Memristors Based on Many-Layer Non-Stoichiometric Germanosilicate Glass Films

Round 1

Reviewer 1 Report

Excellent paper, really well written. Very interesting topic.

Only a few minor comments:

table 1 - there is no explanation for the presence of C, that's a huge fraction - what's going on?

 

Figure 3 description "Height Sensor" is clipped / cropped and doesn't present well.

line 214 axe should read axis

figure 5 Y-axis units (% or absolute?)

figure 6, 8 etc below Si says metal, a bit vague

figure 9 as above but also font size for Current and Voltage compared to eg 6, 8 and 10

Conclusions seem really short! This is a fascinating paper - surely there should be more in the conclusions section.

Author Response

Answers to Reviewer 1

First of all, the authors would like to thank the respected referee for valuable comments. Also, the authors are very grateful for the huge work of the reviewer on correcting errors and typos. We hope that taking into account these comments and corrections will improve the article.

 

table 1 - there is no explanation for the presence of C, that's a huge fraction - what's going on?

Answer: The XPS technique is surface-sensitive, the signal reflects the composition of approximately several nanometers of the structure near the surface. Since the samples are stored in a normal atmosphere, the surface always contains organic materials as an impurity. The organic materials contain carbon and oxygen. In our recent work (T.V. Perevalov, V.A. Volodin, G.N. Kamaev, A.A. Gismatulin, S.G. Cherkova, I.P. Prosvirin, K.N. Astankova, V.A. Gritsenko, Electronic structure of silicon oxynitride films grown by plasma-enhanced chemical vapor deposition for memristor application. Journal of Non-Crystalline Solids, v.598, p. 121925(1-8), (2022). DOI: https://doi.org/10.1016/j.jnoncrysol.2022.121925), we tried to refine the composition of the films by performing a fine analysis of the oxygen peak. In the fine structure of this peak there are components corresponding to oxygen bound to carbon. In this paper, we used the results of our cited recent work. Therefore, we have excluded carbon out of consideration, and also refined the oxygen content. We would like to note that the often-used etching of films with an argon ion beam to analyze the composition of samples depending on the depth also gives an error. When etching with ion beams, the material of the films is etched "incongruently", since different components of the films have different volatility. Some clarifying points are included in the text and highlighted in yellow.

 

Figure 3 description "Height Sensor" is clipped / cropped and doesn't present well.

Answer:

Thank you for your comment. We have made changes to Figure 3 in the manuscript based on Your comments.

 

line 214 axe should read axis

Answer: Thank You for this comment. We have corrected our mistake. We have made changes to the manuscript text (now it is lines 208-215).

 

figure 5 Y-axis units (% or absolute?)

Answer: In IR spectroscopy, either transmission (in percent) or absorbance is usually plotted. This is a dimensionless quantity, equal to the natural logarithm of the transmission (divided by 100) only with a minus sign – A=-ln(T/100). That is, a transmission of 100% corresponds to zero absorption. In our case, in Figure 5 we show absorption in arbitrary units. Some text is added to revised version and highlighted in yellow.

 

figure 6, 8 etc below Si says metal, a bit vague

Answer:

Figures 6,8,9 and 10 show the measurement scheme. For better contact, the back side of the substrate with indium-gallium paste was pressed against the metal sheet, to which the probe of the multimeter was in contact. The ITO contacts were used as the top electrode, to which the other probe of the multimeter was in contact. Some text is added to revised version and highlighted in yellow.

 

figure 9 as above but also font size for Current and Voltage compared to eg 6, 8 and 10 Conclusions seem really short! This is a fascinating paper - surely there should be more in the conclusions section.

Answer:

Thank You for Your comment. We have corrected the font size of the caption in Figure 9. Dear reviewer is right, the conclusions are really quite short.This is due to the fact that multilayer structures are more difficult to analyze, for example, to analyze the mechanisms of conduction.In our previous work (Volodin, V.A.; Kamaev, G.N.; Gritsenko, V.A.; Gismatulin, A.A.; Chin, A.; Vergnat, M. Memristor Effect in GeO [SiO2] and GeO [SiO] Solid Alloys Films. Appl Phys Lett 2019, 114, 233104), we tried to establish the mechanism of conduction in one layer from the I–V characteristics.The Shklovsky-Efros model best approximated the experimental data.In this work, the Shklovsky-Efros model is also applicable to sample A, but we did not repeat our previous conclusions.However, we tried to somewhat expand the conclusions, adding an assumption about the influence of the germanium layer and its charge.

