ZigZag - An Architecture-Mapping Design Space Exploration Framework for Deep Learning Accelerator

Explore the ZigZag framework, a rapid design space exploration tool for deep learning accelerator architecture and mapping, in this 26-minute talk from the tinyML EMEA 2021 conference. Discover how ZigZag's innovative approach, including uneven mapping capabilities and smart search strategies, can lead to more energy-efficient solutions for embedded deep learning systems. Learn about the framework's key components, benchmarking experiments, and case studies demonstrating its reliability and effectiveness. Gain insights into the latest research outcomes, including applications to analog-in-memory-computing architectures and the Loop Order based Memory Allocation method for temporal mapping search.

Introduction
Overview
Hardware
Unified Design Space
Search Engines
Results
Extensions
Conclusion
QA