1-7hit |
High speed simulation of neural networks can be achieved through parallel implementations capable of exploiting their massive inherent parallelism. In this paper, we show how this inherent parallelism can be effectively exploited on parallel data-driven systems. By using these systems, the asynchronous parallelism of neural networks can be naturally specified by the functional data-driven programs, and maximally exploited by pipelined and scalable data-driven processors. We shall demonstrate the suitability of data-driven systems for the parallel simulation of neural networks through a parallel implementation of the widely used back propagation networks. The implementation is based on the exploitation of the network and training set parallelisms inherent in these networks, and is evaluated using an image data compression network.
Hiroaki TERADA Kenichi KAGOSHIMA
This paper presents a view on coming photonic network in which machines are potential customer to the network. The network will be providing unlimited number of virtual free spaces in which point to point and broadcasting modes of information interchanges are taking place simultaneously. It is also pointed out that the Asynchronous Transfer Mode (ATM) should be evolved to support this type of network by using true photonic switching technology.
Fumiyasu ASAI Shinji KOMORI Toshiyuki TAMURA Hisakazu SATO Hidehiro TAKATA Yoshihiro SEGUCHI Takeshi TOKUDA Hiroaki TERADA
This paper details a unique VLSI design scheme which employs self-timed circuits. A 32-bit 50-MFLOPS data-driven microprocessor has been designed using a self-timed clocking scheme. This high performance data-driven microprocessor with sophisticated functions has been designed by a combination of several kinds of self-timed components. All functional blocks in the microprocessor are driven by self-timed clocks. The microprocessor integrates 700,000 devices in a 14.65 mm14.65 mm die area using double polysilicon double metal 0.8 µm CMOS technology.
Kazuhiro KOMATSU Shuji SANNOMIYA Makoto IWATA Hiroaki TERADA Suguru KAMEDA Kazuo TSUBOUCHI
The self-timed pipeline (STP) is one of the most promising VLSI/SoC architectures. It achieves efficient utilization of tens of billions of transistors, consumes ultra low power, and is easy-to-design because of its signal integrity and low electro-magnetic interference. These basic features of the STP have been proven by the development of self-timed data-driven multimedia processors, DDMP's. This paper proposes a novel scheme of interacting self-timed (clockless) pipelines by which the various distributed and interconnected pipelines can achieve highly functional stream processing in future giga-transistor chips. The paper also proposes a set of elementary coupling control modules that facilitate various combinations of flow-thru processing between pipelines, and then discusses the practicality of the proposed scheme through the LSI design of application modules such as a priority-based queue, a mutual interconnection network, and a pipelined sorter.
Kazuhiko KINOSHITA Tetsuya TAKINE Koso MURAKAMI Hiroaki TERADA
We propose a new network architecture nemed Holonic Network for personalized multimedia communications, which is characterized by distributed cooperative networking based on autonomous management and all-optical transport networks. We than propose autonomous routing method. Moreover, an information searching method and a route generation method with network maps, which are essential for this network, are proposed. Lastly, we evaluate the proposed network performance by theoretical analysis and system emulation.
The data-driven model of computation is well suited for flexible and highly parallel simulation of neural networks. First, the operational semantics of data-driven languages preserve the locality and functionality of neural networks, and naturally describe their inherent parallelism. Second, the asynchronous data-driven execution facilitates the implementation of large and scalable multiprocessor systems, which are necessary to obtain considerable degrees of simulation sppedups. In this paper, we present a dynamic data-driven multiprocessor system, and demonstrate its suitability for the paralel simulation of back propagation neural networks. Two parallel implementations are described and evaluated using an image data compression network. The system is scalable, and as a result, the performance improved proportionally with the increase in number of processors.