In 2001, the Logic Systems Laboratory launched, together with the universities of York, Barcelona (UPC), Lausanne, and Glasgow, a project called "Reconfigurable POEtic Tissue" funded by the Future and Emerging Technologies programme (IST-FET) for the European Community. This project, which concluded in September 2004, aimed at defining a novel programmable circuit specifically designed for the implementation of systems inspired by all three axes of bioinspiration (P = phylogenesis; O = Ontogenesis; E = Epigenesis) in digital hardware. As such, the POEtic tissue is an excellent substrate for the realization of cellular systems such as the ones that are at the core of the CARG's interests.


Processor structure

Among the many original contributions of the POEtic project, two are particularly relevant for the activities of our group. The first is a definition of the general structure of a processor for bioinspired systems. Such a processor can be seen as a three-layer structure, where each layer is dedicated to the implementation of one of the axes of bio-inspiration. The bottom layer, or genotype layer, stores the genetic information of the cell (the genome). In this layer, the genetic operations associated with evolutionary approaches can be easily implemented, with the aid of an on-chip microcontroller. The middle layer, or mapping layer, implements developmental algorithms to realize processes analogous to ontogenetic growth. The top layer, or phenotype layer, is used for the actual execution of the application. In the case of a POE system, the applications to be executed are based essentially on neural networks, but in theory the architecture is versatile enough to be used for applications that are not directly bio-inspired (in fact, the structure can be adapted to implement any combination of the POE axes).


Obviously, this subdivision is purely logical, as the implementation of the processors on a 2D silicon substrate implies that the three layers are not physically disjoint. Nevertheless, the logical separation can, in many cases, reflect a structural separation of the different units of a processor and we explicitly seek this this property within the cellular processors developed for our projects.


Routing

The second contribution is more directly technological: by implementing in the circuit a dynamic routing network, it becomes unnecessary to explicitly define the connections between cells. Communication channels are set up dynamically at runtime using an address-based mechanism: a channel can be created (or destroyed) during the operation of the circuit by setting an address register to some value (stored in memory or computed by the cell) and launching the routing process. This (relatively) simple mechanism has major consequences for the implementation of ontogenetic processes, since it allows cells to be created and connected to the rest of the network (or destroyed and removed from the network) at any time during the circuit's operation and anywhere within the circuit's surface.


The impact of such a network for implementing growth and self-repair in a cellular network should be obvious and, in developing novel architectures and systems within our research activities, we generally assume the availability of a dynamic routing mechanism.


Contacts:

http://www.poetictissue.org

Key publications: