Is this a best approach- FPGA ANN

Objective : To build a neural network - lets say a trained ANN in a FPGA Available Resources : Xilinx Virtex-II Pro Hi all I have been trying to build a ANN for quite sometime. I need your comments and criticism on the methods i followed and will be following

  1. I tried to build a single neuron first ( 4 bit input ) , and i used MAC core and LUT for implementing activation function( a managable
1024 values in LUT ) Then i continued with building 8 bit neuron, but got stuck at implementing the LUT ( 2^20 values becoz the output of MAC is 20 bit number )

  1. Lately I came across Xilinx System Generator - A highlevel modelling tool targetted for DSP application integrated with MATLAB Simulink. Its kind of cool so far, I dont have to worry abt entering 2^20 values into LUT, just initialize ROM with the acitivation functino expression and implementing MAC is with a multipier and adder blocks. In fact i made it to the extend of completing the input layer of ANN with 5 neurons.( keep in mind, its a trained network so i knew the weights )

  2. hmm, hmm System generator is cool , but I have a more cool device Virtex-II pro with powerpc processor, now i have the choice of implementing few things in hardware and others in software. But not sure how hard it would be integrate components with PPC bus.

  1. Are there any xilinx cores for a neuron????????!!!

waiting on your comments Thanks Ram

Reply to
Loading thread data ...

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.