Scientists from IBM are developing a new computer architecture that will be better suited to handle increasing volumes of data coming from algorithms of artificial intelligence. They draw their inspiration from the human brain, and their developments have already greatly exceed traditional computers in the comparative studies. The results were published in the latest issue of the Journal of Applied Physics. Modern computers built on the von Neumann architecture, developed in the 1940-ies.
Computer system von Neumann includes a Central processor that controls logic and arithmetic, memory, storage, input devices and conclusions. Instead of this rather primitive “industrial” scheme scientists suggest the use of computers, created by the type of brain where centers of memory are combined.
Computer type brain
Abu Sebastian, the author of the work explains that the performance of certain computational tasks in the computer memory can improve system efficiency and save energy.
“If you look at people, we produce calculations, spending 20-30 Watts of energy, while AI-based supercomputers, require kilowatts or megawatts of power,” says Sebastian. “In the brain synapses at the same time calculate and store information. In the new architecture, which depart from the von Neumann, memory plays a more active role in computing.”
The IBM team three times borrowed ideas from the human brain. The first level involves the dynamic state of a memory device to perform the calculations directly in the memory, just as memory and information processing work together in the brain. The second level borrows the structure of synaptic networks of the brain to create a memory with exchangeable phase (PCM) to accelerate the training of deep neural networks. Finally, the dynamic and stochastic nature of neurons and synapses inspired the team to create a powerful computational substrates for breakthrough neural networks.
What do you think we should take the human brain? Tell us in our chat in Telegram.
New computer type architecture of the brain can improve data processing methods
Ilya Hel