Organic transistors containing platinum nanoparticles showed a reported memory window of more than 20 volts, according to a laboratory preprint. The devices also showed large hysteresis — different current responses when voltage was swept in opposite directions — while the control without particles showed none. The paper describes that contrast as characteristic of nonvolatile-memory-like behavior.
The electrical effect was tested alongside step-like current transport, light-based writing, electrical erasing and paired-pulse facilitation, a temporary change in response after two closely spaced pulses. Researchers also fed conductance changes measured from the devices into a neural-network simulation. The recognition results therefore came from a model using experimental data, not from a full physical computing array.
The device design
The laboratory devices used DPP-DTT and PVDF-HFP in a bottom-gate, top-contact organic-transistor design. At the interface between those materials, the researchers placed platinum nanoparticles between ultrathin layers of aluminum oxide. The fabricated stack consisted of a 12-nanometre Al2O3 layer, the particles and an ALD-grown Al2O3 tunneling layer that was either 2 or 3 nanometres thick.
The study included control transistors without platinum particles and compared devices with the two tunneling-layer thicknesses. Channel lengths ranged from 50 to 125 micrometres, and the devices used p-type transport. The work combined electrical and optical measurements with high-resolution transmission electron microscopy, current-voltage modeling and MNIST neural-network simulations.
Microscopy measurements differed with the particle-deposition time. After 10 seconds of deposition, the reported average particle size was 0.66 ± 0.26 nanometres, with an areal density of about 10¹² particles per square centimetre. After 20 seconds, the average size was 1.30 ± 0.43 nanometres. Those measurements describe the material used in the devices; they do not establish which particle size would perform best across a wider set of transistors.
Current moved in unusual steps
In reverse voltage sweeps, the platinum-particle devices displayed plateaus — stretches where the current changed in a more step-like pattern — along with regular current spikes. The control lacked those features. The authors describe the behavior as Coulomb-blockade-like because it resembles transport associated with discrete charge movement, but they also state that the device is not a conventional single-electron transistor. The result is therefore a reported resemblance, not proof of conventional single-electron-transistor Coulomb blockade.
The first two plateaus were approximately 2.3 volts and 5.7 volts wide. The paper compared those widths with the electron-affinity and ionization-potential energies of a Pt13 cluster, a model cluster containing 13 platinum atoms. This comparison forms part of the authors’ interpretation of the transport pattern, rather than an independent validation of the mechanism.
A phenomenological model, meaning a mathematical description fitted to the observed pattern, reproduced the oscillatory current with a reported root-mean-square residual below 1% for all devices. That result shows that the model tracked the measured curves closely. It does not, by itself, determine whether the proposed transport explanation is correct or rule out competing mechanisms.
Light could write the device state
Transfer sweeps gave positive threshold voltages of +4.62 volts for the particle device with the 2-nanometre tunneling layer and +6.62 volts for the device with the 3-nanometre layer. Threshold voltage is the gate-voltage level used to describe a change in the transistor’s response. The supplied analysis notes that a supplementary table reports a different value for the 2-nanometre device.
The researchers used pulses of green light at a wavelength of 532 nanometres and a power density of 0.03 watts per square centimetre. Drain current rose progressively during optical stimulation and then fell when electrical pulses were applied to erase the stored state. A similar writing-and-erasing response was observed with 405-nanometre light.
The reported current-voltage characteristics remained almost unchanged for more than five months after storage in a glove box, with measurements carried out in air. This is stability under the stated storage and measurement conditions. It does not establish how the devices would behave under different environments, extended cycling or production-scale use.
The recognition test was simulated
To examine a neuromorphic-computing use, the researchers fitted conductance updates measured from the devices and used them in a neural-network simulation. The simulation processed 10,000 images per training epoch and reached approximately 83% accuracy after 125 epochs for the condition combining 532-nanometre illumination with electrical depression, or conductance reduction. Under the 405-nanometre optical-writing condition, the reported accuracy was approximately 36% after the same 125 epochs.
The two extracted update behaviors therefore produced different results in the model. But the comparison does not show that a physical transistor array would reach either accuracy unchanged. It also does not establish that the simulated result would transfer to other data sets or to a complete hardware system.
A temporary response to paired pulses
The researchers also measured paired-pulse facilitation, a short-lived increase in the response to a second pulse after a first pulse. Under paired 532-nanometre optical pulses, the PPF index reached as high as 210%, with the second response approximately 2.1 times the first. The analysis identifies this as a representative-device result; the number of devices and the variability between them were not reported.
A double-exponential fit separated the decay into fast and slow components, with fitted time constants of 0.71 ± 0.27 seconds and 2.34 ± 0.89 seconds. In practical terms, the fit describes the response relaxing on more than one timescale after paired stimulation. The supplied analysis does not report the fitting sample size or the method used to calculate the intervals.
The evidence remains limited to laboratory devices
The number of fabricated or tested transistors is not reported. For the memory window, retention and paired-pulse results, the supplied analysis notes missing device-level denominators, variability or complete distributions. The optical and electrical programming plots are also described as mainly representative. Those gaps make it difficult to judge how consistently the reported behaviors would appear from one device to another.
Most outcomes lack formal statistical uncertainty or between-group testing in the reported analysis. That matters because the central comparisons involve controls, different tunneling-layer thicknesses, different deposition times and different illumination conditions. The results show what was observed in the tested devices, but they do not provide a full estimate of variation across a defined device cohort.
The transport explanation is especially cautious. The paper calls the effect Coulomb-blockade-like and says the device is not a single-electron transistor. Additional transport measurements would be needed to distinguish that interpretation from competing mechanisms. The comparison with Pt13 cluster energies is suggestive within the paper’s framework, but it does not settle the question.
The optical and simulation results also have boundaries. The supplied analysis notes that the reported 405-nanometre power density is inconsistent across sections, while the MNIST accuracy figures were generated from extracted conductance updates rather than a demonstrated physical array. The study does not show fully electrical long-term potentiation under the tested conditions.
Further work would need to test a larger reported device cohort, prolonged cycling, arrays and varied environmental conditions. It would also need to separate the effects of particle size, particle density, particle crystallinity and tunneling-layer thickness, and determine whether fully electrical potentiation can be achieved while preserving the memory window. These are open questions identified by the analysis, not results established by this preprint.
An early device result
Taken together, the experiments describe an organic-transistor platform that combines a reported memory window of more than 20 volts with optical writing, electrical erasing and short-term paired-pulse facilitation. The same devices also showed current features the authors characterize as Coulomb-blockade-like. The evidence supports those observations in the tested architectures and conditions, but not production-scale reliability, population-level reproducibility or a conventional single-electron-transistor effect.
The work is therefore best read as a laboratory demonstration of several functions in one device architecture, with a simulated computing result alongside the measurements. Its next test is not whether the concept can be described, but whether the memory, transport and synaptic behaviors hold across more devices, more cycles and physical arrays.
Paper data and sources
Original title: Coulomb blockade-like transport and enhanced memory in organic transistors embedded with sub-nm Pt nanoparticles for neuromorphic computing
Authors: Arash Ghobadi, Thomas B. Kallaos, Abhi Abhijeet et al.
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text