학술논문
Implementation of Self-Organizing Map with Content Addressable Memory-based Massive-Parallel SIMD Matrix Processing Core
Document Type
Conference
Author
Source
2022 5th World Symposium on Communication Engineering (WSCE) Communication Engineering (WSCE), 2022 5th World Symposium on. :100-104 Sep, 2022
Subject
Language
Abstract
Recently, several multimedia applications, such as digital image compression, digital video compression, and digital audio processing, are being executed on embedded devices. Then, processing cores in such embedded devices are required to have high performance and programmability. In general, multimedia applications and machine learning algorithms consist of repeated arithmetic or logic operations and table-lookup coding operations. In order to improve the processing speed of these two operations, a CAMX (Content Addressable Memory-based massive-parallel SIMD matriX core) has been proposed. The role of the CAMX is to be an accelerator for a CPU core. The CAMX has several processing elements for highly parallel processing capability and consists of two CAM modules for fast table-lookup processing. In this paper, we have implemented the self-organizing map algorithm and compared it with Raspberry Pi. The CAMX is about 2.6 times faster than an Arm processor with NEON at the same frequency. In addition, the CAMX can reduce its operating frequency to one-third, if the CAMX and Arm processor with NEON run at the same processing speed.