Presented by self-learning neuromorphic chip capable of composing music
IMEC ITF2017 at a recent forum the representatives of the Belgian Research and Scientific Center have demonstrated the first of its kind, self-learning neuromorphic chip, the principles of which imitate the principles of the human brain works. The chip has the ability to self-learning, which is enough for the "works" of simple chip music.
At the heart of the chip is OxRAM technology. IMEC research center experts created the "building blocks" of hardware and software systems, the simplest of artificial intelligence. The combination of these tools has allowed to place on a small chip chip consumes very little power, function of machine learning and self-study. Using these functions, the chip independently carries out data analysis, and the greater the number of associations is found in the data, the stronger will be the connection between the individual chip elements. Thus, the "inside" of the chip is formed something like neuronal connections.
chip manufactured by Crystal 65nm technology. It has elements of conventional CMOS-logic and memory arrays of MRAM (Metal Oxide Resistive RAM). This memory is 100 times more efficient in terms of energy than the "conventional" types of memory. Due to the large volume of the memory chip can create and use 40 stable associative relations. During the demonstration chip "learned" to compose music, receiving data from several pieces of music. In this case, the rules of composition and genre of "invented" the chip itself. As one of the employees of the center IMEC Paravin Regavan,
"Right now, we already have the necessary hardware, software and online programming environment that will promote far forward area neuromorphic computing. And now we have the ability to create on the basis of the new chips complex intelligent systems. In this system the level of intelligence will grow in a quadratic or exponential function of the number of neuromorphic chips included in it. "