Symposium Organizers
Yoshihisa Fujisaki Hitachi Ltd.
Rainer Waser RWTH Aachen University
Tingkai Li Sharp Laboratories of America Inc.
Caroline Bonafos CEMES/CNRS
H1: Advanced Flash I
Session Chairs
Gerard BenAssayag
Fabrice Gourbillleau
Tuesday PM, April 14, 2009
Room 2007 (Moscone West)
9:30 AM - **H1.1
Fundamental Limits and Trade-Offs for Flash Memory.
Victor Zhirnov 1 , Ralph Cavin 1
1 , Semiconductor Research Corporation, Durham, North Carolina, United States
Show AbstractThe remarkable progress in semiconductor memory technology has enabled an amazing array of consumer products. While desirable attributes of memory devices are high speed, high density, long retention, and low-voltage operation, the device physics sets fundamental limits and forces trade-offs in our efforts to integrate all desirable attributes in one memory element. In particular, what can be called “The Voltage-Time Dilemma” appears to characterize the limits of semiconductor memory scaling and performance. At the most basic level, for an arbitrary memory element, there is interdependence between operational voltage, the speed of operation and the retention time. This paper will examine the limits of scaling and performance for semiconductor memories with an emphasis on Flash.
10:00 AM - H1.2
Si Nanocrystals Synthesis in HfO2/SiO/HfO2 Multilayer Structures.
Michele Perego 1 , Gabriele Seguini 1 , Claudia Wiemer 1 , Marco Fanciulli 1 2 , Caroline Bonafos 3
1 , Laboratorio Nazionale MDM, CNR-INFM, Agrate Brianza Italy, 2 Dipartimento di Scienza dei Materiali, Università degli studi di Milano Bicocca, Milano Italy, 3 , nMat group CEMES-CNRS, Toulouse France
Show AbstractHigh dielectric constant (high-k) materials have been suggested as a suitable solution to overcome the fundamental trade-off between programming speed and data retention characteristics in Si nanocrystals (ncs) based memory devices.[1] In principle a dielectric with a higher-k than SiO2 allows to use a thicker tunnel oxide reducing leakage currents and increasing charge retention. At the same time due the smaller conduction band off-set between the high-k and the silicon substrate the goal of a low voltage-high speed non-volatile memory device can be achieved.[2] During the last ten years several approaches have been developed to synthesise 2D-layers of Si ncs embedded in SiO2, such as ion beam synthesis, chemical vapor deposition, molecular beam epitaxy, SiOx/SiO2 multilayer deposition followed by high temperature thermal treatment. The latter approach is extremely interesting because, by properly tuning the thickness and the stoichiometry of the SiOx film, an independent control of the dimension and of the density of the ncs can be achieved.[3] The synthesis of Si ncs in high-k materials is more challenging than in SiO2.[4] The super saturation conditions that are necessary to induce nc nucleation and growth are difficult to be achieved in high-k materials since silicon diffusivity in these materials is usually much higher than in SiO2. Moreover moisture absorption as well as oxygen diffusion in the high-k dielectrics represent a potential source of oxidizing agents that induce a strong reduction of the Si supersaturation in the SiO layer, limiting in this way the formation of Si ncs in the high-k matrix.[4]In this work we investigate the synthesis of Si ncs embedded in an HfO2 matrix by e-beam evaporation of HfO2/SiO/HfO2 multilayer structures and subsequent high temperature (1100°C) thermal treatment. Comparing these results with those obtained in analogous SiO2/SiO/SiO2 structures, we demonstrate that the synthesis of a 2D array of Si ncs embedded in HfO2 matrix is strongly inhibited by Si out diffusion phenomena occurring at the SiO/HfO2 interfaces during the thermal treatment. To overcome this problem a compensation mechanism for the reduction of Si concentration in the SiO layer during high temperature annealing is required. By increasing the thickness of the SiO layer the Si supersaturation can be preserved and the Si ncs synthesis is achieved. Other strategies to overcome the reduction in Si concentration during the thermal treatment are explored.[1].D.-W. Kim, T. Kim and S.K. Banerjee, IEEE Transactions on electronic devices 50, 1823 (2003)[2].C. Monzio Compagnoni, D. Ielmini, A.S. Spinelli, A. L. Lacaita, IEEE Transaction on electronic devices 52, 569 (2005)[3].M. Zacharias, J. Heitmann, R. Scholz, U. Kahler, M. Schmidt, J. Blasing, Appl. Phys. Lett. 80 (4) (2002) 661.[4].M. Fanciulli, M. Perego, C. Bonafos, A. Mouti, S. Schamm, G.Benassayag, Advances in Science and Technology 51, 156-166 (2006)
10:15 AM - H1.3
Ultra-Low energy Ion Implantation of Si into HfO2-based layers for Non Volatile Memory Applications
Caroline Bonafos 1 , Kristel Chan Shin Yu 1 , Gerard Benassayag 1 , Sylvie Schamm 1 , Abdou Slaoui 2 , Sahu Bhabani 2 , Marzia Carrada 2 , Sandrine Lhostis 3
1 nMat group, CEMES-CNRS, Université de Toulouse, Toulouse France, 2 , InESS-CNRS - ULP, Strasbourg France, 3 , ST Microelectronics, Crolles France
Show AbstractThe use of nanocrystals (NCs) embedded in a SiO2 matrix as charge storage elements in novel non-volatile memory (NVM) devices has been widely explored in the last decade. The replacement of the polysilicon layer of a flash memory by a 2D NC array presents several advantages but the fundamental trade-off between programming and data retention has not yet been overcome. In principle, low voltage NVM can be achieved by using high-k dielectrics as gate oxide. Several approaches have been explored to fabricate ordered arrays of Si-NCs in SiO2 but the transfer to high–k materials is not obvious due in particular to fast O2 diffusion in these high-k layers1. The fabrication of the NCs is carried out using an innovative method, ultra low energy (≤5 keV) ion implantation (ULE-II) into thin (7-9 nm) HfO2–based layers in order to form after subsequent annealing a controlled 2D array of Si NCs. This technique has been successfully used in the past for the controlled fabrication of a 2D-array (plane) of Si-NCs within very thin (<10 nm) SiO2 matrix2. The implantation of Si into HfO2 leads to the formation of SiOx–rich regions at the projected range due to the oxidation of the implanted Si atoms. This anomalous oxidation that takes place at room temperature is due to humidity penetration in damaged layers and is also present, to a lower extent, when Si is implanted into SiO2 3. Different solutions are investigated here in order to avoid this oxidation process and stabilize a pure Si-phase. They concern the enrichment of the HfO2 layers with N by ULE-II before Si implantation, the Si implantation into nitrurated HfSiO matrices and the implantation of Si through thin Si3N4 capping layers. Structural and chemical studies are carried out at the atomic scale by High Resolution Electron Microscopy and Electron Energy Loss Spectroscopy in scanning mode by using a nanometer probe (STEM-EELS). They are correlated to the electrical properties and memory performance through testing of MIS capacitors and transistors. Unexpected structures as HfO2 NCs embedded with SiO2 matrix are obtained and show interesting memory characteristics. A memory window of 0.6-1 V has been achieved at relatively low sweeping voltage of ± 5 V for the samples implanted at low energies and annealed at an optimum temperature of 950 oC.1 M. Fanciulli, M. Perego, C. Bonafos, A. Mouti, S. Schamm, G.Benassayag, Advances in Science and Technology 51, 156-166 (2006).2 C. Bonafos, M. Carrada, N. Cherkashin, H. Coffin, D. Chassaing, G. Ben Assayag and A. Claverie, T. Müller and K. H. Heinig, M. Perego and M. Fanciulli, P. Dimitrakis and P. Normand J. Appl. Phys. 95, 5696 (2004).3 A. Claverie, C. Bonafos, G. Ben Assayag, S. Schamm, N. Cherkashin,V. Paillard, , P. Dimitrakis, E. Kapetenakis, D. Tsoukalas, T. Muller, B. Schmidt, K. H. Heinig, M. Perego, M. Fanciulli, D. Mathiot, M. Carrada and P. Normand, Diffusion in Solids and Liquids, 258-260, 531 (2006).
10:30 AM - H1.4
Ge Nanocrystals Embedded in Si3N4/HfO2 Stack Gate Dielectrics by Low-energy Ion Implantation for Nonvolatile Memory Application.
Bhabani Sahu 1 , Abdelillah Slaoui 1 , Jean-Jacques Grob 1 , Marzia Carrada 1 , Caroline Bonafos 2 , Sandrine Lhostis 3
1 InESS, InESS-ULP-CNRS, Strasbourg, Strasbourg, France, 2 Groupe Nanomat – CEMES, CNRS – Université de Toulouse, Toulouse, Toulouse, France, 3 ST Microelectronics, ST Microelectronics, Crolles, Crolles, France
Show AbstractGe nanocrystals (nc-Ge) have attracted considerable attentions because of their potential usefulness in the fabrication of nonvolatile memory structure and integrated optoelectronics, as well as prospect of discovering physical phenomena. We have recently reported on the fabrication of nc-Ge is the ion implantation method followed by annealing, which is simple and compatible with current metal-oxide-semiconductor (MOS) device technology1. Replacement of the conventional SiO2 with hafnium based high-k dielectrics for the tunneling layer in nanocrystal floating gate memory can improve the trade-off between data retention and program efficiency due to unique band asymmetry of the high-k dielectrics in the programming and retention modes. In the present investigation, Ge nanocrystals (nc-Ge) embedded in the Si3N4/ HfO2 stack gate dielectrics near the gate of a MOS capacitor have been synthesized with low energy ion implantation and subsequent annealing at different temperatures. The top Si3N4 and bottom HfO2 layers have been deposited with electron cyclotron resonance-chemical vapor deposition (ECR-CVD) and atomic layer deposition (ALD), respectively. The formation and evolution of nc-Ge have been investigated using transmission electron microscopy (TEM), Rutherford backscattering spectroscopy (RBS), capacitance-voltage (C-V) and conductance-voltage (G-V) measurements. TEM images revealed that nc-Ge with an average diameter about 3.5 nm are well distributed within the Si3N4 matrix close to the gate after a post-implantation annealing treatment at 800-900 oC. The nanocrystals are well isolated and almost spherical in shape, thereby giving rise to an enhanced charge confinement effect, which can be suitable for non-volatile memory application. C-V and G-V measurements exhibit significant effects of the implant energy, dose, and annealing temperature on the electrical performance of the MOS capacitor. A large memory window of 1.6 V has been achieved at relatively low sweeping voltage of ± 5 V for the samples annealed at an optimum temperature of 800 oC. These results indicate that the combination of suitable energy Ge implants, annealing temperature, and choice of Si3N4/HfO2 stack can be suitable for the fabrication of low operating voltage memory devices.1 S. Duguay, J.J. Grob, A. Slaoui, Y. Legall, M. Amann-Liess, J. Appl. Phys. 97, 104330 (2005); S. Duguay, S. Burignat, P. Kern, J.J. Grob, A. Souifi and A. Slaoui, Semicond. Sci. Technol. 22 (2007) 837-842
10:45 AM - H1.5
Performance Enhancement of Ti Silicide Nanocrystal Memory Device.
Huimei Zhou 1 , Jianlin Liu 1
1 , UCR, Riverside, California, United States
Show AbstractNonvolatile memory devices with semiconductor or metal nanocrystals as storage elements in metal-oxide-semiconductor field effect transistors (MOSFET) have attracted much attention from researchers in both industry and academic institutions.1–2 Compared with semiconductor nanocrystals, metal nanocrystal device should have better performance due to the higher density of states3. However the drawback of the poor thermal budget problem makes the application of the metal nanocrystal device difficult. Metal silicide nanocrystals possess high densities of states and are more thermally stable than metal nanocrystals. Here we report novel fabrication process of self-aligned silicide nanocrystals and enhanced performance of the self-aligned Ti silicide nanocrystal MOSFET device.The novel silicide nanocrystal fabrication process begins with a thermal oxide of about 5 nm, which was grown at 850 °C. Si nanocrystals were grown at 610 °C for 15s with the pressure of 400 mtorr in a low pressure chemical vapor deposition (LPCVD) system. The Ti silicide nanocrystals were fabricated with a two-step annealing silicidation method. First a 10-nm-thick metal Ti layer was deposited onto the sample. Then the first annealing was performed in nitrogen at 775 °C for 60 s. The unreacted Ti metal on top of nanocrystals as well as in between nanocrystals was removed in selective etchant NH4OH:H2O2:H2O=1:1:5. The second annealing was performed at 880 °C for 30 s after the metal removal to form more thermally robust Ti silicide dots. The sample was then capped with control oxide with thickness of about 15 nm in a low-temperature oxide CVD furnace. Normal MOSFET process was performed afterwards to form MOSFET device. The devices were then characterized by HP 4284A LCR meter and pulse generator at room temperature. Memory effect was clearly found for the Ti silicide nanocrystal memory device after programming. Compare to the reference si nanocrystal memory device, the Ti silicide nanocrystal devices show faster writing, erasing speed and longer retention time. The better performance indicates that CMOS compatible silicidation process to fabricate silicide nanocrystals and Ti silicide nanocrystals memory shows higher promise in memory device applications.REFERENCE:1 Q. Wan, C. L. Lin, W. L. Liu, and T. H. Wang, Appl. Phys. Lett. 82, 4708 2003.2S. Choi, S. S. Kim, M. Chang, H. S. Hwang, S. H. Jeon, and C. W. Kim, Appl. Phys. Lett. 86, 123110 20053Z. Liu, C. Lee, V. Narayanan, G. Pei, and E. C. Kan, IEEE Trans. Electron Devices 49, 1606 2002.
11:30 AM - **H1.6
Nanocrystal Memories.
