Linking Conductive Filament Properties and Evolution to Synaptic Behavior of RRAM Devices for Neuromorphic Applications
Open accessThe DOI matches its Crossref record; 18 of the 19 references it cites were found in a registry or an index; 2 of the 10 sources its claims cite were checked for support.
Abstract
We perform a comparative study of HfO2and Ta2O5resistive switching memory (RRAM) devices for their possible application as electronic synapses. By means of electrical characterization and simulations, we link their electrical behavior (digital or analog switching) to the properties and evolution of the conductive filament (CF). More specifically, we identify that bias-polarity-dependent digital switching in HfO2RRAM is primarily related to the creation and rupture of an oxide barrier. Conversely, the modulation of the CF size in Ta2O5RRAM allows bias-polarity-independent analog switching with multiple states. Therefore, when the Ta2O5RRAM is used to implement a synapse in multilayer perceptron neural networks operated by back-propagation algorithms, patterns in handwritten digits can be recognized with high accuracy.
Overview
CoChat Summary
TL;DR
A comparative study of HfO₂ and Ta₂O₅ RRAM devices links their digital versus analog switching to the properties and evolution of the conductive filament (CF), using electrical characterization plus simulations. The Ta₂O₅ device's CF-size modulation yields bias-polarity-independent analog switching with multiple states, which the authors use as synapses in a multilayer perceptron for MNIST digit recognition.
Key findings
- HfO₂ RRAM shows bias-polarity-dependent digital switching: SET is abrupt and RESET gradual, and identical SET pulses at fixed Vg = 1.2 V produce binary states with current saturation after the first pulse; only varying Vg from 0.6 to 1.2 V splits the binary states into multiple states.
- Simulations attribute this to the creation and rupture of an oxide barrier (δ ≈ 1.5–2 nm after RESET); current depends exponentially on a large δ (> 1 nm) but linearly on CF size, and the field after CF restoration (~1.7 MV/cm) is far below the ~4–5 MV/cm needed to break Hf–O bonds.
- Ta₂O₅ (TiOx/Ta₂O₅) RRAM shows bias-polarity-independent analog switching: 100 identical positive pulses (2–3.5 V, 10 µs) and 100 negative pulses (−3 to −5 V, 100 µs) monotonically tune current, because RESET modulates CF width rather than forming a barrier — aided by oxygen diffusion in Ta₂O₅ with activation energy ~1.2 eV.
- In MNIST pattern recognition with a multilayer perceptron (528 input neurons from a cropped 22 × 24 image, back-propagation), HfO₂ binary-state synapses reached 94% accuracy in off-chip learning, while Ta₂O₅ linearly modulated multiple states enabled 88% accuracy during on-chip learning.
Methods
- Comparative device study: Ti TE (15 nm)/HfO₂ (6 nm)/TiN BE (400 nm) connected to a 0.35 µm transistor, versus bilayered TiOx/Ta₂O₅ sandwiched between TiN BE/TE with TiOx as an oxygen reservoir.
- Electrical characterization (I-V cycling, identical and non-identical pulse trains, compliance-current ramping) combined with simulations using the Ginestra™ package, which self-consistently models charge transport, power dissipation, temperature rise, and oxygen ion/vacancy generation, diffusion, and recombination.
- Neural-network evaluation: MNIST handwritten digits in a multilayer perceptron (input, hidden, output layers) with synaptic weight taken as the conductance difference of a pair of RRAM devices, updated by back-propagation.
Limitations
- The paper states no explicit limitations.
- Evident from the text: the two accuracy figures come from different learning schemes (94% for HfO₂ in off-chip learning versus 88% for Ta₂O₅ in on-chip learning), so they are not a like-for-like comparison, and the Ta₂O₅ result is numerically lower despite being presented as the improvement.
