Bipolar RRAM Research Articles

Low Frequency Noise Behavior in Resistive Memory Devices with Hf/HfO2 Stack

Resistive-random-access-memory (RRAM) devices are considered to be one of the most potential candidates of next generation emerging memory technologies due to their excellent device properties, such as high density, low power, low voltage, and high speed. RRAM features simple metal–insulator–metal (MIM) structures and is fully compatible with traditional CMOS fabrication processes. Low-frequency noise (LFN) measurement is a technique to characterize electrically the trap-assisted conduction processes in dielectrics. In this work, we report the LFN characteristics of Hf/HfO2-based bipolar RRAM and investigate the current conduction mechanism and internal physics.

Copper oxide memristor as artificial synapses emulating Hebbian symmetric and asymmetric learning behavior for neuromorphic computing beyond von Neumann architecture

Beyond von Neumann’s architecture, artificial neural network-based neuromorphic computing in a simple two-terminal resistive switching device is considered the future potential technology for simultaneous data processing and storage. These are also compatible with low-power consumption nanoelectronic devices and, thus, suitable for applications such as image recognition toward solving complex pattern recognition problems. Herein, motivated by the human biological brain, we successfully synthesized low-cost RRAM devices using the thermal oxidation of Cu, i.e., CuO as the active material together with Cu as the top electrode and FTO as the bottom contact for a two-terminal resistive switching device, and investigated characteristics for neuromorphic computing. Cu/CuO/FTO-based devices showed excellent bipolar analog RRAM characteristics with 150 repeatable cycles, retention for 11 000 s, and DC pulse endurance for 5000 cycles. Moreover, devices exhibit a remarkable mimicking ability, demonstrating spike time-dependent plasticity (STDP), pulse-paired facilitation (PPF), synaptic weight, and learning and forgetting characteristics, substantiating the recognition ability. Furthermore, the artificial neural network synaptic membrane exhibits excellent long-term (LTP) and short-term (STP) potentiation for six consecutive cycles. Thus, the present work on Cu/CuO/FTO-based devices provides a detailed understanding of CuO active material-based resistive switching with a potential for neuromorphic computing beyond the von Neumann architecture.

Low-temperature characteristics of magnesium fluoride based bipolar RRAM devices.

This study investigates the temperature-independent switching characteristics of magnesium fluoride (MgFx) based bipolar resistive memory devices at temperatures ranging from 300 K down to 77 K. Filament type resistive switching at the interface of Ti/MgFx and the trap-controlled space charge limited conduction (SCLC) mechanism in the bulk MgFx layer are confirmed. The experimental results indicate that the operating environment and temperature critically control the resistive switching performance by varying the non-stoichiometry of the amorphous MgFx active layer and Ti/MgFx interface region. The gaseous atmosphere (open air or vacuum) affects device performances such as the electroforming process, on-state current, off-state current, on/off ratio, SET/RESET voltage and endurance of resistive-switching memory devices. After electroforming, the device performance is independent of temperature variation. The Ti/MgFx/Pt memory devices show promising data retention for >104 s in a vacuum at room temperature and 77 K with the DC endurance property for more than 150 cycles at 77 K. The devices have great potential for future temperature-independent electronic applications.

Enhanced switching ratio of sol–gel-processed Y2O3 RRAM device by suppressing oxygen vacancy formation at high annealing temperatures

Sol–gel-processed Y2O3 films were used as an active-channel layer for resistive switching memory (RRAM) devices. The influence of post-annealing temperature on structural, chemical, and electrical properties was studied. Y2O3-RRAM devices comprising electrochemically active metal electrodes, Ag, and indium tin oxide (ITO) electrodes exhibited the conventional bipolar RRAM device operation. The fabricated Ag/Y2O3/ITO RRAM devices, which included 500 °C-annealed Y2O3 films, exhibited less oxygen vacancy and defect sites, reduced the leakage current, increased the high-/low-resistance state ratio of more than 105, and provided excellent nonvolatile memory properties without significant deterioration for 100 cycles and 104 s.

Logic-in-memory application of CMOS compatible silicon nitride memristor

Resistive switching memory (RRAM) performs its role as a next-generation memory that overcomes the limitations of von Neumann computing. RRAM can be stated as set or reset state by applying a certain voltage and has advantages in maintaining the states with non-volatile characteristics. That is, both the change of state and storage of operations occur in a single RRAM device. Both bipolar RRAM which operates on two opposite voltage polarity, and unipolar RRAM, which operates on a single polarity, can be combined with the switch to effectively implement the Boolean function. In this paper, both unipolar and bipolar characteristics are conducted in CMOS compatible Ni/SiN/Si RRAM devices. We succeed in demonstrating 14 logical operators out of 16 by applying pulses to implement the Boolean function. In addition, we prove that the other two logical operators cannot be represented by the existing system, and we also suggest ideas to solve this problem.

