Abstract:
Support Vector Machines are widely used in pattern recognition, being the newest achievements in neural network structures. This paper presents an implementation example of an SVM classification function using a Spartan3 FPGA device. A Block Ram based implementation is compared versus a distributed LUT-based RAM one. Aspects regarding memory geometry and instantiation are presented. The number of required clock periods and the maximum clock frequency is calculated and a speed comparison of the implemented system with software running on a PC targeting the same application is also made.