- Evident from the text: the pattern-recognition results are simulation-based, and the device comparison covers only two material stacks without reported endurance, variability, or retention statistics beyond the 100 switching cycles shown.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
1 Valid · 1 Not valid · 5 Not checked · 15 Own results and interpretations
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The data could be recognized more precisely using the linearly modulated multiple states in the Ta2O5 RRAM based synapses in neuromorphic systems, thereby enabling improved accuracy (88%) of pattern recognition.”
hedgedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In this case, the synapses with binary states had limitations in improving the recognition accuracy.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Therefore, when the Ta2O5 RRAM is used to implement a synapse in multilayer perceptron neural networks operated by back-propagation algorithms, patterns in handwritten digits can be recognized with high accuracy.”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Simulation results clearly show that the RESET operation leads to a modulation of the CF size rather than to the creation of the δ , see Fig. 5(e).”
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
In the paper
“Once the CF is recreated, the electric field is approximately equal to VP/tOX ∼ 1.7 MV/cm (being tOX the HfO2 thickness), which is far lower than the critical field of ∼ 4-5 MV/cm needed to break Hf-O bonds [15].”
What the sources say
- Not checked — no full text of the sourcecites [15] [15] M. Alayan, E. Vianello, B. De Salvo, L. Perniola, A. Padovani, and L. Larcher, 'Correlated Effects on Forming and Retention of Al Doping in 16. HfO2-Based RRAM,' IEEE Design & Test , vol. 34, no. 3, June 2017. DOI: 10.1109/MDAT.2017.2682246 - source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
8 more claims
Create a free account to see all 22 claims and every source verdict.
- Probably not a substantial claim. 1
- It may state two separate claims as one. 9
- The quote only partly states this claim. 1
Citations
Cites
A million spiking-neuron integrated circuit with a scalable communication network and interface
Merolla, P. A., Arthur, J. V., Alvarez-Icaza, R., Cassidy, A. S., Sawada, J., Akopyan, F., Jackson, B. L., Imam, N., Guo, C., Nakamura, Y., et al. (2014) · Science · doi:10.1126/science.1254642 · Cited by 4,147
Neuromorphic electronic systems
Mead, C. (1990) · Proceedings of the IEEE · doi:10.1109/5.58356 · Cited by 2,368 · cited 1×
“They follow a simple weight update rule through synapses, similar to how humans remember information [2].”
BackgroundBurr, G. W., Shelby, R. M., Sidler, S., di Nolfo, C., Jang, J., Boybat, I., Shenoy, R. S., Narayanan, P., Virwani, K., Giacometti, E. U., et al. (2015) · IEEE Transactions on Electron Devices · doi:10.1109/ted.2015.2439635 · Cited by 936
Self-citationWoo, J., Moon, K., Song, J., Lee, S., Kwak, M., Park, J., & Hwang, H. (2016) · IEEE Electron Device Letters · doi:10.1109/led.2016.2582859 · Cited by 512 · cited 1×
“A 15-nm-thick Ti top electrode (TE) was then deposited, resulting in Ti TE/HfO2/TiN BE stack [7].”
BackgroundSelf-citationSpin-Transfer Torque Magnetic Memory as a Stochastic Memristive Synapse for Neuromorphic Systems
Vincent, A. F., Larroque, J., Locatelli, N., Ben Romdhane, N., Bichler, O., Gamrat, C., Zhao, W. S., Klein, J.-O., Galdin-Retailleau, S., & Querlioz, D. (2015) · IEEE Transactions on Biomedical Circuits and Systems · doi:10.1109/tbcas.2015.2414423 · Cited by 360
Optimization of Conductance Change in Pr1–xCaxMnO3-Based Synaptic Devices for Neuromorphic Systems
Jang, J.-W., Park, S., Burr, G. W., Hwang, H., & Jeong, Y.-H. (2015) · IEEE Electron Device Letters · doi:10.1109/led.2015.2418342 · Cited by 310
Self-citationAmbrogio, S., Balatti, S., Milo, V., Carboni, R., Wang, Z.-Q., Calderoni, A., Ramaswamy, N., & Ielmini, D. (2016) · IEEE Transactions on Electron Devices · doi:10.1109/ted.2016.2526647 · Cited by 256 · cited 1×
“However, many studies have reported that only two discrete states are demonstrated during potentiation, whereas depression demonstrates multiple states [9]–[11].”