Thierry Baron 1 , Barbara De Salvo 2 , Pierre Mur 2 , Abdelkader Souifi 3 , Michel Gendry 3 , Sylvie Bodnar 4
1 LTM, CNRS, Grenoble France, 2 D2NT, CEA-Leti, Grenoble France, 3 INL, CNRS , Lyon France, 4 , ATMEL, Rousset France
Show AbstractDiscrete storage nodes memories have serious potential for pushing further the scaling limits of conventional NAND Flash. Si nanocrystals memories have reached the maturity to be integrated in conventional CMOS production lines. Due to their barrier height with respect to Si and SiO2, the use of the III-V semiconductor or metallic nanoparticles, to realize the granular floating gate of the memory should improve the retention time as compared to Si nanocrystals floating gate technology. Moreover their process technology could be fully CMOS compatible. Another improvement could be reach in using organized nanoparticles which guarantee the accurate control, in a reproducible way, of the electrical properties of the devices (programming windows). We are presenting a general overview of the nanocrystal memories based on Si nanocrystals, InAs and metallic nanoparticles. The potential and limitation of the introduction of these nano-materials in an industrial memory CMOS process is discussed. The challenge of organizing the nanoparticles by using a based copolymer diblock materials self-assembling technology is exposed.
12:00 PM - H1.7
Pt Nanocrystals with Remote Plasma Atomic Layer Deposited HfO2 for Nonvolatile Memory Devices.
Honggyu Kim 1 , Sanghyun Woo 1 , Hyungchul Kim 1 , Hyeyeong Chung 1 , Yongchan Kim 1 , Daesik Choi 1 , Hyeongtag Jeon 1
1 Materials science & engineering, Hanyang University, Seoul Korea (the Republic of)
Show AbstractNonvolatile memory devices employing semiconductor and metal nanocrystals have been extensively studied. Nanocrystal floating gate shows improved data retention and endurance characteristics compared to conventional flash memories and allows more aggressive scaling of the tunneling oxide. Among various candidates for the storage media, metal nanocrystals have many advantages over semiconductor nanocrystals such as Si and Ge. The metal nanocrystals provide a wide range of work functions, higher density of states around the Fermi level, and smaller energy perturbation due to carrier confinement. High-k dielectrics have been studied to replace SiO2 as a gate oxide in nanocrystal floating gate memories for further improvement of retention time and low voltage operation. Among many high-k dielectrics, HfO2 have been widely investigated due to its high dielectric constant, high density (9.68 g/cm3), large bandgap (5.68 eV) and good thermal stability in contact with silicon. For these reasons, we deposited the HfO2 gate dielectric material by remote plasma atomic layer deposition (RPALD). RPALD is expected to minimize the impurities of film and damage of substrate, increase the reactivity, and widen the process window. HfO2 films were deposited using TEMAH as the Hf precursor with Ar carrier gas. O2 plasma was used for reactant gas and the thickness of HfO2 film was controlled by the number of process cycles.In this study, Pt nanocrystals deposited with an electron beam evaporator were used for floating gates in the flash memory structure due to its high work function and thermal stability. The formation of Pt nanocrystals was confirmed using field emission scanning electron microscopy (FESEM) and high resolution transmission electron microscopy (HRTEM). In addition, the electrical characteristics, such as high-frequency C-V plots and C-t measurements, were examined to investigate the memory window for flash memories and retention time, respectively.
12:15 PM - H1.8
Non-volatile Memory Devices with in-situ Atomic-layer-deposited Ru Nanocrystals on the SiO2/Al2O3 Bilayer Tunnel Oxide.
Do-Joong Lee 1 , Sung-Soo Yim 1 , Ki-Su Kim 1 , Soo-Hyun Kim 2 , Ki-Bum Kim 1
1 Department of Materials Science and Engineering, Seoul National University, Seoul Korea (the Republic of), 2 School of Materials Science and Engineering, Yeungnam University, Gyeongsan-si, Gyeongsangbuk-do, Korea (the Republic of)
Show Abstract Memory structures with nanocrystal floating-gate have been extensively researched as a promising solution for the next-generation non-volatile memory devices. For the scalability of the device, appropriate methods of depositing high spatial density (> 1012 cm-2) nanocrystals are strongly required. In addition, a deposition of high-quality high-k dielectric as tunnel/blocking oxide is one of key processes to achieve the reliability, especially the robust retention characteristics, of devices. In these aspects, atomic layer deposition (ALD) can be a powerful tool for the deposition of reproducible, high density nanocrystals and high-quality high-k oxides such as Al2O3 and HfO2. High-k oxides deposited by ALD have been successfully demonstrated to show the low leakage current and the excellent conformality. In particular, we reported that ALD can be used as a promising method for the reproducible deposition of high density metal nanocrystals by controlling the initial nucleation and growth stage of the film formation, which occurs by the digitally-controlled surface-saturated reaction mechanism.[1] In this presentation, we will demonstrate the electrical properties of the nanocrystal memory using Ru nanocrystals and Al2O3 tunnel/blocking oxide deposited by in-situ ALD process. For the electrical characterization of Ru nanocrystal memory, we fabricated MOS capacitor structures as follows; p-type Si substrate/SiO2/Al2O3 tunnel oxide/Ru nanocrystals/Al2O3 blocking oxide/Al top gate. We used the bilayer of SiO2/Al2O3 as a tunnel oxide; 2-nm thick thermally-grown SiO2 was used for the prevention of interfacial reaction between the Si substrate and the Al2O3, while 3-nm thick ALD-Al2O3 was used as a nucleation promoting layer of ALD-Ru nanocrystals.[2] Ru nanocrystals were in-situ deposited on the Al2O3 using bis(ethylcyclopentadienyl)ruthenium as a precursor and oxygen as a reactant. Deposited nanocrystals have the high spatial density of 1.8×1012 cm-2 and the average size of 2.6 nm. For the electrical characterization, fabricated MOS capacitor structures were analyzed by high-frequency capacitance-voltage measurement. Detailed features of Ru nanocrystal memory including the program/erase characteristics and the retention behaviors will be given. [1] S. -S. Yim et al., Appl. Phys. Lett., 89, 093115 (2006).[2] D. -J. Lee et al., Electrochem. Solit-State Lett., 11, K61 (2008).
12:30 PM - H1.9
MOSFET Memory with NiSi Nanocrystals as Floating Gate.
Bei Li 1 , Jianlin Liu 1
1 Electrical Engineering, University of California Riverside, Riverside, California, United States
Show AbstractNanocrystal memory as one of the alternatives to replace the conventional flash memory in the future VLSI was first introduced by Tiwari [1]. In this presentation we report the use of vapor-liquid-solid (VLS) mechanism to grow NiSi nanocrystals as the floating gate for nonvolatile memory application for the first time. Compared to semiconductor nanocrystal memory, such as Si, metallic NiSi nanocrystal memory has a larger storage capability owing to its higher density of state. The deeper quantum well formed by NiSi embedded in SiO2 not only enhances the retention performance, but also offers uniform programming/erasing characteristics. Moreover, not like metals that easily react with oxide at high temperature process, silicides are more thermally stable.NiSi nanocrystals are synthesized by VLS-growing Si into the Ni catalyst dots for a short time. Controlling factors including growth temperature, growth time and catalyst size effect on the growth have been investigated. VLS is a well-known method for nanowire growth, such as Si [2], Ge [3], GaN [4], and etc. By reducing the growth time, NiSi nanocrystals are formed rather than Si nanowires. At the early stage of the growth, Ni nano-liquid droplets are first formed on SiO2 tunnel oxide, then SiH4 is introduced, Si atoms are absorbed into Ni droplets and form NiSi nanocrystals. NiSi nanocrystal memory device is fabricated using standard MOSFET processes starting with a p-Si (100) substrate. Memory consists of a 5nm thermally grown oxide, NiSi nanocrystals and a 20nm thick control oxide. The device is of 1um x 1um feature size. Memory characterizations including programming, erasing, retention and endurance are carried out by operating the device under both Fowler-Nordheim (FN) and hot carrier injection (HCI) charging mechanism. These NiSi nanocrystal memories show excellent memory performance.References1.S. Tiwari, R. Rana, K. Chan, L. Shi and H. Hanafi, Appl. Phys. Lett. 69, 1232 (1996).2.T. Baron, M. Gordon, F. Dhalluin, C. Ternon, P. Ferret, and P. Gentile, Appl. Phys. Lett. 89, 233111 (2006)3.Hemant Adhikari, Paul C. McIntyre, Ann F. Marshall, and Christopher E. D. Chidsey, J. Appl. Phys. 102, 094311 (2007)4.B. S. Simpkins, P. E. Pehrsson, M. L. Taheri, and R. M. Stroud, J. Appl. Phys. 101, 094305 (2007)
H3/FF3: Joint Session: Magnetic Resistive RAM
Session Chairs
Tuesday PM, April 14, 2009
Room 2007 (Moscone West)
4:30 PM - **H3.1/FF3.1
Spin Torque Effects in Nanoscale Magnetic Tunnel Junction Structures.
Robert Buhrman 1
1 Applied and Engineering Physics, Cornell University, Ithaca, New York, United States
Show AbstractThere has been remarkable progress in advancing the fundamental understanding of spin torque effects in magnetic nanostructures, and in successfully moving this phenomenon towards technological implementations, particularly for spin-torque MRAM and spin-torque microwave oscillator applications that have the potential for broad impact. In this presentation I will discuss some recent work that has sought to contribute to this rapidly advancing spin-torque research effort, to better understand quantitatively the details of the phenomenon, and to understand and enhance the efficiency of the effect, particularly in systems that utilize magnetic tunnel junctions. I will summarize results from studies of spin-transfer switching and microwave excitation in magnetic tunnel junctions (MTJs), which include the use of the spin transfer phenomenon to quantitatively determine, through spin-torque-excited ferromagnetic resonance (ST-FMR), both the bias dependent efficiency of the spin torque in high quality MTJs and the magnetic damping of individual free layer nanomagnets. I will also describe some recent three-terminal spin torque nanodevice experiments where a low impedance spin valve contact is employed to reverse a nanomagnet by a non-uniform injection of spin currents while a high impedance magnetic tunnel junction contact is employed to read-out the nanomagnet’s orientation. Finally I will present some recent results that have been obtained by employing several different magnetic materials engineering approaches to substantially reduce the current density required for ns-scale reversal of a thermally stable free layer, including the development of high TMR tunnel junction technology that utilizes magnetic free layers with low saturation magnetization.
5:00 PM - **H3.2/FF3.2
Material and Device Properties of MgO-Based Magnetic Tunnel Junctions for Spin Torque MRAM.
R. Dave 1 , N. Rizzo 1 , F. Mancoff 1 , P. Mather 1 , K. Smith 1 , B. Butcher 1 , T. Andre 1 , J. Slaughter 1 , S. Tehrani 1
1 , Everspin Technologies, Inc., Chandler, Arizona, United States
Show AbstractSpin torque (ST) switching is a candidate programming method that may enable higher density and lower power operation for future magnetic random access memory (MRAM). In this talk, we discuss the optimization of MgO-based magnetic tunnel junction (MTJ) material to meet requirements for ST-MRAM, including: efficient and repeatable ST switching, high magnetoresistance ratio (MR) and low resistance-area-product (RA). Different oxidation processes of Mg were evaluated to obtain key properties of high-quality MgO tunneling barrier. We discuss the evaluation of CoFeB as a suitable free layer for optimum device properties, including high MR for high spin-torque efficiency, moderate to low magnetization, and low magnetic damping. MTJ devices were fabricated on 200 mm Si wafers using optical lithography to form bits with sizes as small as 100 nm×180 nm. Low-bias MR values over 100% were obtained in patterned bits with CoFeB free layers with RA~6 Ω−μm2 using MgO tunnel barriers formed from natural oxidation of Mg. Quasistatic switching current densities of 3 MA/ cm2 were obtained with thermally-stable bits having a thermal energy barrier of 50kT, which corresponds to high-speed switching current density Jc0≈ 5 MA/ cm2 (extrapolated to 1 ns). We also present results on the performance of ST-MRAM arrays integrated with CMOS, including within-die read distributions, MgO barrier reliability under spin torque writing conditions, and distributions of switching voltage. Arrays with a switching to breakdown voltage ratio Vsw /Vbd ≈0.5, and switching distributions ≈5% have been achieved. The ST switching distributions are tighter than the field-switching distributions for the same bits, which are >10%, and exhibit less shape dependence.
5:30 PM - H3.3/FF3.3
Presence of B Oxide in the MgO Barrier in CoFeB/MgO/CoFeB Magnetic Tunnel Junctions and Its Effect on Tunneling Magnetoresistance.
Judy Cha 1 , J. Read 2 , R. Buhrman 1 , David Muller 1
1 School of applied and engineering physics, Cornell University, Ithaca, New York, United States, 2 Department of Physics, Cornell University, Ithaca, New York, United States
Show AbstractMagnetic tunnel junctions (MTJs) are an essential component in magnetic random access memories (MRAMs). Achieving a high tunneling magnetoresistance (TMR) ratio while maintaining a low resistance-area (RA) product is crucial for MRAMs. Since the theoretical prediction of over 1000% TMR [1], MgO-based MTJs became an excellent candidate for non-volatile memory devices and have been extensively studied. The steady increase in TMR [2] has led to a commercial MRAM 4 Megabit memory [3].
Despite the remarkable experimental progress, there is a discrepancy between the theoretical predictions and the experimental results. Theoretically, the high TMR value is attributed to a slowly decaying, coherent tunneling pathway available only to a majority spin state, which is made possible by lattice-matching the interface between the MgO barrier and the ferromagnetic electrodes [1]. Experimentally however, RF-sputtered MTJs, whose MgO/electrode interface is not controlled at atomic precision, produce high TMR values [4]. Moreover, using an amorphous CoFeB electrode and annealing the device afterward is found to produce higher TMR values. With the CoFeB electrodes, we observe that B diffuses into the RF-sputtered MgO barrier and becomes oxidized [5, 6]. Furthermore, we observe a decreasing TMR with increasing barrier thickness in as-grown junctions, opposite to the theoretical prediction. These observations suggest that the tunneling mechanism through the MgO layer cannot be fully accounted for by the current theoretical model [1].
