The human brain is arguably an intriguing and fascinating computer. Despite the slower speeds of neurons as compared to modern transistors, the brain still outperforms the best supercomputers in common tasks such as picture recognition. A challenge yet to be met is implementing systems that, inspired by the brain’s working mechanism, are capable of efficiently solving complex problems.

Researchers from Stanford University and Sandia National Laboratories have realized a device capable of mimicking brain synapses that operates with remarkably low power consumption. Their system, which relies on inexpensive and flexible materials, turned out to be particularly efficient when implemented in neural network simulations. This work, carried out by the teams of Alberto Salleo at Stanford and Alec Talin at Sandia Labs, was published in Nature Materials.

In the quest to achieve brain-inspired computation, complementary metal oxide semiconductor (CMOS)-based neural architectures and memristors have been intensively pursued. Although compelling, these approaches have concomitant limitations, which ultimately preclude them providing efficient interconnectivity, high information density, and energy efficiency. CMOS architectures, for example, are significantly limited by the volatility of their states, and require complex designs and high operating voltages. Memristors, on the other hand, offer nonvolatility at the expense of high power switching. Salleo and co-workers have developed a novel electrochemical device that, enabled by the nonvolatile control of the conductivity of an organic ionic/electronic conductor, circumvents these limitations. Their neuromorphic organic device, which they named ENODe, combines a low-voltage operation (<1 V) and a high density of nonvolatile states and is compatible with flexible electronics.

Salleo says, “From an experimental standpoint, the devices proved exceptionally easy to make and robust. I think the biggest challenge for us was to figure out how they fit within the landscape of other memristive-type devices. Other technologies are so different (phase-change memories, filament-forming oxide layers) and usually 2-terminal (our device has 3 terminals), that a meaningful and fair comparison was sometimes difficult to construct.”

Paul Meredith, a materials expert from Swansea University and a leading scientist in the area of bioelectronics, who was not involved with this study, says, “This is an excellent work—beautifully executed.”

When asked how this work opens the door in advancing brain-computer interfaces, Salleo is cautious, “I am not certain that interfacing with the brain is necessarily our next step….One thing our devices have going for them in this respect is that they work at very low voltages, compatible with neuron action potentials.” Salleo is particularly excited about the new avenues enabled by this advance, and identifies as a crucial next step scaling up this technology by taking advantage of the power offered by microlithography.

Although far from actually mimicking the operation of a full brain, these findings are particularly interesting for crucial areas of research such as machine learning, artificial intelligence, and pattern recognition, which could dramatically transform the way people interact with intelligent electronic devices. Perhaps efficient computer–brain interfacing might not after all be science-fiction. New ventures such as Elon Musk’s Neuralink point to that direction.

Read the abstract in Nature Materials.