This post is also available in:
עברית (Hebrew)
As artificial intelligence systems grow more powerful, they’re also consuming vast amounts of energy. To address this growing energy gap, researchers are turning to nature’s most efficient processor: the brain.
A team at the University at Buffalo is working on neuromorphic computing—a brain-inspired architecture that blends physics, engineering, and quantum science to rethink how machines process and store information. Unlike traditional computers, which separate memory and processing units, neuromorphic systems aim to combine them—just like biological brains, where neurons and synapses store and process data together. This approach minimizes the energy loss involved in shuttling information between separate components.
The researchers are developing neuromorphic hardware based on phase-change materials (PCMs)—compounds that can switch between conductive and resistive states when hit with electrical pulses. These materials behave much like biological synapses by adjusting their conductivity based on repeated activation. This allows them to store “memory” of their past states, a key requirement for mimicking brain-like behavior, according to TechXplore.
The goal is to construct artificial neurons and synapses capable of synchronizing electrical oscillations, similar to the rhythmic signaling observed in human brains. Such neuromorphic chips could enable computers to perform complex tasks more efficiently and flexibly—such as recognizing patterns, adapting to changing environments, or making real-time decisions.
Unlike conventional systems, which follow rigid logic, neuromorphic models may process information more fluidly, responding to vague or incomplete data more like a human would. This could be particularly valuable in edge-computing scenarios like autonomous vehicles, where decisions must be made instantly without relying on cloud servers.
While full-scale neuromorphic machines remain on the horizon, specialized chips tailored for specific tasks—like navigation or real-time sensor processing—may arrive much sooner. As AI continues to scale, neuromorphic computing could be key to building systems that are not only smarter, but far more energy-conscious.
