Ultrahigh drive current and large selectivity in GeS selector

Selector devices are indispensable components of large-scale nonvolatile memory and neuromorphic array systems. Besides the conventional silicon transistor, two-terminal ovonic threshold switching device with much higher scalability is currently the most industrially favored selector technology. However, current ovonic threshold switching devices rely heavily on intricate control of material stoichiometry and generally suffer from toxic and complex dopants. Here, we report on a selector with a large drive current density of 34 MA cm−2 and a ~106 high nonlinearity, realized in an environment-friendly and earth-abundant sulfide binary semiconductor, GeS. Both experiments and first-principles calculations reveal Ge pyramid-dominated network and high density of near-valence band trap states in amorphous GeS. The high-drive current capacity is associated with the strong Ge-S covalency and the high nonlinearity could arise from the synergy of the mid-gap traps assisted electronic transition and local Ge-Ge chain growth as well as locally enhanced bond alignment under high electric field.


The paper investigates GeS materials for OTS applications. It claims record for ON current density, high device selectivity and key insights into the unique amorphous atomic structures and electronic band structures of chalcogenides. GeS materials are not new in memory field but not yet reported in literature for OTS applications.
Reply: We thank you for the thorough review of our manuscript and appreciate your valuable comments. We tried to implement all suggestions and comments to improve the manuscript further.
-Concerning the claimed record for the ON current density, and performances in general, they do not seem the core of the paper. We can find for example in reference " doi: 10.1109doi: 10. /JEDS.2018.2856853" B-Te alloys reported for OTS applications with about 55MA/cm2 ON current density. I would not rely on records, but more on device behavior analysis. Being difficult to definitely state on the real formed region (and its surface) in the large device considered in this work, the calculation of the ON current density becomes difficult, also in the light of final given hypothesis about "conductive local paths". In the light of this comment, sentence at line 46 becomes maybe too ambitious.
Reply: Thank you for your valuable suggestions. Indeed, the purpose of this work is to propose a new GeS OTS material, and then find the underlying mechanism by combing a variety of techniques including electron diffraction technique, XPS, Raman, photothermal deflection spectroscopy and DFT calculations. According to your suggestions, we added and discussed the B-Te OTS result in the Figure 2d, and deleted the term "record" in our manuscript. We also deleted the sentence at line 46 in the abstract.
-Not clear the cumulative distribution reported in Fig.1d. How many devices were tested? The statistic of the results is not reported because of single device testing?
Reply: To show the performance repeatability of GeS-based OTS device, we repeatedly operated the same cell, the I-V curves of which is presented in Figure 1c. The cumulative distribution was obtained from these I-V curves.We also measured many other GeS-based OTS devices (more than 30 cells), the I-V curved are presented in Figure R1a. From this result, you can find that GeS-based OTS devices shows large I on with the I off mainly ranging between 0.1 nA to 10 nA, as shown in the cumulative distributions of the current in Figure R1b. The large distribution of OFF current may be due to the rough W bottom surface in our devices.
According to your suggestion, we show the repeated I-V curves from one OTS cell and also cumulative distribution of various cells in our manuscript. We replaced Figure 1d by Figure R1b in the revised manuscript. Figure R1 a, Repeatable DC I-V sweeps obtained from various cells with uniform compliance current (10 mA) and low leakage current (10 nA). The inset shows the schematic diagram of applied cell arrays. b, Cumulative probability of OFF current and ON current for various cells measured at 1/2 V th and V th , respectively.
-TEM image of Fig. S6 gives rise to doubts concerning the possible high variability of the device, being high the roughness of the W electrode surface. Being the surface of the device particularly high, how the author can be sure about the no contamination of the GeS layer, which could appear in a localized region of the electrode?
Reply: The used device with 190 nm-diameter W bottom electrodes were obtained by using 130 nm CMOS technology. After deposited the W into the 190 m-diameter hole by chemical vapor deposition (CVD), chemical mechanical polishing was used to remove the W film on the SiO 2 insulator. Since the polish speed of W and SiO 2 are very different, resulting in the rough W electrode surface found in some devices. Not all the devices have this issue.
Before deposited the GeS film on the W electrode, ultrasonic cleaning technique was used to clean these devices. As a result, no contamination of the W was found in the GeS cells, as proved by the EDX mapping of the device in Figure R2b. This figure is added in the supplementary materials ( Figure S6).

