|
Talk Abstract
Gate Array, Configure Thyself
Nicholas J. Macias, 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.
|
Reconfigurable hardware brings the flexibility of software to the hardware
domain, by allowing the behavior of physically-fixed hardware to be
modified by a post-manufacturing configuration step. This allows the costly
ASIC design and fabrication steps to be performed once, with inexpensive
software configuration subsequently used to tailor the device to a
particular need.
Unfortunately, traditional reconfigurable
devices (FPGAs) share a common characteristic of being externally
configured, meaning that configuration information is generated and
transmitted to the device by something outside the device itself. This
creates a configuration bottleneck, leading to an increase in configuration
time as the size of the FPGA grows. Furthermore, configuration operations
based on the current state of the FPGA's circuits (for example, an
architecture-tuning system or a run-time fault handling system) require
information about the FPGA's state to be transmitted to this external
controller, creating additional bottlenecks. These bottlenecks grow from
minor annoyances to serious showstoppers when considering Avogadro-scale
(10^23 gates) machines.
In this talk, the speaker will describe a new type of self-configurable
device called the Cell Matrix ™ which, unlike FPGAs, possesses an
intrinsic and very natural mechanism for configuring its own circuitry.
This and other architectural features of the Cell Matrix will be described,
along with the benefits they present, including rapid autonomous
configuration, fault tolerance, ease of manufacture, and infinite
scalability. The speaker will then describe applications of the Cell Matrix
to evolvable hardware, including the construction of autonomous,
massively-parallel genetic algorithms. The talk will conclude with a
discussion of how the Cell Matrix might be applied to the field of
artificial brain building, by employing the architecture to implement
massively-parallel, self-evolving neural networks.
|
|