Q-MEEN-C of UC San Diego is developing brain-like computers by simulating neurons and synapses in quantum materials. Recent discoveries in non-native interactions represent a critical step towards more efficient AI hardware that could revolutionize AI technology.

We often think that computers are more efficient than humans. After all, computers can instantly solve complex math equations and remember names we might forget. However, the human brain can process complex layers of information quickly, accurately and with almost no energy expenditure. Examples of such tasks are recognizing a face seen only once or distinguishing a mountain from an ocean. These seemingly simple human functions require significant processing and energy from computers, and even then the accuracy of results can vary.

In search of mental computing

Creating brain-like computers with minimal energy requirements could revolutionize nearly every aspect of modern life. Quantum Materials for the Nationwide Energy Efficient Neuromorphic Computing (Q-MEEN-C) Consortium led by the University of California at San Diego and funded by the Department of Energy is at the forefront of this research.

Alex Fragno, associate professor of physics at UC San Diego, is co-director of Q-MEEN-C and is taking over the work of the center step-by-step. In the initial phase, he worked closely with UCLA Honorary Chair and Professor of Physics Robert Dines and Rutgers Professor of Engineering Sriram Ramanathan. Together, their teams have successfully found ways to create or mimic properties of a single brain element (such as a neuron or synapse) in a quantum material.

New discoveries and milestones

Now in its second phase, a new Q-MEEN-C study has been published. Nano Lettersindicates that electrical stimuli transmitted between adjacent electrodes can also affect non-adjacent electrodes. This discovery, known as nonlocalization, is a major milestone on the road to new types of devices that mimic brain function, known as neuromorphic computing.

“The brain understands that these nonlocal interactions are nominal—they happen frequently and with minimal effort,” said Fragno, one of the paper’s co-authors. “It’s an important part of how the brain works, but similar behaviors replicated in synthetic materials are rare.”

Like many research projects currently bearing fruit, the idea to test whether nonlocality in quantum materials was possible was born during the pandemic. The physical lab facilities were shut down, so the team performed calculations on arrays containing multiple devices to simulate multiple neurons and synapses in the brain. While running these tests, they discovered that nonlocality was theoretically possible.

From theory to practice

When the labs reopened, they developed the idea and hired Associate Professor Duiga Kuzuma of the UC San Diego Jacobs School of Engineering whose work in electrical and computer engineering helped turn the simulation into a real device.

This involved taking a thin film of nickelate, a ceramic “quantum material” that exhibits rich electronic properties, placing hydrogen ions, and then placing a metal conductor on top of it. A wire is attached to the metal so that an electrical signal can be sent to the nickel. The signal causes the gel-like hydrogen atoms to switch to a particular configuration, and when the signal stops, the new configuration remains.

“Actually, this is what a memory looks like,” Franjo said. “The device remembers that you messed up the material. Now you can fine-tune where these ions go to create more conductive and easier ways for electricity to flow.”

to a simplified design

Traditionally, building networks that carry enough electricity to power something like a laptop requires complex circuits with permanent ports, which are inefficient and expensive. The design concept of the Q-MEEN-C is much simpler because the non-local behavior in the experiment means that not all the wires in the circuit need to be connected together. Consider a net where the movement of a part of it can be felt across the web.

This is similar to how the brain learns: not linearly, but in a complex way. Each piece of learning creates connections in many areas of the brain, allowing us to distinguish not only trees from dogs, but also an oak tree from a palm tree or a golden retriever from a poodle.

Pattern recognition problem

Today, these pattern recognition tasks, which the brain does very well, can only be simulated using computer software. AI programs like ChatGPT and Bard use advanced algorithms to mimic brain activities like thinking and typing. And they do it very well. But without the appropriate advanced hardware to support it, software will reach its limits at some point.

Next stage and result

Franjo is excited about the hardware revolution that will be similar to what is currently occurring in software, and he brings scientists one step closer by showing that it is possible to reproduce non-native behavior in a synthetic material. The next step will involve creating more complex arrays with more electrodes in complex configurations.

“This is a very important step in our efforts to understand and model brain function,” said Dines, who is also a co-author. “Demonstrating a system with nonlocal interactions takes us further in the thinking direction of our brains. Our brains are of course much more complex than that, but a physical system capable of learning would need to be highly interactive, and this is a necessary first step. It now has long-term coherence in space and time. we can think

“It’s common knowledge that for this technology to really explode, we need to find ways to improve the hardware – the physical machine that can perform tasks along with software,” Fragno said. Said. “The next step is to create efficient machines that learn physical properties. This will give us a new paradigm in the world of artificial intelligence.” Source