Computer Chips That Imitate the Brain


Using electrical pulses, a novel microelectronics device can program and reprogram computer hardware on the fly.

What if a computer could learn to rebuild its own circuits based on the data it receives?

A multi-institutional cooperation led by the US Department of Energy's (DOE) Argonne National Laboratory has developed a material that can be utilized to make computer chips that can do precisely that. It accomplishes this by simulating brain activities using "neuromorphic" circuitry and computer architecture. Shriram Ramanathan, a Purdue University professor, headed the team.

“Human brains can actually change as a result of learning new things,” said Subramanian Sankaranarayanan, a research co-author with dual appointments at Argonne and the University of Illinois Chicago. ​“We have now created a device for machines to reconfigure their circuits in a brain-like way.”

With this capabilities, artificial intelligence-based computers may be able to do challenging tasks more rapidly and precisely while consuming significantly less energy. Analyzing complex medical pictures is one example. Autonomous autos and space robots that can rewire their circuits based on their experiences are more future examples.

Hydrogen ions in the nickelate enable one of four functions at different voltages (applied by platinum and gold electrodes at the top). The functions are artificial synapse, artificial neuron, capacitor, and resistor. The capacitor stores and releases current; the resistor blocks it. Credit: Argonne National Laboratory

The essential material in the new gadget is perovskite nickelate, which is composed of neodymium, nickel, and oxygen (NdNiO3). The researchers filled this material with hydrogen and connected electrodes to it, allowing them to apply electrical pulses of varying voltages.

“How much hydrogen is in the nickelate, and where it is, changes the electronic properties,” Sankaranarayanan explained. ​“And we can change its location and concentration with different electrical pulses.”

“This material has a many-layered personality,” Hua Zhou, an Argonne scientist and article co-author, remarked. ​“It has the two usual functions of everyday electronics — the turning on and blocking of electrical current as well as the storing and release of electricity. What’s really new and striking is the addition of two functions similar to the separate behavior of synapses and neurons in the brain." A neuron is a single nerve cell with which other nerve cells communicate via synapses. Neurons initiate sensory perception of the outside environment.

The Argonne team contributed by doing computational and experimental characterization of what happens in the nickelate device at various voltages. They used DOE Office of Science user resources at Argonne to do this, including the Advanced Photon Source, Argonne Leadership Computing Facility, and Center for Nanoscale Materials.

The results of the experiments showed that just changing the voltage affects the flow of hydrogen ions within the nickelate. A certain voltage concentration concentrates hydrogen at the nickelate core, resulting in neuron-like activity. A separate voltage transports the hydrogen from the core, resulting in synapse-like activity. At higher voltages, the resultant locations and concentrations of hydrogen cause computer chip on-off currents.

“Our computations revealing this mechanism at the atomic scale were super intensive,” said Sukriti Manna, an Argonne scientist. The team used not just the Argonne Leadership Computing Facility, but also the National Energy Research Scientific Computing Center, a DOE Office of Science user facility at Lawrence Berkeley National Laboratory.

Experiments at the Advanced Photon Source's beamline 33-ID-D helped confirm the process.

“Over the years we have had a very productive partnership with the Purdue group,” Zhou added. ​“Here, the team determined exactly how atoms arrange within the nickelate under different voltages. Especially important was tracking the material’s response at the atomic scale to the movement of hydrogen.”

Scientists will use the team's nickelate gadget to build a network of artificial neurons and synapses that can learn and adapt based on their experiences. This network would expand or contract when new information is supplied, allowing it to function with extraordinary energy efficiency. And lower operational expenses are a result of the increased energy efficiency.

Brain-inspired microelectronics, using the team's technology as a foundation, may have a promising future. This is especially true given that the device may be built at room temperature using approaches that are consistent with semiconductor industry norms.

The DOE Office of Basic Energy Sciences, as well as the Air Force Office of Scientific Research and the National Science Foundation, supported Argonne-related research.

Reference: “Reconfigurable perovskite nickelate electronics for artificial intelligence” by Hai-Tian Zhang, Tae Joon Park, A. N. M. Nafiul Islam, Dat S. J. Tran, Sukriti Manna, Qi Wang, Sandip Mondal, Haoming Yu, Suvo Banik, Shaobo Cheng, Hua Zhou, Sampath Gamage, Sayantan Mahapatra, Yimei Zhu, Yohannes Abate, Nan Jiang, Subramanian K. R. S. Sankaranarayanan, Abhronil Sengupta, Christof Teuscher and Shriram Ramanathan, 3 February 2022, Science.

DOI: 10.1126/science.abj7943

Previous Post Next Post