Energy-Efficient Annealing Process of Ferroelectric Hf0.5Zr0.5O2 Capacitor Using Ultraviolet-LED for Green Manufacturing

Thermal annealing process plays an important role in the formation of ferroelectric phase in Hf0.5Zr0.5O2 (HZO) thin films. In this study, the annealing process of the HZO capacitors is demonstrated using ultraviolet (UV)-LED, for the first time. Since the absorptance of the HZO films with TiN electrodes is highest in UV region, the UV-LED annealing process is promising to achieve a much more energy-efficient annealing process than a conventional halogen lamp RTA method. It was experimentally confirmed that UV-LED annealing reduces the energy consumption by nearly half compared to the conventional method. The ferroelectric characteristics obtained by this method are comparable to those achieved by the halogen lamp RTA process at 400-450°C. Grazing incidence X-ray diffraction (GIXRD) pattern shows that no monoclinic phase is formed and only the tetragonal and orthorhombic phases are confirmed. It is also confirmed that there is the in-plane tensile stress remaining after the UV-LED annealing process, which is necessary for the formation of the ferroelectric orthorhombic phase.


I. INTRODUCTION
Since the first report of ferroelectricity in HfO 2 -based thin film, it has been drawing much attention for emerging memory devices because of its CMOS process compatibility and thickness scalability [1], [2], [3], [4].In HfO 2 -based material, thermal annealing process promotes the formation of the polar orthorhombic phase (o-phase, Pca2 1 ), which shows the ferroelectricity [5], [6].The as-deposited thin film is in amorphous state.During thermal annealing process, crystal nucleation and growth occurs and the tetragonal phase (t-phase, P4 2 /nmc) becomes dominant at high temperature.During the cooling down step, the t-phase transforms to the stable monoclinic phase (mphase, P2 1 /c), or the metastable o-phase.It is reported that the tensile stress along the c-axis direction of the t-phase facilitates transition from the t-phase to the ophase [7], and the residual in-plane tensile stress is observed in the ferroelectric film after annealing process [8], [9], [10].
Conventionally, halogen lamp (HL) has been used for RTA process.However, HL has broad spectrum and wastes energy in annealing process depending on the absorption spectrum of the film.As shown in Fig. 1, absorption in ferroelectric TiN/Hf 0.5 Zr 0.5 O 2 (HZO)/TiN capacitors on Si is only about 30% in IR region, and it increases at shorter wavelength, reaching the highest value of over 80% at around 400 nm.Therefore, thermal annealing with shorter wavelength than  HL, such as UV-LED annealing, can enable highly energyefficient process toward green manufacturing.
In this study, thermal annealing of the HZO capacitors using UV-LED is demonstrated [19].The obtained ferroelectric characteristics, including remanent polarization (P r ), coercive voltage (V c ), and switching speed, are compared with those obtained by the conventional HL annealing process.Also, X-ray diffraction analysis is performed to evaluate the composition of the m-, o-, and t-phases, and the residual stress in the film after annealing.

II. EXPERIMENTAL METHOD
Metal-ferroelectric-metal (MFM) capacitors were fabricated on Si substrate, as shown in Fig. 2. The capacitors consisted of 30-nm-thick TiN layers for both top and bottom electrodes deposited by RF sputtering, and a 10.2-nm-thick HZO film in between them.The HZO films were deposited by atomic layer deposition (ALD) at 250 • C. The films were annealed by HL or UV-LED annealing in N 2 ambient.During the annealing process, the sample was set on a Si carrier wafer.Temperature of the backside of the carrier wafer was monitored using a thermocouple.For both HL and UV-LED annealing, annealing temperature was 350-500 • C, as shown in Fig. 3.The temperature profile was matched for both annealing by careful monitoring and PID optimization.After annealing, 100-nm-thick Al was deposited and patterned with the top TiN electrode.