Figure R2
a, The HAADF image of GeS-based device. b-e, Corresponding EDS element mappings of W, Ti, Ge and S respectively. f-h, Cross-section HRTEM images. Insets are corresponding Fast Fourier transform images of GeS layer.
-OTS materials and devices are known for presenting a firing step, needed to "initialize" the device, and to drive it to its stationary behavior. No comments or data are reported here concerning this important aspect. Moreover, all the considerations and explanations concerning the device functionality, should take in account this step.
Reply: This is a very interesting point. As your said, many OTS materials, like GeSe, were reported to have a firing process, characterized by much higher required voltage than V th in the first I-V scanning. We also found the firing process in the S-rich GeS OTS device (Ge 38 S 62 ) as shown in the Figure R3. Initially, ~9 V voltage was needed to operate the Ge 38 S 62 based cell, which then decreased to ~5 V in the subsequent operation. Interestingly, no obvious firing process was observed in the Ge-rich GeS OTS device (Ge concentration≥50 at.%). The origin for this phenomenon is still under investigation. We have added this discussion in our manuscript. -The interesting description of the evolution of the GeS system structure leading to the lengthened Ge chains and the more connected network by the formation of over-coordinated Ge, is used to explain the higher material conductivity at high fields, in presence of Ge vacancies. However, the temperature in the "activated" material, in the light of the extremely high current densities proposed, should reach high values. Now, the model proposed takes in account a structural change under "excited state by hole addition", in a solid to solid-like structural transition. How the author comments the capability of the system to sustain such high temperatures, without any structural change? Is this model valid only for the device before firing or even after?
Reply: Thank you for raising these questions regarding the proposed threshold switching mechanism that synergizes both electronic and structural transitions. Conventionally, it has been believed that ovonic threshold switching (OTS) is purely electronic, given its high switching speed. This has also led to different materials selection criteria from those for phase change materials (PCM) which undergo a solid to solid-like structural transition. To enable solid to solid-like structural transition, PCMs are required to be poor glass former. On the contrary, OTS materials are better glass formers that have slower atomic transition and remain in the amorphous state to higher working temperatures. From an atomic bonding perspective, OTS materials should have stronger bonds to survive high currents or high working temperatures, retarding electromigration or the breaking of network bonds. This means using lighter, shorter bond-length elements like Se and S, instead of Te. Due to the stronger atomic bonds, the splitting of the bonding and antibonding states is more significant, giving rise to much larger bandgap values for OTS materials (e.g., GeSe: 1.1 eV, GeS: 1.5 eV) compared with those for PCMs (e.g., a-GeTe: 0.55 eV). In the first-principles simulations, however, we indeed observed transient structural transition upon carrier excitation under high field, indicating that OTS may not be purely electronic but may be assisted by structural changes. Nevertheless, this type of structural transition is only volatile, not permanent; in other words, the network will return to its pristine amorphous state in the absence of driving field. Our proposed mechanism should be more applicable to devices after firing which are cycled by significantly lower voltages than the firing voltages. This ensures that permanent structural changes do not occur as in the firing step.
- Fig. 2a shows a 200Mohm resistance at the output of the pulse generator. Is it correct?

Reply:
We appreciate for pointing out this mistake. There is no resistance at the output of the pulse generator. We have corrected the figure, which is presented in Figure R4.

Reviewer #2 (Remarks to the Author):
Authors investigated on amorphous GeS environment-friendly and earth-abundant surfide binary selector material that show large drive current density and meet IRDS standards. Besides, they also showed stochastic integrate-and-fire neuron behavior using GeS device. Most of the experimental and theoretical proofs were conducted almost identically to the GeSe analysis, and are quite reliable. It is considered to be an important paper that can be applied to a future selector device.
We are very grateful and pleased to read your positive evaluation of our manuscript and the suggestions and comments to further improve it. Fig. 1 (f).

Please show the error bar according to the repeated experiment for
Reply: Thank you for this suggestion. We have added the error bar in the Figure 1f, as shown in Figure R5.

