Implementation of the conscience mechanism using single-electron transfer in competitive learning

This paper presents a single-electron tunneling (SET) based implementation of the conscience mechanism in order to improve the Kohonen learning by accelerating its convergence rate. This conscience mechanism biases the competition process so that all the processing elements can be brought into desired solutions quickly. By utilizing the electron transfer characteristics of novel single electron devices (SEDs), the proposed circuit architecture can realize the conscience mechanism in a more compact way. System-level simulations are presented to verify the effectiveness of the proposed implementation. Transistor-level simulations (including Monte Carlo simulations) are also performed with analytical models where the behavior of devices is described in Verilog modeling language. Extensive simulation results show the advantages of proposed architecture in terms of area and power cost.