Hardware/Software Codesign

• Describe the principles and elements of HW/SW codesign
• Describe the workflow from specification to deployment of machine learning for embedded devices
• Describe the fundamentals of neural network architectures and training for embedded devices
• Compare various software and hardware options for deployment of machine learning models in embedded systems
• Design an embedded system for machine learning applications
• Train and evaluate machine learning models for microcontroller based embedded systems
• Deploy machine learning models on microcontroller based embedded systems
• Analyze trade-offs between different designs using multi-parametric optimization
• Optimize machine learning applications for embedded systems

The course gives a high-level introduction to topics in hardware/software codesign and the elements of a codesign process. The theoretical foundations and practical implementation of machine learning models on resource constrained devices (commonly denoted embedded AI) are explored with lectures, hands-on exercises, and project work. Through a series of lectures and hands-on exercises, students are expected to be able to train, evaluate, and deploy a machine learning model to a microcontroller-based system at the end of the course. Methods for exploring trade-offs in pure software vs hardware-accelerated implementations are explored including multi-parametric optimization to account for various metrics including accuracy, inference latency, memory use, and energy consumption.

• 6 – 20 (Thurs 13-17)

Students are expected to have prior knowledge in Python and C-programming. Microcontroller and machine learning experience is preferable. Knowledge of digital design and computer architecture is also preferable.

Lectures, exercises and project.