Stress

Stress, or non-destructive readout, is also important for nonvolatile memory

also a basic demand for nonvolatile memories. On the other hand, if random access memories have the property of non-destructive readout, the control circuit will not have to refresh which takes additional clock cycles to write the data back, giving rise to the increase in the operation speed.

The stress tests were implemented also by Agilent 4155C. After switched to on or off state, the device was applied with 0.1 V bias for a long period to observe if any data loss would happen. For the time period of data not modified by bias, it is equivalent that the memory states would not be affected by the same period of total pulse widths when operating under pulses.

Figure 2.1. Sample structure.

Figure 2.2. Experiment flow.

Figure 2.3. Sketch of DC sputter system components.

Figure 2.4. Sketch of thermal oxidation furnace system component.

Chapter 3

Results and discussion

The frame of experimental results and discussion are shown in Fig.3.1. At first, The Ti/CuO/Pt structure with different oxidation conditions are analysis to find the optimum oxidation. Secondary, the different top electrodes, Ti/CuO/Pt, Pt/CuO/Pt, and W-probe/CuO/Pt, with the same oxidation conditions are studied to confirm filament formation and rupture in the bulk of the CuO. Then, the Ti/CuO/Pt device is measured by controlling different current compliance and different reset stop voltage.

At application, multiple level storing can be achieved in Ti/CuO/Pt structure by controlling current compliance at on-state or reset stop voltage at off-state. The device has a potential for non-volatile multiple-valued memory device.

3.1 CuO film with different thermal oxidation condition

After the Cu thin film deposited by DC sputtering, the devices were thermal oxidation at different oxidations. The oxidation temperature conditions are 300^oC, 400^oC, 500^oC, and the oxidation time conditions are 15min, 30min, 60min. Therefore, there are nine oxidation conditions to study. In different oxidation conditions to find optimum, we use Ti to be the top electrodes.

3.1.1 Electrical properties

In different thermal oxidation conditions, the resistive switching characteristics are observed by I-V curve. The operation voltages, including V

on and V

off, are both

less than 3V and suggest that the devices are appropriate for the low voltage applications. The resistive switching localized in small region. There are some properties to note, the turn-on voltage variance more than turn-off variance and the switching gradual stable when switching cycles increase.

The devices yield and endurance are showed in Fig 3.2. The oxidation conditions, 400^oC, 60min, have the best device yield and endurance than other conditions. The resistive switching is nonpolar switching, the nonpolar switching can be seen for either positive or negative polarity of V or I; our samples show turn-on and turn-off when alternating the polarity of V or I as seen in bipolar switching. Thus, the nonpolar and bipolar actions are coexisting in a quite unique fashion.

3.1.2 Material analyses

Fig. 3.3 shows XRD results of device with oxidation condition 400^oC-60min, and 500^oC-60min. It was found that the CuO films have a polycrystalline structure.

Fig. 3.4 and Fig. 3.5 show SIMS result of oxidation condition 400^oC- 15min, and 400oC-60min, respectively. It was found that the Cu film was completely oxidation.

3.2 Compare different top electrodes

The devices of the oxidation temperature at 400^oC, 60 min, were investigated that provide higher yield and endurance.

3.2.1 Relationship of ratio and voltage

The on/off ratio and voltage bias relationship is showed in Fig. 3.24. The on/off ratio is lager at small bias. Due to this reason and small read bias more damage the memory state, the read bias magnitude is setting 0.2V. The reason of the read bias taken 0.2V is 0.2V is ohmic conduction region by current fitting as show in Fig. 3.18, 3.19, 3.21, 3.22.

3.2.2 Operation mode and polarity

After forming process, the memory device remains at ON-state. Then, we need a voltage to switch the memory device back to the OFF-state and this is what we call turn-off process. If the memory is switched back to the ON-state, this is a turn-on process. In ours research, Pt/CuO/Pt devices belongs to nonpolar switching, and the I-V curve is symmetrical as show in Fig. 3.10-13. On the other hand, even though

Ti/CuO/Pt devices belong to nonpolar switching as show in Fig. 3.6-9, the best mode is located at positive turn-on and negative turn-off mode. From now on, in this thesis, we only sweep for this best mode to investigate the electrical properties. The range of the operation voltage is located at 2V ~ -1.5V, but depicts a larger variation. In order to further confine the resistive switching phenomenon can occur without electrode of active metal, we measure CuO thin film directly by W-probe. The result of measure also show nonploar switching as show in Fig. 3.14-16. Fig. 3.17 shows that compare with Ti top electrode and W-probe, the W-probe can avoid the interface effect and small area of diameter ~ 5um.