III. RESULTS AND DISCUSSION
First, we measured static characteristics of the HZO capacitors after wake-up electrical cycles (±3 V, 100 kHz 10 5 times).Figs. 4 and 5 show the PV hysteresis curves of the MFM capacitors annealed by HL and UV-LED and the corresponding 2P r values.Ferroelectric behavior was not observed in the samples annealed at 350 • C for both annealing method, which indicates that annealing temperature is not high enough to form the ferroelectric phase.When the annealing temperature was between 400 and 500 • C, ferroelectric behavior was observed and 2P r was around 50 μC/cm 2 .Fig. 6 shows the endurance properties.The endurance of more than 10 7 cycles was obtained for the samples annealed at 400 and 450 • C.However, the endurance was reduced when the annealing temperature increased to 500 • C, which can be explained by the increase of defect concentration, diffusion of defects, and migration of electrode material into HZO films due to higher temperature annealing [20].
Although the endurance was the same within the measured range and 2P r values were nearly the same for the samples annealed at 400 and 450 • C, there was difference in   coercive field (2E c ). 2E c of the samples annealed at 400 • C were 1.9 MV/cm, whereas those annealed at 450 • C were 1.7-1.8MV/cm.Therefore, the samples annealed at 450 • C is preferable for low voltage operation to those annealed at other temperatures.Fig. 7 compares the leakage currents and the breakdown voltages of the pristine samples annealed by HL or UV-LED at 450 • C. The IV characteristics were measured by the waveform shown in the inset.The measured currents include both leakage currents and polarization switching currents.The polarization switching currents are mainly observed when the voltage is less than 2 V.The leakage currents were nearly identical for two annealing methods.The breakdown voltages were 4.16 ± 0.14 V and 4.08 ± 0.10 V for the samples annealed by HL and UV-LED, respectively.Therefore, the static characteristics of the samples annealed by HL and UV-LED were comparable.
Next, switching speed of the MFM capacitors were investigated [21].The switching speed is strongly affected by imprint effect, which is the shift of the PV hysteresis curves along the voltage axis [22], [23], [24].Therefore, the imprint effect was first measured.Fig. 8 shows the waveform for imprint measurement and the measured voltage shift V shift of the hysteresis curves, defined as V shift = (V c+ +V c− )/2, where V c+/− are positive and negative coercive voltages.It was confirmed that the voltage shift occurred when the interval time between the set pulse and the measurement was the order of 100 μs or more.When the interval was 30 μs, V shift was around −0.1 V for both polarity, which indicates that the bias voltage is not negligible even when the imprint effect is eliminated.This would be caused by the asymmetry in the capacitor structure [21], [25], [26].Although the capacitor structure (TiN-HZO-TiN) is nominally symmetric, the interface between HZO and the bottom electrode can have higher oxygen vacancy concentration and thicker interfacial layer than the top electrode because it is subjected to a larger thermal budget.
In order to eliminate the imprint effect from the switching speed measurement, the waveform of the switching speed measurement was carefully designed as shown in Fig. 9(a): after the previous imprint effect is erased by the reset pulses, the pulse width and interval between each pulse is set shorter than 100 μs, except for the set pulse.Fig. 9(b) shows the measured switching curves with various applied electric field (E FE ) and the width of the set pulse (t set ).
In order to compare the switching speed of the MFM capacitors annealed by the two annealing methods, the contour plots of the normalized polarization for both positive and negative set pulses are replotted from the switching curves as shown in Fig. 10.For both positive and negative set pulses, there is no significant difference in the isopolarization lines between the two annealing methods, which indicates that the switching speeds of the samples annealed by the two methods are nearly the same.However, for both annealing methods, switching speed with the positive set pulse is slightly faster than that with the negative set pulse.The difference in the switching speed can be ascribed to the remaining bias voltage after the reset pulses.Since the negative bias voltage remains after the imprint effect is eliminated by the reset pulses, the effective voltage of the negative set pulse imposed on the ferroelectric layer becomes lower than that of the positive set pulse.
Next, we investigated the effect of soak time of the annealing on the energy consumption for both HL and UV-LED annealing.In this experiment, the input power was fixed to the maximum power during the ramp-up time, and then controlled to keep the target temperature during the soak time.Therefore, the temperature profile of LED annealing does not match with that of HL annealing.The maximum input power of HL was 780 W, and that of UV-LED was 780 W or 1200 W. The target temperature was 450 • C. Fig. 11(a) shows the energy consumption during the annealing process for each soak time relative to that of HL annealing with a soak time of 15 s.It was confirmed that UV-LED annealing reduces the energy consumption by nearly half.When the maximum input power was the same, the rampup time of the UV-LED annealing was much faster and the power consumption was lower than those of HL annealing, whereas the PV characteristics were similar.Therefore, UV-LED annealing is preferable for energy-and time-efficient annealing process to HL annealing.Also, it was indicated that the energy consumption was not largely affected by the maximum input power because the ramp-up time was roughly in inverse proportion to the maximum input power.
Next, X-ray diffraction analysis was performed for the samples annealed by UV-LED at 450 • C for 15 s.First, we measured the grazing incidence x-ray diffraction (GIXRD) pattern and performed Rietveld refinement, which can be utilized to analyze the composition of the m-, o-, and t-phases in the HZO thin film [27], [28], [29], [30], [31].Fig. 12 shows the measured GIXRD pattern and the fitting curve obtained by Rietveld refinement.The incidence angle was fixed at 0.5 • , and 2θ was ranged between 15 and 90 • with 0.05 • step.The measured pattern was analyzed using a Rietveld refinement software, RIETAN-FP [32].In order to eliminate the diffraction peak ascribed to the Si wafer, the data in the range of 50.7-51.5• and 53.5-53.75• were omitted from the analysis.
The obtained composition was 1.2±0.5% of the m-phase, 17.4±2.1% of the o-phase, 22.8±2.5% of the t-phase and 58.6±6.4% of TiN.Therefore, only an insignificant amount of m-phase was formed during the UV-LED annealing process.
Finally, we compared the GIXRD pattern and the in-plane XRD pattern to investigate the residual in-plane stress in the HZO film after the annealing process.In both methods, the diffraction peak corresponding to o(111) or t(011) was clearly observed at around 30 • , as shown in Fig. 13.It was observed that the diffraction angles of the o(111) and t(011) peak of the in-plane XRD pattern was lower than those of the GIXRD pattern.The corresponding lattice spacing is 2.914 Å and 2.959 Å for GIXRD and in-plane XRD, respectively.Therefore, the lattice spacing measured by in-plane XRD is 1.5% larger than that by GIXRD, which indicates that the in-plane tensile stress remains after the UV-LED annealing process.Since the Young's modulus of HZO is 185-224 GPa [33], the remaining tensile stress can be on the order of GPa, which is sufficient to promote the transition from t-phase to o-phase [7].Therefore, this result is consistent with the observed ferroelectric behavior of the samples annealed by UV-LED.

