Abstract
Adaptive methods for growing electronic circuits on an imperfect synthetic matrix
Nicholas J. Macias and Lisa J.K. Durbeck
 
from BioSystems 73(3), March 2004, pp 173-2004 Copyright© 2004 Elsevier
 
We owe thanks to the organizers of the September 2002 Evolvability and Individuality Conference in St. Albans, Hertfordshire, U.K., Chrystopher L. Nehaniv, Paul Marrow, René te Boekhorst, Kerstin Dautenhahn, and Martin Loomes, for arranging this BioSystems publication of the research presented at the conference.
Living systems can adapt to injuries and even heal themselves, an ability desirable also in synthetic systems. A method is presented for dynamically adapting the construction of an electronic circuit to hardware defects by formulating the process as a series of interactions between identical but specialized structures called supercells. The circuit components, including wires, can occupy any place in the hardware that has been determined to be free of defects. The circuit specification is reduced to a connected graph, with no positional information, and provided as a code repeated in each supercell. Using the code, supercells differentiate into circuit components in a late stage of the process, with highly adaptable physical location and organization; supercells also form the wires between circuit components. The structure and function of the system at three major levels is presented, the lowest cellular level, the supercell, and the target circuit level. Adaptation of circuit construction to defective hardware was observed for this method. Results obtained from this development process on simulated and real hardware with a variety of defect types and defect patterns are presented, as well as higher level simulations of the algorithm and its response to a wider range of defect patterns, amounts of hardware, and hardware to fault ratios.