BackgroundBio-Inspired Stochastic Computing Using Binary CBRAM Synapses
Suri, M., Querlioz, D., Bichler, O., Palma, G., Vianello, E., Vuillaume, D., Gamrat, C., & DeSalvo, B. (2013) · IEEE Transactions on Electron Devices · doi:10.1109/ted.2013.2263000 · Cited by 229 · cited 1×
“phase change memory [3], [4], spin-based memory [5], and resistive switching memory (RRAM) [6]–[11].”
BackgroundMicroscopic Modeling of HfOx RRAM Operations: From Forming to Switching
Padovani, A., Larcher, L., Pirrotta, O., Vandelli, L., & Bersuker, G. (2015) · IEEE Transactions on Electron Devices · doi:10.1109/ted.2015.2418114 · Cited by 219
Self-citationBinary neural network with 16 Mb RRAM macro chip for classification and online training
Yu, S., Li, Z., Chen, P.-Y., Wu, H., Gao, B., Wang, D., Wu, W., & Qian, H. (2016) · 2016 IEEE International Electron Devices Meeting (IEDM) · doi:10.1109/iedm.2016.7838429 · Cited by 219 · cited 1×
“6(b) [19].”
Yu, S., Gao, B., Fang, Z., Yu, H., Kang, J., & Wong, H.-S. P. (2012) · 2012 International Electron Devices Meeting · doi:10.1109/iedm.2012.6479018 · Cited by 213
RRAM-based synapse for neuromorphic system with pattern recognition function
Park, S., Kim, H., Choo, M., Noh, J., Sheri, A., Jung, S., Seo, K., Park, J., Kim, S., Lee, W., et al. (2012) · 2012 International Electron Devices Meeting · doi:10.1109/iedm.2012.6479016 · Cited by 191 · cited 1×
“phase change memory [3], [4], spin-based memory [5], and resistive switching memory (RRAM) [6]–[11].”
BackgroundSelf-citationGao, B., Bi, Y., Chen, H.-Y., Liu, R., Huang, P., Chen, B., Liu, L., Liu, X., Yu, S., Wong, H.-S. P., et al. (2014) · ACS Nano · doi:10.1021/nn501824r · Cited by 188
Diffusion of oxygen in amorphous Al2O3, Ta2O5, and Nb2O5
Nakamura, R., Toda, T., Tsukui, S., Tane, M., Ishimaru, M., Suzuki, T., & Nakajima, H. (2014) · Journal of Applied Physics · doi:10.1063/1.4889800 · Cited by 132 · cited 1×
“2 eV [17]) and by a smaller field acceleration”
BackgroundWoo, J., Moon, K., Song, J., Kwak, M., Park, J., & Hwang, H. (2016) · IEEE Transactions on Electron Devices · doi:10.1109/ted.2016.2615648 · Cited by 96
Self-citationLee, H. D., Kim, S. G., Cho, K., Hwang, H., Choi, H., Lee, J., Lee, S. H., Lee, H. J., Suh, J., Chung, S.-O., et al. (2012) · 2012 Symposium on VLSI Technology (VLSIT) · doi:10.1109/vlsit.2012.6242506 · Cited by 68
Self-citationCorrelated Effects on Forming and Retention of Al Doping in HfO2-Based RRAM
Alayan, M., Vianello, E., de Salvo, B., Perniola, L., Padovani, A., & Larcher, L. (2017) · IEEE Design & Test · doi:10.1109/mdat.2017.2682246 · Cited by 11 · cited 1×
“7 MV/cm (being tOX the HfO2 thickness), which is far lower than the critical field of ∼4-5 MV/cm needed to break Hf-O bonds [15].”
BackgroundSelf-citationKim, B., Kim, W., Kim, H., Jung, K., Park, W., Seo, B., Joo, M., Lee, K., Hong, K., & Park, S. (2013) · Japanese Journal of Applied Physics · doi:10.7567/jjap.52.04cd05 · Cited by 7 · cited 1×
“Forming results show that the CF is created only in the thin Ta2O5, whereas the overlying TiOx acts as a reservoir for the generated oxygen ions (O−) [12], [16], Figs.”
