Researchers at the Technical University of Munich (TUM) have developed an innovative AI chip that works completely without relying on cloud servers or internet connections. Named AI Pro, the new processor uses a brain-inspired design to process information locally, which could revolutionize how AI is deployed in everyday devices while significantly reducing energy consumption.
Unlike traditional AI chips, sending data to a remote server for processing, AI Pro performs all calculations directly on the device, eliminating privacy issues while cutting power requirements. The chip is ten times more energy efficient than comparable processors, a key advantage for battery-powered devices and power-limited applications.
The breakthrough is the acceleration of AI integration across the industry, with many current solutions requiring ongoing internet connectivity and improving privacy concerns regarding handling of sensitive data in the cloud. Can this new approach trigger a shift to a more secure, independent AI system to keep data localized?
How brain-inspired chips work
“Although NVIDIA has built a platform that relies on cloud data and promises to solve all problems, we have developed an AI chip that enables customized solutions. There is a huge market there,” explained Hussam Amrouch, chairman of TUM’s AI processor design.
The key innovation of AI Pro is its neuromorphic structure, which mimics how the human brain processes information. Unlike traditional chips that separate compute and memory cells from traditional chips, AI Pro integrates them together, which significantly improves efficiency.
The chip applies “high-dimensional computing” – a calculation method that recognizes patterns and similarities rather than requiring large amounts of data sets to learn. This means that chips can make informed decisions without the extensive training data required by most AI systems.
Professor Amrouch pointed out: “Humans also make inferences through similarities and learn through similarities.
Energy efficiency and real-life applications
During testing, the new chip showed significant energy efficiency. For example tasks, AI Pro consumes only 24 trace amounts of energy, while comparable chips require 10 to 100 times more – “record value”.
This efficiency makes the chip very suitable for the following applications:
- Process health data from wearable devices without sending sensitive information to the cloud
- UAV navigation system, no internet connection required
- IoT devices that require wise decisions with minimal strength
- Edge computing in environments where internet access is unreliable or unavailable
By processing data locally, the chip also reduces the carbon footprint of AI applications by eliminating the need for energy-intensive server computing and data transfer.
Security and Privacy Benefits
In addition to energy benefits, AI Pro also offers huge security and privacy benefits. Since the data never leaves the device, it effectively eliminates issues with Internet connection, network security vulnerabilities and data privacy issues.
Professor Amrouch emphasized: “The future belongs to people with hardware.” The professor emphasized how local processing maintains users’ control over the data.
While currently, the single-sided millimeter chip costs about €30,000 and has about 10 million transistors (much more than Nvidia’s chips, with 200 billion transistors – its professional design makes it efficient for a specific application rather than trying to be a general-purpose solution.
The original prototype has been produced by the Global Foundry, a semiconductor manufacturer in Dresden, demonstrating the technology’s realistic production survivability.
As AI continues to integrate into more aspects of everyday life, chips like AI Pro represent a potential shift we are approaching artificial intelligence – from centralized, cloud-based systems to distributed, energy-efficient devices that can make local and secure locally while making informed decisions at the edge.
Related
Discover more from Neuroweed
Subscribe to send the latest posts to your email.