To further investigate the effect of B oxide within the Mg-B-O layer on TMR, we examined Mg-B-O-based MTJs with varying tunnel barrier thicknesses and different metal alloy electrodes. The MTJs consist of a Si/SiOx / seed layer / 25 nm IrMn / 4nm CoFeB / Mg-B-O / 3nm CoFeB / capping layer. The capping layer is 8 nm Ta / 7nm Ru and the seed layer is 5 nm Ta / [20 nm CuN / 3 nm Ta]x4. We measured TMR and RA products of our samples and characterized them using electron energy-loss spectroscopy. Spectroscopic imaging on a MgO / CoFeB single interface was performed to get a 2D map of B oxide concentration in the Mg-B-O layer. We report the relative concentration of B oxide in the Mg-B-O layer with respect to the O content. We will discuss our TMR results and correlate them with the amount of B oxide found in the Mg-B-O layer.
[1] W. H. Butler, X. G. Zhang, T. C. Schulthess, et al., Phys. Rev. B 63, 054416 (2001).
[2] S. Ikeda, J. Hayakawa, Y. Ashizawa, et al., Appl. Phys. Lett. 93, 082508 (2008).
[3] Freescale, News Release, http://media.freescale.com/phoenix.zhtml?c=196520&p=irol-newsArticle&ID=880030 (2006).
[4] S. S. P. Parkin, C. Kaiser, A. Panchula, et al., Nat. Mater. 3, 862 (2004).
[5] J. J. Cha, J. C. Read, R. A. Buhrman, et al., Appl. Phys. Lett. 91, 062516 (2007).
[6] J. C. Read, P. G. Mather, and R. A. Buhrman, Appl. Phys. Lett. 90, 132503 (2007).
5:45 PM - H3.4/FF3.4
Novel Magnetoresistive Structures Using Self-Assembly and Nanowires on Si.
Mazin Maqableh 1 , Xiaobo Huang 1 , Liwen Tan 1 , Beth Stadler 1
1 , U Minnesota, Minneapolis, Minnesota, United States
Show AbstractAnodic Aluminum Oxide (AAO) was grown both as free-standing membranes and as integrated layers on Si as templates for arrays of magnetoresistive nanowires. These structures will be useful for applications such as current-perpendicular-to-plane giant magnetoresistance (CPP-GMR) sensors, magnetic random access memory (MRAM) and microwave oscillator arrays. As the AAO was formed, using a two-step anodization process, columnar nanopores self-assembled inside the oxide to form a close-packed array. The pore diameters could be varied from 10-60nm by changing the anodization conditions. As the diameter of the AAO nanopores decreased, the distance between the nanopores also decreased. The free-standing membranes had pores with lengths of 17um. The pores that were grown directly onto Si with an adhesion layer of Ti were 600nm in length. In addition to growing these latter pores directly onto Si, they were also grown onto Co/Cu/Co thin films that were evaporated onto the Si. Au nanocontacts were electroplated into these nanopores to study point-contact magnetoresistance and microwave response. For the magnetoresistive nanostructures, multilayered Co/Cu nanowires were fabricated via electrochemical deposition. The samples were measured with vibrating sample magnetometry (VSM) and also using ac and dc magnetotransport systems. The highest magnetoresistance was found in nanowires that had hysteresis loops that were identical as measured in plane and perpendicular to the plane. The highest measured MR (Delta R/R = 11%) of the multilayers was calculated as 33% by subtracting the resistance of the Cu leads on either side of the multilayers from the denominator. Shorter wires are currently under construction to avoid this effect. Spin transfer torque (STT) was also measured in the samples. For 10-60nm diameter nanowires, the change in resistance due to STT was around 6% which represents the full magnetoresistance of the larger wires, but only half that of the smaller nanowires. It is therefore concluded that the 10-nm Co layers do not align antiparallel to parallel as fully at the switching current density of JAP-P = 2.7 x 108 A/cm2 compared to the larger wires which switch at JAP-P = 3.2 x 107 A/cm2. With diameters in the 10-60 nm range and integration with Si, these nanostructures have great potential for future nanosensors, MRAM and microwave oscillator arrays.
Symposium Organizers
Yoshihisa Fujisaki Hitachi Ltd.
Rainer Waser RWTH Aachen University
Tingkai Li Sharp Laboratories of America Inc.
Caroline Bonafos CEMES/CNRS
H7: Organic Memory I
Session Chairs
Robert Mueller
Jea-Gun Park
Wednesday AM, April 15, 2009
Room 2007 (Moscone West)
9:30 AM - **H7.1
Recent Progress on Polymer Memory Devices and Their Switching Mechanism
Yang Yang 1 , Wei Lek Kwan 1 , Bao Lei 1 , Yue Shao 1 , Guanwen Yang 1 , Liping Ma 1
1 Department of Materials Science and Engineering, UCLA, Los Angeles, California, United States
Show AbstractPolymer memory devices remain a promising candidate for future non-volatile storage because of their low cost, easy fabrication process, as well as the potential for achieving 3D integration through a stacked structure. However, for further improvements on performance, the details of their operation mechanism should be better understood. We have designed various experiments for this and they include admittance spectroscopy, temperature dependent IV characteristics, data retention test, cycling test, cross sectional SEM and TEM, Energy Dispersive Spectroscopy, etc. The results suggested a switching mechanism which is related to localized conduction path(s).
10:00 AM - H7.2
Enhancing the Reliability of Polymer Memory Devices Using Dynamic Programming Circuit
Bao Lei 1 , Wei Lek Kwan 1 , Yue Shao 1 , Yang Yang 1
1 Department of Materials Science and Engineering, UCLA, Los Angeles, California, United States
Show AbstractAs multiple characterization experiments suggested the localized switching mechanism for the single polymer based non-volatile memory devices, we have applied controlled cycling tests on the devices to further explore the formation and destruction of localized conduction paths. Based on the relation between cycling response and programming parameters, a comprehensive model has been developed to explain the microscopic switching process. We suggest that the non-uniformity of the film plays an critical role in the switching cycles. Hence we designed a novel dynamic programming approach which manipulates the localized paths through feedback circuit and optimization algorithm. It has been experimentally demonstrated that the new operation circuit apparently enhances the reliability of the memory devices.
10:15 AM - H7.3
Observation of Localized Conductive Pathways in Polymer Memory Device.
Wei Lek Kwan 1 , Bao Lei 1 , Yue Shao 1 , Yang Yang 1
1 Department of Materials Science and Engineering, UCLA, Los Angeles, California, United States
Show AbstractIt has been observed that the current flowing through polymer memory devices is not homogeneous in the high conductivity state. In order to fully understand the origins of this non-uniformity, we prepared TEM cross-sections of the device using a dual beam focused ion beam (FIB/SEM). We found that in certain parts of the device, irregularity in the electrodes may result in larger than expected electric field in the device. Understanding the role of such non-uniformities in the device is crucial in elucidating the switching mechanism in polymer memory devices.
10:30 AM - H7.4
Resistive Electrical Switching of CuTCNQ Based Memory with a Dedicated Switching Layer.
Robert Mueller 1 , Nicole Thomas 1 , Christoph Krebs 1 2 , Ludovic Goux 1 , Dirk Wouters 1 , Jan Genoe 1 , Paul Heremans 1 3 , Sabina Spiga 4 , Marco Fanciulli 4
1 , IMEC v.z.w., Leuven Belgium, 2 , RWTH Aachen, Aachen Germany, 3 ESAT, KULeuven, Leuven Belgium, 4 Laboratorio Nazionale MDM, CNR-INFM, Agrate Brianza Italy
Show AbstractResistive electrical switching memories [1] are currently investigated for non-volatile memory applications, targeting the low-cost mass storage media market as replacement of traditional silicon based Flash memories. Although a huge variety of materials displays bistable resistive electrical switching, only a few of them achieve good endurance in terms of the number of write/erase (WE) cycles. Cu\CuTCNQ\Al devices (where TCNQ denotes 7,7',8,8'-tetracyanoquinodimethane) for example are known since 1979 to display resistive electrical switching [2]. In contrast to early assumptions of switching to be a bulk mechanism, recent results revealed that the change in conductivity is an interfacial effect occurring at a (native) aluminum oxide layer at the CuTCNQ\Al interface and not related to a bulk phenomenon. According to the latest research the Al2O3 layer acts as "switching layer" (SL) which is bridged by conductive channels formed by electrochemical reduction of mobile Cu+ cations of the solid ionic conductor CuTCNQ when the Al contact is polarized negatively with respect to the Cu contact [3]. The bridged SL constitutes the ON state of the CuTCNQ memory cell. Polarity reversal resulting in dissolution of the conductive channels switches the memory element back to the OFF state.Traditional Cu\CuTCNQ\Al memory cells with native Al2O3 SL can typically be prepared with just moderate yield and display only limited endurance (often less than 100 WE cycles). In this contribution, we present improved CuTCNQ based memory cells made by (i) inverting the CuTCNQ\SL stack and (ii) incorporating a well controlled, dedicated SL. The corresponding stack - bottom electrode (BE)\SL\CuTCNQ\top electrode (TE) - was prepared using Pt and n+Si as BE, Al2O3 as well as other clean-room compatible metallic and non-metallic oxides as SL, and Cu and Au as TE. The best memories cells obtained so far were those with BE and SL are directly prepared on 200 mm Si-wafers in the production line. Compared to traditional CuTCNQ memories with native Al2O3, our new memory cells exhibit a significant increase in both yield of operational memory elements and WE endurance of up to several ten thousand WE cycles. These improvements are attributed to the better control of the SL layer thickness and the uniformity of the SL\CuTCNQ interface.These results are extremely relevant for future experiments aiming at the integration of highly reliable CuTCNQ based memory cells, especially since BE and SL materials are often already present in standard industrial CMOS production lines.We gratefully acknowledge L. Lamagna and M. Alia for metal oxide deposition on Pt, and A. Lamperti for characterization of the films. This research was performed within the framework of the EMMA project of the European Commission (FP6-033751).[1] R. Waser and M. Aono, Nature Mat. 6 (2007) 833[2] R.S. Potember et al., Appl. Phys. Lett. 34 (1979) 405[3] J. Billen et al., Appl. Phys. Lett. 91 (2007) 263507
10:45 AM - H7.5
Manipulation the Current of the Polymer/nanoparticle Nonvolatile Memory Device.
Jianyong Ouyang 1
1 Materials Science and Engineering, National University of Singapore, Singapore Singapore
Show AbstractElectronic devices made of polymer and nanoparticles exhibit abrupt current switching during the voltage scan. The device is stable after switching to the high current state, and can be returned to the low current state by applying a negative bias. Thus, the polymer/nanoparticle device can be used a lost-cost flexible nonvolatile memory device. Though many papers have been reported on the polymer/nanoparticle memory devices, there is no report on the manipulation of the electrical current. Here, we report the manipulation of the current by controlling the organic ligands on the nanoparticle. The organic ligands also affect the stability of the device.
11:30 AM - H7.6
A Self-assembled Electrically Driven Molecular Rotor for Non-volatile Memory Application.
Mei Xue 1 , Kang L. Wang 1 , Sanaz Kabehie 2 , Jeffrey I. Zink 2
1 Electrical Engineering, UCLA, los Angeles, California, United States, 2 Chemistry and Biochemistry, UCLA, Los Angeles, California, United States
Show AbstractA novel electrically driven molecular rotor was developed and experimentally demonstrated to have non-volatile memory characteristics. The size of the device can be precisely controlled to the nanoscale regime by self-assembling process. The basic molecular rotor is composed of three components: a stator, a metal axle and a rotator. The stator is a bifunctional bidentate ligand used for both immobilization onto a solid support and for binding to a metal center. The axle is a copper (I) or copper (II) metal center. The rotator is 3, 8-di-ethynyltrityl-1, 10-phenanthroline, about 2 nm in length, and was designed as a rigid phenanthroline ligand containing pi-conjugated arms. The geometry of the Cu (I) system is tetrahedral and that of the Cu (II) system is square planar, resulting in a redox controlled molecular rotor with 90 degree rotation. The interconversion of the two molecular geometries provides the basis for controlled switching. Absorption spectroscopy was used to demonstrate the geometry change of the system as it was suspended in Dimethyl Sulfoxide (DMSO). The metal-to-ligand charge transfer band at 450 nm for the Cu (I) system was observed, proving the electron charging and discharging during the intramolecular rotation of the ligand (3, 8-di-ethynyltrityl-1, 10-phenanthroline) around the copper axle. Cyclic Voltammetry showed that 0.84V is needed to switch between the Cu (I) and Cu (II) bis (3,8-di-ethynyltrityl-1,10-phenanthroline) systems. The use of different stators (i.e. bisphosphine vs diiime) allows the threshold energy to become tunable. Electrical characteristics of switching were obtained to confirm this new concept for nano-memory applications.* The work was in part supported by the Focus Centers Research Program on Functional Engineered Nano Architectonics -- FENA.
