Evolving circuit designs and reconfigurable hardware for adaptation and optimization
Speaker: Dr Adrian Stoica, JPL, Pasadena USA
When: 2003-12-05 11:00:00
Venue: 78-622
Host: Prof. Neil Bergmann
Abstract:The holly grail of design automation would be to formulate
requirements for a device or system specifications and obtain the
design by the push of a button. For a deployable device, the vision
is to communicate to it the new requirements and have it
autonomously reconfigure/adapt to satisfy them optimally.
One of the techniques taking us a step forward toward these visions
is evolvable hardware. Evolutionary algorithms are applied as a
search/optimization technique to generate designs that gradually
move toward a design target. Evolved designs may lead to new,
patentable circuits, could offer trade-offs for various
multi-criteria optimization problems, or new configurations for a
reconfigurable device that needs to adapt in-situ at a new mission,
damage/degradation of own capability, or change in the
environment. The goal of NASA/JPL work in evolvable hardware is to
provide flexible, self-healing, adaptive and evolvable HW resources
for long-life, survivable spacecraft enabling unprecedented missions
at distant locations and in harsh environments.
The talk will introduce the main aspects of evolvable hardware, and
will present a set of examples of evolved circuits. Some of these
circuits were evolved in simulations and then were fabricated in
silicon. Other circuits were evolved directly on a reconfigurable
transistor array chip, converging to satisfactory solution in only
seconds. The talk will also present the potential implications for
designing circuits that can use components coming from a less than
perfect fabrication, with deviations in parameters and faults,
etc. as well as implications for future intelligent reconfigurable
devices.
Biography:Adrian Stoica is a Principal Member of Technical Staff in the
Biologically Inspired Technology and Systems (BITS) Group and
Manager for Computing Devices in the Space Information Systems
Technology Office at the Jet Propulsion Laboratory, California
Institute of Technology. He obtained his MSEE degree from the
Technical University of Iasi, Romania, ranking the first among the
graduates in Applied Electronics Specialty. He received his Ph.D. in
EECS from Victoria University of Technology in Melbourne, Australia,
with a thesis titled "Motion learning by robot apprentices - a fuzzy
neural approach", which was the earliest work on humanoid robot
learning by imitation. Since early 1986 his engineering and research
work revolved around adaptive and learning techniques for autonomous
intelligent systems. He joined JPL in 1996 where he has been working
along two main themes: adaptive hardware for space autonomous
systems, including evolvable/reconfigurable hardware,
adaptive/learning hardware and sensor fusion hardware, and
next-generation robots, focusing on rover intelligence and robot
sensory-motor control. His current projects in reconfigurable and
evolvable hardware addresses reconfigurable analog/digital
field-programmable computing, automated hardware/software co-design
and novel search/optimization techniques. Adrian Stoica initiated
and has chaired yearly since 1999 the NASA/DOD Conference in
Evolvable Hardware, the main conference in the field judged by
number of attendants and references to published papers. He was
Conference keynote speaker at ISMVL, ANNIE, gave tutorials at GECCO
2001 and CEC2003, and taught the first short course on Evolvable
Hardware in the summer of 2003 at UCLA Extension. He has published
over 70 papers in the areas of evolvable hardware, reconfigurable
computing, fuzzy logic, neural networks, robot learning and is
serving in the editorial board of several journals. He has received
the 1999 JPL Lew Allen Award for Excellence (highest JPL award for
excellence in research) and the Tudor Tanasescu Prize of the
Romanian Academy in 2001
Type: ITEE Seminar
Contact:Prof. Neil Bergmann, seminar host (n.bergmann@itee.uq.edu.au)
or Guido Governatori (ITEE seminar co-ordinator)
(guido@itee.uq.edu.au)