System Simulates Neuron Activity in Real-TimeCategory: Science & Technology
Posted: April 29, 2013 09:47AM
The most powerful computational machine man has access to is not some supercomputer in a laboratory, but the human brain. The reason the organ is superior is that its billions of neurons have the ability to operate in parallel, unlike supercomputers which largely operate sequentially. This makes it so difficult for a supercomputer to simulate the human brain that what would take a brain a second to do, the computer requires hours to complete. As reported by NSF though, researchers have developed a news system with the ability to simulate the brain in real-time, while being far less power-hungry than a supercomputer.
Called Neurogrid, the system is made up of only 16 chips with 65,000 simulated, silicon neurons each, and each neuron is networked to thousands of others. While this does emulate the structure of the brain, what really gives Neurogrids its simulation superiority is how the neurons behave. Within the brain, neurons send signals to each other in a binary way, like computers, but within each neuron signals are processed non-linearly. Essentially neurons communicate digitally but think in analog, and supercomputers have a hard time compensating for this, but Neurogrid's neurons have been designed to operate similarly.
Along with being significantly faster than a supercomputer, Neurogrid is also much more efficient at mimicking the brain. While a supercomputer like Sequoia at Lawrence Livermore National Laboratory can require 8 megawatts of electricity to run, Neurogrid can operate with just 5 watts, making it much more accessible for future research into the brain.
Source: National Science Foundation