Author Response File: Author Response.pdf

Reviewer 2 Report

In this research, Fourier-transformed infrared spectroscopy, atomic force microscopy, X-ray photoelectron spectroscopy, and Raman spectroscopy were used to examine the as-deposited materials. Memristor metal-insulator-semiconductor (MIS) structures were created using a transparent indium-tin-oxide (ITO) contact as the top electrode. Studies have been done on the MIS's current-voltage characteristics (I-V) and resistive switching cycles. The paper has been presented I think it can be accepted after minor revisions:

-A convincing comparison with the related methods will be helpful for readers

-The labels in some figures have a big size! for example, see fig 9

-add a link for simulations; this will be so helpful for the reader;

-add a remark about the potential design improvement based on advanced machine learning techniques such as: Dynamic programming strategy based on a type-2 fuzzy wavelet neural network

 

Author Response

Answers to Reviewer 2

First of all, we would like to thank the highly respected reviewers for a careful reading of the article and valuable comments. We tried to take into account all the comments of the reviewers, and we hope that after the revision the article has become more understandable and better organized. For the convenience of reviewers and the editor, all changes made to the text of the articles are highlighted in yellow.

-A convincing comparison with the related methods will be helpful for readers

Answer: Dear reviewer is absolutely right, there are many approaches to simulate neural networks and neural computing. For example, not only multilayer structures are used for this, but also structures containing quantum dots (https://dx.doi.org/10.1021/acs.chemrev.9b00730 Chem. Rev. 2020, 120, 3941−4006).Several literature references have been added to the text.

-The labels in some figures have a big size! for example, see fig 9

Answer: Thank You for Your comment. We have corrected the size of the drawings in the manuscript.

-add a link for simulations; this will be so helpful for the reader;

Answer: We apologize, but it is not entirely clear what simulations the respected reviewer means. If this is about numerical modeling and calculations of I-V characteristics in various models, then we (in a slightly different team of authors) published an article in the same Special Issue, which discusses about a dozen different models - Andrei A. Gismatulin, Gennadiy N. Kamaev, Vladimir A. Volodin, and Vladimir A. Gritsenko. Charge Transport Mechanism in the Forming-Free Memristor Based on PECVD Silicon Oxynitride. Electronics, 12, 598 (2023). DOI: https://doi.org/10.3390/electronics12030598. However, it seems to us not entirely correct to cite this work. If the author tells us what other links would be useful for readers to include, we will be grateful. Several literature references have been added to the text.

-add a remark about the potential design improvement based on advanced machine learning techniques such as: Dynamic programming strategy based on a type-2 fuzzy wavelet neural network

Answer: We cannot say about the improvement of the design based on modern machine learning methods, since we are studying a device that will be further profiled and used as a neuron in non-isomorphic calculations, and if changing its resistance is similar in behavior to changing the weight of a synapse in neural networks, then and any machine learning methods that work on virtual neurons will work on the devices explored in this manuscript.

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors present a study focusing on developing analog memristor basing many-layer non-stoichiometric germanosilicate layers. For both one-layer and many-layer structures, reversible resistive switching was detected. In one-layer structure, device shows resistive switching with good endurance and acceptable ON/OFF ratio. While, in many-layer device, analog switching behaviour can be obtained, offering good device platform for neuromorphic computing. The paper is clear in logic, and very informative in content. Despite I am positively impressed by structural innovation and experiment result, several critical points should be addressed to enhance the manuscript for publishing.

1.      The review recommends the authors to include “Chem. Rev. 2020, 120, 3941” as example of neuromorphic computing in the introduction.

2.      Altogether how many devices were prepared and tested? Authors should give the statistical analyses of device performance, rather than measuring the so-called hero device that features the best performance. Reviewer recommends the authors refer to “Matter 2021, 4, 1702; Appl. Phys. Rev. 2022, 021417” for evaluating cyclic repeatability.

