Particle.news

Download on the App Store

TUM Unveils AI Pro, a Brain-Inspired Chip for Secure, Energy-Efficient Edge AI

The neuromorphic AI Pro chip combines memory and processing, consuming record-low energy while operating offline for privacy-focused applications.

Image
Image

Overview

  • Developed by researchers at the Technical University of Munich, the AI Pro chip integrates memory and computation in a neuromorphic architecture inspired by the human brain.
  • The chip achieves energy efficiency with a record-low 24 microjoules per sample, up to ten times less than conventional AI chips.
  • AI Pro uses hyperdimensional computing, enabling it to infer patterns by recognizing similarities rather than processing vast datasets.
  • Prototypes, fabricated by Global Foundries in Dresden, contain approximately 10 million transistors and are tailored for edge devices like smartwatches and drones.
  • The €30,000 prototype is under review for ESSERC 2025 and demonstrates the potential for secure, cloud-free AI but is not yet ready for mass production.