Thermogravimetric analysis (TGA)

To evaluate effect of low-temperature pyrolysis process we developed,

thermogravimetric analysis (TGA) was utilized to measure the weight of carbon-residue inside the ALD/MLD deposited superlattice. In this measurement, it was conducted by TA Instruments SDT-Q600 with a 100 sccm (standard cubic centimeter per minute) O2

flow.

Chapter 3

Metal oxide superlattice

3.1 Selection of dopants

For verifying the effects of Ti, Zr and Hf doping on thermoelectric properties of ZnO, we deposited four types of doped ZnO films by conventional process, i.e. con 49:1, con 24:1, con 19:1 and con 9:1 for each dopant, and the characteristics of them are summarized in Table 3.1.

Starting with the carrier concentration, it is noteworthy that the carrier

concentration in Ti:ZnO system was higher than those in the other two systems. This phenomenon could be originated from the actual dopant concentration and the distance between ionized dopants. The elemental composition measured by XPS and calculated atomic number density are listed in Table 3.2. Although we chose the precursors with the same functional groups for Ti, Zr and Hf, the reactivity of these precursors were still different owing to the formation energy difference of chemisorbed precursors. Because the formation of a chemisorbed precursor involved breaking a metal/ligand bond within precursor itself and then forming a metal/oxygen bond between precursor and surface hydroxyl group. Therefore, the reactivity of a precursor might be described by the difference between the bond energies (ΔBE) of metal/ligand and metal/oxygen. The bond energy of each bond is shown in Table 3.3. It could be discovered that the actual

dopant concentration indeed followed the trend of ΔBE, i.e. Hf > Zr > Ti, and so did the number density of each dopant. Based on effective field theory, the close-packed dopants are not able to donate free electrons freely owing to the Coulomb repulsion force between adjacent ionized dopants^68,97,98; thus, Ti:ZnO system had a higher doping efficiency. As for the decreasing electrical mobility for all structures, it was attributed to dopant scattering effect and the suppression of crystallinity. The XRD patterns of undoped ZnO and all con 9:1 doped ZnO are shown in Figure 3.1. The reduction of intensity for all peaks implies the suppression of crystallinity with the incorporation of dopants. Especially, (002) peak, which is parallel to direction of film growth and periodicity of dopants, decreases obviously and shifts to higher angle due to lattice shrinkage in z-axis caused by the smaller ionic radius of dopants compared to Zn, which are 56 pm, 73 pm, 72 pm and 74 pm for Ti, Zr, Hf and Zn, respectively. With great radius difference, the degree of change in the diffraction pattern for Ti:ZnO was most severe, so the electrical mobility reduced seriously as well. For the overall influence on electrical conductivity, it could be found that the electrical conductivity rose as the dopant concentration increased except for con 9:1 structure. Also, this result could be explained by effective field theory. Therefore, the carrier concentration only rose slightly even decrease when the concentration of dopant rose beyond a certain level, such as con 9:1 Ti:ZnO.

In the case of Seebeck coefficients, they basically followed Mott formula, which states the trade-off characteristic of the absolute value of Seebeck coefficient and electrical conductivity. Furthermore, it could be discovered that the Seebeck coefficient was less sensitive to the electrical mobility than the carrier concentration by comparing con 24:1 Zr:ZnO with con 19:1 Hf:ZnO and con 9:1 Ti:ZnO and con 9:1 Hf:ZnO.

Overall, the highest PF value occurred at con 49:1 structure in these three systems, which was only slightly doped, the carrier concentration was around 10 × 10¹⁹ and the electrical mobility was maintained as high as possible.

