Threshold resistance switching in silicon-rich SiOx thin films
Chen Da, Huang Shi-Hua†,
Physics Department, Zhejiang Normal University, Zhejiang 321004, China

 

† Corresponding author. E-mail: huangshihua@zjnu.cn

Project supported by the Open Project Program of Surface Physics Laboratory (National Key Laboratory) of Fudan University, China (Grant No. KF2015_02), the Open Project Program of National Laboratory for Infrared Physics, Chinese Academy of Sciences (Grant No. M201503), Zhejiang Provincial Science and Technology Key Innovation Team, China (Grant No. 2011R50012), and Zhejiang Provincial Key Laboratory, China (Grant No. 2013E10022).

Abstract
Abstract

Si-rich SiOx and amorphous Si clusters embedded in SiOx films were prepared by the radio-frequency magnetron cosputtering method and high-temperature annealing treatment. The threshold resistance switching behavior was achieved from the memory mode by continuous bias sweeping in all films, which was caused by the formation of clusters due to the local overheating under a large electric field. Besides, the IV characteristics of the threshold switching showed a dependence on the annealing temperature and the SiOx thickness. In particular, formation and rupture of conduction paths is considered to be the switching mechanism for the 39 nm-SiOx film, while for the 78 nm-SiOx film, adjusting of the Schottky barrier height between insulator and semiconductor is more reasonable. This study demonstrates the importance of investigation of both switching modes in resistance random access memory.

1. Introduction

The first report about the reversible resistance switching (RS) effect in binary metal oxides was published about fifty years ago. With the decrease of the gate oxide thickness, the traditional flash memory is facing challenges in large storage, low power consumption, fast speed, long retention, best endurance, and three-dimensional integration due to its larger leakage current.[1] Therefore, the technique based on the reversible memory RS effect has attracted wide attention all over the world, which is one of the most possible candidates applied in the next generation nonvolatile memory because of its above features.[2] However, there is another RS effect called threshold RS, which has not been observed widely in materials. The difference between the two kinds of RS effects is whether they have two stable resistance states without external applied bias.[37] Furthermore, investigating the RS mechanism of different RS methods is significant to clarify the RS behaviors.

Several materials, such as NiO,[3,4,8] amorphous Ta2O5 film,[9] SiO2,[10] and amorphous NbOx film,[11] have been reported to show the coexistence of bi-stable memory RS and mono-stable threshold RS in one device by changing some fabricating or testing conditions. Silicon-rich oxide (SRO) films, also known as semi-insulating polysilicon films doped with oxygen atoms, have been reported in the using of passivating material for silicon electronic devices and integral part of nonvolatile.[1218] Some researchers have found that SRO films are composed mostly of amorphous Si and SiOx in a two-phase network, which means that there exist a number of Si–O–Si bonds and some Si–Si bonds.[19] If a bias is applied to the film, the formation and modification of silicon nanocrystal in the SiOx matrix will take place through breaking Si–O bonds and gathering Si atoms to form Si clusters.[17] At the same time, a lot of Si atoms cannot bond with other atoms and, as a result, the films contain a high concentration of silicon dangling bonds (Si-DBs).[15] Therefore, RS based on the SRO films is mostly ascribed to the formation and rupture of nonmetallic conduction filaments consisted of Si nanocrystals or Si-DBs.[1517] On the other hand, if SRO is firstly annealed at high temperature, the atoms will permute and combine again. The silicon clusters or particles will form, and the degree of silicon crystallization is depend on the annealing temperature and time.[2022] The leakage current will be smaller due to the lower concentration of defects, which is beneficial for improving the resistance and the on/off ratio for the application in RRAM. Therefore, SRO films are employed to form metal/insulation/Si structures to study their RS effects. Moreover, the threshold resistance switching effect was firstly found in the SRO films in the electrical measurement. And the threshold voltage and current are smaller than those of the other reported materials.[3,1012]

In this paper, we reported the threshold RS effect in silicon-rich SiOx and annealed SiOx. Firstly, we demonstrated the evolutionary process of the transformation from the memory to the threshold effect. Then, the effects of the film thickness and annealing temperature on the threshold RS behaviors were investigated. Finally, the conduction mechanism was employed to analyze the switching mechanism through fitting the current–voltage curves and combining with the previous researches about silicon-rich SiOx films. Furthermore, the transformation process from memory mode to threshold mode was also investigated.

