Talk Abstract
An Approach to Designing Extremely Large, Extremely Parallel Systems
Lisa J.K. Durbeck, Cell Matrix Corporation
talk given at The Conference on High Speed Computing, Salishan Lodge, Gleneden, Oregon, U.S.A., April 26 2001. Conference sponsored by Los Alamos, Lawrence Livermore, and Sandia National Laboratories.
Building computing systems on par with the complexity and parallelism of the human brain requires extremely high switch counts and massive parallelism, both at many orders of magnitude beyond what is readily available today. New technologies for building nanometer-scale switches are expected to provide the means for constructing much denser logic and thinner wires. These low-level fabrication technologies provide a mechanism for the construction of a useful Avogadro computer; that is, inexpensive, extremely dense "hardware" that makes efficient use of on order of, say, 10^23 switches to perform computations.
     Nanofabrication techniques provide the means, not the map: they do not inform the design of such a device, beyond the manufacturing and operating constraints they impose. We believe that the Cell Matrix™ computing architecture provides a design for Avogadro computers, as well as a useful platform for conducting circuit, system, and algorithm research toward Avogadro computers. The Cell Matrix hardware is a conspicuously buildable target for these new technologies. Its manufacture can be based entirely on the replication of a single unit, a cell. The simplicity of a cell further lends to buildability: it is a digital logic circuit composed of around a thousand switches, plus wires.
     This talk will characterize the Cell Matrix in terms of the issues involved in constructing a nanometer-scale computer and handling manufacturing errors. To the topic of Avogadro computers, I will focus on the ways that the architecture's distributed, local, parallel control and fine-grained configuration are conducive to the design and deployment of extremely large, massively parallel circuits and systems. I will solidify the discussion with a concrete example, providing a recently developed algorithm and circuit design with which an Avogadro-scale Cell Matrix can be used to efficiently traverse a huge search space.