Developing Aqueous Volatile Memristors for Neuromorphic Computing: Bridging Biological and Artificial Information Processing

Significance 

The human brain’s remarkable information processing capabilities have long inspired researchers to explore brain-inspired, or neuromorphic, computing paradigms. The brain operates with extraordinary efficiency, far surpassing conventional computing systems in both speed and energy consumption. This efficiency is largely attributed to the brain’s unique architecture, which employs neurons connected by synapses and utilizes ions in an aqueous environment as information carriers. This starkly contrasts with traditional computing systems that rely on solid-state devices and electronic signals. As modern computing faces challenges related to energy consumption and processing speed, the quest to develop devices that mimic the brain’s processing capabilities has become increasingly urgent. One significant challenge in developing neuromorphic computing devices is replicating the brain’s ability to utilize both electrical and chemical signals within an aqueous environment. Conventional memristors, or memory resistors, have shown promise as artificial synapses due to their ability to emulate synaptic behavior. However, most existing memristors are solid-state devices that rely solely on electronic signals, limiting their ability to mimic the brain’s complex signaling mechanisms. Additionally, these devices often involve intricate fabrication processes and materials that do not naturally interface with biological systems. Thus, there is a pressing need for developing memristors that can operate in an aqueous environment, leveraging ion transport to more closely replicate the brain’s functionality.

New study published in PNAS and led by Professor René van Roij  from the Utrecht University in the Netherlands and conducted by Tim Kamsma, Willem Boon alongside Jaehyun Kim, Kyungjun Kim, Cristian Spitoni, and Professor Jungyul Park from Sogang University developed an aqueous volatile memristor. This device aims to emulate the brain’s short-term synaptic plasticity through ion transport in water, creating a more faithful representation of neural processes. The researchers hypothesized that by utilizing an aqueous medium and ions as signal carriers, they could create a memristor that better mimics the brain’s natural environment and signaling dynamics. Their goal was to design a device that is not only efficient and stable but also easy to fabricate, thus providing a practical pathway for integrating such devices into neuromorphic computing systems.

This study’s primary motivation was to bridge the gap between the brain’s fluidic ion transport mechanisms and the solid-state nature of conventional memristors. The researchers sought to develop a theoretical model to support the design and operation of the aqueous memristor, providing insights into its behavior and guiding its practical implementation. They aimed to demonstrate that an aqueous environment could enhance the memristor’s ability to perform complex computing tasks, particularly those involving temporal signal processing, which is critical for applications like reservoir computing. By addressing these challenges, the researchers hoped to pave the way for more advanced neuromorphic systems that closely mimic the brain’s efficiency and processing capabilities.

The authors’ first experiment aimed to validate the steady-state behavior of the memristor. The researchers observed the current-voltage (I-V) relationship, which demonstrated current rectification—a key characteristic of memristors. This rectification was driven by salt concentration polarization, induced by an inhomogeneous ionic space charge density between the colloidal particles. The experimental I-V curves closely matched the theoretical predictions, confirming that the inhomogeneous ionic space charge was responsible for the observed current rectification. This finding highlighted the device’s potential to mimic synaptic behavior by utilizing ion transport in an aqueous environment. Next, the researchers investigated the dynamic behavior of the memristor by applying a sinusoidal voltage and measuring the resulting I-V diagram. They observed a pinched hysteresis loop, a hallmark of memristive behavior, indicating a pronounced memory effect. The dynamic response of the device aligned well with the theoretical model, which predicted a quadratic dependence of the memory retention time on the channel length. This surprising discovery revealed that the memory retention time, despite being driven by a voltage, behaved similarly to a diffusion-like process. This insight provided a straightforward method for designing channels with specific timescales, crucial for neuromorphic computing applications. To further explore the memristor’s potential for neuromorphic computing, the researchers examined its ability to mimic short-term synaptic plasticity features, such as facilitation and depression. They applied consecutive positive and negative write-pulses to the device and measured the resulting conductance changes with read-pulses. The experimental results demonstrated both facilitation and depression, closely matching the theoretical predictions. This finding confirmed that the fluidic memristor could replicate key aspects of neuronal synaptic behavior, essential for information processing in neuromorphic systems. The stability and reproducibility of the memristor were critical for its application in reservoir computing. The researchers conducted experiments to test the device’s response to repeated voltage pulse trains. They found that the device consistently produced similar conductance changes over multiple cycles, with minimal variability. This stability ensured that the memristor could reliably distinguish between different time series, a fundamental requirement for reservoir computing.

Building on these findings, the researchers implemented the memristor in a reservoir computing framework to classify temporal signals. They used voltage pulse trains representing 4-bit strings and observed distinct conductance signatures for each input pattern. The experimental results closely matched theoretical predictions, demonstrating the device’s ability to encode complex temporal data. This capability was further validated by classifying handwritten digits from the MNIST database. The memristor achieved an accuracy of 81% on a test set of 2,000 samples, comparable to more conventional solid-state platforms.