3.2.3 Current fitting

Fig. 3.18, 3.19, 3.21 and 3.22 show that the ON-state is ohmic conduction at all

electrodes in both turn-on and turn-off process. It can be noticed that the current has a linear relationship with voltage of ON-state, which indicates that ohmic conduction is dominant in ON-state. It is suggested that conductive filament paths are formed as a bridge between top electrode and bottom electrode after the turn-on process.

Frenkel-poole emission model is extensively used investigate carrier transport behavior in insulator and semiconductor, which describe in section 1.3.4.

Fig. 3.20 and 3.23 show the logarithmic plot of I versus V^1/2 of OFF-state. The experiments data obeys a good linear relationship at high electrical field region. From the ln (I/V)  V^1/2 relation and Frenkel–Poole constant α calculation, carrier trapping and de-trapping of Frenkel–Poole effect is thought of as the main conduction mechanism of OFF-state. Fig. 3.20 shows that I-V curves were well fitted by the formula of Poole–Frenkel emission model with Pt top electrode. Fig. 3.23 shows that I-V curves were well fitted by the formula of Poole–Frenkel emission model with Ti top electrode.

The Pt is deep work function material and Ti is shallow work function. CuO is p-type semiconductor. In generous, the work function is 5.3eV and band gap is about 1.4eV at 400^oC thermal oxidation. A deep work function metal can give an Ohmic contact to a p-type semiconductor, while a Schottky barrier is formed at the interface between a shallow work function metal and a p-type semiconductor. Pt is deep work function which can is formed ohmic contact to a CuO film, but Ti is shallow work function which can give a Schottky barrier contact to a CuO film.

3.2.4 Size effect

Fig. 3.25-28 show I-V curve of Ti/CuO/Pt structure with top electrode area 10um*10um, and diameter 150um, 250um, and 350um. For constant current

compliance value, the ON-state current is independent size, but the OFF-state current is increased with the size of top electrode. The OFF-state is homogenous current conduction, but the ON-state is the filament of inhomogeneous current conduction.

3.2.5 Endurance

The tests of endurance of Ti/CuO/Pt devices are shown in Fig. 3.31-32, and endurance of Pt/CuO/Pt devices are shown in Fig. 3.33, and endurance of W-probe/CuO/Pt devices are shown in Fig. 3.34. The Fig. 3.31 show the Ti/ CuO/Pt devices of turn-on voltage and turn off voltage at different switching cycles, while the Fig. 3.32 show ON-state and OFF-state current at different switching cycles. It is found that the Ti/CuO/Pt devices can be switched over than 500 cycles under the operation of DC sweeps, and turn-on voltage variance are large than turn-off voltage.

The turn-on process is a random process. Because the DC sweeps produce more stress on the semiconductor devices than pulse switching, the Ti/CuO/Pt device expect to have more than 500 cycles with pulse operation.

3.2.5 Retention

Fig. 3.35 depict the retention of Ti/CuO/Pt devices examination. Both results show an excellent retention property of the device, in which the data of ON or OFF-states are retained after 10⁵ at room temperature.

3.2.5 Stress

The stress test of Ti/CuO/Pt devices displays that both ON and OFF state are not modified after stressed under 0.2 V for 6500 s, as shown in Fig. 3.36.

The stress test of Pt/CuO/Pt devices displays that both ON and OFF state are not modified after stressed under 0.2 V for 700 s, as shown in Fig. 3.37.

The stress test of Ni/CuO/Pt devices displays that both ON and OFF state are not modified after stressed under 0.2 V for 12000 s, as shown in Fig. 3.38.

3.3 Different current compliance

The Ti/CuO/Pt devices are measured by changing current compliance in order to investigate the relation between current compliance and resistive switching property.

3.3.1 Measurement of different current compliance

Fig. 3.39-46 show I-V curve of switching cycles at each current compliance;

1mA, 3mA, 5mA, 10mA, 20mA, 30mA, 40mA, 50mA by an Agilent 4155C Semiconductor Parameter Analyzer. The detail methods of the measure are illustrated as follow. After forming process, the current compliance is set at 1mA. The turn-on process make device to ON-state. Continuously, the turn-off process are increase negative bias step by step until the device turn off to OFF-state. This method can avoid unnecessary damage in turn-off process. Repeat turn-on and turn-off process until more than ten times. The current compliances are increased to 3mA, 5mA, 10mA, 20mA, 30mA, 40mA, 50mA. The turn-on and turn-off process are repeat at each current compliance value.