Referring to thermal conductivity, we initially assumed that Hf:ZnO might has a lower thermal conductivity attributed to the large mass difference. However, this hypothesis only held for Zr:ZnO and Hf:ZnO. With the almost same ionic radius, the effect of dopants on crystallinity was almost about the same; thus, the mass difference played a crucial factor in suppressing thermal conductivity. Thus, the thermal

conductivity of Hf:ZnO was lower than Zr:ZnO regardless of doping concentrations. As for Ti:ZnO, the effect of reducing crystallinity overwhelmed the influence of mass difference, so the thermal conductivity of Ti:ZnO showed a lower value when doping concentration rose. Especially, con 9:1 Ti:ZnO had the lowest thermal conductivity even the concentration of dopant was lower than Zr:ZnO and Hf:ZnO. For the overall

enhancement on ZT, although the degree of reduction for thermal conductivity was

almost 15-fold, the decreasing PF compensated the outstanding effort; thus, the ZT was only enhanced by the factor of 6.44 in con 9:1 Ti:ZnO.

In summary, these three tetravalent dopants were able to promote the electrical conductivity of ZnO by the factor of around 2.5 and suppress thermal conductivity by the factor of near 15, but the decreasing absolute value of Seebeck coefficient retarded the further enhancement in ZT, which was just about 6-fold enhancement. Although the degree of enhancement in ZT was about the same in these three system, we could still conclude that Hf might be the best dopant among them. Because the incorporation of Hf didn’t decrease the electrical mobility severely. This advantage ensured high electrical conductivity without increasing carrier concentration greatly, and thus the high absolute value of Seebeck coefficient could be held. Besides, the great mass difference

suppressed the thermal conductivity effectively.

Table 3.1: The thermoelectric characteristic of undoped ZnO, Ti:ZnO, Zr:ZnO and Hf:ZnO. n: carrier concentration and the unit is ×10¹⁹ cm^-3; μ: electrical mobility and the unit is cm² V^-1 s ^-1 ; σ: electrical conductivity, S cm^-1 ;κ: thermal conductivity, W m^-1

Table 3.2: The concentration of dopant measured by XPS and atomic number density of dopants. The number density is calculated by using composition ratio, film thickness and density^41,99.

O:Zn:dopant Number density of

dopant (nm^-2)

Con 9:1 Ti:ZnO 40.33 : 56.57 : 3.30 3.51

Con 9:1 Zr:ZnO 41.14 : 55.29 : 3.57 4.52

Con 9:1 Hf:ZnO 41.75 : 54.46 : 3.79 4.82

Table 3.3: Bond energy of metal/ligand and metal/oxygen and the difference between bond energies (ΔBE= BE metal/oxygen−BE metal/ligand).

Bond energy of

Figure 3.1: The XRD patterns of undoped ZnO and all con 9:1 doped ZnO.

3.2 Distribution Patterns of dopants

3.2.1 Conventional versus mixed ALD doping processes

Previous section disclosed that a close-packed dopant distribution was

unfavorable to enhance the electrical conductivity, so we developed a new ALD doping process, i.e. mixed ALD doping process. In the mixed ALD doping process, the dopant precursor, TDMAHf, was introduced into reaction chamber along with DEZn

simultaneously while depositing dopant layers. By this way, two precursors were going to compete with each other to react with surface hydroxyl groups, so the decrement of Hf number density in dopant layers was obtained. The schematic illustration of

depositing dopant layer by conventional process and mixed ALD doping process is shown in Figure 3.2. We also deposited types of mixed ALD doping process Hf:ZnO (HZO), the thermoelectric properties of them are summarized in Table 3.4.

First, it could be discovered that the carrier concentration of HZO deposited by mixed ALD doping process was much higher than that deposited by conventional process at the same repetition rate of dopant layers. Since the actual dopant

concentration was quite low in mixed ALD doping process, the number density of Hf in mixed ALD doping process was only around 0.6~0.7 nm^-2, which was much lower than that in conventional process, i.e. ~ 4.80 nm^-2. The dopant concentration measured by XPS and calculated number density is summarized in Table 3.5. In this situation,

