A brain-inspired computer chip is gaining attention for cutting AI energy use by up to 70% while enabling more natural and adaptive machine learning.
The work comes from scientists at the University of Cambridge, who have developed a nanoelectronic component that mimics how the brain processes and stores information. Their findings were published in the journal Science Advances.
Today’s AI systems rely heavily on traditional computer chips. These chips separate memory and processing into different units. As a result, data must constantly move back and forth between them. This repeated transfer consumes a large amount of electricity, especially as AI systems grow larger and more complex.
READ ALSO: Rice University Study Solves Decades-Old Mystery in Light Crystals
The new approach takes inspiration from the brain. In the human brain, memory and processing happen in the same place. This allows neurons to communicate efficiently with very little energy. Scientists call this concept neuromorphic computing.
At the center of this research is a special component known as a memristor. Unlike standard electronic components, memristors can store and process information simultaneously. This makes them ideal for building energy-efficient AI systems.
The Cambridge team created a new memristor using hafnium oxide, a modified material. By carefully engineering this material, they built a device that operates with very low energy while remaining stable and reliable.
Dr. Babak Bakhit, the lead author of the study, highlighted the importance of energy efficiency in AI. He said, “We need devices that use extremely low currents, stay stable over time, and can switch between many states. That’s how we make AI hardware more efficient.”
WATCH ALSO: Engineers from Australia have introduced a spray that enhances the fire-resistance of clothing
Most existing memristors work by forming tiny conductive paths, known as filaments, inside the material. These filaments can behave unpredictably and often require high voltage to function. This limits their usefulness in real-world applications.
The new design avoids this problem. Instead of relying on filaments, the device uses carefully controlled interfaces within the material. The researchers added small amounts of strontium and titanium and used a two-step manufacturing process. This created tiny electronic barriers called p-n junctions.
These junctions allow the device to change its resistance smoothly and consistently. As a result, the memristor can switch states more reliably and with greater control.
Tests showed impressive results. The new devices operated at currents about a million times lower than some traditional memristors. They also supported hundreds of stable conductance levels, which is important for advanced computing tasks.
In addition, the devices demonstrated behaviors similar to those of the brain during learning. One example is spike-timing dependent plasticity. This process allows connections between neurons to strengthen or weaken based on timing. It is a key mechanism behind learning and memory in biological systems.
Bakhit explained this clearly. He said, “These are the properties needed for hardware that can learn and adapt, not just store data.”
READ ALSO: China Fires Back-to-Back Rockets in 24 Hours, Scaling Space Internet Power Play
The devices also showed strong durability in laboratory tests. They remained stable through tens of thousands of switching cycles and could hold their programmed states for about a day.
Despite these promising results, challenges remain. One major issue is the high temperature required to manufacture the devices. The current process needs temperatures of around 700 degrees Celsius, which is higher than what standard chip production allows.
Bakhit acknowledged this limitation. He said, “This is the main challenge right now. We are working on lowering the temperature so it fits better with existing manufacturing methods.”
If the team succeeds, the technology could move from the lab into real-world applications. It could be integrated into future AI chips, making them far more energy efficient.
The research is the result of years of effort. Bakhit spent nearly three years refining the process. Progress came after many failed attempts. A key turning point happened when he changed how oxygen was added during fabrication.
WATCH ALSO: Called Ascentia, the platform is developed by American company Collins Aerospace
He recalled the moment. He said, “We saw the first strong results at the end of November. It took a lot of trial and error to get there.”
The project received support from several organizations, including the Swedish Research Council, the Royal Academy of Engineering, the Royal Society, and UK Research and Innovation. A patent has also been filed through Cambridge Enterprise.
While still in its early stages, this brain-inspired chip points to a future where AI systems are not only more powerful but also far more energy efficient.
