Balancing electronic and ionic charge carriers in polymer nanofilms significantly improves physical reservoir computing, a brain-inspired approach that harnesses material dynamics instead of traditional circuits.
(Nanowerk Spotlight) A polymer film barely ten nanometers thick can perform computations by exploiting the interplay between two fundamentally different charge carriers, a capability that challenges conventional single-carrier approaches to reservoir computing.
Reservoir computing sidesteps the von Neumann bottleneck, where conventional computers waste time and energy shuttling data between separate memory and processing units. The approach treats a physical material as a reservoir, a dynamic medium that transforms simple input signals into rich, high-dimensional outputs through its natural behavior. Engineers need only train a simple output layer to interpret these transformed signals. The material itself performs the complex computational work, much as a pond’s surface encodes information about every pebble thrown into it through overlapping ripples.
Researchers have demonstrated this approach in substrates ranging from silver nanowire networks to carbon nanotube hybrids, nanoparticle assemblies, liquid crystals, ferroelectric films, and biological membranes.
Any material functioning as a reservoir computer must exhibit three properties. Nonlinearity ensures outputs do not simply scale with inputs. High dimensionality provides many internal degrees of freedom to represent complex information. Fading memory allows the system to retain traces of past inputs that gradually decay.
Organic mixed ionic-electronic conductors offer particular promise because they transport both electrons and ions. This dual capability already serves applications in supercapacitors, batteries, biosensors, and neuromorphic devices. Yet a fundamental obstacle has limited their use in reservoir computing: electron mobility typically exceeds ion mobility by two to ten orders of magnitude. One carrier type consequently dominates, preventing the cooperative dynamics that might enrich computational performance.
Demonstrating Brain-Like Computing Performance in S-PEDOT Nanofilms. This figure illustrates how the S-PEDOT nanofilm functions as a physical reservoir, a type of hardware that mimics the complex signal processing found in biological systems. By utilizing both protons (ionic carriers) and holes (electronic carriers) simultaneously, the material transforms simple electrical inputs into complex, high-dimensional patterns suitable for computation. Experimental Setup (a–d): A simple 10 Hz sine wave enters the nanofilm, which connects to seven output electrodes. The material’s internal physics distort this signal, creating higher harmonics (multiples of the input frequency) and Lissajous curves (distorted loops) that reveal the nonlinear transformations occurring within. Waveform Generation (e): The device successfully reconstructs five different target waveforms, including square and sawtooth waves, from the single sine wave input. The Role of Humidity (f): Accuracy depends strongly on environmental conditions. The device performs best at 60% to 80% relative humidity, where protons and holes contribute comparably to conduction, generating the richest dynamics for computation. (Image: Reproduced from DOI:10.1002/advs.202520270, CC BY) (click on image to enlarge)
S-PEDOT differs from conventional PEDOT formulations in a crucial way. Standard PEDOT requires external dopant molecules to achieve conductivity. S-PEDOT contains sulfonate groups attached directly to its side chains, enabling internal proton transfer to dope the polymer backbone. This architecture yields electrical conductivity around 1000 S cm⁻¹ while permitting water solubility and uniform film formation at approximately 10 nm thickness.
The researchers modified these nanofilms through chemical dedoping using tris(2-aminoethyl)amine, or TREN. This treatment reduced conductivity by roughly four orders of magnitude, from about 500 S cm⁻¹ to approximately 0.03 S cm⁻¹. Spectroscopic measurements confirmed that TREN abstracted protons from the polymer backbone, converting charged species into neutral states while forming ion pairs with residual sulfonate groups.
Impedance spectroscopy under controlled humidity then revealed how the dominant charge carrier changed systematically with moisture levels. At relative humidity of 50% or below, holes provided the primary conduction pathway. Between 60% and 80% humidity, hole conductivity and proton conductivity became comparable, creating a mixed conducting state. Above 80% humidity, proton conduction dominated.
Grazing-incidence X-ray scattering exposed the structural basis for this transition. Under dry conditions at 25 °C, S-PEDOT films adopted an ordered lamellar structure with 2.06 nm spacing. At 80% humidity, water molecules penetrated between layers, expanding the spacing to 2.44 nm and forming nanochannels for proton transport. This water uptake enabled proton dissociation from sulfonate side chains, generating mobile ionic carriers.
The team evaluated reservoir computing performance across these regimes by applying 10 Hz sinusoidal voltage signals and measuring outputs at seven electrodes. Output signals displayed phase shifts and nonlinear distortions essential for reservoir computation, while power spectral analysis revealed higher harmonic generation confirming high dimensionality.
Wave generation tasks provided quantitative benchmarks. The system learned to reproduce cosine, triangle, square, sawtooth, and doubled-frequency sine waveforms from simple sinusoidal input. Symmetric waveforms achieved accuracies exceeding 90% regardless of humidity.
Complex waveforms requiring even-order frequency components showed strong humidity dependence: sawtooth waves reached 93.4% accuracy and doubled-frequency sine waves reached 89.2% accuracy in the 60% to 80% range, with performance declining outside this window. Optimal results occurred at frequencies between 10 and 100 Hz, matching the RC time constants derived from impedance measurements.
Nonlinear autoregressive moving average tasks, abbreviated as NARMA, test a system’s ability to predict time-series values based on past inputs and outputs. NARMA2 testing yielded normalized mean squared errors as low as 0.031 at 80% humidity. Memory capacity, which quantifies how well a system recalls previous inputs, peaked at 3.1 in the mixed conducting regime and degraded at both lower and higher humidity levels.
The connection between cooperative carrier dynamics and computational performance is direct. When holes and protons both contribute significantly, their different mobilities and response timescales create richer internal dynamics that enhance the material’s ability to transform inputs nonlinearly and retain temporal information.
The S-PEDOT devices achieved accuracy comparable to or exceeding other material-based reservoir computers while using only seven output electrodes, far fewer than the numbers employed in many competing approaches.
This work constitutes the first demonstration of reservoir computing that explicitly exploits an intrinsic ion-electron mixed conducting state. Previous approaches treated ionic and electronic processes as separate phenomena rather than cooperative contributors. The findings establish that engineering materials to support multiple carrier types with matched conductivities offers a promising pathway for advancing physical reservoir computers.
Extending this framework to other mobile ions such as lithium could multiply the available timescales and nonlinear responses, potentially enabling a single material to tackle computational tasks of increasing complexity.
For authors and communications departmentsclick to open
Lay summary
Prefilled posts
Nanowerk Newsletter
Get our Nanotechnology Spotlight updates to your inbox!
Thank you!
You have successfully joined our subscriber list.
Become a Spotlight guest author! Join our large and growing group of guest contributors. Have you just published a scientific paper or have other exciting developments to share with the nanotechnology community? Here is how to publish on nanowerk.com.
