Speaker
Description
One of the striking results of the artificial-intelligence revolution is the energy consumption associated with it, especially compared to the energy consumed by a human brain. Among many differences between classical computing and brain functioning, the architecture and the carriers are among the most important. In a computer, the data are stored and processed in different location, while these tasks are not spatially differentiated in a brain. Bridging this gap could be possible with a new electronical device called a memristor, whose resistance depends on the current that crossed it in the past. Regarding the carriers, brains use ions (iontronics), which can be of different natures, while computers are limited to electrons. Multiplying the possible carriers could open new routes for more complex operations. In the past recent years, many examples of nanofluidic devices proved to be memristors, due to different physical phenomena. Most of them work thanks to tailor-made nanopores with specific geometries, and thanks to the control of the nanopore surface charge.
In this work, we present a microfluidic system based on a commercial ion-selective membrane which behaves as a resistor, a diode, or a memristor depending on the typical time of the writing/reading processes. The simplicity of the device allows us to precisely pin-point the physical phenomena at stake in these processes, and to open new routes for simple design of iontronic memristors.
The experimental setup consists of a unique ion-exchange membrane (Nafion) separating two similar electrolytes. Selective membranes are known to create concentration polarization under current. The geometry of the setup is hence designed such as the dominating resistances are located in these concentration polarization zones. This is achieved using masking windows creating an access resistance to the membrane [1]. A 1-Volt sinusoidal voltage is imposed on the system at different frequencies, and the I-V response is measured to characterize the memristor effect. A model based on the Poisson-Nernst-Planck (PNP) equations is developed to rationalize the experimental results.
Results shows that the symmetry between the two sides of the membrane yields two different memristor signatures: the hysteresis loops in their I-V curves are self-crossing (asymmetrical configuration) or not (symmetrical configuration), as predicted by the model of Kamsma [2]. Moreover, the system behaviour has been mapped according to the configuration, frequency and radius of the mask opening. Three different regions of behaviours were determined for a symmetric and an asymmetrical configuration: diode, memristor and resistor. These mapping is very close to the results of our theoretical model based on the concentration polarization at the membrane.
The experimental characterization of our Nafion-based memristor is well described by our PNP model considering concentration polarization and membrane access resistance. In the future, this model system will enable comparison with tailor-made nanopores [3], and motivates miniaturization and parallelization of the present device.
[1] Derkenne, T. et al. (2024). ACS Applied Energy Materials, 7(15), 6621–6629.
[2] Kamsma, T. M. et al. (2024). Chaos, Solitons and Fractals, 186, 115320.
[3] Emmerich, T. et al. (2024). Nature Electronics, 7(4), 271–278.