Who is the most sophisticated computer in the world? The answer to this question is the human brain. Now ... he can learn new things quickly with extreme ease in the presence of very little information. Can recognize objects, to recognize speech and react to change the situation and to interpret context. Since the dawn of digital technology, scientists are trying to build a computer to mimic the way the brain works. The technology giant IBM is among these enthusiasts by launching such an ambitious project several years ago. Now there are first results. The company announced that they have developed a chip whose architecture is inspired by the structure of the human brain.
In addition to the technical development of a computer imitation of the most complicated human body is extremely difficult, it is complicated by the fact that even now there are white patches in the knowledge of scientists and doctors for some functions and processes. Therefore, attempts recently in imitation of brain activities are focused more in the creation of software programs and solutions. Simply were not available technologies to create hardware with similar architecture and capabilities. In recent years, scientists and engineers yet started to develop computer chips that are different from the traditional concept of processor architecture. They rely on the construction of a system similar to that of the nerve cells (neurons) in the brain. The human body is composed of more than 100 billion. Neurons that are associated with each other.
Information of pulses
Result achieved by IBM's processor TrueNorth (compass heading that indicates the geographical north - note. Eds.), Which has 4,096 cores, and its overall structure is equivalent to one million neurons and 256 million. Synaptic connections. Namely neurons and synapses are the main elements that make up the human brain and give most of his abilities. The total number of transistors is 5.4 billion., Making it the largest components like chip manufactured by IBM ever.
"In 2001 we had a core processors.
Now able to increase this number up to 4096 cores and also to reduce the size of each of them 15 times and 100 times the energy consumption", told the magazine. Forbes Dr. Dharmendra Modha, Chief researcher at IBM, which manages the project.
Like complex biological body of the chip IBM uses a special technology called pulsed neurons. The idea is that the processor can encode the information as a series of pulses in the same way as neurologists believe that the brain store memories and data. All this architecture is significantly different from the traditional approaches to the construction of computational processors. We rely on two separate components - calculation (CPU) and a graphics processor (GPU). The first is the heart of the computer and performs normal operations while the second got on the provision and rendering images and more difficult tasks. None of them is not good at coding of information in the way that makes the human brain. Therefore, IBM's chip can be very useful for more specific tasks. He does not have a separate CPU or GPU modules, so the possibilities are different from those of traditional computers.
"This really is a successful experiment in processor architecture," said Carver Mead, professor emeritus of engineering and applied science at the Technological Institute of California, quoted by the magazine Wired. Mead is considered a "grandfather" systems inspired by the human brain. "Traditional processors are not good at storing information in the way that makes the brain. Presentation of the data as a nerve impulse is unfamiliar to current computers."
parallel calculation
The new processors are very useful in analyzing the data. New learning machines will be able to solve problems related to the calculation of the movement or information from different sensors. Will be able to recognize people, objects, human speech and others. Currently in similar tasks modern computers sometimes give very strange results. The reason is that the architecture of TrueNorth allows to process data in parallel. For example, in face recognition one core only deals with search matches the nose, ears another third of the distance between the eyes, etc. To increase the efficiency of the development of IBM, the company makes it possible to string together dozens of chips in a system. It has created a motherboard, which can be installed up to 16 new processors and they act as a single network, similar to neural. An important feature is that the new chips are produced with currently available components. Just changed the architecture and mode of interaction between them. Thus, one can easily start to mass production.
IBM program is part of the American SyNAPSE Advanced Research Agency DARPA, which includes a number of universities. Its purpose is to break the pattern of von Neumann microprocessor architecture by creating new hardware that does not require continuous transfer of data from main memory to GPU and vice versa. To that end, IBM has received funding from DARPA for 53 million. Dollars since 2008.
Von Neumann model does not allow efficient parallel processing and requires a lot of energy for its transfer from RAM to the CPU. By using the "pulsed neurons" substantially decreased considerably and the power consumption. When streaming video at 30 frames per second TrueNorth consumes only 63 mW power as process information in real time. This is only a small fraction of the energy it consumes processor in the latest models of laptops that typically use about 15W.