There are some resistive switching parameters be defined in order to convenient follow discuss. Fig. 1.2 shows those parameters in I-V curve. The turn-off voltage ( Voff ) is the voltage of the ON-state to OFF-state, the turn-off current ( Ioff ) is the current of the ON-state to OFF-state, and the stop voltage ( Vstop ) is the maximum voltage value of the sweep voltage range when turn-off process.

3.3.2 Discussion of current compliance

Fig. 3.47 shows the relation between current compliance and current at –0.2 V.

The upper error bars are ON-state current variation under switching cycles, while lower error bars are OFF-state current variation under switching cycles. The ON-state current increased with current compliance, and saturate at current compliance value 20mA. The ON-state current increased with current compliance, imply that when current compliance increased, the filaments are stronger than low current compliance value, and conduction cross area more large lead to ON-state resistance decrease.

Fig. 3.48 shows that the turn-off voltage increased with current compliance, and the variation of turn-off voltage is seldom. The insert shows the relationship of current compliance and turn-off voltage is not observed any power law, and the turn-off voltage is weak dependent current compliance.

Fig. 3.49 shows that the turn-off current increased with current compliance, and the variation of turn-off current is seldom. The dotted line mark current compliance equal to turn-off current. The insert shows the relationship of power law between current compliance and turn-off voltage, and the turn-off current is strong dependent current compliance. The current compliance influence turn-off current is more sensitive than turn-off voltage can be explained as follow. The increasing power

contribute from two term

2

2

P IR I I R Δ = Δ + Δ

Because ON-state is low resistance, first term can be drop. The current mainly contribute to power rather than voltage. This is suggested that power or current dominate turn-off process.

We define that turn-off power is the product of turn-off voltage and turn-off current ( Poff = Ioff*Voff). The turn-off power is the electric power to make device from ON-state to OFF-state. Fig. 3.50 shows that the turn-off power has linear relationship with current compliance, and the variation of turn-off current is seldom. The linear relationship between turn-off power and current compliance can be explained as follow. We make some assumption to explain the linear relationship. First, current density ( J ) is constant current density in filament, and weak dependent current compliance. Second, the effect switching thickness ( ds ) is almost constant and independent current compliance, because the filament has small voltage drop. Third, the defect concentration per volume is independent current compliance.

The effect switching region is modeling a cylinder, is composed with defect, with effect cross section area ( As ). At turn-on process, the current compliance ( Icomp ) is

comp s

I = ⋅ J A

On the other hand, the defect maybe is Cu atoms, Cu ions, oxygen vacancies or CuOx (x<1). The total defect number in cylinder filament is

Therefore, the number of defect proportion to current compliance. In another point of view, the current pass filament in turn-off process. This locally enhances electric field and current density, hence the Joule dissipation and the temperature, which in the further accelerates the dissolution process. This positive feedback is at the basis of the sharp current drop at turn-off process [42]. When the device switched turn-off, the defect will combine with oxygen to form high resistive CuO at the effect switching region. The turn-off power should proportion to number of defect as follow.

The turn-off power would proportion to current compliance.

comp off

I ∝ P

To further confirm Joule heating effect caused the filament rupture model, the various temperature measure to observe turn-off process. Fig. 3.51 shows that turn-off process with different temperature. There is a tendency towards turn-off power decay, as a result of temperature raise.

3.3.3 Relation between temperature and current

At ON-State, as shown in Fig. 3.52, the current is decreasing with raising measurement temperature. This trend indicates that the property of ON-state is metal-like behavior. It is owing to the formation of metallic filaments in thin film. At OFF-state, as shown in Fig. 3.53, an obvious trend is investigated. The current is

can indicate that the property of OFF-state is semiconductor-like or insulator behavior.

Fig. 3.54 shows that double log I-V curve of turn-on process at RT and 150^oC, and Fig.

3.55 shows that double log I-V curve of turn-off process at RT and 150^oC. The ohmic region of OFF-state is larger at 150^oC. The mechanism of ohmic region of OFF-state is different from the ohmic region of ON-state. The ohmic region of OFF-state is due to thermally-generated carrier, rather than injection carrier of metal-like.

3.4 Different stop voltage

The Ti/CuO/Pt devices at voltage bias more than reset voltage, and switching cycles increase, the transition region of the turn-off process would enhanced until stable saturate.

3.4.1 Measurement of different stop voltage

This phenomenon can be explained that over drive voltage induce defect near the effect switching region.

The device operate stop voltage is more than turn-off voltage. In few switching cycles, the sharp current abruptly drop from ON-state current to OFF-state current in turn-off process. The switching cycle increase the current drop gradually. The transition region observed in turn-off process. The Ti/CuO/Pt devices observed the behavior of transition region in turn-off process at bipolar switching operation mode.

