Dynamical analysis, applications, and analog implementation of a biologically realistic olfactory system model

The Freeman model is a biologically plausible dynamical neural network that mimics the complicated wave patterns collected from the electroencephalogram (EEG). Freeman has studied the model to explain how biological systems interact and extract information from the environment. The Freeman model has hundreds of free parameters that need to be fine tuned either manually (through trial-and-error), adaptation algorithms, or numerical simulations to match experimental data.; This work investigates the Freeman model and its dynamical building blocks as the basis for novel information processing systems based on nonlinear dynamics. We considered three issues. First, we performed a theoretical analysis of the reduced KII (RKII) basic building block, where simpler dynamics are observed. We showed, using the center manifold theorem, how the RKII dynamics are controlled by an Hopf bifurcation and designed the parameters analytically. Second, nonlinear dynamics of coupled RKII sets were studied analytically using the synchronization of output channels to define computational primitives based on specific dynamics of the network. The design of coupling coefficients in the network for obtaining a desired output response is based on our theoretical analysis of the RKII set dynamical behavior. The computational power of the RKII network is demonstrated by two applications that address logic computation and associative memory. Thirdly, we designed analog VLSI chips to implement this continuous dynamical system. Analog design faces the challenges of power consumption and limited chip area. To build a more efficient integrated circuit, the nonlinear function in the KII architecture is eliminated by redesigning the summing node. In the new design the summation block serves simultaneously as a current adder and nonlinear activation function preserving the original functionality. Chip tests show the good performance of the new design.