU Public License, we decided to cooperate with the Linux community in order to enhance the operating system for the requirements of large servers such as Express5800/1160 or the TX7/i9000 series. In this paper, we describe the enhancements and distinctive features of the NEC Linux operating system for IPF.
Nowadays, we are able to obtain DEM (Digital Elevation Model) using airborne interferometric SAR (InSAR). However, radar shadowing and layover interfere with generation of the DEM in mountainous areas with steep terrain. To generate the complete DEM in the area, it is necessary to observe with different incident angles and directions, and to compound the DEMs obtained from difference locations. In addition, if we use a small airplane, which tends to be affected by turbulence, the generated DEM contains some errors caused by intense fuselage sway. Therefore, in case of map projection, location errors are caused, and it is very difficult to compound the DEM with accuracy. The volcanic eruption of Miyake Island in Japan, which occurred in July 2000, caused a large-scale cave-in. We observed the crater by an airborne InSAR with two different incident angles from four directions in order to generate the DEM of the inside of the caved crater. After acquisition of the data, we tried to improve the accuracy of the DEM. The method is to repeat the following process, namely motion compensation in consideration of the rough target height in the azimuth compression in time domain and the InSAR data processing. As a result, we were able to obtain comparatively accurate DEM covering the whole of Miyake Island with the inside height information of the crater.
In this paper, the concept of parallel processing on PC-cluster systems is outlined first. Then, issues of high performance PC-cluster systems are discussed, and the features of high performance PC-cluster systems, which NEC provides, are explained. In brief, they consist of high-performance and highly reliable business PC servers, which are connected by high-throughput inter-connecting networks to achieve the absolutely high performance of conventional supercomputers. As a large number of nodes are implemented in the systems, NEC fully uses its system integration technologies to minimize dead time on system introduction and utilization.
Detection of man-made phenomena appearing on the sea surface, such as ships' wakes and oil spills, is a very important issue in radar systems. Synthetic Aperture Radar (SAR) is the most suitable tool for such applications, because the radar backscatter is fully dependent on the surface conditions of the target, so it can distinguish between the surfaces of natural seawater and man-made phenomena. SAR images also include large amount of noise, especially the distinctive sea clutter from the sea surface, in comparison with optical images. This paper introduces some studies adopting different kinds of processing algorithms to optimize the detection of man-made phenomena from noisy SAR images.
Quantum Fourier Transform (QFT) is a key ingredient of a number of quantum algorithms related to Shor's factoring algorithm. We demonstrate that a quantum Fourier transform followed by measurement in computational basis can be implemented with fiber-optic devices.
We demonstrate the generation of correlated photon pairs using a four-wave-mixing process in a photonic crystal fiber. A correlated photon source based on this technique can be useful for a quantum cryptography system.
Although temperature cycling tests are performed as a technique for evaluating the long-term reliability of chip scale packages (CSP) and other electronic parts since a temperature cycling test requires a period of approximately three months to complete 1,000 cycles, from the standpoint of strengthening product development capability, shortening of the test period is in order. Accordingly, in this paper we investigated the thermal fatigue mechanisms of solder in temperature cycling tests with the goal of developing a shortened technique for temperature cycling tests. By analyzing the crystal orientation of solder using Electron BackScatter diffraction Pattern (EBSP) methods as the technique for solder joint analysis, it became clear that repetitive stress from thermal expansion and thermal shrinkage influences the Thermal Fatigue mechanism of Sn-Ag-Cu lead-free solder joints of CSP in a temperature cycling test, and that heat itself has little influence on a solder joint.
We have been researching and developing reflow and wave soldering technologies for Pb-free solder. In this research, the effect of the reheating process, which consists of both reflow and wave soldering, was investigated. Consequently, it was understood that a quad flat package (QFP) solder joint previously formed by reflow soldering delaminated at the interface in wave soldering. In order to control the delamination, the delamination mechanism was researched. As a result, it became clear that the delamination factors were Pb mixing with the lead plating, increase of temperature in the reheating process and part and/or substrate warping. Delamination occurred only when all these factors were present.
A high-density packaging technology is necessary to realize miniaturization and better performance of mobile electronics products like cellular phones and personal digital assistants. One of the most promising technologies is three-dimensional (3D)-stacked packaging technology. This paper describes a newly developed ultra high-density 3D-stacked package called a flexible carrier folded real chip size package (FFCSP). The stacked-FFCSP is fabricated by stacking very thin single chip packages (single-chip FFCSP) of real chip size on top of one another. Each single chip FFCSP consists of an LSI chip and a small piece of flexible printed circuit (FPC), which has an insulating layer made of thermoplastic resin. The advantages of the stacked-FFCSP are its small size, thinness, highly flexible assembly, and good test yield compared with conventional stacked multi-chip packages (S-MCPs). The FFCSP will enable future miniaturization and raise the functionality of the next trend of mobile electronics products.
We have developed self-extinguishing epoxy-resin compounds that contain no environmentally hazardous flame retardants such as halogen or phosphorous derivatives, and used them to develop glass epoxy laminates for printed wiring boards (PWBs). The new epoxy-resin compounds, which mainly consist of phenol-aralkyl-type epoxy resins and hardeners, formed stable foam layers that retarded heat transfer, and thus were self-extinguishing. A laminate consisting of one of the new epoxy-resin compounds with safe metal hydroxides, which absorb heat during combustion, showed high flame retardancy. We achieved excellent flame retardancy (UL94V-0) of the new laminate by adding far less aluminum hydroxide (less than 50% by weight) than in laminates containing conventional epoxy-resin compounds. Especially beneficial is the fact that the new laminate with the hydroxide exhibits good practicality and can withstand the higher temperature required for lead-free soldering. Furthermore, the laminate can be safely disposed of by incineration or reclamation, and it can also be thermally recycled.
Quantum entanglement is the most striking feature of quantum mechanics and is considered an important and valuable resource in quantum information processing. Intensive challenges to harness the power of the quantum entanglement are going on. In this paper, some of the fundamental concepts and applications of the quantum entanglement are reviewed.
The performance of a 320 x 240 bolometer-type uncooled infrared (IR) detector is described. Vanadium oxide thin film is adopted for the bolometer material, having a sheet resistance of similar to10kOmega/square. It is patterned so that the bolometer resistance is by a factor of 10 larger than the sheet resistance. This process can increase responsivity. The on-chip readout integrated circuit is designed to reduce signal drift induced by a fluctuation in operation temperature, by referring signals of pixels for which incident IR radiation is not introduced. A thermal image with an NETD (noise equivalent temperature difference) value smaller than 0.1K is shown for F/1 optics and at 60Hz frame rate.
