TUM Develops Groundbreaking AI Chip with Brain-Inspired Design for Energy-Efficient, Cloud-Free Processing

TUM Develops Groundbreaking AI Chip with Brain-Inspired Design for Energy-Efficient, Cloud-Free Processing

(IN BRIEF) The Technical University of Munich has developed a revolutionary AI chip, the AI Pro, designed by Prof. Hussam Amrouch. Unlike traditional AI chips, the AI Pro performs computations locally, eliminating the need for cloud servers or internet connections, ensuring faster processing and full cybersecurity. Using a neuromorphic design and ‘hyperdimensional computing,’ the chip is up to ten times more energy-efficient than existing chips. Its ability to recognize patterns with fewer data records sets it apart from conventional AI models, making it ideal for customized, efficient applications. The chip’s energy consumption is remarkably low, and it excels in fields such as health, environmental, and space technologies, with the potential to reduce AI’s carbon footprint and enhance data security.

(PRESS RELEASE) MUNICH, 19-May-2025 — /EuropaWire/ — A groundbreaking new AI chip developed at the Technical University of Munich (TUM) promises to revolutionize the field by eliminating the need for cloud servers or internet connections, making it a significant leap forward in AI technology. The AI Pro chip, designed by Prof. Hussam Amrouch, is modeled on the human brain and features an innovative neuromorphic architecture. This allows the chip to perform calculations locally, enhancing both its speed and cyber security, while being up to ten times more energy-efficient than conventional AI chips.

The AI Pro’s design sets it apart from traditional chips, which often rely on vast cloud data processing. In contrast, this chip brings together computing and memory units, utilizing ‘hyperdimensional computing.’ This approach allows the chip to recognize patterns and similarities without the need for millions of data records, making it far more efficient. For example, rather than being trained on thousands of images of cars, the AI Pro can understand a car’s essential characteristics—such as having four wheels and typically driving on roads—based on fewer examples. “Humans also learn by drawing inferences and recognizing patterns, just like our chip does,” explains Prof. Amrouch.

The new AI chip is mounted on a circuit board by Prof Hussam’s research group.

Energy efficiency is a key advantage of this new chip. In tests, the AI Pro consumed just 24 microjoules for a sample task, compared to 10 to 100 times more energy required by similar chips. Prof. Amrouch describes this as “a record value,” emphasizing that the combination of modern processor architecture, algorithm specialization, and innovative data processing makes the AI Pro unique in its field.

The chip’s design also differs from general-purpose chips like those from industry leader NVIDIA. While NVIDIA’s platform depends on cloud-based data and aims to address a wide array of applications, the AI Pro is built for customized solutions, offering significant potential in specific markets. Prof. Amrouch believes this focus on tailored solutions is the future of AI chip development.

Currently priced at 30,000 euros per one-square-millimeter chip, the AI Pro contains around 10 million transistors—less densely packed than NVIDIA’s chips, which have 200 billion transistors. However, Prof. Amrouch’s primary goal is not raw power but efficiency and local data processing. By keeping the data processing onboard, the chip avoids reliance on cloud infrastructure, saving time, server capacity, and reducing AI’s carbon footprint.

The AI Pro is particularly suited for applications that involve sensitive or personal data, such as health tracking via smartwatches or drone navigation. By processing this data locally, the chip eliminates the need for stable internet connections and mitigates cybersecurity concerns, providing a more secure, efficient solution for a wide range of industries. Prof. Amrouch concludes, “The future belongs to those who control the hardware.”

Publications

• Sandy Wasif, Paul Genssler, and Hussam Amrouch. “Domain-Specific Hyperdimensional RISC-V Processor for Edge-AI Training.” IEEE Transactions on Circuits and Systems I: Regular Papers (2025). ieeexplore.ieee.org/document/10931124
• Soliman, Taha, Swetaki Chatterjee, Nellie Laleni, Franz Müller, Tobias Kirchner, Norbert Wehn, Thomas Kämpfe, Yogesh Singh Chauhan, and Hussam Amrouch. “First demonstration of in-memory computing crossbar using multi-level Cell FeFET.” Nature Communications 14, no. 1 (2023): 6348. https://www.nature.com/articles/s41467-023-42110-y
• Wei-Ji Chao, Paul R Genssler, Sandy A Wasif, Albi Mema, Hussam Amrouch, “End-to-end Hyperdimensional Computing with 24.65 µJ per Training Sample in 22 nm Technology”, under review at the European Solid-State Electronics Research Conference (ESSERC). Preprint available: https://go.tum.de/440497

Further information and links

• Prof. Hussam Amrouch started his engagement at TUM two years ago. The Chair of AI Processor Design was created as part of Hightech Agenda Bayern. Further information: https://www.hightechagenda.de/
• Prof. Hussam Amrouch is also active in the Munich Institute of Robotics and Machine Intelligence (MIRMI). His chip developments are relevant for health, the environment, and space. Further information on MIRMI: www.mirmi.tum.de/mirmi/startseite//

Media Contacts:

Corporate Communications Center
Andreas Schmitz
presse@tum.de

Contacts to this article:

Prof. Hussam Amrouch
Chair of AI Processor Design (AI-Pro)
Technical University of Munich (TUM)
amrouch@tum.de

SOURCE: Technical University of Munich