Source-linked AI summary
Filamentary Switching: Synaptic Plasticity through Device Volatility
Selina La Barbera, Dominique Vuillaume, Fabien Alibart
TL;DR
Obtaining synaptic density is a major neuromorphic-engineering challenge. This paper uses independently tunable dendritic-path density and diameter in filamentary devices to reproduce varied plasticity, including STP and LTP, through different volatilities.
Problem
Obtaining synaptic density has been a major challenge in neuromorphic engineering.
Method
The paper tunes dendritic-branch density through Ic and diameter through burst excitation, using excitation strength to control Gmax.
Results
Independently tunable branch density and diameter reproduce various STP/LTP combinations, while conductance states exhibit different volatilities associated with plasticity properties.
Takeaways & Limitations
A single memristive element can emulate different synaptic plasticity properties through filamentary devices with independently controlled dendritic-path features.
Takeaways & Limitations
The experiments cannot establish a one-to-one correspondence between biological processes and filament growth or relaxation.
Abstract
from arXiv · showhide
Replicating the computational functionalities and performances of the brain remains one of the biggest challenges for the future of information and communication technologies. Such an ambitious goal requires research efforts from the architecture level to the basic device level (i.e., investigating the opportunities offered by emerging nanotechnologies to build such systems). Nanodevices, or, more precisely, memory or memristive devices, have been proposed for the implementation of synaptic functions, offering the required features and integration in a single component. In this paper, we demonstrate that the basic physics involved in the filamentary switching of electrochemical metallization cells can reproduce important biological synaptic functions that are key mechanisms for information processing and storage. The transition from short- to long-term plasticity has been reported as a direct consequence of filament growth (i.e., increased conductance) in filamentary memory devices. In this paper, we show that a more complex filament shape, such as dendritic paths of variable density and width, can permit the short- and long-term processes to be controlled independently. Our solid-state device is strongly analogous to biological synapses, as indicated by the interpretation of the results from the framework of a phenomenological model developed for biological synapses. We describe a single memristive element containing a rich panel of features, which will be of benefit to future neuromorphic hardware systems.
RESULTS AND DISCUSSION
Filament morphology and stimulation history provide separate controls over conductance, volatility, and the transition between short- and long-term plasticity. These controls enable a single memristive device to reproduce diverse synaptic behaviors while remaining only phenomenologically analogous to biological synapses.
- Filament formation and relaxation: Positive bias forms Ag dendritic filaments, while negative bias partially destroys conducting paths and leaves traces that guide subsequent switching.Filament relaxation is attributed to Ag+ diffusion in Ag2S and reverse oxidation-reduction of Ag filaments.
- Filament formation and relaxation: Increasing compliance current Ic produces larger, denser dendritic filaments and raises the ON conductance state.The observed correlation extends to the fractal geometry of the dendritic filaments.
- Volatility control: 100 nA to 50 µA conditioning yields strongly volatile ON states, whereas Ic above 50 µA produces stable states with RESET at negative bias.Linear I-V behavior in the ON state indicates that filaments bridge the electrodes.
- Volatility control: The device spans 1 MΩ at 100 nA with switching power <100 nW to 1 kΩ at 1 mA with switching power 300 µW.The dynamic range is attributed to changes in bridging-filament morphology rather than tunnel-barrier modulation.
- Synaptic plasticity: Increasing Gmax raises the relaxation time constant and stabilizes filaments, while Ic changes the sharpness of the STP-to-LTP transition.At high Ic the transition is sharp; at lower Ic it becomes smoother as Gmax increases.
- Biological analogy and scope: The device shows a strong analogy to biological synapses, but the experiments cannot establish one-to-one correspondence between biological processes and filament dynamics.Additional in situ filament-shape observations would be needed for a more refined equivalence.
Filamentary Switching: synaptic plasticity through device
Fractal analysis characterizes dendritic filament geometry through fractal dimension and lacunarity, while burst configurations modulate synaptic potentiation. Independently tuning dendritic branch density and dendrite diameter produces varied STP/LTP combinations.
- Fractal geometry: Fractal analysis estimates filament fractal dimension D and lacunarity λ from binary images of selected optical-image regions.The analysis uses box-counting procedures implemented through ImageJ.
- Fractal geometry: D and λ jointly characterize complex filament patterns, although they do not directly describe dendritic branch density and width.Their evolution with Ic is consistent with the proposed filamentary-switching scenario, while further investigation is needed.
- Fractal geometry: λ correlates with Ic, whereas D anticorrelates with Ic during the SET process.Figure S2 presents the evolution of these fractal parameters as a function of Ic.
- Synaptic plasticity: Increasing burst pulse number or pulse amplitude increases potentiation, represented by conductance Gmax at the sequence end.Pulse amplitude is also identified as a parameter proposed for spike-timing-dependent plasticity based on overlapping pulses.
- Synaptic plasticity: Two long-term-plasticity cases and two short-term-plasticity cases are obtained through different burst configurations.The configurations modulate potentiation by varying the pulse-defined plasticity factor.
- Synaptic plasticity: Independent tuning of dendritic branch density and dendrite diameter reproduces various combinations of short- and long-term plasticity.This provides the device with separate geometric controls over the resulting STP/LTP behavior.