12:00 PM - H7.8
Ferroelectric Size-effects in Nanoscale BiFeO3 Films Measured by Ultra-High Vacuum Piezoresponse Force Microscopy
Peter Maksymovych 1 , Stephen Jesse 1 , Sergey Lisenkov 2 , Laurent Bellaiche 2 , Nina Balke 3 , Mark Huijben 3 , Ramamoorthy Ramesh 3 , Arthur Baddorf 1 , Sergei Kalinin 1
1 Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, Tennessee, United States, 2 Physics Department, University of Arkansas, Fayetteville, Arkansas, United States, 3 Department of Materials Science and Engineering, and Department of Physics, University of California, Berkeley, California, United States
Show AbstractFerroelectric size effect remains among the most pursued and controversial topics in nanoscale ferroelectrics driven by both fundamental and applied interest due potential ferroelectric applications in non-volatile memories and spintronics. The size effect encompasses a series of questions including the scaling of the polar distortion, the domain structure in nanoscale ferroelectrics and the dynamics of polarization switching at the nanoscale. Recent hallmark experiments have showed that the polar distortion in PbTiO3 films can be sustained even in films as thin as three unit-cells by forcing the domain structure to have the periodicity proportional to square root the film thickness (in accordance with Kittel law) to minimize the depolarizing field [1]. Other works have reported uniform monodomain polarization in sub-10 nm films, stabilized in part by the surface chemistry [2].BiFeO3 (BFO) is a lead-free, chemically inert alternative to PbTiO3 with comparable spontaneous polarization. We have explored the domain structure of BFO films as thin as 2 nm using Ultrahigh Vacuum Piezo Force Microscopy (UHV PFM). We find that upon transition to vacuum, the films develop a rich out-of-plane domain structure. The observed domain size is as large as 500 nm even for 2 nm films, in drastic violation of the Kittel law. The amplitude of local piezoresponse displays strong correlation between the in-plane polarization vector and the local surface topography: the presence of a topographic island induces ferroelastic rotation of the polarization vector by 90 deg in plane of the surface. These findings are analyzed using a first-principles based model Hamiltonian approach. We have also explored the ability to switch ultrathin BFO films in the localized electric field of a biased metal tip in contact with the ferroelectric surface. Traditional ambient PFM is shown to fail in producing reliable ferroelectric switching due to parasitic electrochemical effects and a diminishing magnitude of piezoresponse. Both of these issues are resolved using UHV PFM combined with the Band-Excitation method of cantilever excitation, which maximizes the measurement sensitivity via tracking of the cantilever resonance during the complete switching cycle. Epitaxial BiFeO3 films grown by pulsed-laser deposition on LSMO electrode and SrTiO3 substrate were found to be repeatedly and reproducibly switched down to a thickness of 6 unit cells (2.4 nm) in ultrahigh vacuum without electric breakdown or topographic damage. The dependence of the nucleation voltage on the film thickness (L) is distinct from previously observed ((1/L), (1/L)^2/3) scaling laws. [1] D. D. Fong et. al., Science 304 (2004) 1650.[2] D. D. Fong et. al., Phys. Rev. Lett. 96 (2006) 127601.Experiments conducted at the Center for Nanophase Materials Sciences, Oak Ridge National Laboratory (ORNL) sponsored by U.S. Department of Energy. PM: Research performed as a Eugene P. Wigner Fellow and staff member at ORNL.
12:15 PM - H7.9
Effect of Interface Property on Small-molecular Nonvolatile Memory-cells
Sung-Ho Seo 1 , Woo-Sik Nam 1 , Yool-Guk Kim 1 , Young-Hwan Oh 1 , Sang-Yi Lee 1 , Jea-Gun Park 1
1 Department of Electrical & Computer Engineering, Tera-bit Nonvolatile Memory Development Center, Hanyang University, Seoul Korea (the Republic of)
Show Abstract We investigated the effect of interface between small-molecular materials and metal nanocrystals on small-molecular nonvolatile memory cells. The 4F2 memory cells were fabricated with the device structure of Al / small-molecular materials / Ni nanocrystals / small-molecular materials / Al. The small-molecular layers were produced by three kinds of organic materials which include aluminum tris (8-hydroxyquinoline) (Alq3), N,N’-bis(1-naphthalene)-1,1’biphenyl4-4’’diamine (α-NPD) and 2-amino-4, 5-imidazole dicarbonitrile (AIDCN). In addition, small-molecular nonvolatile memory cells were fabricated with the device structure of Al / Alq3 / metal nanocrystals / Alq3 / Al. The metal nanocrystals were produced by three kinds of metal which include Al, Fe and Ni. Furthermore, we compared electrical characteristics of six kinds of samples and analyzed chemical bonding in the interface between small-molecular layer and metal nanocrystals by XPS analysis. The result showed most excellent interface property in the interface between Alq3-Ni nanocrystals more than the interface between other small-molecular materials and metal nanocrystals. Ni2O3(binding energy: 856.6 eV) and NiO(binding energy: 854.5 eV) was formed by just effect of O2 plasma oxidation in the interface between α-NPD-Ni nanocrystals and AIDCN-Ni nanocrystals. However, in the case of the interface between Alq3-Ni nanocrystals, NiCO3 (binding energy: 855.5 eV) was formed because effectively oxidized Ni chemically reacted with small-molecular Alq3. The result of XPS analysis also confirmed only Al2O3 (binding energy: 74.5 eV) and Fe2O3 (binding energy: 710.4 eV) in the interface between Alq3-Al nanocrystals and Alq3-Fe nanocrystals without chemical reaction. In conclusion, multi-level small-molecular nonvolatile memory cells were developed with a sandwiched device structure of Al / Alq3 / Ni nanocrystals / Al. These device presented a memory margin (Ion /Ioff ratio) of 1.2x103, a retention time of 105 s, an endurance of 5×102 erase-and-program cycles, and multi-level cell (MLC) operation, being a tera-bit nonvolatile memory cell. Acknowledgement *This research was supported by “The National Research Program for Terabit Nonvolatile Memory Development” sponsored by the Korean Ministry of Knowledge Economy
12:30 PM - H7.10
Effect of Tunneling Barrier Capping Au Nanocrystals on Nonvolatile Memory Characteristics for Polymer Memory-cells
Hyun Min Seung 2 3 , Jong-Dae Lee 1 3 , Kyoung-Cheol Kwon 1 3 , Chang-Hwan Kim 1 3 , Jea-gun Park 1 2 3
2 Nanoscale Semiconductor Engineering, Hanyang Univ., Seoul Korea (the Republic of), 3 , Tera-bit Nonvolatile Memory Development Center, Seoul Korea (the Republic of), 1 Electrical & Computer Engineering, Hanyang Univ., Seoul Korea (the Republic of)
Show AbstractWe investigated the effect of tunneling barrier capping Au nanocrystals on nonvolatile memory characteristics such as memory margin, retention time, and endurance cycles of program and erase for polymer memory-cells. The 4F2 memory-cells were fabricated with the device structure of conductive polymer (PVK : Poly-n-vinyl carbazole) embedded with Au nanocrystals capped by tunneling barrier between upper and bottom electrodes. In particular, The Au nanocrystals capped by various tunneling barriers were produced by curing at 300°C for 2 hrs after the deposition of laminated layer of bottom PVK, tunneling barrier, Au middle layer, tunneling barrier, upper PVK. In the case of a polymer memory-cell embedded with Au nanocrystals without capping tunneling barrier, they showed bad nonvolatile memory characteristics such as the memory margin of (Ion / Ioff) of 1.9×102, the retention time of less than ~1×105 sec, and less than 10 endurance cycles of program and erase. Otherwise, in the case of a polymer memory-cell embedded with Au nanocrystals with capping tunneling barrier, they showed better nonvolatile memory characteristics such as the memory margin of (Ion / Ioff) of 7.2×102, the retention time of more than ~1×105 sec, and more than 100 endurance cycles of program and erase. In our study, we will review the mechanism by which the tunneling barrier capping Au nanocrystals embedded in conductive polymer improves nonvolatile memory characteristics and present the effect of the tunneling material properties (e.g., Al2O3, TiO2 and SiO2) on nonvolatile memory characteristics.Acknowledgement* This research was supported by "The National Research Program for Terabit Nonvolatile Memory Development” sponsored by the Korean Ministry of Knowledge Economy.
Symposium Organizers
Yoshihisa Fujisaki Hitachi Ltd.
Rainer Waser RWTH Aachen University
Tingkai Li Sharp Laboratories of America Inc.
Caroline Bonafos CEMES/CNRS
H10: Resistive Switching RAM II
Session Chairs
Thursday PM, April 16, 2009
Room 2007 (Moscone West)
2:30 PM - **H10.1
Research Progress in the Resistance Switching of Transition Metal Oxides for RRAM Application: Thin Film Growth, Electrode Materials, Switching Mechanism and Properties Optimization.
Xiaomin Li 1 , Weidong Yu 1 , Lidong Chen 1
1 , Shanghai Institute of Ceramics, ,Shanghai China
Show AbstractElectrical controlled resistance switching (ECRS) based on transition metal (TM) oxides, such as perovskite manganites (Pa1-xCaxMnO3, La1-xCaxMnO3) and titanate (SrTiO3), binary oxides (NiO, TiO2 and CoO) etc, have attracted great interest for potential applications in next generation nonvolatile memory known as resistance random access memory (RRAM). Compared with other nonvolatile memories, RRAM has several advantages, such as fast erasing times, high storage densities, and low operating consumption. Up to date, the switching mechanism, property improvement and new materials exploitation are still the hotspots in RRAM research. In this report, we will summarize the main results from the resistance switching investigation in Shanghai Institute of Ceramics. The TM oxides including La1-xCaxMnO3, SrTiO3 (STO), and TiO2 were deposited on Pt/Ti/SiO2/Si substrates by pulse laser deposition. Based on the materials analysis, the influences of the growth parameters on the ECRS properties were studied detailedly. In view of the importance of the top electrodes (TEs) on the ECRS, we have prepared the TE materials (Ag, In, Al, Ti,W and Ag-Al alloy) on La1-xCaxMnO3 films. The effects of the TEs on the interface characteristics between TEs and La1-xCaxMnO3 were examined by considering the physical (work function) and chemical (oxidizability) nature of these metals. Among them, Ag-Al alloy shows the combined advantages of Ag and Al electrodes to enhance the stability and reproducibility of the ECRS. Especially, it can induce the ECRS with relative low original resistances, which can improve the uniformity of the ECRS in large-area production. For the La1-xCaxMnO3 thin film, we believe the ECRS mechanism is related to the modulation of the space charge limitation conductivity at the interface between TE and the La1-xCaxMnO3 film. However, for STO thin films, we think the modification of the Schottky barrier in grain boundaries by oxygen vacancies migration is the main incentive to realize ECRS in nanocrystalline STO films. And interface controlled filamentary mechanism was proven to be the dominant explanation for ECRS in nanocrystalline TiO2 thin films. By inspecting the influences of the measurement parameters on the resistances at different resistance states, it was found that the resistance maintains a stable value at the saturation state. In addition, it was confirmed that the I-V sweeps in vacuum environment, asymmetry pulse pairs and oxygen annealing of films can improve the endurance of the ECRS property. Our researches will provide some meaningful clues to understanding the ECRS mechanism and some useful pathways to promote the development of RRAM devices.
3:00 PM - H10.2
Improved Electrical Resistive Switching of MnOx Based Nonvolatile Memory Devices.
Min Kyu Yang 1 2 , Tae Kuk Ko 2 , Jeon Kook Lee 1
1 Materials Science and Technology Research Division, Korea Institute of Science and Technology, Seoul Korea (the Republic of), 2 Electrical and Electronic Engineering, Yonsei University, Seoul Korea (the Republic of)
Show AbstractMany metal-insulator-metal systems show electrically induced resistive switching effects and have been proposed at the future non-volatile memories. 100-nm-thick MnOx thin films with polycrystalline structure were fabricated on Pt/Ti/SiO2/Si single crystal substrates at 20oC by radio-frequency sputtering. MnOx exhibits irreversible memory switching from a low resistance state(LRS; on state) to a high resistance state(HRS; off state), with the off to on resistance ratio of greater than 100.To improve the device yield and the process-voltages-distribution, the titanium top electrode and the NiOx buffer layers are used. From room temperature current-voltage measurements of these structures, reproducible resistive switching behaviors were observed. More than 105 repetitive switching cycles were demonstrated with a high current on/off ratio. The program voltage and erase voltages, Vp and Ve , are below 2V, respectively. I(ON)/I(OFF) is higher than 100. Distribution of voltage, V[set] (Δ/σ) and V[reset] (Δ/σ), are 11 and 8.4, respectively. Distribution of resistance, R[on](Δ/σ) and R[off](Δ/σ) are 13 and 304, respectively. Pulse reset switching time is lower than 20 ns. The dominant conduction mechanisms will also be discussed.
3:15 PM - H10.3
Programmable Resistance Switching in Amorphous Silicon Switches and Crossbar Arrays
Sung Hyun Jo 1 , Kuk-Hwan Kim 1 , Ting Chang 1 , Wei Lu 1
1 , University of Michigan, Ann Arbor, Michigan, United States
Show AbstractWe report our studies on a nanoscale Si-based two-terminal resistance switching device and prototype high-density crossbar arrays. We show for the first time that switching characteristics in the nanoscale Si-based devices are dominated by the formation of a single filament and can lead to several novel operational modes of the device. As a prime example, we clearly demonstrate that the resistance switching is probabilistic in nature and can be controlled by adjusting both the amplitude and the length of the applied programming pulse and, more importantly, the switching time can be reduced exponentially by increasing the bias linearly. In addition, we show that both digital and analog switching can be achieved, and how controlling the filament formation can lead to multi-level resistance values in the logarithm scale. The probabilistic resistance switching nature along with the bias- and time-depend switching characteristics opens new opportunities for the device not only as a digital switch (i.e. single bit memory elements) but also as a reconfigurable interconnect in logic circuits (i.e. synapses in neuromorphic circuits). In addition, these findings likely will be applicable to other resistive random-access memories (RRAMs) currently under research following aggressive device scaling. The filament formation mechanism will also be discussed and our calculations show that the switching is dominated by the region ~ 10 nm close to the bottom electrode. As memory devices, the a-Si based switches exhibit excellent performance metrics in terms of switching speed (< 50ns), endurance (> 1E5), on/off ratio (> 1E7), retention (> 7 years) and multi-bits storage (8 states, 3 bits). Finally, we show that a high-density (2 Gbits/cm2) 1kb crossbar memory array can be built based on the nanoscale Si two-terminal switches with excellent uniformity, yield (> 92%) and ON/OFF ratio.
3:30 PM - H10.4
Predictable RESET Switching in Oxide-Based Unipolar RRAM and Its Application for Error Correction Technique to Enhance Reliability.
