Despite the fact that we often talk about neural networks and artificial intelligence, yet what we call the "electronic brain" is fundamentally different in structure from the structure of the human brain. But there are so-called neuromorphic chips, which are much faster than their electronic counterparts, but much less developed, which hinders the development of this technology. Recently, thanks to the joint work of scientists from the United States and France, it was possible to create a structure very similar to the human synapse (the place of contact between two cells, serving to transmit the nerve impulse). This development can be the first step towards the creation of a real electronic brain.
The artificial synapse created by its properties is almost identical to the natural one. The main property that prevents the artificial brain from being similar to ours is the "neuroplasticity", that is, the possibility to change under the influence of experience, and also to restore lost connections after damage or as a response to external influences. In other words, this is something through which we can learn new information and gain experience.
When developing the structure of the new artificial synapse, scientists created highly accurate mathematical models, on the basis of which detailed calculations of the shape and structure of the microscopic device were made. The artificial synapse is designed to develop autonomous self-learning systems. Based on synapses of a new type, as the researchers hope, it will be possible to create a real electronic brain with an artificial intelligence function at the hardware level, rather than on the software one that is currently used.
Left – the principle of the work of a human synapse, on the right – an artificial synapse
The basis of the neuroplasticity of the synapse is a tunnel transition based on the ferroelectric material. The conductivity of a section increases each time an electric current pulse passes through it. Different actions generate different pulses, which allows the synapse to remember the recurring events and "gain experience". In this case, areas that are not used in conductivity, eventually erased and, as it were, "forgotten". Now the researchers are going to create the first prototype chip, on the crystal of which a complex neural network capable of self-learning process will be formed.