Background
Research works printed but not linked: 1
- 2017, MDLSoft Software
Cited by
Recent progress in analog memory-based accelerators for deep learning
Tsai, H., Ambrogio, S., Narayanan, P., Shelby, R. M., & Burr, G. W. (2018) · Journal of Physics D: Applied Physics · doi:10.1088/1361-6463/aac8a5 · Cited by 225
In-memory Learning with Analog Resistive Switching Memory: A Review and Perspective
Xi, Y., Gao, B., Tang, J., Chen, A., Chang, M.-F., Hu, X. S., Spiegel, J. V. D., Qian, H., & Wu, H. (2021) · Proceedings of the IEEE · doi:10.1109/jproc.2020.3004543 · Cited by 196
“In this section, some device optimizations and their underlying mechanisms are discussed, including electrothermal modulation [54], [141], [145], preferable locations of conductive filaments [147], interface barrier modulation [148], and optimization of cell structure [143].”
BackgroundMethodInfluentialDevice and materials requirements for neuromorphic computing
Islam, R., Li, H., Chen, P.-Y., Wan, W., Chen, H.-Y., Gao, B., Wu, H., Yu, S., Saraswat, K., & Philip Wong, H.-S. (2019) · Journal of Physics D: Applied Physics · doi:10.1088/1361-6463/aaf784 · Cited by 158
Resistive Memory-Based Analog Synapse: The Pursuit for Linear and Symmetric Weight Update
Woo, J., & Yu, S. (2018) · IEEE Nanotechnology Magazine · doi:10.1109/mnano.2018.2844902 · Cited by 137
“Physical descriptions of the filamentary rraM have recently shown how filament evolutions are linked with the synaptic characteristics, as shown in Figure 5 [42], [43].”
BackgroundSelf-citationAnalog‐Type Resistive Switching Devices for Neuromorphic Computing
Zhang, W., Gao, B., Tang, J., Li, X., Wu, W., Qian, H., & Wu, H. (2019) · physica status solidi (RRL) - Rapid Research Letters · doi:10.1002/pssr.201900204 · Cited by 132
Metadata
- Authors
- Woo, J., Padovani, A., Moon, K., Kwak, M., Larcher, L., & Hwang, H.
- Affiliations
- Pohang University of Science and Technology (KR)
- University of Modena and Reggio Emilia (IT)
- Venue
- IEEE Electron Device Letters
- Published
- 2017-09
- Volume
- 38
- Issue
- 9
- Pages
- 1220-1223
- ISSN
- 0741-3106 (ISSN-L)
- 1558-0563
- OpenAlex
- W2737711638
- Semantic Scholar
- f5687a2171bdd95bb40c834f93b7ad2d4b770e14
- Cited by
- 103
- References
- 19
- Citations by year
- 2026: 9 · 2025: 9 · 2024: 6 · 2023: 8 · 2022: 21 · 2021: 9 · 2020: 14 · 2019: 15 · 2018: 10 · 2017: 2
- FWCI
- 4.66
- Type
- journal article
- Language
- en
- Open access
- green
- hdl.handle.net
- Locations
- IEEE Electron Device Letters (publishedVersion)
- IRIS UNIMORE (University of Modena and Reggio Emilia) (submittedVersion, cc-by)
- Open Access System for Information Sharing (Pohang University of Science and Technology) (submittedVersion)
- Funders
- MOTIE (10067794)
- KSRC Support Program for the development of the future semiconductor device
- Topics
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Advanced Memory and Neural Computing
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Ferroelectric and Negative Capacitance Devices
- Physical Sciences › Materials Science › Materials Chemistry › Phase-change materials and chalcogenides
- Keywords
- resistive random access memory, conductive filaments, hafnium oxide, tantalum pentoxide, electronic synapses, neuromorphic hardware, resistive switching, oxide barrier, analog switching, multistate memory, multilayer perceptron, handwritten digit recognition, backpropagation
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.