Investigating bipolar resistive switching characteristics in filament type and interface type BON-based resistive switching memory

(1)

type BON-based resistive switching memory

Hsueh-Chih Tseng

a

, Ting-Chang Chang

a,b,

⁎

, Kai-Hung Cheng

c

, Jheng-Jie Huang

a

, Yu-Ting Chen

c

,

Fu-Yen Jian

a

, Simon M. Sze

a,d,e

, Ming-Jinn Tsai

f

, Ann-Kuo Chu

c

, Ying-Lang Wang

g

a

Department of Physics, National Sun Yat-Sen University, Kaohsiung, 804, Taiwan, ROC

b

Advanced Optoelectronics Technology Center, National Cheng Kung University, 1 Ta-hsueh Road, Tainan 701, Taiwan, ROC

c

Department of Photonics, National Sun Yat-Sen University, Kaohsiung, 804, Taiwan, ROC

d_{Department of Electrical Engineering, Stanford University, Stanford, CA 94305‐4085, USA} e_{Department of Electronics Engineering, National Chiao Tung University, Hsinchu 300, Taiwan, ROC} f

Electronics and Optoelectronics Research Laboratory, Industrial Technology Research Institute, Chutung, Hsinchu 310, Taiwan, ROC

g

Taiwan Semiconductor Manufacturing Company, Hsinchu, Taiwan, ROC

a b s t r a c t

a r t i c l e i n f o

Available online 17 September 2012 Keywords: ReRAM BON BON:Gd Interface type NDR

This paper studies the effect of doping on BON-based resistive switching characteristics. Typical bipolar resis-tive switching behavior can be observed in Pt/BON/TiN and Pt/BON:Gd/TiN devices. The conducresis-tive path(s) of the Pt/BON/TiN is vacancy-dominated while the Pt/BON:Gd/TiN is metal-dominated. Additionally, there is an atypical bipolar resistive switching in the Gd-doping device. This atypical characteristic has not only a size effect, but also a lower operating current. The resistance transitions are due to the variation in conductance of the switching layer, which is clearly inﬂuenced by the different area size. A mechanism is proposed to explain this atypical characteristic.

1. Introduction

Resistive random access memory (ReRAM) has attracted consid-erable interest for use in the next generation of nonvolatile memory devices due to its simple structure, low operation voltage and pro-cess compatibility with the present complementary metal–oxide– semiconductor industry[1]. Many materials have been demonstrat-ed to achieve resistive switching characteristics, such as FeOx,

Cr2O3, InGaZnO, and Al2O3[2–8]. Two dominant resistance switching

mechanisms have been proposed. One is oxygen vacancy nucleation at the metal/oxide interface [9,10]. The other mechanism is the conductivefilament (CF) model, the formation/rupture of a metallic filament with a metal such as Cu or Ag acting as the mobile ions in the oxide[11]. Whether the switching mechanism is vacancy-dominated or metal-vacancy-dominated, the oxygen anions and the metal atoms both play an important role in the ReRAM's switching behav-ior. Recently, rare earth (RE) metals, which are used broadly in the electrical and electronic industry[12,13], have been demonstrated to exhibit resistance switching phenomena[14–16]; furthermore, doping that RE metal can improve the ReRAM's characteristics[17]. Hence, in order to help clarify the influence of the metal contained in the ReRAM, this study uses the non-metal-containing BON as the

base transition layer. Up to now, the application of BON in the resis-tive switchingﬁeld has not been researched.

This work investigates resistance switching characteristics in Pt// BON/TiN and Pt//BON:Gd/TiN structures by using the temperature dependence of resistive states to observe the conductive path charac-teristics. The metal-doping method can provide the other resistance switching modes in the BON-based system to induce the dual switching type ReRAM. In addition, a mechanism is proposed to ex-plain the inﬂuence on the dual switching type ReRAM.

2. Experiment

A BON thin_{ﬁlm of 5 nm thickness was deposited on a TiN/Si} sub-strate by reactive magnetron RF sputtering a BN target in Ar and O2

ambient at room temperature. The RF sputtering power, time and pressure of the sputter system were set to 80 W, 800 s, and 1066 Pa. Next, the Pt top electrode (TE) was deposited and patterned by the liftoff process. In addition, the fabrication of the Pt/BON:Gd (10%)/TiN structure was similar to the Pt/BON/TiN structure, which was formed by co-sputtering the BON and Gd targets with the same deposition conditions as that for the Pt/BON/TiN structure. Fourier transform infrared spectrometry (FTIR) transmission spectra were used to observe the composition of BON and BON:Gd_{ﬁlms by Bruker} VERTEX 70v Fourier Transform Infrared Spectroscopy. The standard sample was a Pt/BON/TiN (PBT) structure, whereas the control sam-ple used as reference was the Pt/BON:Gd/TiN (PBGT) structure, and

⁎ Corresponding author at: Department of Physics, National Sun Yat-Sen University, Kaohsiung, 804, Taiwan, ROC. Tel.: +886 3 5726100; fax: +886 3 5722715.

