color=black height=1> Yahoo! is not affiliated with the authors of this page or responsible for its content.
Biology Goes Digital An array of 5,700 Spartan FPGAs
brings the BioWall to life.
Biology Goes Digital
00
Xcell Journal
Fall 2003
by Gianluca Tempesti, Ph.D.
Assistant Professor
Swiss Federal Institute of Technology
gianluca.tempesti@epfl.ch
Christof Teuscher
Research Assistant
Swiss Federal Institute of Technology
christof@teuscher.ch
Biology Goes Digital
An array of 5,700 Spartan FPGAs
brings the BioWall to life. A unique, scientific research instrument and
art piece, the BioWall models bio-inspired
electronic tissues capable of evolution, self-
repair, self-replication and learning.
Much pure and applied scientific
research has focused on replicating biologi-
cal functions in digital hardware. Here, at
the Logic Systems Laboratory of the Swiss
Federal Institute of Technology in Lausanne
(EPFL), we have utilized 5,700 Xilinx
Spartan FPGAs, in multiple configura-
tions, to build bio-inspired computing
machines that exploit three essential biolog-
ical models:
Phylogenesis the history of the
evolution of the species
Ontogenesis the development of
an individual as directed by his
genetic code
Epigenesis the development of an
individual through learning processes
(nervous system, immune system),
influenced both by genetic code
(the innate) and environment
(the acquired).
Although we have individually and jointly
investigated all three models, we have concen-
trated on the ontogenetic model through the
Embryonics (embryonic electronics) Project.
This project studies the development of
multi-cellular organisms for the purpose of
obtaining in digital hardware some of the fea-
tures of biological organisms, notably growth
and fault tolerance.
Our work has attracted a flattering
amount of interest in the most varied and
sometimes unexpected milieus. Among the
most unexpected sources of funding and
support came from Mrs. Jacqueline Reuge.
Mrs. Reuge decided to fund the construc-
tion of the BioWall to display the principles
of embryonics within a museum built to
honor the memory of her late husband.
Her generous support has allowed us
to maintain our tradition of verifying our
theoretical concepts in hardware. Without
Mrs. Reuges support, we could not have
constructed a computing machine of such
magnificent proportions.
We named this machine BioWall because
of its biological inspiration, as well as its size
effort that went into the construction of
our embryonic BioWall. However, in
developing our machine, we quickly real-
ized that the capabilities of such a plat-
form were not limited to a single
application. In fact, it is an ideal platform
to prototype many different kinds of two-
dimensional cellular systems, which are
systems comprising arrays of small, locally
connected elements.
For example, cellular automata (CA)
are very common environments in bio-
inspired research. The BioWall is ideally
suited to the implementation of CAs, but it
is by no means limited to it. We have only
begun to explore the possibilities of the
BioWall as a research tool.
Xilinx Behind the Scenes
We built the BioWall to demonstrate an
embryonic machine. The structure of such
machines is hierarchical: Organisms (appli-
cation-specific systems) are created by the
parallel operation of a number of cells
(small processors). Each cell is implement-
ed as an array of molecules (programmable
logic elements).
To implement this kind of machine, the
BioWall is structured as a two-dimensional
tissue comprising units (each unit corre-
sponds to a molecule). As shown in Figure 2,
each unit consists of:
An input element (a touch-sensitive
membrane)
An output element (an array of 64
two-color LEDs)
A programmable computing element
(a Xilinx Spartan XCS10XL FPGA).
(5.3m x 0.6m x 0.5m = 3.68m
3
, or 130
cubic feet). The main purpose of creating
the BioWall is to demonstrate the features
of our embryonics systems to the public
through a visual and tactile interaction
(Figure 1).
Bio-Inspired Machines
We implemented for the first time in
actual hardware an organism endowed
with all of the features of an embryonics
machine, as it has often been defined in the
literature. One of the functions of our
organism is the BioWatch, which counts
hours, minutes, and seconds. It demon-
strates the growth and self-repair capabili-
ties of our systems.