Authors suggest that it can also be used for new memory technologies such as ReRAM. The switching time (10ns for on,100ns for off) is longer than Te / Se-based OTS devices. Is this switching time enough to apply to new memory technologies?
Reply: For PCM technology, as summarized in the Figure R6 (C. Zambelli et al., Proceedings of The IEEE, 2017, 9, 1790, although GeTe, Sb-doped GST PCM cells can be switched within 50 ns, most of PCM (like GST, N-GST, Ga-Sb-Ge) shows a Set speed of >100 ns. Thus, with 100 ns switching speed, GeS-based OTS cell is fast enough for PCM applications.
In the case of ReRAM, as summarized in the  IEEE, 2017IEEE, , 9, 1770, the typical switching speed of ReRAM is ~50 ns.
Micro-seconds were needed for some ReRAM cells. Therefore, the application of GeS OTS cell in the ReRAM just slightly slow the switching speed.   , 2017, 9, 1770).

Please explain the detailed DFT calculation results for S 3p lone-pair states and the effect under high field.
Reply: Thank you for suggesting an investigation of the S 3p lone-pair states. It has been believed that the interactions between the lone-pair (LP) electrons on different chalcogen atoms create the localized gap states (S. R. Ovshinsky and K. Sapru, Taylor & Francis, London, 1974, p. 447). The formation of valence alternation pairs (VAP) is a common result of these LP interactions because of the low formation energies of VAPs (M. Kastner et al., Phys. Rev. Lett., 1976, 37, 1504. LP interactions induced gap states, or VAPs induced gap states in particular, have been associated with the ovonic threshold switching (OTS) phenomena (D. Adler et al., J. Appl. Phys., 1980, 51, 3289). This chalcogen LP-based midgap defect model works well for systems with chains or clusters of chalcogen atoms, such as amorphous selenium, which have high-lying chalcogen p-LP states. However, the presence of chalcogen-chalcogen chains in amorphous germanium chalcogenides has been questioned experimentally (P. Jovari et al., Phy. Rev. B, 2008, 77, 035202) and from density-functional-theory (DFT) calculations (S. Caravati et al., Appl. Phys. Lett., 2007, 91, 171906). In agreement with these works, our experiment and DFT simulations do not seem to support the presence of a significant amount of S-S chains or clusters in the GeS samples. In fact, our simulations indicate that the dominating number of three-fold S atoms have deep-lying s-LP states rather than high-lying p-LP states (figure R7b). Another difference between germanium chalcogenides (GeTe) and other chalcogenide glasses for which the chalcogen LP-based mid-gap defect model works properly has been pointed out to be the high-lying LP electrons being localized on Ge [A. V. Kolobov et al., Phys. Rev. B., 2013, 87, 155204]. This still enables VAP formation in germanium chalcogenides [A. V. Kolobov, Sci. Rep., 2015, 5, 13698]. We also plot the isosurface of electron localization function (ELF), which is sensitive to nonbonding electron pairs, for GeS, as shown in figure R7b. It can be seen that the nonbonding LP orbitals are not only located at S atoms but also at Ge atoms (mainly 3-fold Ge). A striking feature from our DFT simulated amorphous GeS is the existence of Ge-Ge chain. Indeed, this atomic feature is not exclusive to GeS but seems to be common for germanium chalcogenides, including GeTe and GeSe  Lett., 2020, 116, 052103] using an amorphous GeTe model generated by atomic distortion. Interestingly, the low formation energy of the Ge-Ge chains can be correlated, again, to a VAP formation mechanism, but through interactions between Ge LP electrons. As shown in Figure 6 and Figure S8, the gap states of GeS are quite localized at the Ge chain and Ge pair structures whose formation is therefore also believed to be due to Ge LP interactions. This discussion is added in the revised manuscript and Figure R7 was shown in the supplementary information as Figure S10. Fig 5 (a), explain whether the ratio between 2p3/2 and 2p1/2 from fitting results fits the theoretical values related to spin-orbit splitting. In the Raman spectra, there is information for peaks resulting from the local structure of Ge 4+ and Ge 2+. Please explain how to correlate with the results of 60% 3-fold, 23% 4-fold in XPS.

