Hysteresis in Lanthanide Zirconium Oxides Observed Using a Pulse CV Technique and including the Effect of High Temperature Annealing

A powerful characterization technique, pulse capacitance-voltage (CV) technique, was used to investigate oxide traps before and after annealing for lanthanide zirconium oxide thin films deposited on n-type Si (111) substrates at 300 °C by liquid injection Atomic Layer Deposition (ALD). The results indicated that: (1) more traps were observed compared to the conventional capacitance-voltage characterization method in LaZrOx; (2) the time-dependent trapping/de-trapping was influenced by the edge time, width and peak-to-peak voltage of a gate voltage pulse. Post deposition annealing was performed at 700 °C, 800 °C and 900 °C in N2 ambient for 15 s to the samples with 200 ALD cycles. The effect of the high temperature annealing on oxide traps and leakage current were subsequently explored. It showed that more traps were generated after annealing with the trap density increasing from 1.41 × 1012 cm−2 for as-deposited sample to 4.55 × 1012 cm−2 for the 800 °C annealed one. In addition, the leakage current density increase from about 10−6 A/cm2 at Vg = +0.5 V for the as-deposited sample to 10−3 A/cm2 at Vg = +0.5 V for the 900 °C annealed one.

and the results of the pulse CV technique and the conventional test method were compared. The focus of the present work is, therefore, on exploring the CV hysteresis of the oxides and the effect of high temperature annealing. An interesting correlation between annealing temperature and oxide traps, which provides a reference for the properties of the high-k thin films, will be discussed in the paper. Figure 1a shows the input signal and the output signal of the pulse CV technique [3]. The CV characteristics shown in Figure 1b are extracted from the results in Figure 1a and delta Vg is denoted in Figure 1b as well. The structure of the tested sample was shown in the inset of Figure 1a. From the observation of Figure 1b, there was a shoulder between −0.8 and −0.4 V. The shoulder was attributed to interface states that respond at lower frequencies and may originate from the La and Zr diffusion and mixing with the SiO2 interlayer during deposition [13,26]. In Figure 1a, we used a functional/arbitrary waveform generator to input a pulse voltage waveform (CH1) on the gate of the sample. The related current through the sample (itotal) was fed into a current amplifier and then was amplified as an output voltage signal (CH2). Both channels of an oscilloscope were used to track the input and output voltages. Delta Vg denoted in Figure 1b Figure 2, it is clear that there is an obvious hysteresis between the CV curves (ramp up and ramp down) measured by the pulse technique while there is almost no hysteresis in the conventional test. The large hysteresis is revealed by rapid characterization provided by the pulse CV technique, (in the order of several hundred micros), and the traps are not recovered completely before the end of the measurement. The pulse CV technique tracks the trap density in the oxide more accurately, or in other words, the conventional test underestimates the ∆V g trap density in the oxide. The verification of the influence of test time on change of delta Vg will be fully discussed later. We firstly give a possible explanation for the horizontal shift of the CV curves measured by the pulse technique. The shift of the curves is attributed to charge trapping/detrapping in oxide [25,27]. In our case, the charges existing in the oxide probably belong to as-grown fixed positive charges (oxygen vacancies) with high energy levels or probably are trapped in as-grow electron traps in the high-k oxide. When the gate is applied with a positive bias voltage (Vg > 0), electrons are de-trapped from the as-grown electron traps to the metal gate through tunneling and the net positive charges in the oxide lead to the CV curves shifting negatively. Similarly, if a negative voltage (Vg < 0) is applied on the metal side, the as-grown positive charges will be compensated by electrons from the metal gate and trapped in the as-grown electron traps. Thus, the negative charges shift the measured CV curves in the positive direction [28][29][30][31]. With regard to the test time, the edge time indicated in the inset of Figure 3a, it is critical for the characterization of traps in the oxide and we will discuss it in detail in conjunction with the experiment results shown in Figure 3a,b in this section. During the measurement, the rising edge gave an initial measurement before any stress was applied. Then, the falling slope of