Shin Buhm Lee 1 , Seung Chul Chae 1 , Seo Hyoung Chang 1 , Sunae Seo 2 , Tae Won Noh 1
1 ReCOE & FPRD, Department of Physics and Astronomy, Seoul National University, Seoul Korea (the Republic of), 2 , Samsung Advanced Insititute of Technology, Suwon Korea (the Republic of)
Show AbstractRecently, the unipolar resistance switching has been investigated intensively because it has several advantages for nonvolatile memory applications such as good scalability, multi-staking, and compatibility with conventional CMOS process technology. Unfortunately, unipolar resistance switching has some critical obstacles for applications, such as wide distribution of the switching voltages, which makes it difficult to fabricate reliable RRAM. However, we have little physical understandings on the switching voltage distributions. Here, we will report on a scaling behavior of the RESET voltage and its understandings based on the percolation theory. In addition, we will show that this behavior can be applied to reduce error by avoiding overlaps of SET and RESET voltages. We fabricated Pt/NiO/Pt capacitors, whose typical current-voltage (I-V) curves shows unipolar resistance switchings. By obtaining more than 250 I-V curves, we showed that there existed wide distributions of switching voltages during successive resistance switchings. Recently, based on the random circuit breaker network model, we explained the RESET and SET processes of the unipolar switching in terms of rupturing and forming of conducting filaments, respectively [2]. We also showed that the RESET process was driven by the Joule heating effects [3,4], whose roles were probed using third harmonic generation [5,6]. In this work, we showed that the Joule heating effects could be also probed in the nonlinearity of the I-V curves; namely, all I-V curves of the low resistance states (LRS) can be well fitted by R = Ro+BoI2, where Ro and Bo are Ohmic resistance and third harmonic generation, respectively. We found that the RESET currents IR can be scaled with Bo, i.e., IR ∝ Bo-x, whose behavior is consistent with the nonlinear behavior of the classical percolation theory. Based on the scaling behavior, we can predict the possible RESET voltage value for a given LRS before actually performing RESET process. We will show that the predicted RESET voltages VR (≡ 2IRRo) are nearly same to the experimental values. Using this prediction, we developed error correction technique, which will solve the problem due to overlaps of SET and RESET voltages, which are currently the most serious obstacles for the RRAM applications. Note that this technique is a nondestructive tool, which need to know only VR value before performing RESET process. By measuring a few experimental points in a I-V curve, we can obtain such VR value, so our technique is quite easy to apply for real RRAM applications. This work demonstrates the importance of understanding physics of the unipolar resistance switching phenomena.[1]R. Waser et al., Nat. Mater. 6, 833 (2007).[2] S. C. Chae et al., Adv. Mater. 20, 1154 (2008).[3]S. H. Chang et al., Appl. Phys. Lett. 92, 183507 (2008).[4]S. H. Chang et al., arXiv:0803.4258.[5]S. B. Lee et al., arXiv:0810.0886.[6] S. B. Lee et al., arXiv:0810.4043.
3:45 PM - H10.5
Stackable All Oxide Based Nonvolatile Memory with Al2O3 Antifuse and p-CuOx/n-InZnOx Diode.
Seung-Eon Ahn 1 , Bo Soo Kang 1 , Ki Hwan Kim 1 , Myoung-Jae Lee 1 , Chang Bum Lee 1 , Changjung Kim 1 , Youngsoo Park 1
1 , Samsung Advanced Institute of Technology, Suwon Korea (the Republic of)
Show AbstractWe developed all oxide based nonvolatile memory for low cost, high density and high performance one-time field programmable (OTP) memories compared with Si based antifuse memory using antifuse technologies over glass substrate. The oxide OTP memory employed the p-n CuO/InZnOx diode for switching element of memory cell and Al2O3 for antifuse as storage node of memory cell. The memory cell is programmed from break down of Al2O3 by applying a program voltage bias which is about 4.5 V. The OTP memory cells show large on/off ratio about 10^6 and small current distributions at programmed and unprogrammed states resulting from perfect uniformity of Al2O3 thin film before and after breakdown. It also showed a fast programming speed of about 20 nanoseconds, excellent uniformity and retention characteristic.
4:30 PM - **H10.6
Nonlinear Thermal Effects in Unipolar Resistance Switching and Their Explanations Based on The Random Circuit Breaker Network Model.
Tae Won Noh 1 , Seung Chul Chae 1 , Seo Hyoung Chang 1 , Shin Buhm Lee 1 , Jae Sung Lee 2 , Byoungnam Kahng 2
1 ReCOE & FPRD, Department of Physics and Astronomy, Seoul National University, Seoul Korea (the Republic of), 2 Department of Physics and Astronomy, Seoul National University, Seoul Korea (the Republic of)
Show AbstractRecently, there is a surge of renewed interests on resistance switching (RS) phenomena due to their possible applications in nonvolatile memory device, called the resistance random access memory (RRAM) [1]. Especially, RRAM based on the unipolar RS is a good candidate for multi-stacked, high density, and nonvolatile memory. Despite these advantages, it is still at a “proof-of-concept” stage and we are facing many difficulties due to poor physical understandings on related issues. Especially, unipolar RRAM has very large fluctuations in SET and RESET voltages, which poses a serious obstacle for practical applications of unipolar RRAM [2].We have investigated unipolar RS behaviors in polycrystalline TiO2, NiO and FeOx thin films. Using conducting atomic force microscope, we observed the existence of the formation and rupture of local conducting filaments during the SET and RESET processes, respectively [3]. To explain such collective behaviors, we recently proposed the random circuit breaker (RCB) network model, which approximates the switching medium as a network of “circuit breakers” with two switchable metastable states. This model could successfully explain the collective behaviors of the conducting filaments during the SET and RESET processes. Later, we also found experimentally that the rupture of conducting filaments should be closely related to the thermal effects [4,5]. We modified the RCB network model to take into account of the thermal effects, especially in the low resistance state (LRS). This upgraded model can explain many nonlinear effects, including the LRS stability [5], the third harmonic generation [6], nonlinear current-voltage curves [6], and 1/f noises. Very recently, we found a new intriguing phenomenon that switching voltages and currents could be scaled with the resistance value in the LRS [7]. These intriguing scaling behaviors in the unipolar RS are quite similar to those of the classical percolating system. In this talk, we will briefly review the nonlinear effects of the RS, especially coupled with the local Joule heating effects. We will explain how our newly proposed percolation model can explain such nonlinear thermal effects. We will show that the RCB network model can provide us insights on occurrence of unipolar RS, its reversible dynamic process, and collective motions of conducting filaments. In addition, we will discuss how to overcome the substantial distribution of switching voltages [3], which is currently considered the most serious obstacle to practical unipolar RRAM applications [2].[1]R. Waser et al., Nat. Mater. 6, 833 (2007).[2]G. I. Meijer et al., Science 319, 1625 (2008).[3] S. C. Chae et al., Adv. Mater. 20, 1154 (2008).[4]S. H. Chang et al., Appl. Phys. Lett. 92, 183507 (2008).[5]S. H. Chang et al., arXiv:0803.4258.[6]S. B. Lee et al., arXiv:0810.0886.[7] S. B. Lee et al., arXiv:0810.4043.
5:00 PM - H10.7
Flexible Solution-Processed TiO2-Based Memory.
Nadine Gergel-Hackett 1 , Behrang Hamadani 1 , Barabara Dunlap 1 , John Suehle 1 , Curt Richter 1 , Christina Hacker 1 , Dave Gundlach 1
1 Semiconductor Electronics Division, EEEL, National Institute of Standards and Technology, Gaithersburg, Maryland, United States
Show Abstract The field of flexible electronics has made headway in the recent years with the development of flexible organic thin film transistors for use in computational logic; however, there exists a great need in the field for a flexible device with memory applications. We have demonstrated a flexible, solution-processed, nonvolatile, low power, inexpensive, TiO2-based flexible memory component. The electrical behavior of this component is consistent with a memristor, an electrical device that has recently been recently touted as the missing fourth circuit element. Our flexible memory device has operation voltage of less than 10 V, on/off ratios greater than 10,000:1, exhibits memory potential that is nonvolatile for over 1.2x106 s, and is operational after 4,000 flexes. This technology has potential advantages over existing flexible memory devices including low power operation, rewritability, and a simple two-terminal room temperature processed device design. In contrast to existing methods for depositing the TiO2 films for rigid memory devices, we performed the deposition of TiO2 through a simple, inexpensive, room-temperature, spin-on sol gel process. The device structure consists of a TiO2-layer that is spun from a solution onto a metal contact on a flexible plastic bottom substrate. The active TiO2 layer is then hydrolyzed and is capped with a top metal electrode to form a simple crossbar structure. Spectroscopic ellipsometry and x-ray photoelectron spectroscopy of the active layer are performed, and the sol gel-based active layer has the expected ratio between the TiO2-associated O 1s peak and Ti 2p peak of 2:1. Electrical characterization of the device demonstrates electrical switching with memory behavior that is consistent with that reported from a memristor device. The device’s resistance switches from a high resistance state to a low resistance state upon the application of an adequate bias (write). This low resistance state holds while low bias is applied (read), and only switches back to a high resistance state with the application of an adequate opposite bias (erase). The device can be operated in a multistate “soft” switching regime or a binary “hard” switching regime, depending on the magnitude of bias applied. Memory potential is also demonstrated through the application of write, read, and erase pulses, which result in the expected output current states. Even after the device is flexed over 4,000 times in a flexing apparatus, it is still operational and demonstrates an adequate on/off ratio. Overall, 50 devices were fabricated over 7 different experimental runs, of which 42 devices exhibited electrical behavior indicative of electrical switching (overall yield of 84%). The advantages in the fabrication and operation of this device make it a promising candidate for filling the large void in viable memory within the flexible electronics community.
5:15 PM - H10.8
Engineering of Electrode Materials for NiO Resistive Switching non Volatile Memories.
Sabina Spiga 1 , Alessio Lamperti 1 , Elena Cianci 1 , Graziella Tallarida 1 , Flavio Volpe 1 , Marco Fanciulli 1 2 , Antoine Demolliens 3 , Christian Turquat 3 , Christophe Muller 4 , Ugo Russo 5 , Carlo Cagli 5 , Daniele Ielmini 5
1 Laboratorio Nazionale MDM, CNR-INFM, Agrate Brianza Italy, 2 Dipartimento di Scienza dei Materiali, Università di MIlano Bicocca, Milano Italy, 3 IM2NP, Université du Sud Toulon Var, La Garde Cedex France, 4 IM2NP, Polytech’Marseille, Université de Provence, Marseille France, 5 DEI, Politecnico di Milano, Milano Italy
Show AbstractResistive switching random access non-volatile memories (ReRAM), based on two terminal electrodes that sandwich a resistive change material, could represent the leading alternative to floating gate technology for post 32 nm technology nodes [1]. Among the currently investigated concepts and materials for ReRAM, transition metal binary oxides are receiving increasing interest. Indeed, they offer high potential scalability, thermal stability, and can be integrated in CMOS device fabrication easier than organic materials and ternary/quaternary compounds. Typically, for oxide-based ReRAM devices, an initial electro-forming step is required to induce current-limited electric breakdown in the fresh samples. Then, the memory element can be reversibly switched between a low resistance set state and a high resistance reset state. Reversible switching is achieved as a conductive filament (CF) is repeatedly formed and dissolved inside the dielectric. Although several oxides have shown bipolar and/or unipolar switching, NiO is currently the most promising resistive material generally exhibiting unipolar switching, as required for cell integration in a cross-point memory array for high-density storage [2]. However, NiO resistive switching has been mainly reported in combination with noble metal electrodes, while for its real implementation in memory cell it is also necessary to qualify proper electrodes compatible with the standard CMOS processes, such as Si, TiN, and W.In this work we investigate the switching properties of polycrystalline NiO films (15-160 nm thick) grown by atomic layer deposition and electron beam evaporation on Pt, n-Si, TiN and W bottom electrodes. The structural/chemical properties of the oxides, electrodes and oxide/electrode interfaces are investigated using various techniques. Metal/NiO/Metal structures have been therefore fabricated to test the switching behavior. NiO layers are initially in the high resistance state, independently of electrode combination, and exhibit reproducible unipolar switching after an appropriate forming stage. The resistive switching parameters are investigated as a function of electrodes, film thickness and film structural/chemical properties. Programming voltage and current are shown to depend on the electrode combination, suggesting that intrusions from the electrodes are included in the CF. Experimental evidences are explained on the basis of an electro-thermal model and related to different CF material parameters, which result in different dissolution kinetics and CF breakage, thus affecting electrical programming parameters. The NiO switching properties as a function of deposition techniques are also discussed and possible future integration schemes are proposed.This work was supported by the European project EMMA “Emerging Materials for Mass storage Architectures” (FP6-033751).[1] R. Waser and M. Aono, Nature Materials 6, 833 (2007)[2] M.-J. Lee et al., Adv. Mater. 19, 73-76 (2007)
5:30 PM - H10.9
Oxidation Behavior of Ni Thin Films: Application to NiO-based ReRAM.
Judit Lisoni 1 , Ludovic Goux 1 , Eveline Verleysen 1 , Xin Peng Wang 1 , Nico Jossart 1 , Malgorzata Jurczak 1 , Dirk Wouters 1
1 , IMEC, Heverlee (Leuven) Belgium
Show AbstractIn this paper we investigated the oxidation behavior of thin (20-100 nm) Ni films. The obtained NiOx films were incorporated as the active memory element in Ti\TiN\Ni\NiO\Ni ReRAM structures. The main parameter under investigation was the influence of the substrate in the Ni microstructure (crystallinity and topography) and therefore in the subsequent oxidation behavior of the Ni films. Finally, we correlated the NiOx formation parameters with the (unipolar) switching characteristics of the NiOx films. Further understanding is obtained by a parallel oxidation study on single crystal Ni substrates and the electrical evaluation of the single crystals Ni\NiO\Ni ReRAM elements.Ni films deposited by sputtering were oxidized using ex-situ thermal anneal carried out at 400-550 °C in oxygen atmospheres. The influence of the substrate was investigated by using different kind of TiN films obtained via physical (sputtering) and chemical (ALD, MOCVD) deposition techniques. We found that Ti\TiN stacks that promote the formation of fine Ni grains, i.e. 20-50nm, result in a large Ni oxide growth rate. In this case switching is obtained only if very high compliance currents of 100 μA are used for programming. Alternatively, for those Ti\TiN stacks that promote the formation of large grains (>100 nm) we did observe a much smaller NiOx growth rate. However, for different Ti\TiN stacks that all resulted in large Ni grains, different NiOx switching characteristics were obtained: some samples showed reproducible switching events at low switching voltages (0.8-1 V) while others showed non-reproducible switching events with small memory window and higher switching voltages (1.3 V). While the NiOx microstructure did not show major difference among these samples, these differences in switching correlated well with the presence of Ni(200) crystalline orientation in the initial Ni film, which seems to be enhanced by Ti-rich TiN films. A better understanding of this observation was achieved by the use of Ni single crystals. We observed that the oxidation rate clearly depends on the Ni crystalline orientation: NiOx grows faster in Ni(100) as compared to Ni(111) single crystals. Electrical evaluations of the single crystals showed that the stability of the switching seems to be related to the thickness of the NiOx film formed: a non-stable switching (with difficult reset) is obtained in Ni(100) single crystals, which simultaneously showed the highest initial resistance values indicative of a thick NiOx; the initial resistance of NiOx grown on Ni(111) single crystals was approximately six order of magnitude lower as compared to the NiOx obtained in Ni(100) single crystals, 3x104 Ω vs. 1x1011 Ω, respectively. However, as the NiOx thicknesses in the polycrystalline thin film samples are all in the same range, Ni:O composition may also play an important role as revealed by ERD, XPS and TEM/EELS/ELNES analysis.