Please provide XPS and Raman fitting parameters. From the
Reply: Thank you for this valuable suggestion. Indeed, the fitting parameters for XPS and Raman did not match well with the theoretical values. We have re-fitting them. The fitting parameters of XPS and Raman results are summarized in Table R2 and   Table R3. Now the ratio between 2p3/2 and 2p 1/2 peaks from fitting results is ~2, matching well with the theoretical values related to spin-orbit splitting. The re-fitting results are show in Figure 5a and b in the revised manuscript (Figure R8a and b).  The results of 60% 3-fold, 23% 4-fold were obtained from melt-quench-relaxation amorphous GeS network. The concentration of 4-fold motif is ~10% lower than that found in the XPS. In the XPS and corresponding results, the fraction of Ge 4+ one (4-fold) is ~34%, almost half of that of Ge 2+ state (3-fold+2-fold). These structural motifs also can be detected by the Raman results, as shown in Figure R8a. From the fitting result in Table R3, we can find that 4-fold motif is the dominated one with higher area ratio. However, we cannot get the exact fraction because the unclear relationship of the different peak area and correspond structural fractions in Raman spectra.

The author described On state as the formation of new metavalent bonds used by GeSe case. Please explain why On-speed is slower than Ge-Se even though Ge-S covalency is very strong.
Reply: Thank you for raising another interesting point. The On state of GeS-based OTS is enabled by the synergy of the mid-gap traps assisted electronic transition and local Ge-Ge chain growth as well as locally enhanced bond alignment under high electric field. The transient growth of Ge-Ge chain needs to break the dominated Ge-S bonds and then adjust the local structure. The stronger Ge-S bonds than Ge-Se bonds means that the local structure is more stable and cannot be easily changed. This results in the relatively slower On-speed of GeS based OTS cell in our work.

Reviewer #3 (Remarks to the Author): S. Jia e al. Prepared and studied a GeS material for the application as OTS selector. The authors could show that GeSe-based OTS selectors yield the best-reported performance for such chalcogenide-based OTS selectors, including device
characteristics such as high drive current density and high nonlinearity. In order to understand these advanced properties, the authors studied the local structure of the GeS by theory and experiment. The article is well written and structured. The results are consistent and feasible. I recommend the publication of the submitted manuscript in the journal Nature Communications after some revisions: We thank you for your detailed review and the encouraging evaluation of our manuscript. We tried to implement all suggestions and comments to improve the manuscript further. Reply: Thank you for recommending these references. They are cited in our manuscript for replacing reference 1. Fig. 1b is not a HRTEm image of a device. The authors should provide a HRTEM image of the interface TiN/GeS/W. In addition, the authors should explain the occurrence of bright contrast within of the W bottom electrode. It seems to be porous.

Reply:
Thank you for pointing out this mistake. Figure 1b was a TEM image of a device. We have corrected it. The HRTEM image of the interface TiN/GeS/W is shown in Figure R9. This figure was added in the supplementary information ( Figure S6). Indeed, as your said, there is a porous within the W bottom electrode. The depth of the W bottom electrode is ~530 nm, as shown in the Figure R10, which was deposited by chemical vapor deposition. Since it is so deep that sometimes holes appear inside the W bottom electrode. Not all the devices have this issue.

Figure R10
Correctional TEM image of our device. (Figure 5d), confirm the coexistence of  in 3-fold pyramidal environment and Ge-Ge bonds (~2.44 Å length) in the tetrahedral". I would like to point out that the Figure 5d shows the appearance of the first RDF maxima at 2.41 Å, which is consistent with the simulated RDF profile. Thus, RDF analysis could not confirm the "the coexistence of  in 3-fold pyramidal environment and Ge-Ge bonds (~2.44 Å length) in the tetrahedral". Consequently, the sentences should be revised and the authors should explain what the see in the RDF analysis. Since TEM is local method, at this point, I would even recommend the authors to perform a RDF analysis based on XRD data in order to see two peaks. Reply: Thank you for another valuable suggestions. We have revised the sentences: Ge-S bonds (~2.45 Å length) in 3-fold pyramidal environment and Ge-Ge bonds (~2.44 Å length) in the tetrahedra have almost the same length 40 , both of which contribute to the first peak in radial distribution function (RDF), obtained by electron diffraction experiment of amorphous GeS (Figure 5d). The RDF result is also used to verify the structural fidelity of employed amorphous GeS network in the discussion part.

