论文信息 - Fast prototyping of artificial neural network: GSN digital implementation

Fast prototyping of artificial neural network: GSN digital implementation

This paper describes a framework for a GSN (Goal Seeking Neuron) Boolean neural network fast prototyping into an user-programmable gate array. This system provides a VHDL language description of the trained network, allowing the direct implementation of the circuit on an academic FPGA (Field-Programmable Gate Array). A GSN software tool was designed to train and simulate a user-defined network, with diverse dimensions and applications. The implemented network presents 60 neurons in four pyramids with four layers. The short propagation time (30 ns) of the network output provides the requirements to deal with real time neural applications.

E. do Valle Simoes | L.F. Uebel | D.A. Barone

[1] Michael Fairhurst,et al. Analysis of saturation problem in RAM-based neural networks , 1992 .

[2] R. H. Freeman,et al. A 9000-gate user-programmable gate array , 1988, Proceedings of the IEEE 1988 Custom Integrated Circuits Conference.

[3] Edson Costa de Barros Carvalho Filho. Investigation of Boolean neural networks on a novel goal-seeking neuron , 1990 .

[4] Igor Aleksander,et al. A probabilistic logic neuron network for associative learning , 1989 .

[5] Michael Fairhurst,et al. A Goal Seeking Neuron for Boolean Neural Networks , 1990 .

[6] J.L. Huertas,et al. Design of Analog Neural Programmable Optimizers in CMOS VLSI Technologies , 1991, ESSCIRC '91: Proceedings - Seventeenth European Solid-State Circuits Conference.