2. Experimental details

Silicon-rich SiOx films were deposited by using the radio-frequency magnetron co-sputtering method. A sector piece of a Si wafer with a resistivity of ∼ 2 Ω·cm was tightly stuck to a SiO2 (99.99%) target with a diameter of 60 mm, and it was used as the co-sputtering target. A boron heavily-doped p-type Si wafer with a resistivity of ∼ 0.001 Ω·cm was used as the substrate and bottom electrode (BE). Before deposition, the Si wafer and the substrate were processed by standard RCA cleaning. After drying by nitrogen gas, they were inserted immediately into the pretreatment chamber and the sputtering chamber of the magnetron sputtering deposition system, respectively. When the background pressure was better than 4.0 × 10−3 Pa, 10-min presputtering was carried out to clean the surface of the target and stabilize the glow discharge of argon gas (99.999%) under a power density of 3.5 W/cm2 for each new deposition. During the deposition, the substrate temperature and the base pressure were maintained at 573 K and 0.5 Pa, respectively. Silicon-rich SiOx thin films with different thicknesses were obtained. A part of the deposited films were thermally annealed for 60 min in high-purity argon atmosphere at various temperatures. Finally, Cu, the top electrode (TE, 500 nm, 1 mm in diameter), was defined by depositing Cu film using a thermal evaporator system to complete the full device structure. Figure 1 shows a schematic diagram of the Cu/SRO/Si device.

Fig. 1. Schematic diagram of the Cu/SRO/Si device.

The electrical characteristics of the Cu/SRO/Si devices were measured by a digital sources tester (Keithley 2601) at room temperature (RT). During the electrical measurement, the bias voltage was applied on the Cu electrode (TE) with the Si substrate (BE) grounded.

3. Results and discussion

First, we illustrate the transformation process of the transition from the memory mode to the threshold mode for the Cu/SRO/Si device with 39-nm thickness SRO film annealed at 1073 K, as shown in Fig. 2. For the first DC voltage sweeping, the bias voltage is swept in the sequence of 0 → 10 → 0 → − 10 → 0 V, as shown in Fig. 2(a). The current fluctuates at the level of nA, and this state is called the high resistance state (HRS). The current then has an abrupt increase to a compliance current of 10 nA when the voltage reaches ∼ 7 V. The switching process is called the SET process. The device keeps in the low resistance state (LRS) when the voltage sweeps back from 10 V to 0 V. During the negative voltage sweeping, the device turns from LRS into HRS at the voltage about −5 V, and this process is called the RESET process. When the voltage sweeps from the negative voltage to 0 V, the device is still in the HRS. From the above description, we ascribe this behavior to the bipolar resistance memory mode due to the coexistence of a HRS and a LRS at zero voltage. Following the first IV measurement, the second measurement is carried out immediately, and the IV curve is similar to that of the first measurement. The threshold voltage is different, while the sequence of resistance states with the voltage sweeping has no change. There is an interesting phenomenon happened during the negative voltage sweeping process in the third measurement, which is retained in the fourth. In this IV measurement, the bias voltage is swept in the sequence of 0 → 0.1 → 0 → − 0.1 → 0 V. Observing from Fig. 2(b), when the voltage reaches 0.08 V, the device switches from HRS to LRS. Then, the device has a RESET process at the voltage about 0.008 V during the voltage sweeping back from 0.1 V to 0 V. And the IV curves in the negative bias are symmetrical to those in the positive bias. The difference between the IV curves in different voltage polarities is that the SET and RESET voltages are −0.068 V and −0.08 V, respectively. This phenomenon should be called threshold RS.[3,11] The details about the IV curves in small scale are shown in Fig. 3, which will be carefully discussed later in this paper. The conversion from the memory to the threshold switching will also be clarified.

Fig. 2. Transformation process from memory switching to threshold switching of the Cu/SRO/Si device with as-deposited SRO film: (a) first three voltage sweepings, (b) after four voltage sweeping.
Fig. 3. Typical IV curves of the threshold RS behavior for Cu/SRO/Si devices without annealing and annealed at 673 K, 873 K, and 1073 K, respectively. The thickness of these SRO thin films is 39 nm.

