Ultra-Low Power Computing in the Era of AI on the Edge
Minimizing Energy Consumption & Increasing Efficiency in Computing
Edge Computing & AI
Lecturer: Dr. Thomas Kämpfe, Group Manager Integrated RF & AI, Fraunhofer IPMS
Title: Ultra-Low Power Computing in the Era of AI on the Edge
Content:
- World Energy Consumption for Computing
- AI as Driver for Rising Energy Consumption
- Hardware for Low Power and High Performance Computing
- In-Memory Computing (IMC) for AI acceleration
- Algorithms & Accuracy of Edge AI with IMC
Energy consumption in conventional computing is enormously high. If applications in data centers are taken over by AI, which is increasingly the case, energy consumption increases further due to even higher hardware energy requirements. At the same time, in image recognition, for example, even the most powerful AIs and computing architectures cannot match the accuracy of the human brain. Neuromorphic networks that are based on biological neural structures and that can be accelerated by in-memory computing, are one way to minimize energy consumption and increase computational efficiency
The strategic research project NeurOSmart which Thomas Kämpfe is part of aims to solve the energy problem of computing by increasing the efficiency of all hardware components.
Source: Think Wireless IoT Day 02-2022: Automotive, Industrial Production & Wireless IoT Technologies
Date: March 16th 2022