H12: Poster Session: Phase Change RAM I
Session Chairs
Yoshihisa Fujisaki
Claudia Wiemer
Friday AM, April 17, 2009
Salon Level (Marriott)
9:00 PM - H12.10
Switching Power Reduction in Phase Change Memory Cell using CVD Ge2Sb2Te5 and Ultra-thin TiO2 Films.
Byung Joon Choi 1 , Seol Choi 1 , Taeyong Eom 1 , Cheol Seong Hwang 1 , Suk Kyoung Hong 2
1 Materials science & engineering, Seoul National University, Seoul Korea (the Republic of), 2 R & D division, Hynix Semicon. Inc., Ichon Korea (the Republic of)
Show AbstractPhase change random access memory (PCRAM) has attracted considerable interest recently for highly integrated non-volatile memory devices. However, the high level of reset current needed to switch the GST material from a crystalline to amorphous state has been the major obstacle to the further scaling of PCRAM. Adopting a thermal barrier layer appears to be an effective method for reducing Ireset although other methods also showed some improvements. It was suggested that the oxide layer with a lower thermal conductivity suppresses the heat loss to the metal plug and improve heating efficiency. The authors recently reported the growth of GST thin films by a combined plasma-enhanced chemical vapor deposition (PECVD) and plasma-enhanced atomic layer deposition (PEALD). The PECVD/PEALD process should be the preferred deposition method for the GST material for use in highly integrated devices (> 1 Gb). However, in contrast to sputtering, this chemical process showed a substrate surface dependent growth behavior. The substrate dependent growth behavior is a crucial demerit for fabricating conventional T-shaped PCRAM cells because the GST thin films should be homogeneously and evenly deposited on both the metallic contact plug surface and inter-layer dielectric, which is usually SiO2, where the metal plugs are embedded. Therefore, in this study, a 2, 4 and 8 nm-thick TiO2 layer grown by plasma-enhanced ALD were interposed between the PECVD/PEALD GST and underlying plug/ILD layers. Understanding the influences of the interposed TiO2 layer on the improved electrical performance of these PCRAM cells was the major objective of this study. In addition, improvements of the nucleation for the GST deposition, adhesion property by intervening TiO2 thin film were also investigated. The test cell structures were fabricated on oxidized Si substrates. The 2 - 8 nm-thick TiO2 layers were deposited by PEALD at 250oC using Ti(OC3H7)4 precursor and plasma-activated O2 reactant on the ILD which embedded the W/TiN plug (300 nm in diameter). The 100 nm-thick GST films were deposited by PECVD at a wafer temperature of 200oC on the TiO2 layer. Ge(i-C4H9)4, Sb(i-C3H7)3, and Te(i-C3H7)2 were used as the Ge, Sb and Te precursors, respectively. A TiN top electrode was deposited by sputtering and lithographically patterned (50 x 50μm2 in area for the top electrode probing). DC current–voltage characteristics were determined using a semiconductor parameter analyzer. The dynamic reset/set characteristics of the fabricated cells were tested using the pulse measurement system. It was concluded that interposing an ultra thin ALD TiO2 layer between the metal contact plug and PECVD GST can improve GST growth, which can in turn improve the manufacturability of the device, as well as improve the adhesion property. It also improved the device performances, such as reducing the reset current and the reset power mainly by the thermal insulating effect.
9:00 PM - H12.11
Influence of The Stoechiometry of GeTe Thin Films on Their Physical Properties for PCRAM Applications
Emmanuel Gourvest 1 4 , Sandrine Lhostis 1 , Jens Kreisel 2 , Marilyn Armand 3 , Bernard Pelissier 4 , Cyril Dressler 3 , Frederic Fillot 3 , Patrice Gergaud 3 , Christophe Vallee 4
1 , STMicroelectronics, Crolles France, 4 CNRS-LTM, CEA/LETI, Grenoble France, 2 LMGP, INPGrenoble Minatec, Grenoble France, 3 DRT, CEA/LETI, Grenoble France
Show AbstractChalcogenide materials were shown to be interesting for Phase Change Random Access Memories (PCRAM) applications thanks to their high contrast of electrical resistivity between their crystalline and amorphous phase [1]. For embedded memories, other device performances are required such as high retention time at working temperature. This especially implies high crystallization temperature [2]. Binary compounds GeTe have attracted great attention to produce PCRAM since they present high phase transition temperatures and a good contrast in terms of electrical resistivity between the two crystalline states [3]. In this study, the influence of GexTe1-x stoechiometry on the physical properties of the film is investigated. 30 nm amorphous GeTe thin films with different compositions were grown by co-sputtering PVD method on 200 mm wafers applying different powers on the Ge and Te targets. Film thickness and composition are characterized using Rutherford Back Scattering (RBS), Spectroscopic Ellipsometry, X-Ray Reflectivity and X-Ray Photoelectron Spectroscopy. The evolution of the phase transition temperature is reported using laser beam reflectivity. The results show an influence of the Germanium ratio in the film on the crystallization temperature. The oxidation rate of the film surface also depends on the film composition. However this oxidized layer is composed essentially on GeO2 whatever the Germanium ratio in the samples. In order to follow the changes in the crystallization state and chemical bonding, nucleation and crystals growth being known to be a kinetic phenomenon, GexTe1-x thin films were submitted to various thermal budget and heat up ramps. A correlation between the results of in situ annealing characterizations set-up that are Ellipsometry, Raman Spectroscopy and X-Ray Diffraction will be presented. The results will be discussed with respect to the evolution of crystallization temperature.[1] M. Welnic and M. Wuttig, Materials Today, vol. 11, n° 6, (2008).[2] M. Chen, K.A. Rubin and R.W. Barton, App. Phys. Lett., vol. 9, 502-504, (1986).[3] M.M. Abdel-Aziz, App. Surf. Science, vol. 253, 2059-2065, (2006).
9:00 PM - H12.12
Multiple Phase-transition in Ge2Sb2Te5 Films During the Cubic to Hexagonal Phase Change.
Riccardo De Bastiani 1 2 , Corrado Bongiorno 3 , Egidio Carria 1 2 , Giuseppe Nicotra 3 , Maria Grazia Grimaldi 1 2 , Emanuele Rimini 2 3
1 , MATIS CNR-INFM , Catania Italy, 2 Dipartimento di Fisica ed Astronomia, Università di Catania, Catania Italy, 3 , IMM-CNR, Catania Italy
Show AbstractGe2Sb2Te5 (GST) is a technologically very important phase-change material because it can be switched rapidly between amorphous and crystalline states for millions of cycles by appropriate heating. Bulk GST and relatively thick films are known to show an amorphous to face-centered cubic (fcc) phase transition at around 150 °C. With increasing temperature the GST films undergo, at about 350 °C, a structural change into the stable hexagonal close-packed (hcp) crystalline state. However, at present, the structural changes behind the cubic to hexagonal phase transition are unknown yet. In this paper, we investigated the cubic to hexagonal multiple phase-transition phenomena of Ge2Sb2Te5 films, 70 nm thick, deposited on a SiO2/Si substrate by RF magnetron sputtering. The optical and electrical properties were analyzed during isothermal treatments in the temperature range 150 - 450 °C by means of in situ time resolved reflectivity (TRR) and four-point probe method respectively. Rutherford backscattering spectrometry (RBS) spectra of the annealed films were identical within experimental error, which showed that the composition was practically unchanged for the used temperature range. The optical and electrical analyses have been related to structural characterisation by X-ray diffraction and TEM, in order to highlight the progressive changes occurring during the phase transition. Experimental measurements revealed that the phase change of GST films proceeds from the metastable fcc to the stable hcp phase through the generation of different crystalline hexagonal structures characterized by intermediate reflectivity and resistance states. The crystal structure of the hexagonal phase depends dramatically on the heating process. Films annealed to 350 °C show clearly the hcp phase but interestingly, increasing the annealing temperature leads to the appearance of new peaks and to a shift in the peak positions of the X-ray diffraction pattern. Finally a strong increase in the (005) peak intensity of the indexed hexagonal structure was observed in the last stage of the transformation. High-resolution transmission electron microscopy (HRTEM) study on the different hexagonal states allowed to determine the atomic arrangements of the hexagonal structured GST and clarify the nature of the phase transition process.
9:00 PM - H12.13
Optical Phase-Change Experiments on and Characterization of Novel Sample Geometries of the Ge2Sb2Te5 Alloy.
David Baker 1 , P. Taylor 1 , Kris Campbell 2
1 Physics, Colorado School of Mines, Golden, Colorado, United States, 2 Physics, Boise State University, Boise, Idaho, United States
Show AbstractThe ternary alloy Ge2Sb2Te5 and similar phase-change materials stand at the forefront of many investigations in the scientific community. Current memory technology such as DVD-RAM and Phase Change RAM (PCRAM) exploits the switching characteristics, both optical and electronic, of this family of materials. In spite of these engineering successes, debate still occurs regarding both the mechanism of switching and the local order in the amorphous phase in the device structure. Examinations of the latter via Extended X-ray Absorption Fine Structure yield conflicting results, as the amorphous phase may have multiple regions of varying morphology. Whilst previous optical switching experiments yield micro- and nano-crystalline regions surrounding the area affected by the laser1, recent laser crystallization and re-amorphization experiments carried out on samples of Ge2Sb2Te5 employ a novel geometry which allows for the examination of the material properties of a fully-switched amorphous state. In concert with these phase-change experiments, results and analysis from Electron Spin Resonance, Infrared, and Photoluminescence spectroscopies provide a comprehensive picture of both the optically crystallized and re-amorphized phases of this important material.1 J. Appl. Phys. 69, 2849 (1991)
9:00 PM - H12.2
Crystallization-induced Stress in Thin Phase Change Films of Different Thicknesses.
Qiang Guo 1 2 , Minghua Li 3 , Yi Li 1 2 , Luping Shi 3 , Tow Chong Chong 3 , Johannes Kalb 1 4 , Carl Thompson 1 4
1 Advanced Materials for Micro-/Nano- Systems, Singapore-MIT Alliance, Singapore Singapore, 2 Department of Materials Science and Engineering, National University of Singapore, Singapore Singapore, 3 , Data Storage Institute, Singapore Singapore, 4 Department of Materials Science and Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts, United States
Show AbstractPhase change materials have been extensively used for optical data storage in commercial rewritable compacts disks (CDs) and digital video disks (DVDs). These materials are also widely considered for next-generation phase change random access memories to replace current Flash memories. In this work, we have studied crystallization-induced stress in phase change films (Ge2Sb2Te5) as a function of thickness and with and without a capping layer, by measuring the deflection of micro-cantilevers. The stress is found to increase with decreasing film thickness. A thin dielectric capping layer leads to a further increase in stress compared to uncapped films. This observation can be explained by the suppression of stress relaxation in the phase change film in the presence of a capping layer. High stress will affect device performance as the size of phase change memory cells decreases.
9:00 PM - H12.3
Effect of Nitrogen Doping and Working Pressure on Crystallization of 1Sb4Te7 Thin Films for PRAM Application.
Hyung Keun Kim 1 , Seung Yoon Lee 1 , Sangwoo Shin 2 , Hyung Hee Cho 2 , Doo-Jin Choi 1
1 Department of Materials Science and Engineering, Yonsei University , Seoul Korea (the Republic of), 2 Department of Mechanical Engineering, Yonsei University, Seoul Korea (the Republic of)
Show AbstractRecently, PRAM (Phase-change RRAM) is attracting its attention among the next generation non-volatile memories due to its scalability and simple cell structure. The On/Off states are distinguished by the electrical resistivity of phase change film. The On state, which is low resistivity crystalline state is gained by maintaining the temperature upper crystallization temperature for relatively long time, and the Off state, which is high resistivity amorphous state is gained by melt-quenching process. The energy to raise temperature is from electrical Joule heating which is presented by Q=I2R, and there needs to increase the resistivity value of crystalline state to reduce the RESET operation power. In this study, Ge1Sb4Te7 thin films were deposited by DC magnetron sputtering method with various nitrogen doping contents at various working pressures.By measuring the sheet resistance values of thin films, we observed that in the case of nitrogen doping, values of sheet resistance measured at crystalline state were higher than that of un-doped film. Therefore, we concluded that the nitrogen doping could be the one of the solutions to increase the resistivity of crystalline state. And with depositing films at higher working pressure, the behavior of sheet resistance reducing was seemed stronger at the temperature range of resistance reducing. As increasing the working pressure, the thickness reduction ratios of the films after annealing were measured higher. It might be because of large ‘gap’ volume due to weak compressive stress, which is from weak ‘atomic-peening effect’. At the higher working pressure, scattering between the target atoms and neutral gas molecules in the chamber is very frequent, so the target atom loses its energy more. Therefore, when the target atoms arrive to the substrate, because of their weak energy, weak ’atomic-peening effect’ occurs at the films. And we measured thermal conductivities by 3-ω method which were thoroughly agreeable with sheet resistance results. Independent of nitrogen doping contents, the thermal conductivity values of the amorphous films were similar. And after annealing, the values of thermal conductivity were increased by crystallization. However, when the nitrogen was doped in films, the value of thermal conductivity was lower than that of un-doped crystalline film. By analyzing XRD (X-Ray Diffraction) data, we observed that the thin films undergo crystallization process with annealing. And by increasing nitrogen doping contents, we observed that the crystallization temperature increases compared with un-doped films. In summary, we investigated the effect of nitrogen doping and working pressure on crystallization process of Ge1Sb4Te7 phase change thin films. It was observed that the crystallization process was disturbed by nitrogen doping. And as increasing the working pressure, the behavior of resistance reducing seemed stronger at the range of temperature where resistance reduces.
