{"id":3323,"date":"2025-11-05T07:57:13","date_gmt":"2025-11-05T07:57:13","guid":{"rendered":"https:\/\/terrarara.com.br\/en\/?page_id=3323"},"modified":"2025-11-05T07:57:13","modified_gmt":"2025-11-05T07:57:13","slug":"novos-neuronios-artificiais-conseguem-reproduzir-fisicamente-o-cerebro-humano","status":"publish","type":"page","link":"https:\/\/terrarara.com.br\/en\/?page_id=3323","title":{"rendered":"New artificial neurons can physically replicate the human brain"},"content":{"rendered":"
\"\"
An integrated spiking artificial neuron, with rich neuron functionality, single-transistor footprints, and low energy consumption for neuromorphic computing systems, can be created by stacking one diffusive memristor and one resistor on top of a transistor. The photograph on the cover shows the chip of an array of these integrated neurons, which are fabricated in the university”s cleanroom and have an active region of around 4 “m2 for each neuron. Credit: The Yang Lab at USC<\/figcaption><\/figure><\/div>
<\/p>

\r\n\tdoi.org\/10.1038\/s41928-025-01488-x<\/a>
Credibility<\/a>: 9<\/font>9<\/font>9<\/font>
#artificial neurons<\/a><\/p>

<\/p>

\r\n\tA major breakthrough in neuromorphic computing could dramatically reduce chip energy consumption and accelerate the path toward artificial general intelligence (AGI) – systems capable of thinking and learning in a way similar to humans<\/strong><\/p>

Researchers at the Viterbi School of Engineering and the School of Advanced Computing at the University of Southern California (USC) have developed artificial neurons that very closely mimic the electrochemical behavior of real brain cells.<\/p>

\r\nThe study, published in the journal *Nature Electronics*, represents a major step toward creating chips inspired by brain function.<\/p>

\r\nThis new approach promises to make circuits much smaller, reduce energy consumption by several orders of magnitude, and bring science closer to creating truly general artificial intelligence.\r\n<\/p>

\r\nUnlike traditional digital processors or silicon neuromorphic chips that only simulate the functioning of neurons, these new artificial neurons physically reproduce biological processes in an analog way.<\/p>

\r\nJust as neurotransmitters activate brain activity, certain chemical compounds can now be used to initiate calculations in brain-inspired hardware.<\/p>

\r\nThis makes the new system fundamentally different from previous models, which relied solely on mathematical representations of neural behavior.\r\n<\/p>

\r\nThe project was led by Professor Joshua Yang of the USC Department of Electrical and Computer Engineering, a pioneer in the creation of artificial synapses.<\/p>

\r\nHis team developed a novel type of artificial neuron based on a “diffusive memristor”-a device capable of processing information in a manner similar to a biological neuron.<\/p>

\r\nWhile conventional chips use the flow of electrons to perform calculations, the device created by Yang works with the movement of atoms, allowing for much more precise imitation of brain processes, offering high energy efficiency and a promising path for the advancement of AGI.\r\n<\/p>

\r\nHow it works<\/b>\r\n<\/p>

\r\nIn the human brain, information is transmitted by both electrical and chemical signals.<\/p>

\r\nWhen a neuron is activated, it generates an electrical impulse that, upon reaching the synapse (the small gap between two neurons), is converted into chemical signals.<\/p>

\r\nThese signals cross the space between the cells and become electrical signals again in the next neuron, continuing the transmission of information.\r\n<\/p>

\r\nYang and his colleagues managed to reproduce this physical process with great precision.<\/p>

\r\nThe major difference is that their artificial neuron occupies the same space as a single transistor-instead of the dozens or hundreds needed in conventional architectures.\r\n<\/p>

\r\nIn biological neurons, ions-electrically charged particles-are responsible for generating electrical impulses.<\/p>

\r\nIn the human brain, this role is played by elements such as potassium, sodium, and calcium.<\/p>

\r\nIn the USC experiment, the researchers used silver oxide ions to create electrical pulses that mimic this behavior and perform calculations associated with processes such as learning, movement, and planning.\r\n<\/p>

\r\n“Even though they are not exactly the same ions used by the brain, the physics behind the movement and ionic dynamics is very similar,” explains Yang.<\/p>

\r\nAccording to him, silver diffuses easily and offers the necessary dynamics to reproduce biological processes with a simple structure.<\/p>

\r\nTherefore, the device was named a *diffusive memristor*, in reference to the movement and diffusion of silver ions.\r\n<\/p>

\r\nYang emphasizes that the team chose to work with ion dynamics precisely because this is the mechanism used by the brain-the most efficient system ever produced by evolution.<\/p>

\r\n“It’s not that our current chips aren’t powerful enough.<\/p>

\r\nThe problem is that they aren’t efficient.<\/p>

\r\nThey consume too much energy,” he states.<\/p>

\r\nThis is particularly relevant given the enormous energy expenditure of artificial intelligence models based on machine learning, which process gigantic amounts of data.\r\n<\/p>

\r\nAccording to the researcher, our computers were never designed to learn on their own from a few examples-something the brain does naturally.<\/p>

\r\nTo improve both the energy efficiency and the learning capacity of machines, it is necessary to build artificial systems that follow the same principles as the brain.\r\n<\/p>

\r\nHe explains that, although electrons allow for fast operations, ions are a more suitable medium to imitate the brain.<\/p>

\r\nElectron-based computers rely on software-based learning, while the brain learns directly from its own “biological hardware”-or, as Yang says, a type of *wetware* (a “wet” system).\r\n<\/p>

\r\nA simple example: a child can learn to recognize handwritten numbers after seeing only a few examples, while a computer needs thousands.<\/p>

\r\nEven so, the human brain accomplishes this learning consuming only about 20 watts of energy-the equivalent of a household light bulb-while modern supercomputers use megawatts to perform the same type of task.\r\n<\/p>

<\/p>

\r\nA more efficient and intelligent future<\/b>\r\n<\/p>

\r\nThe new method represents an important step towards creating systems that mimic natural intelligence.<\/p>

\r\nYang notes that, although the silver used in the experiment is not compatible with traditional semiconductor manufacturing processes, other types of ions can be studied to achieve similar results.\r\n<\/p>

\r\nThe efficiency of *diffusive memristors* is evident in both energy and size.<\/p>

\r\nA typical smartphone contains about ten chips and billions of transistors that alternate between states 0 and 1 to perform calculations.<\/p>

\r\nYang’s proposed technology can use the space of just one transistor for each neuron, which would allow for a reduction in chip size and energy consumption by several orders of magnitude, making artificial intelligence more sustainable and closer to the efficiency of the human brain.\r\n<\/p>

\r\nNow that researchers have demonstrated compact and functional artificial neurons and synapses, the next step will be to integrate large quantities of them to test the extent to which it is possible to replicate the efficiency and capabilities of the brain.<\/p>

\r\n“The most exciting thing,” concludes Yang, “is that systems so faithful to the brain can not only bring us closer to true artificial intelligence, but also help us better understand how the brain itself works.”<\/p>

<\/p>

\r\n

\r\n\r\n

Published in 11\/05\/2025 07h57<\/em><\/p>\r\n

\r\n

Portuguese version<\/a><\/em><\/p>\r\n

\r\n\r\n\t\t

Text adapted by AI (Grok) and translated via Google API in the English version. Images from public image libraries or credits in the caption. Information about DOI, author and institution can be found in the body of the article.<\/p>\r\n\t\t

\r\n\t\t

Reference article:<\/p>\r\n\t\t\r\n\t\t\r\n\r\n\r\n