the pulse applied to the gate provided another measurement post stress. The stress was provided and determined by the width of the pulse. If the width of a pulse induced charge trapping in oxides, there would be a hysteresis between the ramp up and ramp down of the CV curves as described in Figure 1b. In addition, the change of the edge time leads to the change of the test time. Longer edge time means more test time at each voltage bias. When the edge time was increased to a relatively large value, or relative long test time (several s), the pulse CV technique may be considered to be the same as the conventional CV test technique carried out using a LCR meter, (the Agilent 4284A in our case). This hypothesis was consistent with the results shown in Figure 3a,b. In Figure 3a, it is obvious that the extracted trap density, Not, was reduced from 2.48 × 10 12 cm −2 with edge time of 300 μs to 0.7 × 10 12 cm −2 with edge time of 900 μs. The Not was calculated from the formula, Not = ΔVg × Cox/q, assuming these traps were located at the silicon and oxide interface. The above analysis implies that the trapped electrons/holes were recovered with an increase of the measurement time, in turn, leading to the partial reduction of trapped charges. Figure 3b graphically summarizes the relationship between trap density and edge time of a high-k material (LaZrOx) and a silicon dioxide. For the silicon dioxide gate dielectric, it has been shown that there are no as-grown electron traps in the oxide and no hysteresis is observed whatever the edge time is [3]. Thus, the silicon dioxide sample was used as a reference in this experiment. By adjusting the edge time, the time-dependent trapping/de-trapping was tracked correspondingly. For the high-k material in this research, Not was reduced with an increase of the edge time as shown in Figure 3b. Due to the limitations of the function generator (RIGOL DG2041A), the longest edge time used was 900 micro s. However, from the trend of Not, it can be estimated that if the edge time was long enough, say, one second, the trap density measured by pulse CV technique would be equal to that of the conventional CV test as shown in Figure 2. The dashed line in Figure 3b indicates the behaviour of Not with edge time. It can also be concluded that the recovery of trapped electrons/holes are sensitive to the measurement time from the change of Not with variation of test time. To be precise, the short test will give the large Not. Therefore, the time-dependent trapping/de-trapping should be probed by short edge time in the pulse CV measurement, which is able to access shallower traps (traps at a high energy levels within oxide), at least for the timescale considered here. While for the longer values of edge time, the transient shift of the Not is attributed to slow electron trapping/de-trapping [3].

Results and Discussion
After the examination of the edge time, the research focused on the influence of the stress time, or the width of a pulse, on trap density. The stress time (pulse width) is indicated in the inset of Figure 4. Figure 4 shows that charges trapped in the oxide were highly dependent on the stress time with Not of 4.14 × 10 12 cm −2 , 3.17 × 10 12 cm −2 , 2.48 × 10 12 cm −2 for stress time of 4500 μs, 2500 μs and 800 μs, respectively, at an initial state (edge time of 300 μs). It may be that the longer width, or stress time, the larger the electron fluency supplying the charge into traps in the oxide, this in turn, leads to more trapped charges. Also, the figure indicates that the recovery of Not with the edge time for the three curves shares similar trends regardless of the initial charges trapped in the oxide. This result also indicates that the de-trapping process is highly dependent on the test time. Finally, the pulses with various VPP were applied to the gate of the MOS capacitors to investigate the effect of VPP on charge trapping. The pulses used in the experiment did not have any voltage offset. In other words, if VPP was 4 V, then the pulse started from −2 V and ended with 2 V. Figure 5 concludes the relationship between extracted trap density and edge time under different VPP. It shows that for larger VPP stress (4 V), the larger Not (3.58 × 10 12 cm −2 ) was observed because the traps in the deeper/higher energy level in the oxide were also tracked by the higher voltage [3]. Interestingly, for VPP = 4 V, the extracted trap density was saturated with small edge time, 400 μs in this case, which meant that all the traps at the corresponding energy level were fully tracked provided that the edge time was below 400 μs at this voltage level.
