Single carbon nanotubes have been shown to produce computational effects similar to common CMOS based devices. Other nanotube configurations have also been shown to produce useful computational circuits. These devices indicate the possibility of powerful computational devices based on large assemblies of nanotubes. However, design and assembly techniques for such systems still need to be developed, and biological systems may provide good models. The technique described here uses a microtubule based assembly dynamic to allow for a controlled stochastic rearrangement of a developing structure, potentially a network of nanotubes, for creating a useful information processing transform.
Microtubules are one of three cytoskeletal components found in eukaryotic cells. They have a large family of functionality enhancing associated proteins, and assemble from sub-components into a tube, providing structural characteristics similar to those of carbon nanotubes. During microtubule assembly the mass of individual microtubules is constantly changing, but the net population produces a stable mass. This effect is known as "dynamic instability" and can be viewed as a structural search mechanism. This phenomenon also forms the basis of a new learning algorithm developed by the first author. In this algorithm dynamic instability is regulated by adaptive self-stabilization. This is essentially a feedback mechanism that uses structural utility to regulate both individual microtubule stability and the interactivity of the associated proteins with the larger developing structure.
Combining the growth dynamic with the feedback mechanism allows the overall structure to anneal as it comes closer to a solution. To further refine the algorithm an evolutionary mechanism is used that will retain an altered structure only if it is equivalent or superior in utility to the previous structure. This produces an adaptive structure that alters its configuration more slowly as it comes closer to an optimal solution, and allows it to search the state space for a more optimized structure without losing a previously reached level of functionality.
The computer model that simulates this system allows for an abstract wave dynamic to propagate through the developing structure creating a signaling mechanism. Signals are introduced and manipulated by the bound accessory proteins, allowing the whole structure to integrate signals in both space and time. This integration of signals is used to perform an information processing task that can be tested against a desired outcome to determine the system's structural utility. An abstract wave dynamic was chosen to represent any one of several possible signaling mechanisms in the natural system. It may be possible to find a wave dynamic abstraction that will include the electrical activity present in carbon nanotubes, with the activity of the associated bound proteins representing interactivity between nanotubes. Our hope is that this model will be able to inspire techniques for producing assemblies of nanotubes that can adapt to desired computational tasks.
Jeffrey O. Pfaffmann
Department of Computer Science, Wayne State University
5143 Cass, 431 State Hall, Detroit, Michigan 48202 USA
Email: email@example.com http://www.cs.wayne.edu/~kop