Explore the cutting-edge world of ultra-low power machine learning hardware for edge devices in this comprehensive tinyML Asia 2020 conference talk. Delve into the fundamentals of TinyML, its applications, and model architecture optimization techniques. Examine various TinyML hardware examples and trends, including the Arm Cortex-M55 Core IP, Green Waves GAP8 Applications Processor, and Arm Ethos-U Family of low-power MicroNPUs. Investigate the potential of in-memory compute and compute-in-memory technologies, with a focus on mapping CNNs to analog crossbars and the Mythic IPU. Gain valuable insights into Arm's role as the software and hardware foundation for tinyML, and discover how these tiny but powerful solutions are revolutionizing machine learning at the edge.
Tiny but Powerful: Hardware for High Performance, Low Power Machine Learning
Overview
Syllabus
Intro
Outline
Motivation/Background: What is TinyML?
TinyML Applications
TinyML Model Architecture and Optimization
TinyML Hardware Examples
TinyML Hardware Trends
Arm Cortex-M55 Core IP
Green Waves GAP8 Applications Processor
Low-Power MicroNPU: Arm Ethos-U Family
In-memory Compute for Tiny ML
Compute-in-Memory Overview
Mapping a CNN to an Analog Crossbar
Mythic IPU
Closing Thoughts
Arm: The Software and Hardware Foundation for tinyML
Taught by
tinyML
Related Courses
-
Creating End-to-End TinyML Applications for Ethos-U NPU in the Cloud
-
Hardware-Aware Model Optimization in Arm Ethos-U65 NPU
-
Powering Innovation in AI Devices on the Edge with Micro-NPUs - tinyML Asia 2020
-
tinyMLPerf: Benchmarking Ultra-low Power Machine Learning Systems
-
Ultra-Low-Power Command Recognition for Ubiquitous Devices
-
Empowering the Edge - Advancements in AI Hardware and In-Memory Computing Architectures for TinyML