The traps in the oxide characterized by the pulse technique were discussed above, and it was found that more traps were tracked by the pulse CV characterization method compared with the conventional one (e.g., using an LCR meter with an analogue ramp). Also, the time-dependent trapping/de-trapping was traced by adjusting the edge time, width and peak-to-peak voltage of a voltage pulse. If the edge time (test time) was long enough, the result obtained by the pulse CV technique was the same as the conventional method, which provided a calibration of the new technique. Before the discussion of high temperature annealing effect on the oxides, the dielectric constant of the as-deposited sample was calculated. Firstly, the stack thickness of about 22 nm was measured by ellipsometer. Then, based on the thickness and the size of the electrode contacts, diameter of 0. 3  was estimated to be 27 approximately. From our previous work on ALD deposition and other research outputs using water as the oxidizing agent, a thin SiOx film was formed at the film/Si substrate interface [3,19,32]. It is likely that the overall dielectric constant of the stack was limited by the presence of such an interfacial layer. The above analysis was based on the as-deposited sample and the annealed ones will be discussed in the following sections. In order to examine the effect of the high-temperature annealing on the oxides, post-deposition annealing was performed at 700 °C, 800 °C and 900 °C in N2 ambient for 15 s. The CV characteristics for the annealed samples are presented in Figures 6 and 7 in order to illustrate the horizontal shift and vertical change of the CV curves, respectively. Due to the unacceptable distortion of the CV curves caused by the large leakage current for the sample annealed at 900 °C, only CV curves from the as-deposited, 700 °C annealed and 800 °C annealed samples are presented in Figure 6. The delta Vg is indicated in each curve as well. From the delta Vg of the curves in Figure 6a-c, it was obvious that the RTA led to the increase of trap density, from 1.41 × 10 12 to 4.55 × 10 12 cm −2 . The increase of trap density was probably caused by the increase of oxide trap, interface state and border trap density. Other researchers also found similar results, which suggested that as the oxide-trap charge increased after the high temperature annealing, a concomitant increase was also observed in the interface and border-trap densities [33]. From Figure 7, it was discovered that the accumulation capacitance, or CET, of the high-k/IL stacks decreased with the increase of annealing temperature, which was mainly due to the thickness increase of the SiOx interfacial layer after RTA. This phenomenon also occurred in our previous work which showed that the interfacial layer, SiOx, was increased after 900 °C annealing attributed to either internal or external oxidation mechanism [34]. In addition, the CV curve was stretched out with annealing temperatures, which indicated the increase of interface states by either silicate formation or diffusion [33,35]. The two figures (Figures 6 and 7) indicated that annealing conditions should be controlled carefully, otherwise some disadvantages, such as the increase of the oxide traps, interface state and of the IL thickness, can be caused after annealing.  Figure 6a (square symbols) as comparison. The result implies that the delta Vg (or trap density) will increase with the RTA temperature. Figure 8 provides the XRD diffraction patterns (with normal angle) for the as-deposited, 700 °C annealed, 800 °C annealed and 900 °C annealed samples to determine the morphology of the films. From the observation of the pattern for the as-deposited one, the thin films was poly-crystalized with weak reflections. The planes for diffraction peaks were indexed in the figure. With the increase of the RTA temperature, the diffraction peaks became stronger and shaper. In addition, more noticeable diffraction peaks appeared when the RTA temperature reached 900 °C. Noticeably, for the RTA temperature of 900 °C, a diffraction pattern for ZrSiO4 was observed, implying the diffusion of the elements occurred at the interface and zirconium silicide was formed [36,37]. The behaviour would deteriorate the interface and enhance the increase of leakage current, which was consistent with the result in Figure 9.    It is obvious that the leakage current will increase dramatically with the annealing time. The 900 °C annealing one, with 10 −3 A/cm 2 at Vg = +0.5 V, is 1000 times than that of the as-deposited one.

