Neuromorphic Computing

CPUs of modern day computers are great at precise and complex calculations is analogous to a brain, but its nothing like the natural brains found in skulls. But have you wondered how to wire a computer like a human brain? A trending field of artificial intelligence called neuromorphic computing is trying to mimic the working of the neurons within our brain. Intel and IBM researchers are converting true silicon brains to reality. Technical approach for these sort of projects involves neural networks. Neural networks are basically software approaches, mimicing the working of brains. Like machine learning, results of neural networks relies on examples.

Sometimes it may require thousands or even millions of examples to attain thye desired results. IBM brought about their brain inspires chip called TrueNorth in 2014, while Intel introduced its chip, called Loihi in 2017, in an attempt to mimic working of brains at hardware level.

Same silicon transistors are used in two neuromorphic chips, interconnecting like neurons. One million neurons are connected by 256 million synapses in TrueNorth, while 130,000 neurons are connected by over 130 million synapses in Loihi. Both of them combined memory and computation within the same chip. Their neurons fire independently, whose timings depends upon the spikes of such activities. This can be used as an effective way to encode information. This processes allows neuromorphic chips to use less energy and learn quickly and effectively than a conventional CPU. Neuromorphic computers can be useful in tasks involving constraint satisfaction and optimization. Intel wired 768 Loihi chips together to create Pohoiki Springs, ranging 5 servers and boasts about 100 million neurons. It needed under 500 watts of power for operation, almost half of a conventional gaming PC. As the technology develops and matures, complex and tougher problems can be solved which is beyond the reach of our current CPU “brains”. But since its hardware is emerging, its software still needs time to develop.

Leave a Comment

Your email address will not be published. Required fields are marked *