Training the Nanocell Through Neuro-Dynamic Programming

Chris Husband
Rice University

The nanocell is an approach to molecular computing in which small
areas of bi-state molecules and metallic nanoparticles are self-assembled
into a random network and are addressed by a small number of relatively
large I/O leads. Applying voltages across these leads elicits a change in
the conductivity of groups of molecules within the nanocell with respect to
the voltage drop across each molecule. As fabricated, the nanocell is not
expected to perform any useful logic function. Therefore, the challenge is
to program, or train, the nanocell post-fabrication by applying these
voltage pulses repeatedly in order to change the internal molecular
conducting states until the nanocell performs the desired logic function.
This post-fabrication training can be viewed as an optimal control problem
with the voltages pulses as the controls. We use neuro-dynamic programming,
NDP, to solve this optimal control problem. Preliminary results show that
NDP is able to improve training times significantly over a random walk

Back to NANO2002 Workshop I: Alternative Computing