By finding materials that act in ways similar to the mechanisms that biology uses to retain and process information, scientists hope to find clues to help us build smarter computers.
Inspired by human forgetfulness — how our brains discard unnecessary data to make room for new information — scientists at the U.S. Department of Energy’s (DOE) Argonne National Laboratory, in collaboration with Brookhaven National Laboratory and three universities, conducted a recent study that combined supercomputer simulation and X-ray characterization of a material that gradually “forgets.” This could one day be used for advanced bio-inspired computing.
“It’s hard to create a non-living material that shows a pattern resembling a kind of forgetfulness, but the specific material we were working with can actually mimic that kind of behavior.” –
Subramanian Sankaranarayanan, Argonne nanoscientist at ANL teams up with the study’s author.
“The brain has limited capacity, and it can only function efficiently because it is able to forget,” said Subramanian Sankaranarayanan, an Argonne nanoscientist and study author. “It’s hard to create a non-living material that shows a pattern resembling a kind of forgetfulness, but the specific material we were working with can actually mimic that kind of behavior.”
The material, called a quantum perovskite, offers researchers a simpler non-biological model of what “forgetfulness” might look like on an electronic level.
The perovskite shows an adaptive response when protons are repeatedly inserted and removed that resembles the brain’s desensitization to a recurring stimulus.
Quantum simulations performed at the Argonne Leadership Computing Facility (ALCF), a DOE Office of Science User Facility, probed the origin of this adaptive response.
“When scientists add or remove a proton (H+) from the perovskite (SmNiO3 (SNO)) lattice, the material’s atomic structure expands or contracts dramatically to accommodate it in a process called ‘lattice breathing,'” said Badri Narayanan, an Argonne assistant material scientist and co-author of the study. But when this happens over and over again, the material’s behavior evolves such that the lattice breathing is reduced — the proton “threat” no longer causes the material to hyperventilate. “The material’s electronic properties also evolve with this process,” said Narayanan.
“Eventually, it becomes harder to make the perovskite ‘care’ if we are adding or removing a proton,” said Hua Zhou, a physicist involved in characterizing the behavior of the material using X-rays provided by Argonne’s Advanced Photon Source (APS), a DOE Office of Science User Facility. “It’s like when you get very scared on a water slide the first time you go down, but each time after that you have less and less of a reaction.”
As the material responds to protons that scientists add and subtract, its ability to resist an electrical current can be severely affected. This behavior allows the material to be effectively programmed, like a computer, by the proton doping. Essentially, a scientist could insert or remove protons to control whether or not the perovskite would allow a current.
Researchers have recently pushed to develop non-silicon-based materials, like perovskites, for computing because silicon struggles to use energy as efficiently. Scientists may use perovskites in learning machines down the line. But scientists can also take advantage of perovskite properties by using them as the basis for computational models of more complex biological learning systems.
“These simulations, which quite closely match the experimental results, are inspiring whole new algorithms to train neural networks to learn,” said Mathew Cherukara, an Argonne postdoctoral scholar at the APS.
The perovskite material and the resulting neural network algorithms could help develop more efficient artificial intelligence capable of facial recognition, reasoning and human-like decision making. Scientists are continuing the research to discover other materials with these brain-like properties and new ways to program these materials.
Finally, unlike silicon, whose properties can be reliably described using simple computer models, understanding the perovskite material requires computationally intensive simulations to capture how its atomistic and electronic structure reacts to proton doping.
“A classical framework doesn’t apply to these complex systems,” said Sankaranarayanan who, along with Narayanan and Cherukara, modeled the perovskite’s behavior at Argonne’s Center for Nanoscale Materials, a DOE Office of Science User Facility, and the ALCF. “Quantum effects dominate, so it takes very computationally demanding simulations to show how the proton moves in and out of the perovskite structure.”
Habituation based synaptic plasticity and organismic learning in a quantum perovskite
Fan Zuo, Priyadarshini Panda, Michele Kotiuga, Jiarui Li, Mingu Kang, Claudio Mazzoli, Hua Zhou, Andi Barbour, Stuart Wilkins, Badri Narayanan, Mathew Cherukara, Zhen Zhang, Subramanian K. R. S. Sankaranarayanan, Riccardo Comin, Karin M. Rabe, Kaushik Roy & Shriram Ramanathan
Nature Communications 8, Article number: 240 (2017)
DOE/Brookhaven National Laboratory