dopants were much freer to donate free carriers and then became ionized ions without severe repulsion from other ionized dopants in the same dopant layer. For this reason, we could increase the carrier concentration of HZO by increasing repetition rate of dopant layer until dopants were no longer such free to donate free carrier owing to the repulsion from ionized dopants in the adjacent dopant layers, i.e. from mix 4:1 to mix 3:1. With regard to the electrical mobility, the decrement of dopants in mixed ALD doping process cut down the dopant scattering effect and reduced the suppression of crystallinity, so the electrical mobility in mixed ALD doping was higher than that in conventional process. Figure 3.3(a) shows the XRD patterns for comparing crystallinity of the same dopant layer period length deposited by two different process, i.e. con 24:1 versus mix 24:1. Nevertheless, Figure 3.3(b) reveals that densely inserted dopant layers still decreased the crystallinity in (002), especially. Besides, an abundance of free carriers raised the electron scattering, which may retard the transport of each other, so the electrical mobility decreased with carrier concentration, i.e. mix 3:1. Luckily, the overall effect of mixed ALD doping process on electrical conductivity was positive, the best value was almost doubled in comparison to conventional process, i.e. mix 14:1 versus con 24:1.

Referring to Seebeck coefficient, the benefit of mixed ALD doping process on promoting doping efficiency was unfavorable to maintain the high absolute value of

Seebeck coefficient, so the absolute value of Seebeck coefficient decreasd severely and the lowest value was only remain 40% compared to undoped ZnO, -44.62 and -107.49, respectively. Therefore, the PF value was quite low when dopant layers were densely inserted. Luckily, the PF value was greater than undoped ZnO when the period length of dopant layers was long, such as mix 34:1, mix 29:1 and mix 24:1. The highest PF

occured at mix 24:1 with the value of 1.90×10^-4 W m^-1 K^-2. Due to the low degradation to the electrical mobility in these structures, high carrier concentration was inessential for acquiring high electrical conductivity, so decrement on the absolute value of Seebeck coefficient could be suppressed. As the period length of dopant layers kept increasing, e.g. from mix 24:1 to mix 34:1, the PF value decreased and was close to the value of undoped ZnO.

Unlike its benefit to electrical performance, mixed ALD doping process showed a poor effect on reducing the thermal conductivity of ZnO. It could be seen that the degree of reducing thermal conductivity of mixed ALD doping process was much lower than that of conventional process at the same dopant layer repetition rate, i.e. ~2.13 for con 9:1 and ~8.48 for mix 9:1. As for the similar dopant concentration, e.g. mix 9:1 and con 49:1, conventional process was able to suppress thermal conductivity more

effectively even densely inserted mixed ALD doping dopant layers reduced the crystallinity more severely, which can be seen in Figure 3.3(c). This result was

attributed to the outstanding effect of more distinct interface on phonon scattering, so the conventional process could reduce the thermal conductivity much more effectively.

With regard to the overall effect on ZT, the best performance occured at mix 3:1, whose thermal conductivity was the lowest, and the degree of enhancement was about 4-fold.

Summing up two distribution patterns of dopant, it could be realized that close-packed dopants deposited by conventional process were able to form complete

heterogeneous interfaces and then had an advantage in reducing thermal conductivity.

On the other hand, sparsely distributed dopants deposited by mixed ALD doping process were able to increase carrier concentration efficiently and less harmed the electrical mobility. These two characteristics ensured a high PF value, such as mix 24:1.

In order to obtain the advantage of each process simultaneously for further enhancing ZT value, we were going to combine two processes and found out the best distribution pattern of the dopants.

Table 3.4: The thermoelectric characteristic of undoped ZnO, conventional process and mixed ALD doping process HZO. n: carrier concentration and the unit is ×10¹⁹ cm^-3; μ:

electrical mobility and the unit is cm² V^-1 s ^-1 ; σ: electrical conductivity, S cm^-1 ;κ:

Table 3.5: The concentration of dopant measured by XPS and atomic number density of Hf. The number density is calculated by using composition ratio, film thickness and density.

O:Zn:Hf Number density of Hf

(nm^-2)

Con 9:1 41.75 : 54.46 : 3.79 4.82

Con 24:1 45.71 : 52.87 : 1.42 4.83

Con 49:1 45.50 : 53.81 : 0.64 4.76

Mix 24:1 44.63 : 55.17 : 0.20 0.69

Mix 19:1 44.29 : 55.45 : 0.26 0.68

Mix 9:1 45.59 : 53.83 : 0.58 0.72

Mix 4:1 44.75 : 53.99 : 1.26 0.71

Mix 3:1 43.82 : 54.58 : 1.60 0.73

Figure 3.2: The schematic illustration of depositing dopant layer by conventional process and mixed ALD doping process.

