Release date: 2015-06-30 Memristor "neural network" Recently, researchers at the University of California, Santa Barbara, demonstrated a simple artificial neuron line containing 100 human synapses, demonstrating for the first time that this line can perform simple human visual functions—classifying images. It marks a major advancement in artificial intelligence. The human brain has an advantage over the computer Despite the potential flaws in the human brain, the calculations make mistakes, but maintain a powerful and efficient computing model that can perform certain special tasks in less than a second, and a computer needs to complete these tasks. The task takes more time and consumes more energy. What are these features? For example, if you read an article, your brain will make countless instantaneous decisions on the letters and symbols you see, distinguishing their shapes, relative positions, and deriving different levels of meaning based on many background channels. All this happens. You read the article for a short time. Change the font, even the letter direction, you can still read and infer the same meaning. Researchers published a paper in Nature that the lines they developed used basic artificial neural networks to successfully distinguish between the three letters “zâ€, “v†and “nâ€, each letter. There are a variety of styles to present, or to add a variety of "interference." This process is like human beings finding out their friends from a group of people, or picking the right one out of a string of similar keys. Simple neural circuits can correctly distinguish simple graphics. Demetri Stakov, a professor of electrical and computer engineering at the University of California, Santa Barbara, said: "This is a small step, but it is an important step." With further development, the line may eventually be extended to upgrade. Close to the human brain, there are about 100 trillion synaptic connections between human brain neurons. One of the authors, Farnold Merrick Beyate of the School of Electrical and Computational Engineering, said: "Although the line is very small compared to the actual neural network, it is enough to prove the practicality of the concept." Another paper author, Gina Adam, also said that as people's interest in this technology increases, the research will be more dynamic, "more technical problems can be solved, allowing it to enter the market faster." Memory storage The key to this technology is the memristor (a combination of "memory" and "resistance") whose resistance changes depending on the direction of charge flow. Traditional transistors rely on the drift and diffusion of electrons and holes in semiconductor materials. Memristors operate on an ion-based basis, similar to the way human nerve cells produce neural electrical signals. Stakov said: "Memory storage is a special form of erbium concentration distribution that can move back and forth within a memristor." Compared to pure electronic memory, the ion memory mechanism has many advantages and is more suitable. In artificial neural networks. “For example, a variety of different ion concentration distributions will result in a continuous memory state that mimics the memory function.†Ions are heavier than electrons and do not tunnel easily, which allows one to greatly upgrade memristors without sacrificing their simulations. performance. This kind of simulation is better than digital memory: in order to achieve the same functions as the human brain with traditional technology, the device must be large and load a large number of transistors, which also consumes more energy. Merck Pritz Oso, the first author of the paper, said: "People have found that in efficient brain-like computing, the architecture of traditional computers always has inevitable limitations. Membrane-based technology is inspired by the biological brain. , performing the calculation in a completely different way." However, in order to get close to the human brain function, more memristors are needed to build a more complex neural network, in order to do things that humans can do without any effort, such as identifying a different thing, or relying on Other objects in a scene, rather than the target itself, infer whether there are objects that are not found. Future computers have new ideas “The most exciting thing is that this technology is different from most other weird solutions. It is not difficult to integrate it with ordinary processing units, and it greatly promotes the development of computers in the future.†Pritz Oso Say. Currently, there are already areas where such emerging technologies may be used, such as medical imaging, which improves the navigation system to enable navigation based on images. As market demand has grown, digital transistors predicted by Moore's Law have multiplied, and traditional electronic devices will become too cumbersome. Researchers are working on energy-efficient lines, and there is still a long way to go to build high-performance computers and memory storage devices. At present, researchers continue to improve the performance of memristors, upgrade the complexity of the line, and increase the functionality of artificial neural networks. Next, they will integrate a memristor neural network with traditional semiconductor technology to demonstrate more complex functions, making this early "artificial brain" more complex and subtle. Ideally, this “artificial brain†consists of the vertical integration of trillions of such memristor devices. One of the authors, material scientist Brian Hoskins, said: “They have many potential applications. No doubt it gives us a whole new kind of thinking." Source: China Science and Technology Network - Technology Daily Sexual Enhancement Materials,Health Care Material,Icariin Rich Foods,Forsythia Suspensa Extract Xi'an XJeson Biotech Co., Ltd , https://www.xjesonbio.com