New IV curves with higher measuring accuracy for the devices annealed at various temperatures are shown in Fig. 3. It is clear that the as-deposited device has a set process at a voltage of about 0.1 V and a reset process at a voltage of about 0.01 V. Compared to the as-deposited device, the other devices annealed at various temperatures have larger threshold voltages and smaller currents in both resistance states. With the annealing temperature decreasing from 1073 K to 673 K, VSET and VRESET increase from 0.48 V to 1.08 V and from 0.17 V to 0.34 V, respectively. However, the leakage current increases with the decrease of the annealing temperature, which is opposite to the fact that we know.[23] It should be pointed out that all devices can keep the reversible threshold RS behavior for more than 100 cycles and 105 s. The retention characteristics of the Cu/SRO/Si device with 39 nm interlayer annealed at 1073 K are shown in Fig. 4. The resistance states and switching processes of every device are quite stable.

The thickness of the interlayer is changed to investigate the thickness dependence of the IV characteristic of the Cu/SRO/Si devices annealed at 1073 K, as shown in Fig. 5. With the increase of the interlayer thickness, the currents in HRS and LRS are kept in a lower magnitude and the buck resistance is larger. On the other hand, VSET increases from 0.48 V to 3.72 V and VRESET increases from 0.17 V to 0.84 V. Besides, the IV curve turns from symmetry to asymmetry with the interlayer thickness increasing from 39 nm to 78 nm. On other hand, the lowest point of the IV curve shifts from the zero voltage to negative bias, which means that the Schottky junction is dominate or the Schottky barrier height is high enough to show up.[24] Characteristics of Si clusters formed in SiOx have been successively studied by Kanzawa et al.,[25] Furukawa et al.,[26] and Yi et al.,[27] including the infrared absorption, photoluminescence spectra, component, and structural characterization.

Fig. 4. Retention characteristics of the Cu/SRO/Si device with 39 nm interlayer annealed at 1073 K.
Fig. 5. The IV characteristics of Cu/SRO/Si devices with different thicknesses of interlayers annealed at 1073 K.

With increasing annealing temperature from 673 K, the phase separation process of SiOx can be divided into three states due to the following two reactions:

where (SiOx) and SiO(g) represent a nonstoichiometric SiOx monomer at the void perimeter and a gas-phase SiO molecule, respectively. Stage 1, for annealing at 673 K in Ar atmosphere, Ar atoms can diffuse into the film, which results in the decrease of the distortion for the bonding network in the film and the deviation from stoichiometry in the local structure. However, the Ar atoms will be removed when annealing at around 773 K, which was observed by Furukawa et al.[26] in the transmitted infrared (IR) absorption and x-ray photoelectron spectroscopy (XPS). Stage 2, for annealing at 873 K, reaction (1) may occur in the thin film and gaseous SiO molecules will generate. Stage 3, the SiOx species separate into SiO2 and Si phases on the basis of reaction (2) when the annealing temperature reaches 1073 K. However, the Si clusters are amorphous in this stage. If the annealing temperature is higher than 1073 K, the crystallization of the amorphous Si clusters will take place. The three devices annealed at the three temperatures correspond to the three states, respectively. The decreases of VSET and VRESET with the increase of the annealing temperature shown in Fig. 3 can be ascribed to the better regular crystal structure, which is good for the formation of the conduction channel in the grain boundaries.[28]

Comparing the Raman spectra of the samples annealed at various temperatures with those of the samples with different Si concentrations, Kanzawa et al.[25] noted that the spectral changes caused by annealing were very similar to those caused by the increase in the excess Si concentration. Therefore, it is believed that amorphous Si clusters have formed in the film annealed at 1073 K.

Fig. 6. Fitting analysis of the IV curves. (a) Double-logarithmic plot of the as-deposited device in HRS. The inset of panel (a) is a Schottky emission plot. (b) Double-logarithmic plot of the as-deposited device in LRS. (c) and (d) The positive bias IV characteristic of the 39 nm-SiOx film annealed at 1073 K in coordinates corresponding to Fowler–Nordheim tunneling in HRS and LRS, respectively. (e) and (f) Schottky emission presentations of the IV curves of the 78 nm-SiOx film annealed at 1073 K in HRS and LRS, respectively.