IV. CONCLUSION
The ferroelectric characteristics of the HZO capacitors annealed by conventional HL and UV-LED were investigated.Both static and dynamic characteristics obtained by UV-LED annealing were comparable to those obtained by HL annealing.The GIXRD pattern and its Rietveld refinement result show that only the t-and o-phases were formed during the UV-LED annealing process.Also, the difference in the peak position of GIXRD and in-plane XRD shows that there is the residual in-plane tensile stress in the HZO film after the annealing process, which is important to form the o-phase.
Since the energy consumption of UV-LED annealing was much lower than that of HL for the HZO capacitors with TiN electrodes, UV-LED annealing process is promising as energy-efficient annealing process toward green manufacturing.

FIGURE 1 .
FIGURE 1. Absorption spectrum of TiN/HZO/TiN capacitors on Si substrate, (solid line) measured by spectrophotometry, (dashed line) calculated using the transfer-matrix method.

FIGURE 3 .
FIGURE 3. Temperature profile of HL and UV-LED annealing.

FIGURE 5 .
FIGURE 5. Remanent polarizations of the HZO films annealed by HL and UV-LED.

FIGURE 7 .
FIGURE 7. Leakage currents and breakdown voltages of the samples annealed by HL or UV-LED at 450 • C (10 samples for each).

FIGURE 8 .FIGURE 9 .
FIGURE 8. (a) Waveform for imprint measurement.(b) Measured V shift of the HZO film annealed by UV-LED at 450 • C.

FIGURE 10 .
FIGURE 10.Contour plots of normalized polarization for positive set pulse with the samples annealed by (a) UV-LED (b) HL, and for negative set pulse with the samples annealed by (c) UV-LED, and (d) HL.Iso-polarization lines where normalized polarization charge is 0.5 are drawn by black lines.

FIGURE 11 .
FIGURE 11.(a) Energy consumption of HL and UV-LED annealing with different soak time relative to that of HL with a soak time of 15 s.(b) Temperature profile when the soak time is 15 s.(c) PV hysteresis curves when the soak time is 15 s.

Fig. 11 (
Fig. 11 (b) and (c) show the temperature profile and PV hysteresis curves when the soak time is 15 seconds, respectively.When the maximum input power was the same, the rampup time of the UV-LED annealing was much faster and the power consumption was lower than those of HL annealing, whereas the PV characteristics were similar.Therefore, UV-LED annealing is preferable for energy-and time-efficient annealing process to HL annealing.Also, it was indicated that the energy consumption was not largely affected by the maximum input power because the ramp-up time was roughly in inverse proportion to the maximum input power.Next, X-ray diffraction analysis was performed for the samples annealed by UV-LED at 450 • C for 15 s.First, we measured the grazing incidence x-ray diffraction (GIXRD) pattern and performed Rietveld refinement, which can be utilized to analyze the composition of the m-, o-, and t-phases in the HZO thin film[27],[28],[29],[30],[31].

FIGURE 12 .
FIGURE 12. GIXRD pattern of the sample annealed by UV-LED at 450 • C for 15 s, and the fitting curve by Rietveld refinement (R wp =11.8%,S=1.8).

FIGURE 13 .
FIGURE 13.GIXRD and in-plane XRD patterns of the sample annealed by UV-LED at 450 • C for 15 s.