Прочитайте цю статтю вголос
Увімкніть мікрофон і читайте у своєму темпі. Слова підсвічуються на ходу, і ніхто не зупиняє вас через одну помилку.
Ваш звук не потрапляє до MovaReader: розпізнає його ваш браузер. Політикою конфіденційності
Увійдіть, щоб ця сесія збереглася в історії читання.
Neuromorphic Computing and Brain-Inspired Architectures
Elena wanted one distinction kept straight. The pieces she was reading called almost any low-power neural chip a brain, and she needed to know what IBM's TrueNorth actually was. She was not hunting for a slogan about the end of ordinary computers. She wanted the architecture: where the memory sat, and what moved.
TrueNorth, as IBM described it, holds one million programmable spiking neurons and 256 million configurable synapses on a single chip. The synapses are not stored off to the side for a processor to fetch, one weight after another. They sit with the neurons, across a large grid of cores. A conventional machine in the von Neumann pattern keeps memory apart from the arithmetic. For data-heavy work, a large share of the effort is the trip between the two. TrueNorth's bet is to shorten that trip.