E-mail address:[email protected](H.-C. Tseng).

(2)

will be referred to as PBT and PBGT samples hereafter. Additionally, both the PBT and PBGT samples needed to be activated by electroforming with a positive bias treatment about 5 V. All electrical characteristics were measured over an 8μm×8 μm to 2 μm×2 μm cell size with an Agilent B1500 semiconductor parameter analyzer. During these measurements, bias was applied to the TiN bottom elec-trode (BE) while the Pt top elecelec-trode (TE) was ground.

3. Results and discussion

Many studies have indicated that the ReRAM characteristic is due to the redox process by cation or anion migration, which is strongly relat-ed to oxygen or metal ions. In order to clarify the inﬂuence between ox-ygen and metal ions for ReRAM switching behaviors, this study uses the BON thinﬁlm with and without Gd doping as the insulator.Fig. 1shows the FTIR transmission spectra analysis which can be examined for the composition of the PBT and PBGT device transition layers.Fig. 1(a) shows that the bonds around 792 and 1064 and 1083 (cm−1), respec-tively, are due to the H-BN, the C-BN and the B-O-B stretching vibration for the PBT device[18,19]. The inset ofFig. 1(a) shows that there are no chemical bond signals around the Far-IR spectrum region. Furthermore,

Fig. 1(b) shows that the bonds around 440 and 542 (cm−1) represent the Gd–O vibration of cubic Gd2O3for the PBGT device[20]. The inset

ofFig. 1(b) also shows the C-BN and B-O-B bonds for the PBGT device.

Fig. 2(a) shows the typical bipolar switching behavior for the PBT and PBGT devices under identical operation conditions. The transitions between low resistance state (LRS) and high resistance state (HRS) are observed for 100 cycles by using dc voltage sweeping mode, as shown in insets (i) and (ii) ofFig. 2(a). Both PBT and PBGT devices require ac-tivation by the electroforming process. The resistance can be switched to HRS by applying negative bias of about−2 V. In a subsequent

sweep, the RS can be switched again to LRS by applying a predetermined positive bias of about 2 V with a 10 mA current compli-ance (Icom). The temperature dependence of the LRS for the PBT and

PBGT devices is shown inFig. 2(b), which indicates that the LRS of the PBT device decreases as ambient temperature increases, as is typical of semiconductor behavior properties[14]. This implies that the CF is dominated by vacancies. In contrast, the LRS for the PBGT device in-creases as ambient temperature rises, which indicates a typical metallic behavior[14]. These results support that doping BON with Gd can trans-form the CF from vacancy-dominated to metallic-Gd dominant.

Recent studies have indicated that the bipolar resistance switching is related to the redox reaction near the anode–electrode/oxide interface, which can be also deﬁned as the switching layer (SL) [14,21,22].

Fig. 3(a) and (b) shows the size effect of the PBT and PBGT devices, re-spectively. The CF cannot connect the top electrode (TE) and the bottom electrode (BE) undergoing the reset process. Hence, the leakage can be proportional to the device area. In contrast, there are no obvious trends when comparing leakage and area during the set process because the TE and BE have been connected by the CF. No matter whether the CF is vacancy-dominated or metallic-Gd dominant, the size effect of the off-current is due to the SL area size; moreover, the set process is a local be-havior causing there to be no clear area dependence for the on-current. According to the PBGT device results found inFig. 2(b), using the multilevel measurement for LRS re-conﬁrms the metallic-Gd domi-nant CF.Fig. 4(a) shows that applying different Icomduring the set

process, such as 1 mA, 5 mA, 10 mA, and 15 mA, induces the multi-on states. Accordingly, inset ofFig. 4(a) shows that the ature dependence of the multi-on state increases as ambient temper-ature rises. Because the CF's diameter or conductance is proportional to the Icom[23], a higher Icomaccompanies higher conductance for the

CF[24]. In addition, Fig. 4(b) shows that the PBGT device has an

Fig. 1. Fourier transform infrared spectrometry (FTIR) transmission spectra of (a) BONfilm in M-IR region, and (b) BON:Gd film in far-IR region. Insets of (a) and (b) show BON film in Far-IR region and BON:Gdfilm in M-IR region.

Fig. 2. (a) Typical bipolar resistive switching I–V curves for the BON and BON:Gd devices. (b) Temperature dependence of resistance in LRS for the BON and BON:Gd devices, with a 0.1 V reading voltage.

