Researchers at the Autonomous University of Barcelona (UAB) have developed a magnetic material that can simulate the way the brain stores information. This material makes it possible to mimic the synapses of neurons and, for the first time, to mimic the learning that occurs during deep sleep.
Neuromorphic computing is a new computing paradigm in which brain behavior is simulated by mimicking the original synaptic functions of neurons. Among these functions, we can mention neural plasticity: the ability to store or forget information depending on the duration and repetition of electrical impulses that sanction neurons. This function is related to learning and memory.
Among the materials that mimic neuron synapses, we can mention resistive materials, ferroelectrics, phase change memory materials, topological insulators and ionic magnetic materials. In the last case, the use of an electric field causes the displacement of ions and changes in the magnetic properties of the material.
In these materials, the magnetism is tuned when an electric field is applied, but it is difficult to control the evolution of the magnetic properties when the voltage is stopped. This complicates simulating some brain-inspired functions, such as maintaining learning efficiency even when the brain is in a deep sleep state.
A substance that simulates learning by the brain
Now researchers have presented a new method to control the magnetic evolution in excited and post-stimulated state. They made a material from a thin layer of cobalt mononitride (CoN) in which, by applying an electric field, the accumulation of N ions at the interface between the layer and the liquid electrolyte in which the layer is placed is controlled.
“This new material works with the movement of ions controlled by electrical voltage in a similar way to our brain and at speeds similar to those produced in neurons, on the order of milliseconds. “We have created an artificial synapse that in the future may be the basis of a new computing paradigm, an alternative to the one used by current computers.”
By applying voltage pulses, processes such as memory, information processing, information retrieval and, for the first time, controlled information updating without applied voltage, can be simulated in a controlled manner. This control is achieved by modifying the thickness of the cobalt mononitride layers (which determines the speed of the ions) and the frequency of the pulses.
It is stated in the study of scientists:
“When the thickness of the cobalt mononitride layer is below 50 nm and with the voltage applied at a frequency of more than 100 cycles per second, we have succeeded in simulating an additional logic function.”
The results obtained with this new material open a wide range of possibilities for neuromorphic computing functions and create more efficiency in perception, learning and memory by using neural networks.