What is Neuromorphic Computing?

August 4, 2020

Traditional computers work linearly with strict instruction sets and logic gates. As demands on computing power continue to increase, such rigid design has limitations. Neuromorphic computing uses nodes and analog circuitry to create artificial neurons and synapses to mimic the complex neural network the human brain. A neuromorphic computer is faster and more energy-efficient than traditional computers and may be the future of artificial intelligence.

Neuromorphic Computing Projects

IBM revealed its neurosynaptic processor, TrueNorth, in 2014. The processor was built as part of DARPA’s SyNAPSE program, which seeks to develop neuromorphic technology. IBM continues to improve the processor’s capabilities, though it has not been released for use. In 2017, TrueNorth was demonstrated in a gesture-recognition system that can recognize 10 hand gestures with 96% accuracy within a tenth of second of the gesture’s start. Intel introduced its Loihi neuromorphic chip in 2017 – also a research project not yet available for commercial use. Since then, the Loihi chip has learned to identify hazardous chemicals by smell. The chip learned 10 smells from a small sample of each, outperforming traditional deep learning algorithms that required thousands more samples.
Intel has also combined 768 Loihi chips in a system that contains over 100 million neurons. The system was created to explore the capacity of neuromorphic computing to handle workloads that run slowly on traditional systems.

SpiNNaker, a neuromorphic supercomputer capable of simulating up to a billion neurons, was first introduced in November 2018. Though the goal of the project is to help neuroscientists better understand how the human brain works, the project also has implications in robotics. SpOmnibot uses SpiNNaker chips to make real-time decisions in navigating the world around it.

MIT released a neuromorphic chip design in 2018. The design is unique in its ability to control the flow of ions through single-crystalline silicon versus typical amorphous materials through which the flow of ions is more chaotic. Recently, MIT downsized the chip to smaller than a piece of confetti. However, the chip isn’t a physical product and currently only works in simulation.

Future of Neuromorphic Computing

The goal of neuromorphic computing is to achieve supercomputer levels of computation while operating at the power level of the human brain, which runs at the equivalent of 20 watts. Super computers run on around 4 megawatts (millions of watts) of power while Intel’s Loihi chip system runs under 500 watts.

Neuromorphic computing also has the potential to advance current artificial intelligence technologies, enabling faster real-time decisions with accurate predictive computations – with obvious applications in smart cars, robotics, defense and healthcare.

SHARE
TwitterFacebookLinkedIn