A Survey of Hardware Self-Organizing Maps.

Area of Science:

• Computer Science
• Artificial Intelligence
• Hardware Architecture

Background:

• Self-organizing feature maps (SOMs) are vital for unsupervised clustering and dimensionality reduction.
• Their topology preservation makes them suitable for Internet of Things (IoT) and big data (BD).
• High computational costs hinder real-time online processing, limiting SOMs to offline applications.

Purpose of the Study:

• To survey existing hardware (HW) implementations of SOMs.
• To analyze common computing blocks, architectures, and design choices in HW SOMs.
• To identify challenges and trends for adopting HW SOMs as accelerators.

Main Methods:

• Literature review of hardware SOM implementations.
• Analysis of reported computing blocks, architectures, and optimization techniques.
• Overview of challenges and future trends in the field.

Main Results:

• Identified widely used computing blocks and architectures for hardware SOMs.
• Detailed various adaptation and optimization techniques for hardware implementations.
• Provided insights into challenges and trends for ubiquitous adoption.

Conclusions:

• Hardware implementations are crucial for overcoming SOM computational limitations.
• Further research in HW SOM accelerators can enable real-time processing for IoT and big data.
• This survey serves as a valuable resource for researchers in AI, hardware architecture, and system design.