=================================================================
TITLE: Automated Synthesis of Multi-Core Processor Systems
Researchers at the Friedrich-Alexander University (FAU) in
Germany are conducting a research project to automate the
synthesis of multi-core processor systems.
The PARO (Piecewise Linear Algorithm and Architecture Research
and Optimization) project is based on a class of algorithms
called piecewise linear algorithms that can be mapped very
efficiently on massively parallel architecures. These
architecures are known as VLSI processor arrays or systolic
arrays whose principal properties are:
* high degree of parallelism
* regular structure
* local interconnect
In the area of systolic arrays, a lot of research was performed
in the eighties and nineties. Algorithms from many application
areas were implemented as systolic arrays, e.g., RADAR data
processing (Kalman filtering) or linear algebra (matrix and
vector operations).
Unfortunately, the class of systolic arrays is too rigid to allow
complex algorithms to be implemented efficiently. However,
semiconductor technology will allow the implementation of 10-100
32-bit processors on a single die within the next few years.
Mobile telephony will require several thousands Mips/Watt
efficiency in order to save battery power and allow internet
technology to become ready for mobile devices. Current DSP
(digital signal processor) technology cannot cope with these
requirements. In the above and many other areas of embedded
computing, such systems must be carefully designed in order to
meet cost, power efficiency and speed requirements. Therefore,
FAU researchers plan to provide support for the automatic
synthesis of such tightly coupled multiprocessor systems that may
be implemented on a single chip. They seek to provide solutions
to the following unsufficiently solved problem areas:
* affine mapping methodology
* design space exploration of space-time transformations
* regular hardware synthesis
Testing of the methodology will be conducted using reconfigurable
logic since ASICs are too inflexible and expensive.
For additional information, access: PARO
=================================================================