(Invited) Current and Voltage Control of Intermediate States in Bipolar Rram Devices for Neuristor Applications

Nowadays, there is an increasing interest in the fabrication of neuristors in which memristors are used to create a neuron-like behavior (1). Memristors based on resistive switching memories (RRAM) are promising candidates to implement artificial synaptic devices for their use in neuromorphic systems, due to their high number or reachable conductance levels.In a previous work (2) we show that TiN/Ti/ HfO2/W capacitors exhibit resistive switching behavior and that intermediate conductance states can be obtained by varying the voltage applied to the device (Voltage-control mode, VCM). We demonstrated that the conductance values vary near linearly when applying voltage depression pulses. During depression process, the conductance state depends on the amplitude and the duration of the voltage pulses length. Moreover, this process is accumulative: the conductance decreases when applying successive pulses of same amplitude and length. In contrast, the potentiation characteristic is not linear, as for other synaptic devices, as PRAMs: the transitions from high resistance state (HRS) to low resistance state (LRS) is very sharp and intermediate conductance state cannot be controlled. Decreasing the voltage pulse length or amplitude was not a choice. Additionally, we tried to use ramps where the voltage linearly increases and once again the characteristics remain very nonlinear. In summary, a very poor control of the potentiation state is obtained when using voltage as synapse stimulus.In the present work, we demonstrate that the potentiation process can be linearly controlled when using current as the synapse stimulus (Current-control mode, CCM).The devices used in this work are TiN/Ti/10-nm HfO2/W MIM capacitors. The high-k dielectric was deposited by the atomic layer deposition (ALD) technique at 225 ºC using TDMAH and water as hafnium and oxygen precursors respectively. Nitrogen was used as carrier and purge gas. The bottom electrode consists of a 200 nm W layer, and the top electrode consists of a 200 nm TiN layer and a 10 nm Ti layer. Metal electrodes were deposited by magnetron sputtering.An HP 4155B Semiconductor Parameter Analyzer was used to perform the voltage and the current measurements. In order to perform the pulsed measurements, we used an Agilent 33500B Series waveform generator which provides voltage pulses and ramps. The voltage waveforms are converted to mimic current waveforms using a home-made voltage to current converter. All the measurements were carried out at room temperature and the entire experimental setup was computer controlled.In figure 1 we show that the conductive path formation (electroforming) requires a very few nA current value. As the filament is formed, the devices reaches the low resistance state (LRS) and the voltage falls to a very low value. Once the filament is formed, we have measured the voltage-current (V-I) characteristic instead of the more usual current-voltage (I-V) characteristic. Figure 2 shows 50 consecutive bipolar switching V-I loops. It is important to point that in the CCM mode, a voltage compliance in the reset transition has to be used. Otherwise, irreversible oxide breakdown takes place due to the huge increase of the structure dissipated power. Departing from the HRS state we have applied current pulses of increasing amplitude. In Fig. 3 we plot the conductance of the device measured at 0.1 mA as a function of the amplitude of the current pulse previously applied. We see that the synaptic potentiation linearly depends on the pulse amplitude in the CCM mode. In Figure 4 we demonstrate that potentiation occurs at the very first current pulse. Successive pulses do not modify the conductance value. In summary, the potentiation process in CCM mode is very fast, non-accumulative and nearly linear. To confirm that this process is very fast we have obtained current transients in the VCM mode (Figure 5) which confirms that when a positive voltage is applied the current increases occurs at very short times (lower than the experimental setup time response).In contrast, the depression process is better controlled by applying voltage pulses (VCM mode) as described in ref. 2. Moreover, the depression process is slower and accumulative. That can be observed in Fig. 6 in which a train of -0.8 V voltage pulses is applied. The conductance decreases from pulse to pulse and show very long time rates (tens of ms for the showed voltage)In conclusion, CCM mode for potentiation and VCM mode for depression characteristic is the optimal combination to control the synaptic behavior of the studied bipolar RRAM devices.References Timmer, Nature Materials, 2012. DOI: 10.1038/NMAT3510García et al., Micr. Eng. 215, 110984 (2019). Figure 1

(Invited) Current and Voltage Control of Intermediate States in Bipolar Rram Devices for Neuristor Applications

The control of the intermediate conductance levels in HfO2-based MIM capacitors for neuromorphic applications is presented in this work. Using voltage or current control signals shows significant differences. The potentiation levels can be controlled in a linear way using current as control signal. These levels are achieved in very short times and only depends on the current pulse amplitude magnitude. On the contrary, depression levels cannot be controlled by current but with voltage pulses. A proper control of synaptic behavior requires the combination of both types of signals.