Lines 248-250: the authors wrote: "…obtained by electron diffraction experiment of amorphous GeS
Also, since the bond lengths of Ge-S bonds and Ge-Ge are almost the same, 2.45 Å and 2.44 Å, respectively, X-ray diffraction technique cannot distinguish them, as reported by N. Fueki et al. (Figure R11, Journal of the physical society of Japan, 1992Japan, , 6, 2814. Reply: Thank you for suggesting an investigation of GeS-based cell under different temperatures. We annealed two devices at 200 o C and 300 o C for 30 min, respectively, and then measured their I-V curve at the room temperature. The results are presented in Figure R12. Although the V th and V hold are a ~0.5 V different, no obvious performance degradation of the selector is found.  Reply: According to your suggestion, we have performance DC stress test of GeS-based OTS cell by applying 1 V (~1/3 V th ), 1.5 V (~1/2 V th ) and 4 V (~1.3 V th ) bias, as shown in Figure R13. No obvious degradation of ON and OFF performances is observed.   Figure R1. Clearly, although high current has had passed through the device, the GeS film remained in the amorphous structure after that ( Figure R1 f-h). This means that the thermal accumulation was still insufficient to induce the crystallization of GeS ( Figure R2a). Based on these results, it is reasonable to ignore the thermally driven permanent solid-to-solid transition in our DFT calculations before any performance degradation occurs (less than 1E8-cycle operations).
After 1E8-cycle operations, the OTS device performance began to deteriorate (Figure 2e), which indeed suggested the change of the structure or composition in the GeS film. This may be due to the gradual diffusion after long-time operation or high electrical field-induced electromigration. We cannot agree more that the analysis of device failure mechanism is important; however, it is another research subject which cannot be fully addressed in this manuscript focusing on the switching mechanism of GeS based OTS device. Nevertheless, we would like to underscore the our philosophy of choosing GeS as the OTS material with respect to the minimization of permanent solid-to-solid structural changes: to avoid permanent solid to solid-like structural transition, OTS materials should be better glass formers that have slower atomic transition and remain in the amorphous state to higher working temperatures, in contrast to PCM materials such as GeTe. From an atomic bonding perspective, OTS materials should have stronger bonds, preferably forming saturated covalent (fully connected) networks, to survive high currents or high working temperatures, retarding electromigration or the breaking of network bonds. This means using lighter, shorter bond-length elements like Se and S, instead of Te.
Thanks for your consideration. Figure R1 and Figure R2 are shown in Figure S6 and Figure S1 in the supplementary information, respectively. Figure R1 | a, The HAADF image of GeS-based device. b-e, Corresponding EDS element mappings of W, Ti, Ge and S respectively. f-h, Cross-section HRTEM images. Insets are Fast Fourier transform image of GeS layer. This GeS device has undergone repeated electrical operations. Homogenous element distributions of GeS film, without metal filament, are observed in the repeatedly operated selectors, as shown in Figure S6 a-e. Also, the GeS film maintains its amorphous state ( Figure S6 f-h), which proves that the threshold switching of these OTS selectors is the result of electronic processes, unlike the Conductive bridge threshold switch (CBTS) and Phase change memory (PCM).

Fig. R2 | XRD patterns of annealed a, GeS and b, Ge 38 S 62 films at different temperatures. 200
nm-thick GeS and Ge 38 S 62 were deposited on SiO 2 /Si substrate annealed at different temperatures for 0.5 h. A small amount of GeS crystalline phase appears in the film annealed at 400 o C, while the sample at 350 o C is completely amorphous. Hence, the crystallization temperature is higher than 350 o C, which is higher than reported mature selection material systems-GeSe. A crystallization temperature higher than 450 o C is observed for Ge 38 S 62 .

Reviewer #2 (Remarks to the Author):
We thank the authors for faithfully answering all the answers. All answers were logically well answered, so it is recommended that the paper be published in nature communications.