Brain-inspired AI chip uses 100 times less energy to power devices; works offline

Brain-inspired AI chip uses 100 times less energy to power devices; works offline

traditional chips, the chip integrates computing and memory in a single
unit, enabling faster, more efficient on-device processing.

German scientists have developed a groundbreaking AI chip modeled after the human brain that performs real-time calculations directly on the spot, while ensuring complete cybersecurity without relying on cloud services or internet connectivity.

The research team at the Technical University of Munich (TUM) in the state of Bavaria, designed the chip, called AI Pro, to combine memory and processing units in a compact neuromorphic architecture that mimics how humans recognize patterns and make decisions.

Developed by Hussam Amrouch, PhD, an AI processor design professor at the university, the one square millimeter chip mimics how the human brain processes information, allowing it to learn and function independently, all while consuming up to ten times less energy than conventional AI chips.

Smarter, safer computing

Already in the prototype stage and produced by semiconductor manufacturer Global Foundries in Dresden, the chip contains approximately 10 million transistors. While this may seem worlds apart from the 200 billion transistors packed into NVIDIA’s high-performance processors, the chip delivers complete data privacy, exceptional energy efficiency, and true on-device intelligence, an advantage that industry giants often lack.

Amrouch emphasized that the chip achieves this using a concept known as hyperdimensional computing, an approach that groups and associates different data points to draw inferences. In simple terms, the chip learns by spotting similarities and patterns, not by processing vast amounts of data.

Hence, rather than processing countless images of cars, the chip understands a car by associating it with features like four wheels, road use and typical shapes. But, according to Amrouch, what makes the new chip truly unique is that, just like humans, it draws inferences and learns by recognizing similarities.

Moreover, the brain-like thinking comes with a major perk as it uses far less energy. During a trial run, the AI Pro consumed just 24 microjoules, while competing chips used between 10 and 100 times more. Calling it a record value, Amrouch credits the breakthrough to a combination of specialized algorithms, innovative processor architecture, and a novel approach to data processing.

Amrouch said,

This mix of modern processor architecture, algorithm specialization and innovative data processing makes the AI chip something special,

Cloud-free intelligence

The team stated that the AI Pro is tailored for specific applications. Its architecture tightly integrates memory and processing units, enabling real-time decision-making directly on the spot. This has vast implications for industries like healthcare, robotics, and autonomous navigation, especially in environments where internet access is unreliable or where data privacy is paramount.

“That makes them very efficient,” Amrouch claimed, adding that smartwatches fitted with the chip could process sensitive heart rate or blood oxygen data on the spot, eliminating the need to transmit it to external servers, while drones could analyze navigation patterns mid-flight without relying on satellite links. Because data stays local, both latency and cybersecurity risks drop dramatically.

Priced at around $33,700 (€30,000), the prototype isn’t ready for mass production just yet. Still, the team isn’t concerned about competing with big tech companies. Their goal is to create a new kind of AI chip, one that’s efficient, secure, and built for specific tasks instead of raw power.

READ the latest news shaping the AI Chips market at AI Chips News

Brain-inspired AI chip uses 100 times less energy to power devices; works offline, source