Tens of thousands of brain synapses on a chip

Source
MIT engineers have designed something like a "brain on a chip," made up of tens of thousands of devices known as silicon-based memristors that mimic brain connections or synapses.

These memristors would be the equivalent of transistors in conventional chips but would function similarly to how neurons do, in what is called neuromorphic computing.

Basically all of this amounts to a "hardware" neural network, that is, a trained neural network that can be housed in a chip the size of a confetti.

Source
Until now most of the neural networks are lines of code running on a more or less powerful computer or in the cloud, which represents an energy and communication cost.

Although these memristors have long been in use, the merit of this team of MIT engineers is to change their composition, the previous ones made of silicon and these are made of silver, copper and silicon, which seems to significantly improve their performance.

Source
With this system, a properly trained neural network could be implemented in a physical circuit that could be installed in a mobile phone for any task that comes to mind.

Language translators, image recognition, medical diagnostic devices and everything you can imagine with a chip, with minimal energy consumption and without having to be connected anywhere.

Simply amazing ...

More information:
https://news.mit.edu/2020/thousands-artificial-brain-synapses-single-chip-0608

Versión en español