9:00 PM - H12.5
Electrical Resistance and Structural Changes on Crystallizaiton Process of Amorphous Ge-Te Thin Films.
Yuta Saito 1 , Yuji Sutou 1 , Junichi Koike 1
1 Department of Materials and Science, Graduate School of Engineering, Tohoku University, Sendai, Miyagi, Japan
Show AbstractGeTe-Sb2Te3 pseudobinary compounds are attracting considerable attention as phase change materials for optical disk and phase change random access memory (PRAM). In these compounds, Ge2Sb2Te5 (GST) has been used for an optical disk memory such as DVD-RAM because the crystallization by laser beam heating is very fast (~20ns). Recently, the GST has been much considered as material for PRAM and, therefore, the electrical resistance change due to crystallization and the phase change by applying an electrical current have been widely investigated. On the other hand, although GeTe compound has been suggested as the phase change material for the optical disk by Chen et al in 1986, the study focusing on the phase change material for PRAM is limited. Since GeTe is known to show the phenomenon of electrical switching, this compound has a potential of PRAM. In this study, the electrical resistance and crystalline structural changes on crystallization process in Ge-Te thin films were investigated.Films of amorphous Ge100−xTex (x : 45-85) with 200 nm thickness were deposited by sputtering of GeTe alloy target or co-sputtering of GeTe and Te targets on SiO2/Si substrates. In-situ electrical resistance measurements during heating process of these films were performed by two point probe method in a heating rate of 2~45 K/min. X-ray diffraction (XRD) analysis was employed for the structural identification of thin films using X-ray diffractometer with Cu-Kα. Transmission electron microscope (TEM) analysis was carried out to investigate the microstructure and to identify crystalline structure. The compositions of these films were confirmed by energy dispersive X-ray spectroscopy (EDS) attached TEM. All as-deposited Ge-Te films were confirmed amorphous by XRD and TEM. From the in-situ electrical resistance measurements, it is found that resistance change with crystallization process depends on the composition and the stoichiometric GeTe compound shows abrupt electrical resistance change at around 460 K. The crystallization temperature of GeTe was higher than that of GST and resistance difference between the amorphous and the crystal was also larger. While the electrical resistance of GST film gradually decreased with increasing temperature after the crystallization at around 430 K, that of GeTe film showed small temperature dependence after crystallization. It was found by X-ray measurement observation that the amorphous GeTe compound film crystallized first into a cubic state, and then into a stable rhombohedral state by further heating. The crystallization kinetics of Ge-Te thin films will be also presented.
9:00 PM - H12.6
Energy Band Offset of Amorphous Ge2Sb2Te5 on Dielectric or Metal and the Impact of Nitrogen Incorporation on its Alignment.
Lina Wei-Wei Fang 1 , Zheng Zhang 2 , Ji-Sheng Pan 2 , Minghua Li 3 , Rong Zhao 3 , Luping Shi 3 , Tow-Chong Chong 3 , Yee-Chia Yeo 1
1 , National University of Singapore, Singapore Singapore, 2 , Institute of Materials Research and Engineering, A*STAR, Singapore Singapore, 3 , Data Storage Insititue, A*STAR, Singapore Singapore
Show AbstractChalcogenide Ge2Sb2Te5 material has generated significant interest for application in phase change random access memory (PCRAM). The ternary alloy possesses both high switching speed and stability, making it a material candidate of choice over less complex binary components i.e. GeTe and Sb2Te3. On the other hand, PCRAM devices employing nitrogen-doped Ge2Sb2Te5 have been demonstrated to display lower reset current and greater endurance characteristics than using undoped Ge2Sb2Te5. Several structural studies based on XRD have shown that nitrogen-doped Ge2Sb2Te5 contains smaller grains and higher crystallization temperature. In terms of electrical properties, film resistivity was found to increase with nitrogen content, leading to lower reset current. In this paper, we explore the energy band alignment of amorphous Ge2Sb2Te5, on a dielectric (SiO2) and metal material (tungsten), as these are common materials that are in contact with the Ge2Sb2Te5 film in a phase change memory cell. The band alignment was examined through x-ray photoelectron spectroscopy, employing the core-level spectra as well as the valence band spectra in the analysis. Band alignment of nitrogen-doped Ge2Sb2Te5 is similarly investigated. Nitrogen atomic concentration in the Ge2Sb2Te5 film was varied from 0 at. % to 8.4 at. % in this work. Dependence of the band offset on the nitrogen content is determined.
9:00 PM - H12.7
Deposition Study of Binary and Ternary Chalcogenide Layers by Conventional MOCVD in Nitrogen Atmosphere.
Massimo Longo 1 , Claudia Wiemer 1 , Olivier Salicio 1 , Marco Fanciulli 1 2
1 Laboratorio Nazionale MDM, CNR-INFM, Agrate Brianza Italy, 2 Dipartimento di Scienza dei Materiali, University of Milano Bicocca, Milano Italy
Show AbstractChalcogenide layers are very attractive for phase-change applications and, although promising results are reported by Metal Organic Chemical Vapor Deposition (MOCVD) [A. Abrutis et. al., Chem. Mater., 20, 3557-3559 (2008), G. S. Tompa et al., MRS Spring Meeting 2007, Symposium I, I10.8], still modified MOCVD reactors are needed to improve the metalorganic precursor decomposition and lower the growth temperature.In this work, GeTe, Ge-Sb and GeSbTe chalcogenide layers with different compositions were deposited on Si/SiO2 substrates by both conventional chemical vapor deposition (CVD) and by CVD mixed with alternated precursor pulse injection (CVD/ALD-like) mode. The deposition study was performed without any process activation and by the use of pure nitrogen; a 10 nm Ge nucleation layer was deposited prior of the chalcogenide growth. The aim of the study was to explore an alternative approach to improve the 2D nucleation mode, as well as the layer adhesion and morphology. Extensive field emission scanning electron microscopy observations (FEG-SEM) were performed, along with structural and compositional analysis, carried out by X-ray diffraction (XRD) and total reflection X-ray fluorescence analysis (TXRF). Rhombohedric (GeTe) and hcp (Ge2Sb2Te5) phases were achieved, whereas for Ge-Sb a mixing of the separate Ge and Sb phases was also observed. The mixed CVD/ALD-like deposition resulted more effective than the CVD mode in improving the surface morphology of the binary compounds.
9:00 PM - H12.8
Evolution of the Transrotational Structure During Crystallization of Amorphous Ge2Sb2Te5 Thin Films.
Emanuele Rimini 1 2 , Corrado Bongiorno 2 , Egidio Carria 1 3 , Riccardo De Bastiani 1 3 , Maria Grazia Grimaldi 1 3 , Giuseppe Nicotra 2 , Alberto Piro 1 3 , Corrado Spienella 2
1 Dipartimento di Fisica ed Astronomia, Università di Catania, Catania Italy, 2 , IMM-CNR, Catania Italy, 3 , MATIS CNR-INFM, Catania Italy
Show AbstractThe crystallization of 70 nm thick amorphous Ge2Sb2Te5 thin films deposited by RF sputtering at room temperature on SiO2/Si has been studied by transmission electron microscopy and X-ray diffraction. The analysis has been performed on partially crystallized films, with a surface crystalline fraction (fc) ranging from 20% to 100%. XRD analysis indicates the presence, in the partially transformed layer, of grains with average lattice parameters higher than that of the equilibrium metastable fcc phase (from 0.606 nm at fc=0.2 to 0.601 nm at fc=1). The amorphous to crystal transition, as shown by TEM analysis, occurs through the nucleation of face-centered-cubic crystal domains at the film surface. Local dimples appear in the crystallized areas, due to the higher atomic density of the crystal phase compared to the amorphous one. At the initial stage of the transformation, a fast bi-dimensional growth of such crystalline nucleus occurs by the generation of transrotational grains in which the lattice bending gives rise to an average lattice parameter significantly larger than that of the face-centered-cubic phase in good agreement with the XRD data. As the crystallized fraction increases above 80%, dimples and transrotational structures start to disappear and the lattice parameter approaches the bulk value.
9:00 PM - H12.9
Investigation on the Growth Behavior of Ge Doped SbxTey Thin Films Deposited by a Plasma-enhanced CVD.
Seol Choi 1 , Byung Joon Choi 1 , Taeyoung Eom 1 , Cheol Seong Hwang 1
1 material science and engineering, Seoul National University, Seoul, Kwanak-gu, Korea (the Republic of)
Show AbstractPhase change random access memory(PRAM) is one of the nonvolatile memory devices which utilizes the largely different resistances of the crystalline and amorphous phase change materials, most typically Ge2Sb2Te5, where the phase change was achieved by the application of current pulses. For the application of the PRAM in universal nonvolatile memories, set and reset speeds of PRAM must be improved. In the optical memory devices, Ge-doped Sb-rich SbTe(G-ST) has been reported to have a faster crystallization speed and better thermal stability than Ge2Sb2Te5. Accordingly, in this study, the G-ST films are deposited by a cyclic plasma-enhanced chemical vapor deposition(PECVD) at 423K on atomic-layer-deposited TiO2/thermally grown SiO2/Si substrates using a shower-head type 8 in.-scale PEALD reactor(Quros Co.,Plus-200). Ge(i-C4H9)4, Sb(i-C3H7)3, and Te(i-C3H7)2 were used as the Ge, Sb, and Te precursors, respectively. The sequence of precursor injection pulses was Sb-Ge-Te(super-cycle) and 1 super-cycle was composed of 3 elemental sub-cycles. Each sub-cycle was composed of the precursor-precursor purge-reduction gas-process purge. Reduction gas was H2-Ar mixture and carrier gas was Ar. Experiments were conducted under the following conditions; Te source feeding time was fixed(0.6s) and Sb source feeding time was varied from 0.5 to 2.5s. Ge source feeding time was 0, 0.8, 1.2s, respectively. Layer densities of Sb and Te in the deposited films were not influenced by Ge source feeding time. This suggests that the doping of Ge with a small amount(<10at%) hardly affects the deposition behaviors of the Sb rich Sb-Te thin films. The morphologies of films observed by FE-SEM showed that increasing Sb source feeding time resulted in the significant void formation and grain growth while increasing Ge source feeding time did not show any significant effect on the film morphologies. XRD data of as-deposited films showed that crystalline Sb7Te phase was formed and the (1 0 6) peak intensity of films increased with increasing Sb source feeding time while increasing Ge source feeding time slightly inhibits crystallization of Sb7Te. The composition was also confirmed by the XRF. As-deposited films having a higher Ge content or Te content showed higher sheet resistance measured by 4-point probe. The Ge14Sb61Te25 film showed a crystalline temperature of ~ 473K when the heating rate was 3K/min.In summary, Ge-doped Sb-rich Sb-Te phase change films were deposited by a cyclic PECVD technique. The doping of Ge with a low concentration(<10at%) hardly affected the growth behaviors of the Sb-rich Sb-Te films. It was noted that the Ge doping increased the crystallization temperature of the Sb7Te films suggesting that the doped Ge worked as the amorphous phase stabilizer. Meanwhile, Sb-rich phase showed higher crystallinity, lower resistivity, and rough surface morphology with several voids. Ge and Sb showed opposite functionality on the crystallization of these materials.
Symposium Organizers
Yoshihisa Fujisaki Hitachi Ltd.
Rainer Waser RWTH Aachen University
Tingkai Li Sharp Laboratories of America Inc.
Caroline Bonafos CEMES/CNRS
H13: Phase Change RAM II
Session Chairs
Friday AM, April 17, 2009
Room 2007 (Moscone West)
10:00 AM - **H13.1
Recent Advances in Phase Change Memory.
Greg Atwood 1
1 , Numonyx B.V., Santa Clara, California, United States
Show AbstractPhase Change Memory is emerging as one of the most promising technologies to be the mainstream non-volatile memory of the next decade, offering the promise if extending NVM scaling beyond the Floating Gate era and providing new levels of functionality previously not available in NVM.This talk will provide a review of the current understanding of chalcogenide (Phase Change) material physics and optimization for the material for usage in high density electronic memories. Chalcogenide has a long history of use in optical storage devices using lasers to invoke phase change and exploiting its optical properties, but serious study of the material for use in integrated Si memory exploiting its electrical properties is relatively recent. The material properties of many of the chalcogenides show high promise to enable a new level of functionality for non-volatile memories. The second half of this talk will review the device structures and memory architectures currently being explored and recent progress toward the production of high density memories. High density memories have been demonstrated with promising functionality and reliability utilizing several different cell concepts.
10:30 AM - H13.2
Field Induced Crystal Nucleation in Chalcogenide Phase Change Memory.
Marco Nardone 1 , Victor Karpov 1 , ILya Karpov 2 , Mukut Mitra 2
1 Physics and Astronomy, University of Toledo, Toledo, Ohio, United States, 2 , Intel Corporation, Santa Clara, California, United States
Show AbstractWe present a summary of our 2-year experimental and theoretical work on switching in chalcogenide phase change memory as governed by the structural transition form the high-resistive amorphous state to the low-resistive crystalline state. As a significant addition to the well-known experiments, we have studied switching under conditions far beyond the standard, particularly, under considerably lower voltages and elevated temperatures, as well as the statistics of switching events. Our data showed that the switching delay time is exponential in both the reciprocal temperature and reciprocal voltage. While the former dependence agrees with classical crystal nucleation theory, the latter implies the mechanism of electric-field induced nucleation consistent with the original hypothesis by Ovshinsky (1968). Indeed, under low bias we have observed exponentially long switching delay times attributable to the correspondingly long nucleation incubation times under low applied fields. Furthermore, we have verified that the average threshold voltage is inversely proportional to the logarithm of the time of the experiment. Under low applied voltages, the phenomenon of short-lived switching was observed and interpreted in the same framework as a result of the crystal nucleus decay upon its shorting of the induced electric field. Our statistical analysis has revealed the exponentially broad distribution of switching times in combination with a relatively narrow (~ 1% wide) distribution of threshold voltages between nominally identical samples – also consistent with the field induced nucleation mechanism.We have developed an analytical theory of crystal nucleation in strong electric fields (~ 0.1 –1 MV/cm) predicting all the above-observed features and their dependencies on the material parameters. As opposed to the zero-field case, our theory predicts nucleation of significantly asymmetric stretched spheroid (cigarette shaped) crystal particles with the nucleation barrier inversely proportional to the field strength. Such nuclei create the lightning rod effect thereby concentrating the electric field, which facilitates nucleation of other particles at their ends. This ‘lightning rod instability’ results in a cylinder shaped filament shorting the device. In agreement with the experimental observation, the theory predicts ‘under-threshold’ switching over the times many orders of magnitude longer than the standard switching times. We have extended our theory to the case of nucleation in disordered media where the structural parameters fluctuate between different microscopic regions, which explains the observed statistics of switching events.
