Memristor Circuit. Researchers at UC Santa Barbara and Stony Brook University successfully built a neural network to house memristors. The prototype was successful in recognizing small images and may be expanded to develop futuristic computers that simulate the human brain. (via UC Santa Barbara)
A team at the University of California – Santa Barbara and Stony Brook University is on the brink of finally breaking the secret on how to develop memristors on their own neural hardware using old perceptron technology. Memristor research has been a long time coming, but if the researchers are successful, the devices can assist in computer energy consumption management and may allegedly lead to thinking computers that mimic human neurons and synapses.
Memristors, or memory resistors, are thought to be a crucial component to developing computers that can really “think” like human brains. A human brain will build brand new synapses based on an individual’s need for a particular type of information. A mathematician, for example, would have a very different brain, structurally, than a musician, because the part of the brain most used would become more developed over time. Computer scientists think memristors are the key to allowing computers to work in this way, as they can regulate the flow of electrical energy to various circuits, based on which circuits are most frequently used.
Concept Blueprint (via UC Santa Barbara & Nature)
Although memristors are a common topic of conversation for future computer-building, scientists struggle with building a neural hardware to house them. The new study published by UC Santa Barbara and Stony Brook University, however, may change that. The team built a 12 x 12 memristive crossbar assay that functions as a single perceptron, or an early neural network often used for pattern recognition and basic information organization. The team programmed a network of perceptrons to decipher things like letters and patterns and say together, the micro hardware functions as a collection of basic synapses.
The hardware is built using aluminum and titarium, but manufactured under low temperatures to allow for monolithic three-dimensional combination. This allows for the memristor to “remember” the amount of energy and the direction of the previous current for future use, even after the main device has been powered off. This recognition is currently possible using other technology, but it is much more involved. Using memristors means easier functionality while using no power.
In the trial, the memristor model was able to decipher 3 x 3-pixel back-and-white patterns into three types. The model they created had thee outputs, ten inputs and 30 perceptron synapses. In future, the team plans to shrink the current device down to 30nm across, in the hopes to simulating 100 billion synapses per square centimeter.
While some argue computers will never have the real processing power of the human brain, others say memristors will still be useful as analog memory devices or components of logic for larger systems. Since they use no energy, but record energy used, memristors may also be useful for energy management.
See more news at: