- Email: [email protected]
Microscopy study of the conductive filament in HfO2 resistive switching memory devices
Microelectronic Engineering 109 (2013) 75–78
Contents lists available at SciVerse ScienceDirect
Microelectronic Engineering journal homepage: www.elsevier.com/locate/mee
Microscopy study of the conductive filament in HfO2 resistive switching memory devices S. Privitera a,⇑, G. Bersuker b, B. Butcher b,c, A. Kalantarian b,d, S. Lombardo a, C. Bongiorno a, R. Geer c, D.C. Gilmer b, P.D. Kirsch b a
Institute for Microelectronics and Microsystems (IMM), National Research Council (CNR), Zona Industriale VII Strada 5, Catania 95121, Italy SEMATECH, Albany, NY, USA College of Nanoscale Science and Engineering (CNSE), Albany, NY, USA d Stanford University, 450 Serra Mall, Stanford, CA 94305, USA b c
a r t i c l e
i n f o
Article history: Available online 3 April 2013 Keywords: Resistive switching Conductive filament Electron energy loss spectroscopy
a b s t r a c t A detailed physical analysis of the conductive filament electrically formed in HfO2-based resistive switching memory devices with both Hf and Ti metal oxygen exchange layers is presented. The filament, observed by applying transmission electron microscopy (TEM), scanning TEM (STEM), and electron energy loss spectroscopy (EELS) techniques to 50 50 nm2 cells, is a cone-shaped metal-rich region in the HfO2 dielectric of the resistive switching device. Ó 2013 Elsevier B.V. All rights reserved.
1. Introduction Resistive switching (RS) random access memory (RRAM) based on the resistance change of transition metal oxides, such as HfO2, has attracted significant interest due to its low power operation, switching speed, high endurance and dense integration. The conductive filament formation mechanism in RRAMs is not yet fully understood, although it has been shown that the RS properties strongly depend upon the metal electrodes. Proposed models [1– 6] (primarily based on electrical characterization) agree that the switching phenomenon is due to formation and rupture of a conductive filament. However, the filament physical properties remain a controversial issue, in part due to a complexity associated with locating a filament in the device, and preparing a sample for TEM study without affecting the filament composition. Direct microscopic observation of the conductive filament in the memory cell is thus critically important to support RS models. In this study, we have employed scanning transmission electron microscopy (STEM) and electron energy loss spectroscopy (EELS) to observe electrical stress-formed conductive filaments in the HfO2-based crossbar devices with Hf or Ti top metal gettering or oxygen exchange layers (OEL), and TiN electrodes.
film as an oxygen exchange layer for oxygen gettering. TiN bottom electrodes have been employed in both type of devices. Thickness of HfO2 layer is 5 nm. Current–voltage (I–V) characteristics for both fresh and formed devices have been measured as a function of temperature in the 50 °C to 100 °C range using an Agilent HP4156B parameter analyser. The conductive filament formation (the forming operation) has been achieved by DC I–V voltage sweep at room temperature. The structural characterization of the conductive filament by the Transmission Electron Microscopy (TEM) technique has been performed using a JEOL JEM2010 equipped with the electron energy loss spectroscopy (EELS) and with scanning TEM (STEM) imaging capabilities. To study the morphology of the formed conductive filaments, two different techniques have been employed: STEM in dark field configuration and EELS at low energy. The former is sensitive to the local average atomic number while the latter is strictly related to the plasmon losses, determined by the local chemical composition and phase. Resulting micrographs have a lateral resolution of about 1 nm, mainly determined by the STEM electron beam size.
3. Results 2. Experimental procedure Crossbar devices with size of 50 nm 50 nm have been manufactured using either Hf or Ti metal layer over the HfO2 dielectric ⇑ Corresponding author. Tel.: +39 0955968233. E-mail address: [email protected] (S. Privitera). 0167-9317/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.mee.2013.03.145
Fig. 1(a) shows the I–V characteristics measured from 50 °C to 100 °C, for a fresh (prior to forming) 50 nm 50 nm crossbar device. The fresh cell exhibits low conductivity in the entire temperature range, with its conductance (dI/dV) exponentially increasing as a function of temperature (see Fig. 1(b)). Extracted activation energy for the conductance is 0.11 eV.
76
S. Privitera et al. / Microelectronic Engineering 109 (2013) 75–78
Fig. 1. (a) The current–voltage measurements performed on a fresh device in the temperature range of extracted from the data in (a).
50 °C to 100 °C. (b) Conductance versus temperature dependency
Fig. 2. (a) The current–voltage characteristic measured during the forming operation. In the inset: Conductance versus temperature measured after the filament was formed. (b) Current–voltage characteristics after filament formation. The device is initially in the SET state (line 1). It can be reset by applying negative voltage (line 2) and then set again by positive voltage (line 3).
Fig. 2(a) shows I–V sweep performed under the current compliance limit of 1 mA at room temperature, during which the formation of the conductive filament occurred at around 1.4 V. After the filament formation, the device exhibits much higher conductance, which linearly decreases with temperature as shown in the inset of Fig. 2(a), thus suggesting that the conductive filament is of metallic nature [6]. Fig. 2(b) shows typical operation of a formed filament as resistive switch memory. After filament formation the device is in the high conductive ‘‘SET’’ state (line 1). By applying negative voltage it can be reset (line 2) to a low conductive state, and then set again by positive voltage (line 3). To study the morphology and microstructure of the conductive filament formed under these conditions, a number of crossbar devices after the forming have been analysed by TEM. To avoid a filament oxidation, which may occur when the sample cross-section is done through the filament region, the entire device volume was included in the TEM specimen, which was about 60 nm thick, as prepared by using a Focused Ion Beam tool. Fig. 3 reports a STEM dark field micrograph showing a bright conically shaped region in the HfO2 oxide. The contrast in this image is determined by the local electronic density, i.e. by the local atomic number Z per unit volume. Therefore, a brighter area in this image corresponds to a region in the insulating layer with higher average Z. This observation can be explained by considering that a metallic Hf-rich filament was formed in HfO2. The atomic density of the pure metallic Hf is 4.5 1022 atoms/ cm3 while the Hf atomic density in HfO2 is 9.27 1021 atoms/cm3. Therefore, in the 60 nm thick sample the number of Hf atoms in pure Hf and in HfO2 is 2.7 1017 atoms/cm2 and 5.56 1016 atoms/cm2, respectively. If a pure metallic Hf filament with the
Fig. 3. STEM dark field image of a crossbar device.
width and thickness of 5 nm is formed in the HfO2 film, then the local Hf atoms concentration should increase by 30%. Therefore, given a large difference of the Hf atoms density in the metallic and dielectric phases, even a few nm of metallic Hf within the Hf oxide volume can be clearly seen by the Z contrast. This explains well the image in Fig. 3, which indicates the presence of a metallic Hf-rich filament in HfO2. Complementary to a Z-contrast dark field STEM, EELS-STEM imaging has also been performed. The EELS spectra were collected point-by-point (<1 nm diameter electron beam spot with the pixels of 1 nm 1 nm) throughout the entire image area. Electron energy loss at low energy is mainly dominated by the plasmon losses, sensitive to the local chemical composition. Fig. 4 shows EELS spectra collected in the region containing either Hf, or HfO2, or TiN. Each spectrum can be fitted by a sum of Gaussian functions, whose peak position, width and height are characteristic of a specific material. Due to a higher scattering factor of the atoms with higher atomic number, like Hf, the signal intensity acquired in the regions containing pure metallic Hf is correspondingly much lower.
S. Privitera et al. / Microelectronic Engineering 109 (2013) 75–78
77
Fig. 6. STEM dark field image of a crossbar device with the Ti OEL.
Fig. 4. Electron energy loss spectra collected in a region containing TiN, Hf oxide or Hf.
Fig. 7. TiN, HfO2 and Ti fitting coefficients obtained from the 3D spectral image acquired in the region shown in the top figure.
Fig. 5. TiN, HfO2 and Hf coefficients determined from the fitting of the 3D spectral image acquired in the region outlined in the dark-field image in the top figure.
The EELS image is obtained by collecting pixel by pixel the EELS spectra throughout the image area. Each spectrum was fitted as the superimposition of the plasmon losses of the Hf, HfO2 and TiN layers, each with a specific ‘weight’ reflecting the volume fraction of the relative phase in the pixel volume. Fig. 5 shows, from top to bottom, the maps of the normalized coefficients of TiN, HfO2 and metallic Hf, as determined from the fitting of the EELS spectra in each pixel. A lack of the HfO2 signal is observed around the central region of the device dielectric, with
a corresponding increase of the metallic Hf signal. The data shows that the filament is of a conical shape with the estimated top and bottom radii of 5.5 nm and 2.5 nm, respectively. This is in good agreement with the Z-contrast dark-field STEM of Fig. 3. So, in summary STEM analysis with two different techniques indicates the presence of a filament of the metallic Hf in the device after the forming process. This conclusion is in a good agreement with the model proposed in [7], where the conduction in Hf oxide is described by multi-phonon trap assisted tunnelling (TAT) and a conical filament with the narrower end formed near the cathode is expected. Indeed the electron transfer in the TAT process allows more energy to be released at the traps in the dielectric region farther away from the e-injecting cathode. The same type of analysis has also been performed on devices fabricated with the Ti gettering layer. Fig. 6 shows a dark field STEM image of a 50 50 nm2 device, in which the conductive filament has been formed using a current compliance of 0.1 mA. In contrast to the observations in the case of the Hf electrode, this device in dark field/Z-contrast STEM imaging is characterized by the presence of a darker region inside the HfO2 dielectric layer, i.e. by a region with a lower average atomic number. This data might be explained by assuming that a Ti-rich filament has been formed since
78
S. Privitera et al. / Microelectronic Engineering 109 (2013) 75–78
Ti atoms (Z = 22) are much lighter than the Hf atoms (Z = 72). The EELS STEM imaging is consistent with this hypothesis. Fig. 7 shows the maps of the normalized coefficients of TiN, HfO2 and Ti, as determined from the fitting of the EELS spectra taken pixel by pixel. The darker region in the dielectric layer seen in the dark field STEM corresponds to the region where the EELS map shows the highest Ti signal across the dielectric, that is a Ti filament propagating through the HfO2 film. The diameter of this filament is about 2.5 nm, smaller compared to the Hf, since in this case a lower current compliance level has been used for the filament formation The Ti+ ions migration into the HfO2 film may occur due to a high current compliance level used in the forming process resulting in high local temperature around the filament region. 4. Conclusion The conductive filament formed in the HfO2-based RRAM devices has been studied by two transmission electron microscopy techniques (dark field/Z-contrast STEM and EELS STEM), sensitive to either local scattering factor or plasmonic losses. Two different
metal oxygen exchange layers were studied: Hf and Ti. In both the cases, the data indicate the presence of a nm-sized metallic filament in the formed devices, with composition dominated by Hf and Ti, respectively. This finding allows to link electrical and physical filament characteristics opening the way to better understanding of the switching mechanism. References [1] S. Yu, H.-S.P. Wong, IEEE Electron. Dev. Lett. 31 (2010) 1455–1457. [2] E.A. Miranda, C. Walczyk, C. Wenger, T. Schroeder, IEEE Electron. Dev. Lett. 31 (2010) 609–611. [3] D. Ielmini, IEEE Trans. Electr. Dev. 58 (2011) 4309–4317. [4] D. Bocquet, D. Deleruyelle, C. Muller, J.-M. Portal, Appl. Phys. Lett. 98 (2011) 263 507-1–263 507-3. [5] X. Guan, S. Yu, H.-S.P. Wong, IEEE Electron. Dev. Lett. 59 (2012) 1172–1182. [6] G. Bersuker, D.C. Gilmer, D. Veksler, P. Kirsch, L. Vandelli, A. Padovani, L. Larcher, K. McKenna, A. Shluger, V. Iglesias, M. Porti, M. Nafría, J. Appl. Phys. 110 (2011) 124518-1–124518-12. [7] G. Bersuker, D.C. Gilmer, D. Veksler, J. Yum, H. Park, S. Lian, L. Vandelli, A. Padovani, L. Larcher, K. McKenna, A. Shluger, V. Iglesias, M. Porti, M. Nafría, W. Taylor, P.D. Kirsch, R. Jammy, IEDM (2010) 19.6.1–19.6.4.
Contents lists available at SciVerse ScienceDirect
Microelectronic Engineering journal homepage: www.elsevier.com/locate/mee
Microscopy study of the conductive filament in HfO2 resistive switching memory devices S. Privitera a,⇑, G. Bersuker b, B. Butcher b,c, A. Kalantarian b,d, S. Lombardo a, C. Bongiorno a, R. Geer c, D.C. Gilmer b, P.D. Kirsch b a
Institute for Microelectronics and Microsystems (IMM), National Research Council (CNR), Zona Industriale VII Strada 5, Catania 95121, Italy SEMATECH, Albany, NY, USA College of Nanoscale Science and Engineering (CNSE), Albany, NY, USA d Stanford University, 450 Serra Mall, Stanford, CA 94305, USA b c
a r t i c l e
i n f o
Article history: Available online 3 April 2013 Keywords: Resistive switching Conductive filament Electron energy loss spectroscopy
a b s t r a c t A detailed physical analysis of the conductive filament electrically formed in HfO2-based resistive switching memory devices with both Hf and Ti metal oxygen exchange layers is presented. The filament, observed by applying transmission electron microscopy (TEM), scanning TEM (STEM), and electron energy loss spectroscopy (EELS) techniques to 50 50 nm2 cells, is a cone-shaped metal-rich region in the HfO2 dielectric of the resistive switching device. Ó 2013 Elsevier B.V. All rights reserved.
1. Introduction Resistive switching (RS) random access memory (RRAM) based on the resistance change of transition metal oxides, such as HfO2, has attracted significant interest due to its low power operation, switching speed, high endurance and dense integration. The conductive filament formation mechanism in RRAMs is not yet fully understood, although it has been shown that the RS properties strongly depend upon the metal electrodes. Proposed models [1– 6] (primarily based on electrical characterization) agree that the switching phenomenon is due to formation and rupture of a conductive filament. However, the filament physical properties remain a controversial issue, in part due to a complexity associated with locating a filament in the device, and preparing a sample for TEM study without affecting the filament composition. Direct microscopic observation of the conductive filament in the memory cell is thus critically important to support RS models. In this study, we have employed scanning transmission electron microscopy (STEM) and electron energy loss spectroscopy (EELS) to observe electrical stress-formed conductive filaments in the HfO2-based crossbar devices with Hf or Ti top metal gettering or oxygen exchange layers (OEL), and TiN electrodes.
film as an oxygen exchange layer for oxygen gettering. TiN bottom electrodes have been employed in both type of devices. Thickness of HfO2 layer is 5 nm. Current–voltage (I–V) characteristics for both fresh and formed devices have been measured as a function of temperature in the 50 °C to 100 °C range using an Agilent HP4156B parameter analyser. The conductive filament formation (the forming operation) has been achieved by DC I–V voltage sweep at room temperature. The structural characterization of the conductive filament by the Transmission Electron Microscopy (TEM) technique has been performed using a JEOL JEM2010 equipped with the electron energy loss spectroscopy (EELS) and with scanning TEM (STEM) imaging capabilities. To study the morphology of the formed conductive filaments, two different techniques have been employed: STEM in dark field configuration and EELS at low energy. The former is sensitive to the local average atomic number while the latter is strictly related to the plasmon losses, determined by the local chemical composition and phase. Resulting micrographs have a lateral resolution of about 1 nm, mainly determined by the STEM electron beam size.
3. Results 2. Experimental procedure Crossbar devices with size of 50 nm 50 nm have been manufactured using either Hf or Ti metal layer over the HfO2 dielectric ⇑ Corresponding author. Tel.: +39 0955968233. E-mail address: [email protected] (S. Privitera). 0167-9317/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.mee.2013.03.145
Fig. 1(a) shows the I–V characteristics measured from 50 °C to 100 °C, for a fresh (prior to forming) 50 nm 50 nm crossbar device. The fresh cell exhibits low conductivity in the entire temperature range, with its conductance (dI/dV) exponentially increasing as a function of temperature (see Fig. 1(b)). Extracted activation energy for the conductance is 0.11 eV.
76
S. Privitera et al. / Microelectronic Engineering 109 (2013) 75–78
Fig. 1. (a) The current–voltage measurements performed on a fresh device in the temperature range of extracted from the data in (a).
50 °C to 100 °C. (b) Conductance versus temperature dependency
Fig. 2. (a) The current–voltage characteristic measured during the forming operation. In the inset: Conductance versus temperature measured after the filament was formed. (b) Current–voltage characteristics after filament formation. The device is initially in the SET state (line 1). It can be reset by applying negative voltage (line 2) and then set again by positive voltage (line 3).
Fig. 2(a) shows I–V sweep performed under the current compliance limit of 1 mA at room temperature, during which the formation of the conductive filament occurred at around 1.4 V. After the filament formation, the device exhibits much higher conductance, which linearly decreases with temperature as shown in the inset of Fig. 2(a), thus suggesting that the conductive filament is of metallic nature [6]. Fig. 2(b) shows typical operation of a formed filament as resistive switch memory. After filament formation the device is in the high conductive ‘‘SET’’ state (line 1). By applying negative voltage it can be reset (line 2) to a low conductive state, and then set again by positive voltage (line 3). To study the morphology and microstructure of the conductive filament formed under these conditions, a number of crossbar devices after the forming have been analysed by TEM. To avoid a filament oxidation, which may occur when the sample cross-section is done through the filament region, the entire device volume was included in the TEM specimen, which was about 60 nm thick, as prepared by using a Focused Ion Beam tool. Fig. 3 reports a STEM dark field micrograph showing a bright conically shaped region in the HfO2 oxide. The contrast in this image is determined by the local electronic density, i.e. by the local atomic number Z per unit volume. Therefore, a brighter area in this image corresponds to a region in the insulating layer with higher average Z. This observation can be explained by considering that a metallic Hf-rich filament was formed in HfO2. The atomic density of the pure metallic Hf is 4.5 1022 atoms/ cm3 while the Hf atomic density in HfO2 is 9.27 1021 atoms/cm3. Therefore, in the 60 nm thick sample the number of Hf atoms in pure Hf and in HfO2 is 2.7 1017 atoms/cm2 and 5.56 1016 atoms/cm2, respectively. If a pure metallic Hf filament with the
Fig. 3. STEM dark field image of a crossbar device.
width and thickness of 5 nm is formed in the HfO2 film, then the local Hf atoms concentration should increase by 30%. Therefore, given a large difference of the Hf atoms density in the metallic and dielectric phases, even a few nm of metallic Hf within the Hf oxide volume can be clearly seen by the Z contrast. This explains well the image in Fig. 3, which indicates the presence of a metallic Hf-rich filament in HfO2. Complementary to a Z-contrast dark field STEM, EELS-STEM imaging has also been performed. The EELS spectra were collected point-by-point (<1 nm diameter electron beam spot with the pixels of 1 nm 1 nm) throughout the entire image area. Electron energy loss at low energy is mainly dominated by the plasmon losses, sensitive to the local chemical composition. Fig. 4 shows EELS spectra collected in the region containing either Hf, or HfO2, or TiN. Each spectrum can be fitted by a sum of Gaussian functions, whose peak position, width and height are characteristic of a specific material. Due to a higher scattering factor of the atoms with higher atomic number, like Hf, the signal intensity acquired in the regions containing pure metallic Hf is correspondingly much lower.
S. Privitera et al. / Microelectronic Engineering 109 (2013) 75–78
77
Fig. 6. STEM dark field image of a crossbar device with the Ti OEL.
Fig. 4. Electron energy loss spectra collected in a region containing TiN, Hf oxide or Hf.
Fig. 7. TiN, HfO2 and Ti fitting coefficients obtained from the 3D spectral image acquired in the region shown in the top figure.
Fig. 5. TiN, HfO2 and Hf coefficients determined from the fitting of the 3D spectral image acquired in the region outlined in the dark-field image in the top figure.
