In an effort to electronically mimic neurons—the nerve cells of the brain—scientists from the Sensors and Electronics Laboratory of HRL Laboratories, LLC have successfully demonstrated electronic neuron circuits that exhibit as many as 23 known behaviors of biological neurons and the three qualitatively distinctive classes of neuron activation (excitation) that code information between neurons about sensory events, cognitive processes, or motor actions. The neuron-emulating circuits were built using two exotic nanoscale switches called active memristors with a few common circuit components. The importance of the miniature size of the circuit is that memristor-based neurons could be building blocks for an entirely new type of computer that mimics the human brain.
Built from electronic neurons and synapses that use very little power, “neuromorphic” computers that rival the human brain’s size, low power usage, and high computational capabilities could be a major leap beyond our current silicon-chip-based computers that use the classic Von Neumann architecture. Neuromorphic computing has emerged as a revolutionary paradigm, in which parallel and energy-efficient information processing is achieved with biologically inspired models of neurons and synapses (the tiny unions between neurons through which they communicate), circuit-organizing principles, and network learning strategies. It differs entirely from current Von Neumann-architecture computers that run on static algorithms based on the Turing Principle.
It may require two steps to build an electronic brain that can possibly compete with a biological brain in speed and energy efficiency
The active memristors in the study were made with a tiny nanocrystal of vanadium dioxide (VO2) material sandwiched between two metal layers. The combined material can switch between an insulating phase and a metallic phase when a minuscule current is passed through it. The coordinated switching of two active memristors coupled with resistors and capacitors mimic the neuronal electrical impulses (spikes) that encode information in the brain.
A related nanoscale device—the passive memristor—can retain its history in the form of electrical resistance, emulating how synapses work. Memristors do not suffer the limitations of silicon transistors because of their diminutive size, ultra-low power consumption, and realistic mimicry of biological neurons and synapses. This HRL Laboratories breakthrough could enable a neuromorphic computer architecture that uses only memristors, with no silicon transistors needed. Using active and passive memristors, future computers could rival the human brain in flexibility and learning capabilities.
“We are trying to build an intelligent information machine that can achieve some of the computations that the human brain does, such as delicate motion controls, attention, reasoning, association, and decision making,” said Principal Investigator Wei Yi. “For example, humans are much slower than current computers at arithmetic operations, in which all values have to be precisely calculated. But humans are superior to classical digital computers at intelligent tasks, such as fusing sensory data, mining data, filtering attention choices, induction and deduction, and even developing new concepts. Human brains do not run on preprogrammed and fixed algorithms. The brain’s intricate, highly parallel network of 100 billion neurons and 100 trillion synapses is dynamic and adapts as it gains experience. At the end of the day, we are thinking about the possibility of building an adaptive electronic brain that might be able to self-learn. This seems a grand dream, but building biologically plausible neurons is a key to eventual memristive computers that mimic the cerebral cortex.”
The HRL paper, entitled Biological plausibility and stochasticity in scalable VO2 active memristor neurons, was published in Nature Communications on November 7th, 2018.
“It may require two steps to build an electronic brain that can possibly compete with a biological brain in speed and energy efficiency,” Yi said. “The first step is a departure from von Neumann architecture, so that memory and computation are spatially and temporally integrated instead of separated. That will save energy and data-shuffling time. The second step is a departure from the silicon transistors used to build essentially all current computing rigs. Silicon transistors were not created, and are not necessarily the optimal device, for mimicking biological neurons and synapses. Silicon neurons and synapses are not capable of duplicating the rich dynamics of their biological counterparts without sacrificing energy consumption and size. A more promising approach is to use active and passive memristors to unleash the potential of neuromorphic computing.”
The other authors on the paper were Kenneth K. Tsang, Stephen K. Lam, Xiwei Bai, Jack A. Crowell, and Elias A. Flores.
Read more about the article in Nature’s Behind the Paper, written by Wei Yi:
HRL Laboratories, LLC, Malibu, California (hrl.com) is a corporate research-and-development laboratory owned by The Boeing Company and General Motors specializing in research into sensors and materials, information and systems sciences, applied electromagnetics, and microelectronics. HRL provides custom research and development and performs additional R&D contract services for its LLC member companies, the U.S. government, and other commercial companies.
Media Inquiries: media[at]hrl.com, (310) 317-5000