Due to the transition region phenomenon presented during turn-off process, the I–V characteristics of the Ti/CuO/Pt devices were investigated with different spans of voltage scan for OFF-process. Bias voltage was swept as 0 V→1.5 V (current limited

at 5mA) → 0 V→ -0.8 V ( Vstop1 )→0 V→ -1 V ( Vstop2 )→0 V→ -1.5 V ( Vstop3 )→0 V. As the Vstop increased, the higher resistance value of OFF-state increased, which demonstrating the intermediate resistance state ( IRS ) was shown in Fig. 3.56.

In order to further confirm the resistive switching can transfer in different IRS and controlled by stop voltage, bias voltage was swept as 3 modes, 0 V→1.5 V (current limited at 5mA→ -0.75 V ( Vstop1 )→0 V→ -2 V ( Vstop2 ); 0 V→1.5 V (current limited at 5mA→ -0.8 V ( Vstop1 )→0 V→ -2 V ( Vstop2 ); and 0 V→1.5 V (current limited at 5mA→ -1V ( Vstop1 )→0 V→ -2 V ( Vstop2 ) as show in Fig. 3.57-59, respectively. The stop voltage 1 was changed with different model.

Next, I-V curve were repeatable measured at different sweep range; stop voltage ( Vstop ) was varied as -0.8 V, -0.9 V, -1V, and -2 V in turn-off region as show in Fig.

3.60-63. Fig. 3.64 shows that statistics plot of current at -0.2V of ON-state and OFF-state with various stop voltage. As Vstop decreased, hysteresis has shrunk, OFF-state approached to ON-state. This phenomenon can be explained that filament be complete ruptured as decrease Vstop.

3.4.2 Discussion of different stop voltage

In order to elucidate that the Vstop decreased with OFF-state resistance, we proposed a possible model to explain transition region occur in turn-off process. At first, the Vstop increased such that over drive voltage induce defect near the effect switching region. After few switching cycles, the amount of defect would raise near the Ti/CuO interface. When voltage bias is the turn-off voltage, the filaments were rupture by Joule heating. However, the voltage bias continue to increased in negative bias, the electric field effect wound enhanced. In the beginning, the filament would be

ruptured. As the voltage bias continue to increased in negative bias, the electric field would tend to formation more thin filament to form IRS. Following, the voltage bias increased induce current crowding to generate Joule heating, the thin filament rupture again. Repeat foregoing process until voltage bias to Vstop.

According to previous reports, Chen et al. suggested that small changes in oxygen concentration may result in large resistance change and the oxygen stoichiometry change induced by electric field might not be uniformly distributed in the interface region [35]. Lin et al. proposed that in the turn-on process, the biased electrons found one or few conducting paths composed of possible point defects including oxygen vacancies, ionic, and electronic defects, and the OFF-process was caused by the defects would trapped electrons, thus leading to the rupture of conducting paths [76]. Form Lin et.al discusses the resistive switching property of 0.1, 0.2, and 0.3% Mo-doping SZO thin films, the apparent transition region in turn-off process with the Mo-doping percentage increased [97]. This result can support our model. As Mo-doping percentage increased, the defect concentration would be increased to enhance transition region in turn-off process.

If our model is correct, the larger current compliance would cause defect be covered by larger filament. Fig. 3.67 shows the mechanism of turn off process with transition region different current compliance. It is predicted that transition region is unapparent with current compliance increased as show in Fig. 3.68-71.

Fig. 3.72 shows the stress characteristics of IRS. In this figure, the stress of IRS is worse than ON-state and OFF-state, because the IRS is thin filament. When the CuO film was stressed by voltage 0.2 V, the soft breakdown is more easily.

3.5 Possible resistive switching model

According to the literature reviews for the proposed resistance switching mechanisms of the NiOx film in a RRAM cell, it has been most commonly believed that the voltage stress creates multiple filamentary conducting paths in the NiOx film due to the nonnegligible Joule heating effect [34,42,51].They have strongly suggested that the formation and rupture of the filamentary conducting paths in the NiOx film are

According to the literature reviews for the proposed resistance switching mechanisms of the NiOx film in a RRAM cell, it has been most commonly believed that the voltage stress creates multiple filamentary conducting paths in the NiOx film due to the nonnegligible Joule heating effect [34,42,51].They have strongly suggested that the formation and rupture of the filamentary conducting paths in the NiOx film are

在文檔中 利用熱氧化法製備氧化銅薄膜於電阻式轉態記憶體之研究 (頁 65-0)