(3)

unusual trend when comparing the reset-voltage region of the PBT and PBGT devices. As a higher Icomis applied, there is a sudden

obvi-ous transformation from LRS to middle-LRS. The different tempera-ture dependencies of the LRS and the middle-LRS are also shown in the inset ofFig. 4(b). Comparing the on- and off-current values re-veals that the leakages for the PBGT device are higher than the PBT device due to the doping metal effect. Undergoing the reset process, the applied-voltage prefers to act on the metallic-type CF. After the sudden transformation, the conductance of the metallic CF becomes low enough to cause the applied-voltage to act on the vacancy-type CF instead. Regardless of whether the CF is vacancy-dominated or

metallic-Gd dominated, the resistance transition can be deﬁned as a ﬁlament type resistive switching behavior.

However, compared with other ReRAM characteristics[25,26], a particular behavior for the PBGT device during resistance switching exhibits a negative differential resistance (NDR) at the HRS while ap-plying positive bias as shown in the inset (ii) ofFig. 5. This phenom-enon is similar to the standard reset process of the bipolar switching cycles. In order to examine whether the NDR at HRS (NDRHRS)

pos-sesses a memory window, the sweeping voltage is applied from 0 V→ +0.8 V →−2 V → 0 V. There is a clockwise (CW) parasitic switching, and the sub-LRS/sub-HRS switching (sub-LRS is equal to standard-HRS) is below HRS of the standard ReRAM, as shown in

Fig. 5. In addition, the sub-LRS/sub-HRS ratio is suf_{ﬁcient to be} dis-tinguished during 100 dc bias switching cycles, as shown in the inset (i) ofFig. 5. These results also show that this CW resistance switching (CW-RS) mode not only reduces operation voltages such as Vset and Vreset, but also improves the operating current. The

CW-RS mode, therefore, has potential applications in low power portable electronic products.

Conﬁrmation of the path characteristics can be achieved by using the temperature dependence of the resistance state for sub-LRS and sub-HRS.Fig. 6(a) shows that both the sub-LRS and sub-HRS decrease when the ambient temperature increases, which indicates that the CW-RS is typically semiconductor-like. Hence, the switching behav-iors are related to the conductance variation of SL. As the state has not switched to LRS yet during the set process, the electric_{ﬁeld still} slightly drives the mobile oxygen anions. This effect prefers to occur near the SL/metallic-type CF region due to the high conductance of metallic CF. This phenomenon causes few oxygen anions to drift to the SL region; moreover, the oxygen anions may accumulate and re-combine with the oxygen vacancies forming a more insulated SL

[27]. Subsequently, the SL conductance can become lower, causing

Fig. 3. Size effect of LRS/HRS for (a) Pt/BON/TiN device and (b) Pt/BON:Gd/TiN device.

Fig. 4. (a) Typical bipolar resistive switching I–V curves for BON:Gd devices with different current compliance, and (b) reset process for the BON:Gd device. Inset of (a) shows the temperature dependence of the multi-LRS, and inset of (b) shows the temperature dependence of the LRS and middle-LRS.

Fig. 5. Untypical bipolar resistive switching I–V curves (CW-RS) for the BON:Gd device. The inset (i) shows the on/off ratio for standard-RS and CW-RS, and (ii) shows the ND-HRS.

(4)

the leakage current decrease which typiﬁes NDRHRSbehavior.

There-fore, the SL oxidation effect should be related to the area size, as shown inFig. 6(b). The current transformation of the CW-RS mode is attributed to the conductance variation for SL under the small elec-tricﬁeld, so there is indeed a size effect for sub-LRS and sub-HRS. The CW-RS mode can then be deﬁned as interface type resistive switching behavior[28–30].

The results of current fitting indicate that the sub-LRS and sub-HRS for the CW-RS obey the Schottky conduction mechanism, as shown inFig. 7(b) and (c). Gibb's free energy change accompany-ing the solution of the oxygen in the Gd and Ti has been studied, which indicates that the oxygen affinity of Gd is higher than Ti[31]. Since the TiN is regarded as the buffer layer blocking and attracting the oxygen ions, the mobile oxygen ions can accumulate among the SL region of TiN/bulk (residualfilament) which has undergone the positive bias treatment and the_{filament-type RS has not been} trig-gered yet. Then, the accumulated oxygen ions can oxidize the Gd and form a temporary barrier oxide due to the higher Gibb's free en-ergy of Gd. Accordingly, the transition current become lower. More-over, if there is no Schottky barrier formation, the soft break down voltage (set process offilament type ReRAM) should be located at Vset′. However, Gd attracting the oxygen ion induces the barrier for-mation, causing a shift of the set voltage (Vset). Therefore, the sweep-ing condition treated within the Vset offilament type ReRAM induces a reset behavior for the CW-RS. In a subsequent sweep, the reverse

polarity bias can drive the oxygen ions back to the residueﬁlament region due to the temporary barrier annihilation. Owing to the forma-tion and deformaforma-tion of the Schottky barrier, the CW-RS behavior can be achieved.

