A team of specialists from the United States has created a chip that can dynamically transform to perceive new information. This ability is similar to how the human brain works.
The developed chip will help AI to learn further without losing the data already acquired. The researchers' work is published in Science.
In the human brain, the process of learning is responsible for constantly emerging neural connections. Conventional computer chips do not have the ability to form bonds because they are static. That is why it is currently impossible to create an artificial intelligence that can make decisions on its own. However, by incorporating AI into hardware, machines could be made more efficient.
In the course of the work, the researchers developed a hardware that can be reprogrammed using electrical impulses. This ability will enable the device to acquire the functionality required to obtain a computing machine that functions in a manner similar to the brain.
The created neurochip is a small rectangular device. It is made from hydrogen-sensitive perovskite nickelate. The action of an electric pulse in the chip causes a change in the concentration of a substance in nanoseconds. For example, if a larger volume of hydrogen is located in the central part of the neurochip, it can play the role of a neuron, a separate cell of the human brain. If the volume of hydrogen in this area becomes smaller, then the chip turns into a synapse - a connection of neuron cells used by the brain to store information in developed neural chains.
By modeling experimental data, the experts found that the physics of the device is capable of generating a dynamic design for an artificial network of neurons. This network can better distinguish electrocardiogram parameters compared to a static network. Such a neural network uses “reservoir calculations” (a block of various non-linear functions, from which it is possible to make any function), explaining the connection and transmission of information between different areas of the brain. It was also found that the dynamic network is able to choose and select the best schemes to solve new emerging problems.
The microcircuit was subjected to over a million reprogramming cycles, during which it demonstrated stable operation. Already known semiconductors are suitable for the production of such a neurochip. At the moment, scientists are working on creating a demonstration large prototype, which in the future will help develop a computer that works like a human brain.
The developed chip will help AI to learn further without losing the data already acquired. The researchers' work is published in Science.
In the human brain, the process of learning is responsible for constantly emerging neural connections. Conventional computer chips do not have the ability to form bonds because they are static. That is why it is currently impossible to create an artificial intelligence that can make decisions on its own. However, by incorporating AI into hardware, machines could be made more efficient.
In the course of the work, the researchers developed a hardware that can be reprogrammed using electrical impulses. This ability will enable the device to acquire the functionality required to obtain a computing machine that functions in a manner similar to the brain.
The created neurochip is a small rectangular device. It is made from hydrogen-sensitive perovskite nickelate. The action of an electric pulse in the chip causes a change in the concentration of a substance in nanoseconds. For example, if a larger volume of hydrogen is located in the central part of the neurochip, it can play the role of a neuron, a separate cell of the human brain. If the volume of hydrogen in this area becomes smaller, then the chip turns into a synapse - a connection of neuron cells used by the brain to store information in developed neural chains.
By modeling experimental data, the experts found that the physics of the device is capable of generating a dynamic design for an artificial network of neurons. This network can better distinguish electrocardiogram parameters compared to a static network. Such a neural network uses “reservoir calculations” (a block of various non-linear functions, from which it is possible to make any function), explaining the connection and transmission of information between different areas of the brain. It was also found that the dynamic network is able to choose and select the best schemes to solve new emerging problems.
The microcircuit was subjected to over a million reprogramming cycles, during which it demonstrated stable operation. Already known semiconductors are suitable for the production of such a neurochip. At the moment, scientists are working on creating a demonstration large prototype, which in the future will help develop a computer that works like a human brain.
Login or register to post comments
Comments 0