10:45 AM - H13.3
Analysis of Nanoscale Transformation of Phase Change Materials.
Kristof Darmawikarta 1 2 , Bong-Sub Lee 1 2 , Simone Raoux 3 , Albert Liao 4 , Eric Pop 4 , Stephen Bishop 2 4 , John Abelson 1 2
1 Materials Science and Engineering, University of Illinois at Urbana-Champaign, Urbana, Illinois, United States, 2 Coordinated Science Laboratory, University of Illinois at Urbana-Champaign, Urbana, Illinois, United States, 3 , IBM Almaden Research Center, San Jose, California, United States, 4 Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, Illinois, United States
Show AbstractThe development of high-density phase change memory cells requires a precise knowledge of material properties on the scale of real devices, i.e. 10 nanometer dimensions and 10s of nanosecond phase transformations. Here, we present high speed nanocalorimetry measurements of thin film phase change materials, and new device structures that utilize carbon nanotubes for high density memory cells.The high speed nanocalorimeter consists of a 20 nm-thick Ge2Sb2Te5 sample deposited on a Si3N4 membrane, with a thin metal film on the back side that serves both as the resistive heater and thermometer. This configuration achieves a heating rate of 80,000 °C/sec. Compared to literature data obtained on bulk samples at slow heating rates, our data reveal a higher crystallization temperature, a lower melting temperature, the occurrence of glass transition before crystallization, and the absence of slow structural relaxation.We also present the result of switching phase change materials using carbon nanotube heaters. The device consists of 10 nm of Ge2Sb2Te5 film deposited on 2–10 μm long carbon nanotubes of ~ 2 nm diameter. We use conductive AFM measurements to observe the switching, which occurs due to Joule heating of the nanotubes. I–V curves show a sharp current increase, which we attribute to the amorphous-to-crystalline transition of the phase change material. We will show that these devices have the ability to switch nanovolumes of phase change material and the potential for application in high density memory cells.
11:30 AM - H13.4
Compact Thermal Model for Segmented Nanowire Phase Change Memory Cell.
I-ru Chen 1 , Eric Pop 1
1 Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, Illinois, United States
Show AbstractPhase-change memory (PCM) is a promising candidate to eventually replace current Flash memory technology. However, reducing power consumption for PCM cells remains a key challenge, and optimization of write current/power requires a more complete understanding and modeling of the temperature dependences on cell geometry and material properties. Here, we introduce a compact thermal model for the temperature distribution in PCM cells in both transient (0.1-1 ns) and steady-state time scales. The result can be used in circuit simulations, and allows us to efficiently optimize the power consumption of PCM cells.Previous work has focused on numerically solving finite-element (FE) electro-thermal models for write operations of PCM cells [1]. The effects of thermal boundary resistance (TBR) on programming current and temperature distribution have also been explored with FE models [2]. The simple analytic thermal circuit model proposed by Russo et al offers insight into the geometric dependence of required programming currents [3]. However, the transient temperature distribution of PCM cells remains unsolved for analytic models.The transient analytic thermal model developed here is applied to the segmented nanowire PCM cell, which is axisymmetric and is the ultimately desired cell configuration. Transient and steady-state temperature distributions are obtained by solving the heat diffusion equation, including effects of TBR varying from 0 to 100×10-9 m2K/W [2], and taking advantage of the cylindrical cell symmetry. The solution agrees with FE simulations within 10% error, yet the simulation time is reduced by 2-3 orders of magnitude. The analytic time- and position-dependent temperature distribution offer insights into the energy diffusion across materials and boundaries. The required programming power as a function of cell geometry, material properties and TBR is also discussed. We demonstrate how power consumption of PCM write operations can be minimized using calculated dependences. The model proposed here enables the efficient simulation of many PCM cell arrays using circuit simulators such as SPICE.[1] A.L. Lacaita, A. Redaelli, D. Ielmini, F. Pellizzer, A. Pirovano, A. Benvenuti, and R. Bez, IEDM’04 Tech. Dig., 911 (2004).[2] J.P. Reifenberg, D.L. Kencke, and K.E. Goodson, IEEE Elec. Dev. Lett. 29, 1112 (2008).[3] U. Russo, D. Ielmini, A. Redaelli, and A.L. Lacaita, IEEE Trans. Elec. Dev. 55, 506 (2008).
11:45 AM - H13.5
In-situ TEM Study of Phase-change Memory Cells.
Stefan Meister 1 , David Schoen 1 , Yi Cui 1
1 Materials Science and Engineering, Stanford University, Stanford, California, United States
Show AbstractThe great promise of phase-change memory as a future non-volatile memory solution has motivated many theoretical and experimental studies that have helped elucidate the underlying theoretical foundations. However, reliability issues and inconsistent switching in real devices suggest a need for further experimental investigation. In situ transmission electron microscopy (TEM) studies have the potential to reveal the causes of the observed variability of switching behavior by allowing us to directly correlate structural or compositional changes with electrical behavior. Further, the TEM can provide great insight into the phase-change behavior by providing high resolution images and compositional analysis. With the use of a special in situ TEM holder and standard micro-fabrication techniques, we have developed a method to electrically pulse single memory cells under direct observation in a TEM. In this talk we demonstrate the fabrication and measurements of GeTe NW lateral phase-change cells on TEM membranes and compare it to our switching studies on silicon substrates. Surprisingly, the dominant switching mechanism appears to be the formation and closing of voids. Further we discuss the fabrication of sputtered GST line cells on membranes and show our newest results of TEM observation during electrical pulsing.
12:00 PM - H13.6
Nuclei in Various Amorphous States of Phase Change Materials and their Effect on Transformation Speed.
Bong-Sub Lee 1 2 , Simone Raoux 3 , Robert Shelby 3 , Charles Rettner 3 , Geoffrey Burr 3 , Kristof Darmawikarta 1 2 , Stephen Bishop 2 4 , John Abelson 1 2
1 Materials Science & Engineering, University of Illinois at Urbana-Champaign, Urbana, Illinois, United States, 2 Coordinated Science Laboratory, University of Illinois at Urbana-Champaign, Urbana, Illinois, United States, 3 , IBM Almaden Research Center, San Jose, California, United States, 4 Department of Electrical & Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, Illinois, United States
Show AbstractThe speed of phase change memory devices depends strongly on the kinetics of phase transformation, especially crystallization, of the active materials such as Ge2Sb2Te5 or AgInSbTe. It has been empirically found that, for a fixed composition, different processing conditions may result in amorphous states with significantly different transformation rates. Here we systematically characterize the structural and kinetic differences between various amorphous states of phase change materials, i.e. melt-quenched, laser-primed (irradiated but not crystallized), low-temperature annealed, and as-deposited films prepared under different conditions. While these amorphous states all appear amorphous and indistinguishable in conventional diffraction measurements, Fluctuation Transmission Electron Microscopy (FTEM) clearly reveals structural differences. FTEM quantifies the structural order on the 1-3 nm scale by statistically analyzing a set of nanodiffraction patterns; we show that it can detect the existence of subcritical nuclei in the amorphous phase. The crystallization speed from each amorphous state is measured by an in-situ laser pump-probe technique; in the case of AgInSbTe, individual nucleation events can be monitored and the data averaged to determine the nucleation rate. The combined use of FTEM and the laser technique shows that the transformation speed is affected by the presence and number of pre-existing subcritical nuclei: the content of nuclei may be either high or low depending on processing conditions and, accordingly, the incubation delay before crystallization can be shorter or longer. The relationship between nuclei content and processing conditions can be explained using classical nucleation theory; numerical simulations of the size distribution of subcritical nuclei under different thermal conditions will be presented. This study provides a rare experimental proof of the predictions of nucleation theory, including the temperature- and time-dependent behaviors of embedded nuclei in an amorphous matrix.
12:15 PM - H13.7
Crystallization Rate, Glass Transition, and Thermomechanical Properties of Nitrogen-doped Ge2Sb2Te5 Film.
Il-Mok Park 1 , Tae-Youl Yang 1 , Young-Chang Joo 1
1 , Seoul National Univ., Seoul Korea (the Republic of)
Show AbstractPhase change materials are applied to the rewritable optical recording and the electrically programmable non-volatile memories, because these materials can be reversibly transformed between the amorphous and crystalline phase. Nitrogen (N)-doping into GeSbTe system materials is used as effective way in order to enhance the overwrite capability in the optical media and reduce the reset current in the electrical memory, respectively. The effects of N-doping on GeSbTe alloys have been studied extensively, e.g., increase of resistivity and lattice parameter, and suppression of grain growth. N-doping is also known to retard the crystallization time and to increase the activation energy for crystallization. However, detailed explanations on the mechanism for suppression of the crystallization in the N-doped GST film are lacking. The crystallization behaviors can be studied through thermomechanical properties since the crystallization induce inherent volume shrinkage. Furthermore, the viscosity which is thermomechanical material property has been known to be inversely proportional to the crystallization rate, consisting of the nucleation rate and growth velocity. Therefore, it is necessary to measure the viscosity of N-doped GST films, and then it may be able to understand the crystallization mechanism. In this study, we investigated the crystallization behavior of N-doped Ge2Sb2Te5 (GST) films by evaluating the thermomechanical properties. GST films with a thickness of 300 nm were N-doped during the sputtering by adjusting the N2 gas flow rate with the argon gas flow rate fixed. The N contents were prepared by 5, 10, and 15 at.% to examine the dependence of viscosity on doping level. Because the viscosity is time dependent due to structural relaxation, the viscosity can be determined from the mechanical stress change as a function of time. The stress changes in the film were evaluated from the change in the curvature of the film–substrate system. The stress relaxation experiments were done at a constant temperature for 10 hours and the viscosity was measured by fitting viscous flow equation to the stress curves. The temperature dependence of the viscosity of N-doped GST films decreased linearly. The viscosity increased as the doping level increased, implying that the N-doping reduced the nucleation rate and growth velocity. In addition, we could determine the glass transition temperature (Tg) of N-doped GST films through viscosity measurement. The Tg was determined to be the temperature at which no increase of viscosity during stress relaxation was observed any longer. The Tg for 10 and 15 at.% N-doped GST film were about 140 and 170 oC, respectively. The N-doping suppressed not only crystallization but also glass transition, which implies increase of the operation time and enhancement of the thermal stability, respectively. Consequently, this study is helpful for determination of N-doping level and doping materials to improve the performance.
12:30 PM - H13.8
The Influence of Nitrogen Doping on the Chemical and Local Bonding Environment of Amorphous and Crystalline Ge2Sb2Te5.
Joseph Washington 1 , Eric Joseph 2 1 , Jean Jordan-Sweet 2 , Simone Raoux 3 , Chieh-Fang Chen 2 , Adam Pyzyna 2 , Ravi Dasaka 2 , Dolores Miller 3 , Alejandro Schrott 2 , Chung Lam 2 , Ying Zhang 2 , Bruce Ravel 4 , Joseph Woicik 4
1 Department of Physics, North Carolina State University, Raleigh, North Carolina, United States, 2 T. J. Watson Research Center, IBM/Macronix PCRAM Joint Project, Yorktown Heights, New York, United States, 3 , IBM Almaden Research Center, San Jose, California, United States, 4 , National Institute of Standards and Technology, Gaithersburg, Maryland, United States
Show AbstractThere is currently keen interest in utilizing phase change materials (PCMs) in constructing non-volatile memory cells. This has been fueled by recent advances in fabrication techniques and the promise of scalability beyond the limit of conventional DRAM and NAND flash memory [1]. For solid state device applications, Ge2Sb2Te5 (GST), GeSb, and other currently widely accepted PCMs require doping to favorably modify their crystallization properties such as, crystallization speed, crystallization temperature, and thermal stability [2]. In this work, we clarify the chemical and structural role of nitrogen doping (N:) in as-deposited and crystalline (fcc and hcp) GST thin films. We use a combination of X-ray Photoelectron Spectroscopy (XPS), Fourier Transform Infrared Spectroscopy (FTIR), and X-ray Absorption Spectroscopy (XAS) to obtain the chemical and local bonding environment around each of the elements in the sample, pre/post-anneal, at various doping concentrations. Transmission Electron Microscopy (TEM) data was also used to confirm the variation of grain size with dopant concentration. The XPS results shows an apparent shift in the Ge 2p3 electron binding energy spectra of N:GST (1218.5 eV) versus undoped GST (1217.66 eV) due to the formation of Ge-N bonds. No similar shift was observed in XPS electron binding energy data for Sb or Te. This Ge-N bonding is further confirmed in the FTIR absorption spectra of N-GST films of varying dopant concentrations, as we observed Ge-N peaks in the IR absorption spectra at approximately 770 and 720 cm-1. XAFS, an ideal technique for providing local structure and bonding information (such as co-ordination number, bond distance, and mean square disorder) around the constituent elements in our sample, further provides a self consistent analysis confirming the affinity for nitrogen to form Ge-N bonds in both the amorphous and annealed films. Furthermore, in the annealed case, these Ge-N bonds did not enter the cubic crystal lattice (see ref. 3,4). Instead, it is proposed that the dopants remain at the grain boundary of the crystallites observed in the TEM data such that the annealed film is comprised of crystallites with a dopant rich grain boundary.