Experimental Section
The n-type Si (111) wafers with resistivity of 1-10 Ω·cm were used as the substrates of MOS capacitors in this experiment. The wafer was cleaned following the standard Radio Corporation of America (RCA) cleaning procedures and dried by N2 gun. After cleaning, the lanthanide zirconium oxide thin films were deposited by ALD at the substrate temperature of 300 °C using (C2H6N)4Zr and (C11H19O2)3La as precursors. The temperatures for zirconium and lanthanum precursors were 130°C and 100 °C, respectively. The 50 °C deionized water (DI water) was served as the oxygen source. For deposition, nine cycles of water (30 ms)/purge (20 s)/(C2H6N)4Zr (1000 ms)/purge (25 s) were followed by one cycle of water (30 ms)/purge (20 s)/(C11H19O2)3La (300 ms)/purge (25 s). In other words, the cycle ratio of zirconium oxide to lanthanum oxide was 9:1. The whole sequence was repeated 20 times and the total number of cycles was 200. Pure N2 was used as both carrier gas and purge gas. After the deposition of gate dielectric, the wafer was sliced into four pieces. For three of them, the rapid thermal annealing (RTA) at 700 °C, 800 °C and 900 °C, respectively, for 15 s was performed in N2 ambient before the deposition of contact gates. Then, XRD (with normal angle) analysis was carried out using Rigaku miniflex diffractometer (Rigaku, Tokyo, Japan) with CuKα radiation (0.154051 nm, 40 kV, 50 mA), spanning a 2θ range from 30° to 60° at a scan rate of 1°/min for all samples. The thickness of each thin film was measured by ELLIP-SR-1 ellipsometer with the incident angle of 65°and wavelength from 300 to 850 nm with the step of 10 nm. The aluminum electrode contacts with a diameter of 0.3 mm and thickness of 500 nm were deposited by an E-beam evaporation. The backside was deposited with aluminum as well after the treatment of a diluted HF solution to form ohmic back contact. The fabricated structure of the sample was shown in Figure 10. Aglient 4284A precision LCR meter, Keithley 487 picoammeter and a useful developed pulse CV system [3] were employed to investigate the electrical properties of the samples. All of the electrical measurements were performed in the dark at room temperature with the Faraday Cage surrounding the wafer prober.

Conclusions
Lanthanide zirconium oxide thin films were deposited on the n-type Si (111) substrates by Atomic Layer Deposition. Overall, dielectric constants of the stack of 27 were estimated. In order to explore the influence of rapid thermal annealing, post deposition annealing was performed at 700 °C, 800 °C and 900 °C, respectively, in N2 ambient for 15 s. A useful characterization technique, pulse capacitance-voltage technique, was employed to be a powerful tool to track the traps in the oxides. It was found that the traps in the stack was highly dependent on the edge time, width and peak-to-peak voltage of a gate voltage pulse. The recovery of trapped electrons/holes were sensitive to the measurement time and the short test would give the large trap density. For the width time, the longer width, or stress time, the larger the electron fluency supplying the charge into traps in the oxide, this, in turn, led to more trapped charges. With the higher voltage level, more traps were detected since the traps in the deeper/higher energy level in the oxide were tracked. The influence of the high temperature annealing on the thin films as studied as well. It showed that more traps were generated after annealing with the trap density increasing from 1.41 × 10 12 cm −2 for as-deposited one to 4.55 × 10 12 cm −2 for 800 °C annealed one. In addition, the leakage current was also increased. The leakage current density of the 900 °C annealed sample (about 10 −3 A/cm 2 at Vg = +0.5 V was 1000 times larger than that of the as-deposited film (around 10 −6 A/cm 2 at Vg = +0.5 V. The X-ray diffraction was carried to give the comprehensive analysis. The XRD patterns showed that with the increase of the RTA temperature, the diffraction peaks became stronger and shaper. More noticeable diffraction peaks appeared when the RTA temperature reached 900 °C. This behaviour supported the results of the increase of leakage current density.