Tiny droplets can form soft networks for brain-inspired computing

UNIVERSITY PARK, Pa. — Inside the human brain, billions of individual neurons connect to form a vast network capable of rapidly processing information and electrical signals. A team of researchers and engineers at Penn State is taking inspiration from these natural computers to build soft networks of water-based droplets, connected by membranes like those found in the walls of living cells. Their networks lay the foundation for a new kind of soft computer built for wet environments, capable of sensing and processing information with ions much like silicon chips do with electrons, but at far lower energy cost.

The team immerses tiny lipid-containing droplets in oil, where lipid molecules naturally arrange themselves around each droplet. When neighboring droplets touch, these lipids form a thin, two-layer membrane between them. The soft surfaces contain voltage-sensitive channels that allow electrical signals to pass between droplets through the movement of charged atoms or molecules called ions. The membranes store charge, while the channels respond to present and recent signals. When combined, these properties allow the droplets to form a synthetic, interconnected network that the researchers call a “neuromorphic tissue.”

They detailed these tissues’ properties in a paper published in Science Advances, using laboratory measurements and computer simulations to show that interconnecting the droplets’ membranes can significantly improve the material’s ability to process complex and unpredictable signals.

According to Joseph Najem, associate professor of mechanical engineering, associate director of the Convergence Center for Living Multifunctional Materials System and co-corresponding author on the paper, integrating a computer interface or sensor in the body, for example, is not as simple as implanting circuitry because even though both electronics and biology process signals and compute, they carry information differently. Conventional electronics work by moving charge carriers like electrons through conductive materials such as metal wires, while the electrical signaling that happens in the body relies on ions moving through fluids and across cell membranes.

“Scientists and developers must create interfaces in devices that use traditional silicon computing to process and translate between electronic and ionic signals,” said Najem, who holds an additional affiliation in the Intercollege Graduate Degree Program in Materials Science and Engineering at Penn State. “However, these interfaces can delay and distort signals, while adding significant energy costs. To address this problem, we must investigate creating new computing platforms that go beyond just electronics.”

Ionic computing explores approaches to processing information using ions instead of the traditional charge carriers used in electronics. Neuromorphic computing is an approach to computing that draws inspiration from the brain’s structure and information processing functionality. The team’s new approach uses aqueous compartments, lipid membranes and peptide ion channels — structures that mimic the natural architecture of the brain — to bring both the structure and functionality of these ideas together, drawing on the nodes' interconnectivity to create electrical feedback within the material.

With this architecture, the researchers can use reservoir computing, in which a changing input produces a pattern of responses that retains traces of earlier inputs. In this network, membrane charging and ion-channel activity generate responses, and the connections allow the droplets to influence one another. Coupled nodes can accumulate information in the form of electric charge, passing it to the next node in the chain. This behavior, which also occurs in neurons, allows electricity to rapidly transfer through each node without losing overall signal clarity. The membranes between the droplets give the network short-term, fading memory, so its response to a new input depends partly on recent inputs, Najem explained.

“In traditional systems, memory and processing are separated, which consumes a lot of energy to make those actions work together,” Najem said. “In the brain, information processing and memory occur in the same place. That’s why we are seeing interest in this idea of neuromorphic computing, as it could significantly reduce the amount of energy needed for operation.”

Joseph explained that individual droplets can connect to as many as six neighbors in a structure reminiscent of a honeycomb, noting that various arrangements provide different patterns of connectivity. To measure the dynamics of a droplet chain, the team introduced an electrical current, recording voltages at individual nodes. These measurements were compared with a physics-based model of the network, allowing the researchers to explore how connections and differences between droplets influence overall computational performance.

“When the droplets join together this way, you create coupling that enables rich electrical dynamics capable of processing complex signals that change over time,” Najem explained. “In our simulations, these coupled droplets generated more accurate signal predictions than the uncoupled reservoirs we tested.”

The team assessed computational performance using simulations based on its experimentally tested model, comparing interconnected reservoirs with parallel reservoirs whose node’s responded independently. Using benchmark tests that use mathematical equations to model chaotic behavior, the team found that connected networks reduced prediction error by nearly 50% when compared to independent nodes. By reducing these errors, the team can better understand how electricity will flow in extremely complicated systems.

Although external equipment is still used to supply inputs and measure responses, Najem said, the fact that the network’s recurrent connections arise within the material without external feedback circuitry is what makes it so exciting.

“With silicon, circuitry is needed to build a computer,” Najem said. “With our approach, the circuitry is the material itself: the membranes store charge, and ion channels regulate signals between droplets. This draws inspiration from the way biological membranes contribute to information processing and memory.”

The team plans to continue investigating their material, its properties and how it can be further optimized for ionic computing. According to Najem, potential future applications include adaptive sensors, soft robotics and computing at biological interfaces.

“Using this material, we want to create a framework to enable commercially scalable ionic computing,” Najem said. “Our hope is to one day harness this technology to make Jelly-like computer chips to complement and optimize electronics in the body.”

Additional co-first authors on the work include Ahmed S. Mohamed, a postdoctoral scholar at Penn State, and Nicholas X. Armendarez, a postdoctoral fellow at the University of Texas at Austin who conducted the work at Penn State as a doctoral student. Md Sakib Hasan, an assistant professor at the University of Massachusetts Dartmouth, contributed as a co-corresponding author.

This work was supported by the Army Research Office under award number W911NF-23-1-0314 and the U.S. National Science Foundation (NSF) under award numbers CCF-2403560, CMMI-2314559 and CCF-2403559. Additional seed funding was provided by Penn State’s Convergence Center for Living Multifunctional Materials and Systems. The content is solely the responsibility of the authors and does not necessarily represent the official views of the funders.

At Penn State, researchers are solving real problems that impact the health, safety and quality of life of people across the commonwealth, the nation and around the world.

For decades, federal support for research has fueled innovation that makes our country safer, our industries more competitive and our economy stronger. Recent federal funding cuts threaten this progress.

Learn more about the implications of federal funding cuts to our future at Research or Regress.