Carnegie Mellon's Reconfigurable Computer Project addresses the
two most significant problems with current reconfigurable computing
systems:
- traditional FPGAs have hard resource constraints, which makes
it difficult for a compilation tool to consistently and easily
generate applications, and
- there is no mechanism to provide forward-compatibility, causing
the investment in generating applications to be lost for future
generations of silicon.
This project addresses these limitations by virtualizing hardware, which allows a hardware design of any size to execute on a compatible
device with any capacity. Hardware virtualization is accomplished
through extremely high-speed reconfiguration. We are developing
an architecture, called PipeRench, that provides the high-speed reconfiguration necessary for hardware
virtualization, compilation tools for this architecture, and applications
that will demonstrate
both high-performance
and
forward-compatibility.
NEW!
The PipeRench chip is out and
has passed all the tests.
All our DIL programs go
through the entire tool-chain
and execute correctly!
We are expecting the PCI
boards soon.
|