The implementation of the BioWatch
would have been sufficient to justify the
Fall 2003
Xcell Journal
00
Transparent
touch-sensitive
element
Two-color
LED display
FPGA
Interconnections with neighbor cells
Figure 1 The BioWall reacts to touch.
Figure 2 The BioWalls basic building block The circuits are mounted on double
boards. The logic board hosts 25 FPGAs,
and the display board holds the displays
and membranes (Figure 3). The two
boards are rigidly bound together and
connected by a bus to allow two-way com-
munication between the logic and the dis-
play (a dedicated circuit on the logic board
automatically distributes the signals to and
from the displays).
On the logic board, the Spartan devices are
placed in a regular two-dimensional grid. A
subset of the pins of each FPGA (approxi-
mately 20 per side) are used to make a direct
pin-to-pin connection between each circuit
and its four cardinal neighbors. The pins of
the FPGAs placed along the edges of the
board are brought to a set of connectors to
allow the pin-to-pin association to continue
across boards (Figure 4), thus creating per-
fectly uniform surfaces of FPGAs spanning as
many boards as required.
The remaining pins are connected to a
centralized circuit that handles the distribu-
tion of the global signals (the clocks, resets,
and FPGA configurations) arriving from
the outside.
We have built 228 such boards (not
including spare materials) for a total of
5,700 units. The architecture of the boards
implies that they can be seamlessly connect-
ed with each other to form a uniform sur-
face of any shape and size. Throughout the
development phase and lifetime of the
machine, we have so far constructed several
independent machines:
A 3,200-unit machine (Figure 5)
displayed at the Villa Reuge museum
A 2,000-unit machine kept in our labora-
tory to develop and test new applications
A 150-unit machine embedded
together with the necessary control
logic to charge applications into a
suitcase for portability
A 4,000-unit machine that will be on
display at the Telecom 03 conference
in Geneva in October 2003.
This tissue of 5,700 FPGAs represents
an impressive amount of computational
power, coupled with I/O interfaces (com-
prising the membranes and LED arrays)
that allow for large-scale visual and tactile
interaction. The advantage of this solution
is the size of the display, which enables an
immediate interaction with applications
normally limited to software simulation on
a computer screen. Furthermore, the com-
puting power and programmability of the
Xilinx FPGAs enables the prototyping of
new bio-inspired systems.
In the current version of the BioWall,
the Xilinx FPGAs can only be programmed
with the same configuration, which limits
the functionality of the units to the 10,000
equivalent logic gates of Spartan devices.
The considerable delays inherent in propa-
gating a global signal over distances meas-
ured in meters seriously limit the clock
speed. Considering the role of the BioWall
as a demonstration tool, we have not tried
to push the clock to its limits, as the cur-
rent frequency of 1 MHz is more than ade-
quate if coupled with the massive
parallelism of the machine.
Besides the I/O capabilities of the mem-
branes and LED displays, a set of modules
placed on the borders of the machine
allows the tissue to be interfaced with stan-
dard logic either via a PC or directly with
user-defined modules. The modules allow
access only to the borders of the array, but,
if necessary, signal propagation logic can be
programmed in the FPGAs.
The software tools developed for the
BioWall are rudimentary but complete. A
simple interface on a PC allows users to
define a set of files to configure the tissue.
Four kinds of files are currently defined:
the configuration file for the Xilinx
FPGAs, and three different formats used to
send user-defined data on the input pins at
the borders of the tissue (used, for example,
to provide an initial configuration for a cel-
lular automation). The values on the out-
put pins at the borders of the tissue can be
read by the PC and either stored on disk or
used as required.
Applications
The cellular structure of the BioWall is well
suited to the implementation of all sorts of
bio-inspired applications. The BioWall can
exploit the versatility inherent in its pro-
grammable logic and in its architecture to
implement hardware inspired by all the
three models of biological inspiration: phy-
logenesis, ontogenesis, and epigenesis.
BioWatch
To illustrate the implementation of the
BioWatch application on the BioWall, we
will introduce a slightly simplified exam-
ple. Whereas the complete BioWatch is an
organism capable of counting hours, min-
00
Xcell Journal