Researchers at Sandia National Laboratories have shown that neuromorphic computers that synthetically mimic the brain’s logic can solve more complex problems than AI’s.
In a newly published article in the magazine Nature ElectronicsThe researchers detailed their findings showing that neuromorphic simulations using the statistical method called random walks can do all kinds of sophisticated calculations, such as tracking X-rays passing through bone and soft tissue, disease passing through a population, information passing through social networks flows and more .
According to Sandia theoretical neuroscientist and lead researcher James Bradley Aimone, in optimal cases neuromorphic computers can even solve problems faster while consuming less energy than conventional computers. This should be of particular interest to the high-performance computing (HPC) community, as statistical problems are not really appropriate for GPUs or CPUs.
Sandia engineer and the author of the new paper, Brian Franke, provided more insight in a press release about how neuromorphic computers can be more efficient than GPUs in certain scenarios, saying:
“The natural randomness of the processes you list makes them inefficient when mapped directly to vector processors such as GPUs in next-generation computing efforts. Meanwhile, neuromorphic architectures are an intriguing and radically different alternative to particle simulation that can lead to a scalable and energy-efficient -efficient approach to solving problems that matter to us.”
To conduct their tests, the Sandia researchers used the Loihi platform with 50 million chips that they received from Intel a year and a half ago.
While neomorphic computing isn’t meant to challenge other computing methods, there are other areas where the combination of computing speed and low energy costs makes it a better choice, according to Aimone.
At the same time, chips with artificial neurons are cheap and easy to install, unlike the problems associated with adding qubits to quantum computers. However, moving data to or from neurochip processors can get expensive because the more data they collect, the slower a system using them becomes until it eventually stops working altogether. However, Sandia’s researchers were able to overcome this obstacle by configuring a small group of neurons that computed summary statistics that were run instead of raw data.
Like the human brain, neuromorphic chips work by electrifying tiny pin-like structures and adding tiny charges that are emitted from surrounding sensors until a certain electrical level is reached. Then the pin flashes a small electrical burst like a biological neuron.
In the future, the next version of Loihi will increase its current chip scale from 128k neurons per chip to a maximum of one million with large-scale systems combining multiple chips on a board. Ultimately, a technology like Loihi may find its way into a powerful computing platform to make HPC more energy efficient, environmentally friendly and affordable.