Indeed, technological advances in artificial intelligence (AI) are bringing ultra-fast computing thanks to neuromorphic computing. But the question is whether the current infrastructure is ready to handle this workload.
In reality, the codes that humans write are often executed on conventional silicon architectures. However, the latter are not suitable for this task. Researchers from Purdue University, the University of California, San Diego (USCD) and the École Supérieure de Physique et de Chimie Industrielles (ESPCI) in Paris then sought to solve this challenge. They published their results in Advanced Electronic Materials, where they explored an approach to reshape the hardware by imitating the synapses of the human brain.
A promising future for the neuromorphic computer
The neuromorphic computer, which mimics brain behavior, relies on special computer chips. In the brain, neurons transmit information via synapses, which play a key role in memory. Researchers found that vanadium oxides are promising for neuromorphic computing, as they make it possible to create both artificial neurons and synapses.
Neuromorphic computer architectures have a major advantage: lower energy consumption than traditional silicon architectures. This is due to their ability to imitate the basic components of a brain namely neurons and synapses. Unlike silicon, which is efficient for memory storage, neuromorphic materials mimic neuronal behavior.
However, finding suitable materials to create both good synapses and good artificial neurons is a challenge. Only a few quantum materials show promise in this area including vanadium dioxide. Researchers have discovered that memory accumulates throughout the vanadium sample. This discovery opens up new possibilities for controlling this property.
The results of this research have been revealed. However, microscopic videos showed that changes in the metallic and insulating domains of vanadium cause a memory accumulation throughout the sample.
This memory results from local temperature changes during the transition of the material from insulator to metal and vice versa. The preferential diffusion of point defects in the metal domains appears to contribute to this memory accumulation.
The researchers are now considering continuing their work by locally modifying the vanadium and observing the effects. Including the impact of ion bombardment on the material surface.
This could allow to improve the synaptic behavior of this neuromorphic material . This by guiding the electrical current towards areas where the memory effect is most pronounced.
dans un article qui peut être bien classé dans google
#les #chercheurs #ont #découvert #matériel #qui #imite #comportement #cerveau #humain
publish_date] pt]