4. Conclusion

In conclusion, an investigation of the effect of a BON insulator layer with and without doped metal indicates that vacancy-type CF always exists in the Pt/BON/TiN and Pt/BON:Gd/TiN devices, with the metallic-type CF only appearing in the metal-doped device. Fur-thermore, the metal-doping device induces a dual-switching type de-vice causing there to two trends in resistance switching, namely ﬁlament-type and interface-type. The formation and deformation of the Schottky barrier, result in the interfacial-type ReRAM. The interfacial-type ReRAM not only has lower switching voltages such as Vsetand Vreset, but also reduces the operating leakage, which

im-proves the ReRAM's power consumption. Finally, if the size area can be scaled down sufﬁciently, operating current can be further reduced without an increase in switching voltages.

Acknowledgment

This work was performed at the National Science Council Core Facil-ities Laboratory for Nano-Science and Nano-Technology in

Kaohsiung-Fig. 6. (a) Temperature dependence for sub-LRS and sub-HRS, and (b) size effect for sub-LRS and sub-HRS.

(5)

[6] M.C. Chen, T.C. Chang, C.T. Tsai, S.Y. Huang, S.C. Chen, C.W. Hu, S.M. Sze, M.J. Tsai, Appl. Phys. Lett. 96 (2010) 262110.

[7] M.C. Chen, T.C. Chang, S.Y. Huang, S.C. Chen, C.W. Hu, C.T. Tsai, S.M. Sze, Electrochem. Solid-State Lett. 13 (6) (2010) H191.

[8] Y.T. Tsai, T.C. Chang, C.C. Lin, S.C. Chen, C.W. Chen, S.M. Sze, F.S. Yeh (Hung), T.Y. Tseng, Electrochem. Solid-State Lett. 14 (3) (2011) H135.

[9] J.J. Yang, F. Miao, M.D. Pickett, D.A.A. Ohlberg, D.R. Stewart, C.N. Lau, R.S. Williams, Nanotechnology 20 (2009) 215201.

[10] C. Yoshida, K. Kinoshita, T. Yamasaki, Y. Sugiyama, Appl. Phys. Lett. 93 (2008) 042106.

[11] C. Schindler, G. Staikov, R. Waser, Appl. Phys. Lett. 94 (2009) 072109. [12] A. Fissel, M. Czernohorsky, H.J. Osten, Superlattices Microstruct. 40 (2006) 551. [13] L. Marsella, V. Fiorentini, Phys. Rev. B 69 (2004) 172103.

[14] H.C. Tseng, T.C. Chang, J.J. Huang, P.C. Yang, Y.T. Chen, F.Y. Jian, S.M. Sze, M.J. Tsai, Appl. Phys. Lett. 99 (2011) 132104.

[24] Y.F. Chang, T.C. Chang, C.Y. Chang, J. Appl. Phys. 110 (2011) 053703.

[25] P.C. Yang, T.C. Chang, S.C. Chen, Y.S. Lin, H.C. Huang, D.S. Gan, Electrochem. Solid-State Lett. 14 (2) (2011) H93.

[26] S.C. Chen, T.C. Chang, S.Y. Chen, C.W. Chen, S.C. Chen, S.M. Sze, M.J. Tsai, M.J. Kao, F.S.Y. Huang, Solid-State Electron. 62 (2011) 40.

[27] Y.T. Tsai, T.C. Chang, W.L. Huang, C.W. Huang, Y.E. Syu, S.C. Chen, S.M. Sze, M.J. Tsai, T.Y. Tseng, Appl. Phys. Lett. 99 (2011) 092106.

[28] R. Muenstermann, T. Menke, R. Dittmann, R. Waser, Adv. Mater. 22 (2010) 4819. [29] K.P. Biju, X. Liu, S. Kim, S. Jung, J. Park, H. Hwang, Phys. Status of Solidi RRL 5 (3)

(2011) 89.

[30] F. Miao, J.J. Yang, J. Borghetti, G. Medeiros-Ribeiro, R.S. Williams, Nanotechnology 22 (2011) 254007.

[31] T.H. Okabe, K. Hirota, E. Kasai, F. Saito, T. Waseda, K.T. Jacob, J. Alloys Compd. 279 (1998) 184.