Overview
Explore the future of deep-learning accelerators in this 28-minute conference talk from the tinyML Summit 2022. Delve into the challenges and opportunities presented by next-generation accelerators, with a focus on system-level implications for design, integration, and scheduling. Learn about the transformative potential of machine learning across various industries, including computer vision, natural language processing, autonomous driving, and robotic manipulation. Discover how novel deep-learning accelerators are being developed to meet the growing performance and efficiency demands of deep-learning applications. Gain insights into hardware demands, domain-specific systems, and the full-stack approach to accelerator design. Examine user case studies, system-level challenges, and programming considerations for accelerators. Conclude with an evaluation of current progress and future directions in this rapidly evolving field.
Syllabus
Introduction
Hardware Demand
Moores Law
Domain Specific System
Domain Specific Hardware
Full System Full Stack
What are Accelerators
User Use Cases
SOC
Memory
System CPU
System Level Challenges
Programming Accelerator
KOSA
Evaluation
Conclusion
Questions
Sponsors
Taught by
tinyML