Automated high level synthesis of hardware building blocks present in ART-based neural networks, from VHDL-AMS descriptions

This contribution presents a VHDL-AMS model for a building block present in a multi-channel neural network based on the adaptive resonance theory, and its automated synthesis using a VHDL-AMS to HSPICE netlist translator. This building block shows continuous dynamic behavior, and it is complex enough to check the functionality of our translator. Both simulations, the behavioral high-level one based on the VHDL-AMS model and the structural one based on an automatically synthesized SPICE description, have verified the matching between the SPICE netlist synthesized against its VHDL-AMS model.