In order to better understand the origins of the threshold RS effect, fittings to the IV characteristics are carried out to study the conduction mechanism. For the as-deposited device, a double-logarithmic plot for HRS in the positive bias is shown in Fig. 6(a). In the low voltage region, the Schottky emission mechanism is dominant. In the high voltage region, as indicated by the slope ∼ 1.02, the ohmic conduction is dominant. In addition, the ohmic conduction is still dominant in Fig. 6(b) as shown by the near 1.00 slope in the log–log plot. This means that in the two stages, the conduction theory of the variable rang hopping between uncontinuously trapping centers can well explain the phenomenon. For the device annealed at 1073 K, owing to the decrease of the defect density, such as silicon dangling bonds and oxygen vacancies, the process of injected electrons trapping by a-Si clusters will be obvious in LRS, which can be confirmed by the liner fitting about Fowler–Nordheim tunneling for the voltage < 0.5 V. However, as seen from Fig. 6(c) and the inset of Fig. 6(c), F–N tunneling and traps-filled space controlled limited conduction (SCLC) are dominated in the low and the high bias voltage regions, respectively. Moreover, both HRS and LRS are dominated by the Schottky emission for the 78 nm-SiOx film annealed at 1073 K, as shown in Figs. 6(e) and 6(f). In conclusion, a possible switching mechanism for the 39 nm-SiOx films can come from the same origin, i.e., formation and rupture of conduction paths. Note that different types of paths are formed in the films, for the as-deposited sample, electron trapping and detrapping by defects in high concentration is dominated; while for the 1073 K-annealed sample, tunneling between a-Si clusters plays an important role in the threshold RS behavior. On the other hand, for the 78 nm-SiOx film, adjusting of the Schottky barrier height between insulator and semiconductor is more reasonable to describe the switching behavior. In contrast from the mechanism reported by Jiang et al.,[29] formation of a-Si clusters by annealing is random and the distribution of a-Si clusters is uniform. When the electrons are injected from a cathode, a number of a-Si clusters trap electrons, but the possibility of electrons transported to the neighbor cluster is limited. Therefore, the way that electrons are collected by the anode is similar to a funnel.

We also measure the temperature dependence of RLRS at 0.4 V for the Cu/SRO/Si device with 39-nm thickness SRO film annealed at 1073 K, as shown in Fig. 7. It is clear that the resistance increases as the temperature decreases from 295 K to 230 K. This implies that the electronic transport in the LRS of the device is semiconduction rather than the metallic conduction. Therefore, it is reasonable for us to exclude the effect of the Cu electrode.

The switching from the memory mode to the threshold mode of SRO can be explained well by soft breakdown models for SiO2 thin films.[30,31] As sweeping with a large electric field for the initial time, local overheating occurs, which means that the balance between Joule heating and thermal dissipation is broken. This can induce local crystalline structural changes because the atoms will move to the local lattice point with the lowest energy. In addition, the Si–O bonds will be broken to form Si–Si bonds, which then aggregate into clusters. Abundant clusters will present in the SiOx film by increasing the sweeping times and different conduction behaviors will happen between neighboring clusters because of their different morphologies. Therefore, we need a large bias voltage of about 8 V to active the reaction as shown in Fig. 2. In addition, the local overheating will facilitate the atoms to recombine again and the process is similar to current annealing, which is similar to the results reported by other groups; i.e., using compliance current,[10,11] bottom electrode with different thicknesses,[8] or controlling by temperature.[3]

Fig. 7. Temperature dependence of RLRS at 0.4 V for the Cu/SRO/Si device with 39-nm thickness SRO film annealed at 1073 K. The measurement was carried out about one year after completing the device.
4. Conclusion

We find that the threshold switching can be obtained from the memory switching by continuous high voltage sweeping in all fabricated devices. In addition, the IV characteristics show a dependence on the annealing temperature and the interlayer thickness of the Si-rich SiOx films. Fittings to the IV characteristics and features of Si clusters embedded in SiOx prove that different conduction channels are formed in the 39 nm-SiOx films. For the unannealed and 1073 K-annealed devices, electron trapping and detrapping by defects and tunneling between amorphous Si clusters dominate, respectively. On the other hand, the 78 nm-SiOx film annealed at 1073 K is caused by adjusting the Schottky barrier height between insulator and semiconductor. Above all, the conversion between the memory switching and the threshold switching is determined by the formation of clusters due to the local overheating under a large electric field. Our investigation provides a guide to control the two switching modes and to improve the performance of resistance random access memory.

