Researchers at Loughborough University have developed a new chip that could dramatically reduce the energy consumed by artificial intelligence systems. The device is designed to process data that changes over time directly within the hardware itself, bypassing the constant back-and-forth movement of information between memory and processor that characterises conventional AI systems. The team argues the chip could be up to 2,000 times more energy efficient than existing software-based approaches. The findings have been published in a new study led by Dr. Pavel Borisov.
Traditional AI systems operate by repeatedly shuttling data between separate memory and processing units, a method that demands significant energy. The new chip consolidates these functions, allowing data to be handled in a single location. Dr. Borisov described the advance as an opportunity to fundamentally rethink how AI systems are constructed. “By using physical processes instead of relying entirely on software, we can dramatically reduce the energy needed for these kinds of tasks,” he said.
At the core of the device is a memory resistor, a component that retains a record of past signals and adjusts its response to new ones accordingly. This behaviour mirrors the way the human brain forms and uses connections between neurons. The team replicated this principle by engineering pores in nanometre-thin films of niobium oxide, creating complex, random physical connections within an artificial neural network. The result is a device that does not merely follow instructions but draws on prior experience to inform its outputs.
The chip is specifically suited to processing time-dependent data — information that shifts continuously, such as weather readings, stock market figures, or wave patterns. These types of data are sensitive to small fluctuations, and conventional AI systems must expend large amounts of energy to track every change. By analysing historical measurements, the new chip can more efficiently follow and anticipate such chaotic signals, reducing the energy required to do so.
Dr. Borisov highlighted several real-world applications where the technology could prove valuable. Heart rate monitoring, brain electrical activity, and ambient temperature are all examples of continuously changing data streams that current systems handle in an energy-intensive manner, often requiring a stable connection to a remote server. The chip could enable smarter, lower-power devices capable of processing such signals locally and independently.
Looking further ahead, Dr. Borisov outlined a broad range of potential deployment scenarios. “Whether that’s in a car, a robot, a nuclear power plant, or in a smart watch,” he told Decrypt, citing uses such as detecting strokes, monitoring vehicle engine health, or verifying that a nuclear reactor is operating within normal parameters. The research positions the chip as a tool for any application where data is in constant flux rather than static, distinguishing it from AI tools such as chatbots that primarily handle fixed information.
Originally reported by Decrypt.