Resistive switching characteristics of HfO2 based bipolar nonvolatile RRAM cell

Non-volatile memory (NVM) will play a significant role in the progress of the next-generation of electronic products due to the physical limitations of the conventionally used flash memory. In this paper, switching properties of Cu/HfO2/Pt/Si resistive random access memory device were successfully demonstrated. Semiconductor parameter analyzer (model no. B1500A Agilent) was used for the measurement of the resistive switching characteristics of the fabricated device. It has been observed that the device shows the bipolar mode of switching operation. The temperature dependence of electrical conductivity for low resistance state and high resistance state was also measured in the range from 300 K to 450 K. The switching behaviour of the proposed device explained on the basis of conduction filaments formed in HfO2.

Unidirectional threshold switching in Ag/Si-based electrochemical metallization cells for high-density bipolar RRAM applications

We herein present a novel unidirectional threshold selector for cross-point bipolar RRAM array. The proposed Ag/amorphous Si based threshold selector showed excellent threshold characteristics in positive field, such as high selectivity (~ 105), steep slope (< 5 mV/decade) and low off-state current (< 300 pA). Meanwhile, the selector exhibited rectifying characteristics in the high resistance state as well and the rectification ratio was as high as 103 at ± 1.5 V. Nevertheless, due to the high reverse current about 9 mA at − 3 V, this unidirectional threshold selector can be used as a selection element for bipolar-type RRAM. By integrating a bipolar RRAM device with the selector, experiments showed that the undesired sneak was significantly suppressed, indicating its potentiality for high-density integrated nonvolatile memory applications.

A 14-nm FinFET Logic CMOS Process Compatible RRAM Flash With Excellent Immunity to Sneak Path

In this paper, we have demonstrated an oxygen-vacancy-based bipolar RRAM on a pure logic 14-nm-node HKMG FinFET platform. A unit cell of the memory consists of a control transistor (FinFET) and a storage transistor (a second FinFET). The later performs as a bipolar RRAM. This unit cell can be integrated in an AND-type memory array. The memory cell has an ON/OFF ratio equal to 200 and 400 for the n-type and p-type FinFET RRAMs, respectively, endurance larger than 400 and 1000 times for n- and p-type devices, respectively, and the retention test for over 1 month under 125 °C temperature environment. To analyze the array performance of the AND-type FinFET RRAM at the circuit level, we have further discussed the issues of the sneak path and disturbance, in which an active-fin isolation of FinFET in an AND-type array has been suggested to minimize the leakage current induced by sneak paths. The results have shown a large window with up to 103 ON/OFF ratio, 30% standby power reduction, and 90% active power reduction with reference to the conventional AND-type array. As a result, the bipolar FinFET RRAM exhibits great potential for the embedded memory applications, in particular it can be extended to 28-nm HKMG and the FinFET platform beyond 14-nm technology node, to fill the Moore’s gap between the high-performance logic and the embedded memory.

Excellent selector performance in engineered Ag/ZrO2:Ag/Pt structure for high-density bipolar RRAM applications

A high-performance selector with bidirectional threshold switching (TS) characteristics of Ag/ZrO2/Pt structure was prepared by incorporating metallic Ag into the ZrO2 matrix. The bidirectional TS device exhibited excellent switching uniformity, forming-free behavior, ultra-low off current of &amp;lt;1 nA and adjustable selectivity (from 102 to 107). The experiment results confirmed that metallic Ag clusters were penetrated into the ZrO2 matrix during the annealing process, which would function as an effective active source responsible for the bidirectional TS. The volatile behavior could be explained by the self-dissolution of unstable filaments caused by minimization of the interfacial energy and thermal effect. Furthermore, a bipolar-type one selector-one resistor (1S-1R) memory device was successfully fabricated and exhibited significant suppression of the undesired sneak current, indicating the great potential as selector in a cross-point array.