The EELS image is obtained by collecting pixel by pixel the EELS spectra throughout the image area. Each spectrum was fitted as the superimposition of the plasmon losses of the Hf, HfO2 and TiN layers, each with a specific ‘weight’ reflecting the volume fraction of the relative phase in the pixel volume. Fig. 5 shows, from top to bottom, the maps of the normalized coefficients of TiN, HfO2 and metallic Hf, as determined from the fitting of the EELS spectra in each pixel. A lack of the HfO2 signal is observed around the central region of the device dielectric, with
a corresponding increase of the metallic Hf signal. The data shows that the filament is of a conical shape with the estimated top and bottom radii of 5.5 nm and 2.5 nm, respectively. This is in good agreement with the Z-contrast dark-field STEM of Fig. 3. So, in summary STEM analysis with two different techniques indicates the presence of a filament of the metallic Hf in the device after the forming process. This conclusion is in a good agreement with the model proposed in [7], where the conduction in Hf oxide is described by multi-phonon trap assisted tunnelling (TAT) and a conical filament with the narrower end formed near the cathode is expected. Indeed the electron transfer in the TAT process allows more energy to be released at the traps in the dielectric region farther away from the e-injecting cathode. The same type of analysis has also been performed on devices fabricated with the Ti gettering layer. Fig. 6 shows a dark field STEM image of a 50 50 nm2 device, in which the conductive filament has been formed using a current compliance of 0.1 mA. In contrast to the observations in the case of the Hf electrode, this device in dark field/Z-contrast STEM imaging is characterized by the presence of a darker region inside the HfO2 dielectric layer, i.e. by a region with a lower average atomic number. This data might be explained by assuming that a Ti-rich filament has been formed since
78
S. Privitera et al. / Microelectronic Engineering 109 (2013) 75–78
Ti atoms (Z = 22) are much lighter than the Hf atoms (Z = 72). The EELS STEM imaging is consistent with this hypothesis. Fig. 7 shows the maps of the normalized coefficients of TiN, HfO2 and Ti, as determined from the fitting of the EELS spectra taken pixel by pixel. The darker region in the dielectric layer seen in the dark field STEM corresponds to the region where the EELS map shows the highest Ti signal across the dielectric, that is a Ti filament propagating through the HfO2 film. The diameter of this filament is about 2.5 nm, smaller compared to the Hf, since in this case a lower current compliance level has been used for the filament formation The Ti+ ions migration into the HfO2 film may occur due to a high current compliance level used in the forming process resulting in high local temperature around the filament region. 4. Conclusion The conductive filament formed in the HfO2-based RRAM devices has been studied by two transmission electron microscopy techniques (dark field/Z-contrast STEM and EELS STEM), sensitive to either local scattering factor or plasmonic losses. Two different
metal oxygen exchange layers were studied: Hf and Ti. In both the cases, the data indicate the presence of a nm-sized metallic filament in the formed devices, with composition dominated by Hf and Ti, respectively. This finding allows to link electrical and physical filament characteristics opening the way to better understanding of the switching mechanism. References [1] S. Yu, H.-S.P. Wong, IEEE Electron. Dev. Lett. 31 (2010) 1455–1457. [2] E.A. Miranda, C. Walczyk, C. Wenger, T. Schroeder, IEEE Electron. Dev. Lett. 31 (2010) 609–611. [3] D. Ielmini, IEEE Trans. Electr. Dev. 58 (2011) 4309–4317. [4] D. Bocquet, D. Deleruyelle, C. Muller, J.-M. Portal, Appl. Phys. Lett. 98 (2011) 263 507-1–263 507-3. [5] X. Guan, S. Yu, H.-S.P. Wong, IEEE Electron. Dev. Lett. 59 (2012) 1172–1182. [6] G. Bersuker, D.C. Gilmer, D. Veksler, P. Kirsch, L. Vandelli, A. Padovani, L. Larcher, K. McKenna, A. Shluger, V. Iglesias, M. Porti, M. Nafría, J. Appl. Phys. 110 (2011) 124518-1–124518-12. [7] G. Bersuker, D.C. Gilmer, D. Veksler, J. Yum, H. Park, S. Lian, L. Vandelli, A. Padovani, L. Larcher, K. McKenna, A. Shluger, V. Iglesias, M. Porti, M. Nafría, W. Taylor, P.D. Kirsch, R. Jammy, IEDM (2010) 19.6.1–19.6.4.