Figure 3.3: XRD comparison for (a) same period length; (b) densely inserted mixed ALD doping; (c) same Hf content.

3.2.2 Combination of conventional and mixed ALD doping processes

As mentioned before, the combination of conventional process and mixed ALD doping process was potential to obtain the advantage of each way, i.e. heat blocking of conventional process and excellent electrical performance of mixed ALD doping process. From the results of last section, it revealed that the optimal electrical

performance was achieved by forming the dopant layers as a mixture of both ZnO and the HfO2 in a periodicity of 1 mixed monolayer per 24 ZnO monolayer since mix 24:1 showed the highest PF value. Based on this contention, we were going to use mix 24:1 as a basis structure and then insert complete HfO2 guest layers with different periodicity and thickness to enhance the suppression of thermal conductivity for mix 24:1 structure.

First, the optimal periodicity of inserted complete HfO2 layers was studied. Table 3.6 presents the thermoelectric properties of films composed of alternating mix 24:1 host and 5-cycle conventional HfO2 dopant layers, where the mix 24:1 host layers were of 1 period (24 ZnO cycles/1 mixed doping cycle/24 ZnO cycles), 2 periods

(24/1/24/1/24), or 3 periods (24/1/24/1/24/1/24); the total number of cycles of the HZO films with the 3 mix 24:1 period lengths were 540, 553, and 520, respectively. The illustrated structures of 3 films are shown in Figure 3.4. The results confirmed that further the enhancements of ZT were indeed achievable with the combination of two processes. The addition of the 5-cycle conventional dopant layers caused only minor

reductions in power factor and other electrical properties of the mix HZO, while it lowered the thermal conductivity to ~3 from ~10 W m^-1 K^-1, resulting in ~3 fold increase in ZT from that of mix 24:1 HZO. The optimal structure in terms of ZT was 1 period of mix 24:1 alternating with the conventional dopant layer, as the structures with 2 and 3 periods of mix 24:1 contained insufficient numbers of conventional dopant layers to effectively suppress thermal conductivity.

After optimizing the periodicity of the inserted HfO2 layers, the effects of cycle of the conventional HfO2 layers on thermoelectric performance were investigated. Figure 3.5 is the illustration of superlattices composed of 1 period of mix 24:1 HZO and varied cycles of conventional HfO2 layer, and the thermoelectric characteristics are

summarized in Table 3.7.

In order to elucidate the correlation between the deposition cycles of complete HfO2 guest layers and the properties of mix 24 / nH, in-situ QCM was utilized to

monitor the growth mechanism while switching materials. Figure 3.6(a) is the mass gain per cycle (MGPC) of HfO2 on the surface of 24 cycles ZnO which was the deposition procedure of mix 24 / nH. Owing to the less stereo-hindrance of DEZn, the number density of surface hydroxyl groups on ZnO is much higher than that on HfO2, so the MGPC of HfO2 on the ZnO surface was much higher in the initial cycles than that in the latter cycles. However, once a monolayer of HfO2 formed, i.e. 3 cycles in our result, the

MGPC of HfO2 decreased and approached to a constant value, which represented the steady ALD growth for HfO2 on the surface of itself. Because TDMAHf couldn’t react with the residual hydroxyl groups at the interface between ZnO and HfO2 owing to its bulkier structure. On the other hand, Figure 3.6(b) shows the MGPC of one ZnO cycle over the surface of a ZnO film (thickness = 24 cycles) primed with 0 to 30 cycles of a HfO2 layer. The MGPC of the ZnO cycle sharply decreased with increasing thickness of the HfO2 priming layer, again as a result of the substrate’s original ZnO surface being increasingly converted into a HfO2 one. The MGPC did not reach a steady value until ≥ 7 HfO2 cycles, indicating that it required ≥ 7 cycles of the HfO2 priming layer to fully block the incoming ZnO precursors from interfacing with the bottom ZnO layer. This result interprets that: at < 7 HfO2 cycles, the HfO2 layer apparently still contained molecular-scale voids—as can be discerned from the cross-sectional TEM images presented in Figure 3.7—to allow the incoming ZnO precursors (DEZn and H2O) to diffuse through and adsorb directly onto the bottom ZnO surface, which upon

subsequent ZnO deposition would conceivably result in molecular-scale ZnO conduits through the HfO2 layer, connected to the bottom ZnO layer. Such ZnO conduits would serve as electron-conducting channels through the HfO2 layer.

