Despite the fact that the term "neural network" can be applied to both anatomical structure and a computing system, these neural networks still have more differences than similarities. This is primarily due to the extremely complex structure of the neural plexuses of the brain. But that could change thanks to the development of scientists from the US National Institute of Standards and Technology. Their invention may become a new stage in the development of neural network technology.
The fact is that the human brain consists of hundreds of millions of neurons, each of which is connected to tens of thousands of other neurons. This complex structure in a rather simplified form is taken as the basis for the construction of artificial neural networks, with the only difference that the electronics existing today cannot cope with such complex routing and the number of connections between the elements of the neural network has to be reduced tens and hundreds of times, which affects performance … Of course, now there are projects aimed at emulating the brain, but they also run into the limit of the power of existing technologies.
Chip structure diagram. The three-dimensional structure provides the sophisticated routing and signal rate required to simulate the structure of the brain.
The development team from the US National Institute of Standards and Technology offers a slightly different approach. Scientists want to use light instead of electricity as a means of transmitting a signal. For this, a special chip has been developed that distributes optical signals along a miniature grid in different directions. Thanks to the use of a new type of signal, it was possible to overcome the problem of connecting neural network elements by vertically stacking two layers of photonic structures. They confine light to lines to guide optical signals in much the same way that wires transmit an electrical impulse. This approach allows you to create complex routing schemes necessary to simulate the neural structures of the brain, as well as increase both the signal transmission rate and the number of connections between network elements. Besides,the system is very scalable and easy to expand in the future. And if "ordinary" neural networks are making progress in self-learning and pattern recognition, it is difficult to imagine what a structure that imitates the work of the brain will be capable of.
Vladimir Kuznetsov