Reference
1Kim S NAustin TBlaauw DMudge TFlautner KHu S JIrwin J MKandemir MNarayanan V 2003 Computer 36 68
2Lee Y HChen S PWu Y TChen S YWang C CTzeng J PLin H CChen FLien H CTsai J M2008IEEE International Electron Devices MeetingDecember 15–17, 2008San Francisco, America110.1109/IEDM.2008.4796677
3Chang H SLee S JChae C SLee B SLiu CKahng BKim W DNoh W T 2009 Phys. Rev. Lett. 102 026801
4Seo SLee J MSeo H DJeoung J ESuh S DJoung S YYoo K IHwang R IKim H SByun S IKim S JChoi S JPark H B 2004 Appl. Phys. Lett. 85 5655
5Dearnaley GStoneham M AMorgan V D 1970 Rep. Prog. Phys. 33 1129
6Inoue H IYasuda SAkinaga HTakagi H 2008 Phys. Rev. 77 035105
7Chen DHuang H S 2015 J. MicroNanolith. MEMS MOEMS 14 024501
8Chang H SChae C SLee B SLiu CNoh W TLee S JKahng BJang H JKim Y MKim W DJung U C 2008 Appl. Phys. Lett. 92 183507
9Cai Y YSheng C CLiang H C 2013 Appl. Phys. 111 1065
10Sun T HLiu QLi F CLong B SLv B HBi CHuo L ZLi LLiu M 2014 Adv. Funct. Mater. 24 5679
11Pang Y HLi F TLin N WWang Z YGao Y XWu T 2012 Sci. Rep. 2 442
12Matsushita TAoki TOtsu TYamoto HHayashi HOkayama MKawana Y 1976 IEEE Trans. Electron. Devices 23 826
13Matsushita TAoki TOtsu TYamoto HHayashi HOkayama MKawana Y 1976 Jpn. J. Appl. Phys. Suppl. 15 35
14Dori LAcovic ADiMaria J DHsu H C 1993 IEEE Electron Device Lett. 14 285
15Wang F YQian Y XChen J KFang H ZLi WXu J 2013 Appl. Phys. Lett. 102 042103
16Chen T YFowler BWang Z YXue FZhou FChang F YLee C J 2012 J. Solid State Sci. Technol. 1 148
17Yao JSun Z ZZhong LNatelson DTour M J 2010 Nano Lett. 10 4105
18Huang RZhang J LGao J DPan YQin Q STang PCai M YWang Y Y 2011 Appl. Phys. 102 927
19Hamasaki MAdachi TWakayama SKikuchi M 1978 J. Appl. Phys. 49 3987
20Nesheva DNedev NLevi ZBrüggemann RManolov EKirilov KMeier C 2008 Semicond. Sci. Technol 23 045015
21Nesbit A L 1985 Appl. Phys. Lett. 46 38
22Bae JHwang IJeong YKang O SHong SSon JChoi JKim JPark JSeong J MJia QPark H B 2012 Appl. Phys. Lett. 100 062902
23Zhu Z WTamagawa TGibson MFurukawa TMa P T 2002 IEEE Electron Device Lett. 23 649
24Huang J JKuo W CChang C WHou H T 2010 Appl. Phys. Lett. 96 262901
25Kanzawa YHayashi SYamamoto K 1996 J. Phys.: Condens. Matter 8 4823
26Furukawa KLiu YNakashima HGao DUchino KMuraoka K 1998 Appl. Phys. Lett. 72 725
27Yi X LHeitmann JScholz RZacharias M 2002 Appl. Phys. Lett. 81 4248
28Yang J JStrachan P JXia QOhlberg D A AKuekes J PKelley D RStickle F WStewart R DGilberto R MWilliams S R 2010 Adv. Mater. 22 4034
29Jiang RDu X HHan Z YSun W D 2015 Appl. Phys. Lett. 106 173509
30Tomita TUtsunomiya HSakura TKamakura YTaniguchi K 1999 IEEE Trans. Electron Devices 46 159
31Houssa MNigam TMertens W PHeyns M M 1998 J. Appl. Phys. 84 4351