In conclusion, the study conducted by Professor René van Roij and his team holds significant implications for the future of neuromorphic computing. By developing an aqueous volatile memristor that emulates the brain’s short-term synaptic plasticity, the researchers have taken a crucial step towards creating computing systems that more closely mimic the human brain’s efficiency and processing capabilities. This advancement is particularly important given the growing energy demands and limitations of conventional computing architectures. One of the major contributions of this study is its demonstration of a fluidic memristor that operates in an aqueous environment, utilizing ions as information carriers. This approach addresses the limitations of traditional solid-state memristors, which rely solely on electronic signals and often involve complex fabrication processes. By leveraging ion transport in water, the aqueous memristor not only better replicates the brain’s natural environment but also offers enhanced biointegrability, making it a promising candidate for future biomedical applications and brain-computer interfaces. The practical implications of this study are far-reaching. First, the development of a memristor with adjustable memory retention times, achieved through a straightforward and cost-effective fabrication process, opens up new possibilities for designing neuromorphic systems tailored to specific computational tasks. The ability to fine-tune the memory timescale by simply adjusting the channel length or other geometric parameters provides a versatile platform for various applications in artificial intelligence and machine learning. Furthermore, the successful implementation of the memristor in reservoir computing—a brain-inspired machine learning framework—highlights its potential for handling complex temporal data and sequential tasks. The researchers demonstrated that the fluidic memristor could classify handwritten digits with high accuracy, comparable to conventional solid-state platforms. This capability is crucial for applications requiring real-time data processing and pattern recognition, such as speech and image recognition, autonomous driving, and adaptive control systems. In addition, the study’s theoretical model, which accurately predicts the device’s behavior based on continuum transport equations, provides a solid foundation for future research and development. This model not only explains the underlying mechanisms of the memristor’s operation but also guides the optimization of device parameters for specific applications. The predictive power of the theoretical framework ensures that further advancements in aqueous memristor technology can be pursued efficiently and effectively. The integration of multiple fluidic memristors into complex circuits represents another promising direction for future research. The flexible fabrication methods used in this study allow for the creation of intricate networks of interconnected channels, potentially leading to more sophisticated neuromorphic systems that can perform a wider range of cognitive functions. Optimizing the device to operate at lower voltages and integrating it with other neuromorphic components could further enhance its performance and energy efficiency.

Developing Aqueous Volatile Memristors for Neuromorphic Computing: Bridging Biological and Artificial Information Processing - Advances in Engineering
Features and properties of our iontronic memristor through theory and experiment (Figure creditProc Natl Acad Sci U S A. 2024 Apr 30;121(18):e2320242121. doi: 10.1073/pnas.2320242121.  )

About the author

Kamsma MSc
PhD Candidate
Utrecht University

I am a PhD candidate at Utrecht University in Theoretical Physics and Mathematics. I was awarded a personal grant for my own research proposal, a sought after opportunity, only offered to a select handful of candidates. Supported by my double degree programmes at Utrecht University and my research experience at the University of Cambridge and the Netherlands Institute for Neuroscience, I am currently working on neuromorphic computing, with a wider interest in functional intelligent materials. Moreover, facilitated by the substantial research budget offered to me, I have instigated various new fruitful (international) collaborations.

About the author

Prof. dr.  René van Roij  

Utrecht University
The Netherlands.

Research largely involves the solid-liquid interface, typically between water or oil on the one hand and glass, rock, or nanoparticles on the other. In conventional materials the presence of these interfaces only gives small corrections to material properties, but in nanomaterials, with their huge surface area as large as a square kilometre per litre, these solid-liquid interfaces completely change and dominate the properties. This gives rise to a plethora of new physical phenomena with a variety of potential applications in e.g. water desalination, harvesting and storing electric energy, micro- and nano-fluidics, and catalysis.

For instance, we develop and test models for the electric and thermal properties of so-called supercapacitors, which are porous carbon electrodes filled with ionic liquids or water. But we also study transport of water, charge, and salt through narrow channels, which finds not only finds direct applications in harvesting blue energy from mixing river- and sea water, but has also revealed very strong effects of fluid flow on the chemistry and charge of the channel surface. We also run projects on liquid crystals, active matter, the action potential in neurons, and artificial kidneys.

Reference

Kamsma TM, Kim J, Kim K, Boon WQ, Spitoni C, Park J, van Roij R. Brain-inspired computing with fluidic iontronic nanochannels. Proc Natl Acad Sci U S A. 2024 Apr 30;121(18):e2320242121. doi: 10.1073/pnas.2320242121.

Go to Proc Natl Acad Sci U S A.

Check Also

Photonics-Assisted Dual-Functional Radar and Microwave Frequency Measurement System: Enhanced Resolution and Accuracy without High-Speed Electronics - Advances in Engineering

Photonics-Assisted Dual-Functional Radar and Microwave Frequency Measurement System: Enhanced Resolution and Accuracy without High-Speed Electronics