3.      Mechanism study is missing in this paper. Why analog switching behaviour achieved in many-layer device?

4.      Device size is large, can similar study be demonstrated in very small device sizes?

5.      Authors should provide the roughness of the layer in Figure 3, rather than claiming sub-nanometer roughness.

6.      The font is too small to view in Figure 7, Figure 12.

 

Author Response

Answers to Reviewer 3

First of all, we would like to thank the highly respected reviewers for a careful reading of the article and valuable comments. We tried to take into account all the comments of the reviewers, and we hope that after the revision the article has become more understandable and better organized. For the convenience of reviewers and the editor, all changes made to the text of the articles are highlighted in yellow.

1. The review recommends the authors to include “ Rev. 2020, 120, 3941” as example of neuromorphic computing in the introduction.

Answer: Thanks for Your recommendation. We took into account Your comment and included this article in the references.

2. Altogether how many devices were prepared and tested? Authors should give the statistical analyses of device performance, rather than measuring the so-called hero device that features the best performance. Reviewer recommends the authors refer to “Matter 2021, 4, 1702; Appl. Phys. Rev. 2022, 021417” for evaluating cyclic repeatability.

Answer: Four devices were prepared on a p-type silicon substrate. Since not all samples exhibited stable switching, we presented cyclicity studies only for those samples that retained their resistive states for more than 30 cycles. Because of this, cycling studies were only done for samples A and D. Thank you for the good recommended articles. We have included these articles as references.

3. Mechanism study is missing in this paper. Why analog switching behaviour achieved in many-layer device?

Answer: Thanks for Your question. At the moment, we cannot imagine the mechanism of the analog effect in a memristor structure with a germanium layer, since the effect of negative differential resistance has not been fully studied. It is known from the available data that the negative differential resistance is due to the fact that charge carriers can accumulate in the germanium layer. This is discussed in the article. Recently we (in a slightly different team of authors) published an article in the same Special Issue, which discusses about a dozen different mechanisms of electron transport in dielectrics - Andrei A. Gismatulin, Gennadiy N. Kamaev, Vladimir A. Volodin, and Vladimir A. Gritsenko. Charge Transport Mechanism in the Forming-Free Memristor Based on PECVD Silicon Oxynitride. Electronics, 12, 598 (2023). DOI: https://doi.org/10.3390/electronics12030598. However, it seems to us not entirely correct to cite this work in this article. that multilayer structures are more difficult to analyze, for example, to analyze the mechanisms of conduction. In our previous work (Volodin, V.A.; Kamaev, G.N.; Gritsenko, V.A.; Gismatulin, A.A.; Chin, A.; Vergnat, M. Memristor Effect in GeO [SiO2] and GeO [SiO] Solid Alloys Films. Appl Phys Lett 2019, 114, 233104), we tried to establish the mechanism of conduction in one layer from the I–V characteristics. The Shklovsky-Efros model best approximated the experimental data. In this work, the Shklovsky-Efros model is also applicable to sample A, but we did not repeat our previous conclusions. However, we tried to somewhat expand the conclusions, adding an assumption about the influence of the germanium layer and its charge. Some text is added to revised version (in conclusions) and highlighted in yellow.

4. Device size is large, can similar study be demonstrated in very small device sizes?

Answer: It is known that from the dependence of the characteristics of memristors on their area, it is possible to study the mechanisms of conduction and the mechanisms of analysis of filaments; with a small area of the structure, only a small number of filaments can be under contact. The respected reviewer is absolutely right, such studies should be carried out, but unfortunately we could not reduce the lateral dimensions of our structures to nanometer scales, since we did not have the opportunity to use lithography.

5. Authors should provide the roughness of the layer in Figure 3, rather than claiming sub-nanometer roughness.

Answer: We indicated the roughness at Your request and changed some of the text in the manuscript.

6. The font is too small to view in Figure 7, Figure 12.

Answer: Thank you for Your comment. We have increased the font in Figures 7 and 12.

Author Response File: Author Response.pdf