Novel SOM Processor Design using FPGA

Abstract

One of the fastest growing and most demanding areas of computer science is Machine Learning (ML). Personalized Mapping (SOM), classified as unchecked ML, is a popular data mining algorithm widely used in the Artificial Neural Network (ANN) for making high-resolution data maps on low-level feature maps. SOM, as it is very computer-intensive, requires a lot of time and energy to calculate when working with large networks. The acceleration of many computationally intensive algorithms can be achieved using Field- Programmable Gate Arrays (FPGAs) but requires extensive hardware knowledge and long development time using the traditional Hardware Description Language (HDL) format. The Open Computing (OpenCL) language is a standard framework for writing compatible computer programs that run on multiple computer programs. The OpenCL Intel FPGA Software Development Kit (IFSO) is a High-Level Synthesis (HLS) tool that provides an effective alternative to HDL-based architecture. This study demonstrates the open implementation of OpenCL SOM algorithm in Stratix V and Arria 10 FPGAs using IFSO.