Abstract  Natural  neural  networks  greatly  benefit  from  their  parallel  structure  that  makes  them  fault  tolerant  and  fast  in  processing the inputs. Their artificial counterpart , artificial neural networks, proved difficult to implement in hardware where  they  could  have  a  similar  structure.  Although,  many   circuits  have  been  developed,  they  usually  present  problems  regarding  accuracy, are application specific, difficult to pr oduce and difficult to adapt to new applications. I t is expected that developing  a software tool that allows automatic generation of  neural hardware while using high accuracy solves t his problem and make  artificial  neural  networks  a  step  closer  to  the  nat ural  version.  This  paper  presents  a  tool  to  respond   to  this  need:  A  software  tool  for  automatic  generation  of  neural  hardware.  T he  software  gives  the  user  freedom  to  specify  the  number  of  bits  used  in  each  part  of  the  neural  network  and  programs  the  se lected  FPGA  with  the  network.  The  paper  also  presen ts  tests  to  evaluate  the accuracy of the implementation of an automatica lly built neural network against Matlab.   

References

[1] Aguiar L., Reis L., and Morgado-Dias F., Neuron Implementation Using System Generator, in Proceedings of the 9 th Portuguese Conference on Automatic Control , Portugal, 2010.

[2] Arroyo M., Ruiz A., and Leal R., An Artificial Neural Network on a Field Programmable Gate Array as a Virtual Sensor, in Proceedings of the Third International Workshop on Design of Mixed-Mode Integrated Circuits and Applications , Mexico, pp. 114-117, 1999.

[3] Ayala J., Lome a A., L pez-Vallejo M., and Fern ndez A., Design of a Pipelined Hardware Architecture for Real-Time Neural Network Computations, in Proceedings of IEEE Midwest Symposium on Circuits and Systems , USA, pp. 419-422, 2002.

[4] Ferreira P., Ribeiro P., Antunes A., and Morgado- Dias F., A High Bit Resolution FPGA Implementation of a FNN with a New Algorithm for the Activation Function, Neurocomputing, vol. 71, no. 1-3, pp. 71-77, 2007.

[5] Lippmann R., An Introduction to Computing with Neural Nets, IEEE ASSP Mag, vol. 4, no. 22, pp. 4-22, 1987.

[6] Moctezuma-Eugenio J. and Huitzil C., Estudio Sobre la Implementaci n de Redes Neuronales Artificiales Usando XILINX System Generador, in Proceedings of XII Workshop Iberchip , Costa Rica, pp. 1-7, 2006.

[7] Morgado-Dias F., Antunes A., and Mota A., Artificial Neural Networks: a Review of Commercial Hardware, Engineering Applications of Artificial Intelligence-IFAC , vol. 17, no. 8, pp. 945-952, 2004.

[8] Morgado-Dias F. and Mota A., Direct Inverse Control of a Kiln, in Proceedings of the 4 th Portuguese Conference on Automatic Control , pp. 336-341, 2000.

[9] Reis L., Aguiar L., Baptista D., and Morgado- Dias F., ANGE-Automatic Neural Generator, in Proceedings of the 21 st International Conference on Artificial Neural Networks , Berlin, Germany, pp. 446-453, 2011.

[10] Rosado-Mu oz A., Soria-Olivas E., Gomez- Chova L., Vila J., An IP Core and GUI for Implementing Multilayer Perceptron with a Fuzzy Activation Function on Configurable Logic Devices, Journal of Universal Computer Science , vol. 14, no. 10, pp. 1678-1694, 2008.

[11] Soares A., Pinto J., Bose B., Leite L., Da-Silva L., Romero M., Field Programmable Gate Array (FPGA) Based Neural Network Implementation of Stator Flux Oriented Vector Control of Induction Motor Drive, in Proceedings of IEEE International Conference on Industrial Technology , Mumbai, pp. 31-34, 2006.

[12] Taspinar N. and Isik Y., Multiuser Detection with Neural Network MAI Detector in CDMA Systems for AWGN and Rayleigh Fading Asynchronous Channels, the International Arab Journal of Information Technology , vol. 10, no. 4, pp. 413-419, 2012.

[13] Tisan A., Buchman A., and Oniga S., A Generic Control Block for Feedforward Neural Network with On-Chip Delta Rule Learning Algorithm, in Proceedings of the 30 th International Spring Seminar on Electronics Technology , Cluj-Napoca, pp. 567-570, 2007. A Software Tool for Automatic Generation of Neural Hardware 235 Leonardo Reis received his bachelor s degree in electronics and telecommunications engineering from the University of Madeira, Portugal in 2011, and is currently finishing his Telecommunications and Energy Networks Master s thesis at the same university. Lu s Aguiar finished his bachelor s degree in electronics and telecommunications engineering in 2009 and Master s degree in Telecommunications and Networks Engineering in 2011, both at the University of Madeira. He is currently researcher at the University of Madeira, Portugal. Dar o Baptista received his bachelor s degree in electronics and telecommunications in 2007 and his Master s degree in telecommunications and network engineering in 2009 both from the University of Madeira, Portugal. His main research field is artificial neural networks a nd he is currently researcher at the University of Madeir a. Fernando Morgado-Dias received his Master s degree in microelectronics from the University Joseph Fourier in Grenoble, France in 1995 and his PhD from the University of Aveiro, Portugal, in 2005 and is currently Assistant professor at the University of Madeira and Pro-Rect or. His research interests include artificial neural ne tworks and their applications, especially regarding their hardware implementations.