Scalability and reliability issues of Ti/HfOx-based 1T1R bipolar RRAM: Occurrence, mitigation, and solution

Scalability and reliability issues are the most dominant obstacle for the development of resistive switching memory (RRAM) technology. Owing to the excellent memory performance and process compatibility with current CMOS technology of Ti/HfOx-based filamentary type bipolar RRAM, its scalability and reliability issues have been investigated in this document. Towards this goal, we demonstrate that there exists a clear correlation between the transistor and memory cell, which ultimately limits the scaling in terms of operation current and size of the transistor as well and performance of the Ti/HfOx-based 1T1R bipolar RRAM. Due to the resemblance of switching behaviour between complementary resistive switching, i.e., CRS in a single memory stack, and bipolar resistive switching, the Ti/HfOx-based bipolar RRAM suffers from resistance pinning (RP) issues, whereas the minimum resistance during the 1st RESET operation always impeded below 20 kΩ; this occurs through the interaction between the transistor and memory cell during the FORMING process. However, a sufficiently lower FORMING voltage can mitigate the RP issue occurring in Ti/HfOx-based bipolar RRAM and an alternative Ta buffer layer over HfOx dielectrics is proposed to prevent the activation of self-CRS in the memory cell during the FORMING process.

Pulse area dependent gradual resistance switching characteristics of CMOS compatible SiNx-based resistive memory

In this work, we investigated the gradual resistance switching phenomenon of our fabricated silicon nitride-based bipolar RRAM. By positive (set) and negative (reset) pulses applied between top electrode (TE) and bottom electrode (BE), the resistance state of the RRAM cell was delicately controlled. We checked the effect of pulse width, rise and fall time and pulse amplitude on the change of the resistance state. In conclusion, it is demonstrated that change of resistance state is determined by applied pulse area above a certain threshold voltage. The memory cell and gradual resistance change characteristics would be used to implement accurate and reliable synaptic devices in low power neuromorphic system.

3-bit Resistive RAM Write-Read Scheme Based on Complementary Switching Mechanism

While standard bipolar switching RRAM memory devices can be programmed into different resistance states, the complementary switching mechanism allows for two distinct switching locations that each can be programmed to these resistance states. In this letter, we present a technique to discriminate these switching locations and report on a novel scheme allowing for sub- $\mu \text{s}$ pulse write and read of eight different logic states in Pt/Ta2O5/Ta/Pt devices by using only four different resistive states. Thus, in addition to the multilevel capability of bipolar switching devices, double the information can be stored and read in a single complementary switching device.

Research on feasibility of using a Transient Voltage Suppressor as the selection device for bipolar RRAM

In this paper, the Transient Voltage Suppressor (TVS) is provided as the selection device for bipolar RRAM crossbar array applications. Simulation results suggest that, by using the TVS as a selection device, the sneak current in the crossbar array can be effectively suppressed. Compared to the single RRAM structure (1R), the readout margin is sufficiently improved, and the size of the crossbar array can be enhanced to 1Mbit (103×103), indicating that the TVS has potential application as a selection device in bipolar RRAM for achieving high density crossbar array.

(Invited) Simulating the Behavior of a Bipolar Filamentary ReRAM Cell for Upcoming Memory Devices

