CAEMO - A Flexible and scalable high performance matrix algebra coprocessor for embedded reconfigurable computing systems

Many applications in mobile and embedded systems like signal processing, machine learning, kinematics, dynamics, and control depend on computationally expensive matrix operations. However, such systems underlie tight constraints regarding power consumption and physical space, which prohibits the usa...

Full description

Saved in:
Bibliographic Details
Published in:Microprocessors and microsystems 2018-02, Vol.56, p.47-63
Main Authors: Woehrle, Hendrik, Kirchner, Frank
Format: Article
Language:English
Subjects:
Algebra
Computer architecture
Coprocessors
Embedded systems
Energy consumption
Field programmable gate arrays
FPGA
Hardware acceleration
Kinematics
Machine learning
Mathematical analysis
Matrix
Matrix algebra
Matrix methods
Microprocessors
Personal computers
Power consumption
Signal processing
Studies
System on chip
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Many applications in mobile and embedded systems like signal processing, machine learning, kinematics, dynamics, and control depend on computationally expensive matrix operations. However, such systems underlie tight constraints regarding power consumption and physical space, which prohibits the usage of powerful multicore systems. In this paper, we propose a novel scalable and power-efficient architecture for matrix algebra in FPGA-based Systems-on-Chip. The architecture is based on a linear systolic array and has been developed with a focus on flexibility in order to be adapted to different applications. We evaluate the performance, resource utilization and power consumption of different configurations and show that it provides significant speed-ups over a mobile processor and is significantly more power efficient than a standard PC.
ISSN:0141-9331
1872-9436
DOI:10.1016/j.micpro.2017.10.005