Soft Circuitry research in the FAN Club
Overview
This research is founded upon the use of configurable logic
technology, currently provided in the form of Field-Programmable
Gate Array (FPGA) chips. The over-arching question is how
traditional computer design will change with the advent of
soft circuitry - we advocate a move from a one-dimensional
- circuit/architecture/hardware v. program/algorithm/software
co-design point of view to a three-dimensional
- circuit v. program, architecture v. algorithm and hardness
v. softness
co-design point of view. Although configurable logic is the catalyst
for this, the ideas also have relevance to the process of designing
and programming/configuring system-level integrated (SLI) chips too.
A recent publication on the overall philosophy is:
This research has application in our
flexible architecture for
networking applications research.
Underpinning technology
The work has been strongly based upon an FPGA technology rooted in
the Ph.D. work of Tom Kean at Edinburgh, which led to Algotronix Ltd
and its CAL1024 FPGA chip. Tom has a
photo-history
of this. Algotronix became part of
Xilinx Inc in 1993, and the
CAL1024 evolved into "CAL 2", the Xilinx XC6200 series.
Virtual Computer Corporation
(VCC) led the way in making the XC6200
available on a plug-in PCI bus board for PCs. Xilinx discontinued
development of the XC6200 in 1998. Our work has now moved to the
new mainstream line from Xilinx - the Virtex family. We thank Algotronix,
Xilinx and VCC for their support over the years, in the form of
pre-release and no-cost hardware and software, and other goodies.
Some of our tools for using the XC6200 board under Linux are
available for general use. Others are available on request.
Areas of investigation
Introduction
As an introduction, here are two sets of slides from the
Dagstuhl-Seminar on Dynamically Reconfigurable Architectures,
held in February 1998:
A second
Dagstuhl-Seminar on this topic was held in June 2000
- Gordon Brebner was one of the organisers.
The four sections below concerns the four main strands of
work carried out in the past, and all continuing now and in
the future. There is much still to be done.
Novel programming tools
One area investigated extensively has been systems for allowing
programs and programmers direct access to configurable logic
circuitry, enabling intimate interaction not possible when
conventional hardware design tools are used. Some publications:
We are now collaborating with Steve Guccione at Xilinx, aiming at
convergence with his JBits (formerly JERC) tools.
Virtual circuitry
A second area has been systems for supporting "virtual
circuitry" - a circuit analogue of virtual memory.
Here are some recent publications:
Circlets
The third area has been systems for supporting "circlets"
- logic circuit equivalent of program applets. Here is a
recent publication:
Novel architectures
The fourth area has been novel FPGA architectures for supporting
soft circuitry.
We have designed a self-timed FPGA technology, and simulated but not
physically implemented it. Asynchronous technologies offer benefits for
both using and managing soft circuitry, since one is liberated from
worrying about global clock signals. For much more information, see
a
collection that includes a thesis and several papers.
Future work will concern not only architectures of isolated configurable
logic arrays, but also the incorporation of such arrays into system
architectures, with novel interfaces to processor cores and memories.
System-level integrated chip design is central to this project,
as is the use of optoelectronics as an integrated system component.
Back to FAN Club page
Gordon Brebner, questions to <gordon@dcs.ed.ac.uk>,
26 March 1999