Fundamental research on resistive switches for future non-volatile memories has led to the mass production of resistive random access memory devices (ReRAM)1. In order to realize and design these new memory devices the industry relies on behavioral models of the memory cell. Key aspects to describe these memory cells are still debated. For example, a comprehensive charge transport expression that correlates defect density and conductivity is still missing 2 3. The models for setting and resetting the cell are largely dependent on migration of defects or anions as well as thermal effects4 5. However, these models and concepts for each mechanism are sometimes applicable for one mechanism only or mutually exclusive. Finite element simulations of a bipolar filamentary ReRAM cell based on a binary oxide are used to model the cell behavior and highlight missing aspects to reliably evaluate the behavior of such a device. The operation of a ReRAM cell can be separated in 3 main aspects, namely, forming, set and reset. The forming (soft breakdown) entails the creation of oxygen vacancies and the conductive filament which can be expressed by a reaction term. It will be shown that the conductivity definition and the mobility of ions strongly influence the forming event. The behavioral modeling of the reset-event of a conducting filament by oxygen vacancy migration yields a typical I(V) curve. However, the bipolar characteristic of the device cannot be explained by such a migration of oxygen vacancies alone since it can be applied in either direction. Also, the re-oxidation of the filament results in a very large electric field which would result in return in a soft breakdown event and obstructing the reset. The common expressions for forming generally prevent a successful reset simulation and to simulate the bipolarity of set and reset an anisotropic mobility of the oxygen vacancies is required. Joule heating effects strongly influence the forming, set and reset but without a concrete correlation of conductivity and defect density predictions are difficult to make. Figure 1 : Finite element simulation of a forming event. The colors represent the density of oxygen vacancies within the dielectric. The metal oxide is sandwiched between two metal electrodes. A positive bias is applied to the top electrode and swept from 0V – 3V (1V/s). a) 0V, b) ~1V, c) ~2V, d) ~2.1V, c) ~2.2V. Figure 2: a), b), c) normalized color map of oxygen vacancy density. d), e), f) color map of the corresponding electric field. The re-oxidized filament area depicts a very large field that would cause a soft breakdown. Figure 3: a), b), c) I(V) curve of the corresponding reset event in Figure 2. In essence, utilizing the known physical characteristics, behavioral models for the ReRAM cell can be realized; however, a coherent behavioral model that is able to simulate all 3 aspects is still problematic to create with the known physical characteristics. 1. Global, P. N. Headquarters News. 2013–2015 (2015). 2. Chiu, F.-C. A Review on Conduction Mechanisms in Dielectric Films. Adv. Mater. Sci. Eng. 2014,1–18 (2014). 3. Park, S., Magyari-köpe, B. & Nishi, Y. Impact of Oxygen Vacancy Ordering on the Formation of a Conductive Filament in TiO 2 for Resistive Switching Memory. 32,2010–2012 (2011). 4. Waser, R., Dittmann, R., Staikov, G. & Szot, K. Redox-Based Resistive Switching Memories - Nanoionic Mechanisms, Prospects, and Challenges. Adv. Mater. 21,2632–2663 (2009). 5. Larentis, S. et al. Resistive Switching by Voltage-Driven Ion Migration in Bipolar RRAM — Part II : Modeling. 59, 2468–2475 (2012). Figure 1

A Combined Modulation of Set Current With Reset Voltage to Achieve 2-bit/cell Performance for Filament-Based RRAM

A combined operation scheme to realize multibit switching in filament-based bipolar RRAM device is proposed. By combining the modulations of the current compliance in set operations with the stop voltage in reset operations together, the size or the quantity of the filaments in the film bulk can be controlled. An RRAM device with the structure of Ag/HfOx/Pt is fabricated and the 2-bit/cell memory function is achieved by the proposed modulation. Furthermore, the 2-bit/cell data reliability is satisfactory including the high-temperature retention, cycling endurance.

Oxide based two diodes–one resistor structure for bipolar RRAM crossbar array

A new bipolar selector to suppress the sneak current in crossbar array has been proposed using anti-parallel connected oxide diodes. The highly nonlinear I–V characteristics are realized by the anti-parallel connected Ag/TiOx/Ti oxide diodes. By connecting the selector and a bipolar Cu/HfO2/Pt RRAM in series, the sneak current can be effectively suppressed and the readout margin is sufficiently improved compared to that obtained without using the selector. The maximum array size of the crossbar array with the anti-parallel connected diodes selector can be increased to more than 1Mb at a readout margin of 10%. These results indicate that the anti-parallel connected oxide diodes as a selector has great potential for high density bipolar RRAM crossbar array applications.

Progresses in Modeling HfOx RRAM Operations and Variability

This paper reports on recent progresses in modeling bi-polar RRAM devices based on hafnium oxide. The unique modeling environment adopted for the simulation of device operations accounts self-consistently for the charge and ion transport, and the structural device modification occurring during forming and set/reset operations. Reliability mechanisms as well as the major sources of devices variability are included thanks to a multi-scale approach that connects the electrical device performance to the atomic-level material properties. The modeling methodology can be successfully applied to both improve device performances and fabrication process of state-of-the-art RRAM devices, and devise device solutions for future 3D RRAM architectures.

A SELF-RECTIFYING BIPOLAR RRAM DEVICE BASED ON Ni/HfO2/n+-Si STRUCTURE

A bipolar RRAM device based on Ni / HfO 2/n+- Si structure with self-rectifying characteristics is demonstrated for high density cross-bar memory application. Experimental results indicate that Ni conductive filament generated at LRS plays an important role in resistive switching, resulting in the formation of a Schottky junction at the Ni -CF/n+- Si interface which determines the self-rectifying behavior at LRS. These results are very important from the point of view of understanding the self-rectifying switching mechanism and improving the resistive switching characteristics of self-rectifying RRAM devices.