Based on this contention, the decrements of carrier concentration, electrical mobility and electrical conductivity as the number of cycles is less than 7 could be

explained as follows. Although the insulating nature of HfO2 and the much higher conduction band position of HfO2, some conductive ZnO channels inside the inserted HfO2 layers still offered paths for carriers to transmit through insulated barriers instead of tunneling; thus, there were still considerable amounts of free carriers. With regard to the degree of suppression on electrical mobility, mix 24 / nH was made up of a thin insulating layer (＜1 nm) and a thick (~8 nm) conductive domain which consisted of one segment of mix 24:1 and 24 cycles of ZnO. The conductive domain is called as super ZnO in the rest of dissertation for convenience. In this structure, super ZnO was thick enough to build up a high mobility path for carriers, so once electrons transmitted through the thin insulating barrier, their transportation in next super ZnO was about the same as that in the previous super ZnO. Therefore, the electrical mobility was only influenced slightly and then could be maintained at 14~16 cm² V^-1 s ^-1. As for Seebeck coefficients, since the reduction of carrier concentration in mix 24 / nH (n<7) was mainly attributed to the shrinkage of total transmission area, which is

energy-independent. Thus, the distribution of carriers with energy didn’t change and then the Seebeck coefficient was maintained at ~-64 μV K^-1.

As the cycles of inserted HfO2 was more than 7, conducting ZnO channels no longer existed inside the insulated HfO2 layers, so the carriers could only transmit through the barriers by tunneling effect. Thanks to the atomic-scale thickness of HfO2,

most of carriers were able to go across the barriers, so the decrement of carrier

concentration was still moderate. As for the electrical mobility, it was as the same as the case when cycles were less than 7, once electrons tunneled through the thin insulating barrier, their transportation in next super ZnO was almost as the same as that in pervious super ZnO, so the electrical mobility could still be maintained at ~14 cm² V^-1 s ^-1.

Besides, it is noteworthy that the absolute value of Seebeck coefficient increased abruptly when the cycles of HfO2 become 7 and then dropped. This phenomenon might result from energy filtering effect. Typically, energy filtering effect is conducted by a material whose conduction band is located around the Fermi level of the matrix material, which is mentioned in section 1.3.2. However, the great difference of conduction band position between HfO2 and ZnO, i.e. ~1.6 eV, was unfavorable for energy filtering effect. Luckily, 7 cycles of HfO2 was still quite thin, so it didn’t build up a thick barrier. For this reason, only carriers with low energy were blocked intensely but most of carriers with high energy were able to tunneling through barriers, so the energy-dependent tunneling effect may also be referred to energy filtering effect. However, as

Besides, it is noteworthy that the absolute value of Seebeck coefficient increased abruptly when the cycles of HfO2 become 7 and then dropped. This phenomenon might result from energy filtering effect. Typically, energy filtering effect is conducted by a material whose conduction band is located around the Fermi level of the matrix material, which is mentioned in section 1.3.2. However, the great difference of conduction band position between HfO2 and ZnO, i.e. ~1.6 eV, was unfavorable for energy filtering effect. Luckily, 7 cycles of HfO2 was still quite thin, so it didn’t build up a thick barrier. For this reason, only carriers with low energy were blocked intensely but most of carriers with high energy were able to tunneling through barriers, so the energy-dependent tunneling effect may also be referred to energy filtering effect. However, as

在文檔中 原子層沉積之金屬氧化物及高分子/金屬氧化物超晶格複合材料熱電性質及氣體滲透率研究 (頁 87-0)