• 沒有找到結果。

Chapter 1 Introduction

1.4 Thesis Organization

In this thesis, a novel method to fabricate NW-TFT was originally proposed by

Advanced Device Technology Laboratory (ADTL), NCTU [47]. To further improve the

characteristics of the NW devices, the main idea of this thesis is to apply LTPS technology

to form crystalline Si channel. Without involving expensive lithography equipment in the

process, TFT devices with nano-scale channels can be achieved and are compatible with

modern semiconductor manufacturing. Performance comparisons with PRTA MILC and

conventional furnace-annealed SPC, and MILC will be explored and addressed.

The overview of LTPS-TFT and NW devices is mentioned in Chapter 1. In Chapter 2,

we briefly depict the device structure and process flow. In Chapter 3, NW-TFTs fabricated

by PRTA MILC with different annealing conditions are characterized. In Chapter 4,

comparison and analysis of different strategy in crystallizing are described. Finally, we

summarize the conclusions and the suggested future work in Chapter 5.

Chapter 2

Device Fabrication and Measurement

2.1 Device Structure and Process Flow

The fabrication began by depositing a 100nm in-situ doped n+-poly Si gate material on

6-inch (100)-Si substrate capped with a 100nm oxide layer. Then a 35nm-thick TEOS

serving as the gate oxide and 100nm amorphous Si (a-Si) layer were sequentially deposited

by using a low–pressure chemical vapor deposition (LPCVD) system. After ion

implantation by P31+ with a dose of 1015 cm-2 at 15 keV, source/drain (S/D) regions were

defined by photolithography. An isotropic reactive ion etching process was subsequently

employed to remove the a-Si, and the NW channels were simultaneously formed on the

sidewall of the poly-Si gate. A 100nm-thick low-temperature oxide (LTO) was next

deposited. In some devices, either one-sided or two-sided metal-induced lateral

crystallization (MILC) seeding windows were defined on the surface of the heavily doped

regions (i.e., the source or drain regions) and opened through the LTO, followed by the

deposition of a 5 nm-thick nickel (Ni) by a physical vapor deposition (PVD) system

serving as the seeding layer. Wafers were then divided into several splits according to the

crystallization conditions.

(1) NW-TFTs

One of the splits was fabricated by pulsed rapid thermal anneal (PRTA) MILC

crystallization to transfer the a-Si channel into polycrystalline. Using a specific heating

pattern with very short annealing pulses, a much faster crystallization rate at a smaller

thermal budget can be achieved. Wafers were then treated by PRTA in a rapid thermal

processor. During channel MILC annealing, the heating lamp was turned on for 10 seconds

and off for 30 seconds periodically 60 times in N2 ambient. A peak annealing temperature

was set at 650°C, 675°C, or 690°C, respectively. Other samples were annealed in

conventional furnace tube at temperature of 525°C, in N2 ambient for comparison purpose.

The unreacted Ni was subsequently removed in an H2SO4/H2O2 solution at 120°C for 10

minutes, followed by the deposition of a 200nm-thick LTO over all wafers before contact

holes were opened. An additional anneal step in N2 ambient at 600°C for 12 hr was adopted

to ensure the activation of the dopants in the S/D region. The fabrication was completed

after the formation of test pads using standard metallization steps. The main fabrication

flow of NW-TFT is shown in Fig. 2-1.

(2) NW-TFTs with Nitride capping layer on mask oxide

The other split was fabricated by Nitride-capping PRTA MILC crystallization. The

processes were basically identical to that described in the previous section, except the

seeding window opening. During channel MILC annealing, peak annealing temperature

was individually set at 625°C, 650°C, or 675°C in a rapid thermal processor for 5 or 10

seconds, and off for 30 seconds periodically 60 times, respectively.

Figure 2-2 illustrates (a) top view of the NW-TFT and (b) the definition of nanowire

width and thickness used. In this experiment, the NW channels were formed by RIE system.

Their feature size is determined by several factors, including height of the side gate, a-Si

thickness, and over-etching time. In this work, the NW channel width and thickness are 45

nm and 60nm, respectively, as shown in the transmission electron microscopy (TEM)

image in Fig. 2-3 and Fig. 2-4. Also note that a parameter called offset, OS, is defined as

the horizontal distance between the MILC open window and the channel. In this study, the

condition for OS is 1 um.

2.2 Measurement Setup and Electrical Characterization

Electrical characterization on the NWFETs was performed using an HP 4156

semiconductor parameter analyzer system. In all measurements, the temperature was

controlled and fixed at a stable value by temperature-regulated hot chuck. From the

measured ID-VG curve at VG = 0.5V, key parameters including threshold voltage (Vth),

subthreshold swing (SS), and field-effect mobility (μFE) were extracted based on the

following relations.

In this study, a constant-current method was employed to determine the threshold voltage,

in other words, the threshold voltage is defined as the gate voltage as the drain current ID

reaches the value of (W/L)×100 nA, or.

nA

L I W V

Vth = G@ D = ×100 (2-1),

where W, L are the channel width and length, respectively.

The subthreshold swing could be calculated from the subthrshold current in the weak

inversion region by

Finally, the filed-effect mobility (μFE) is determined by using the following equation:

where gm is the maximum transconductance and Cox is the gate oxide capacitance per unit

area.

Chapter 3

NW-TFTs Fabricated by Pulsed RTA MILC

3.1 Material Analysis of PRTA Grain Growth

For the material analysis, a 5 nm-thick Ni seeding layer was deposited onto a selected

a-Si region which is called seeding window with dimension of 90×90 μm2.

Recrystallization of the amorphous silicon was subsequently performed to observe the

PRTA MILC characteristics, where all samples were annealed in a RTA processor (AG heat

pulse 610 system) with N2 ambient for 60 cycles with each cycle consisting of 10 seconds

on-period and 30 seconds off-period. Various annealing temperatures, i.e., 650 °C, 675 °C,

690 °C, 700 °C, or 725 °C, were set within every operational cycle. Figure 3-1 illustrates

the heating pattern of the applied temperature which is measured in the chamber during the

annealing process. However, MILC growth rate during cooling interval at 400 °C is less

than 0.1μm/h [48] and can be ignored.

The length of the MILC region formed by PRTA in different annealing conditions is

shown in Fig. 3-2. The value was measured by an optical microscope and was defined as

the length from the center of seeding window edge to the MILC front. Since SPC also

occurs throughout the entire a-Si film during the annealing process, it was observed that the

growth rate of the MILC region gradually slows down at higher temperature. It can be

explained by competing crystallization mechanisms between MILC and background solid

phase crystallization. While SPC takes place in the film, the laterally-grown cryststalline

grain is blocked and no further lateral growth is observed. At the onset of the crystallization,

an incubation time is required for the SPC of a-Si to take place. At higher annealing

temperature, the SPC takes place earlier and becomes more dominant than MILC process.

Conversely, the effect of SPC diminishes at low annealing temperature. A distinctive

illustration is depicted in Fig. 3-3. Moreover, the lateral saturation region and film

crystallinity are dependent on the temperature of MILC annealing process. To obtain large

grain size with high quality of film crystallinity, it is necessary to retard the background

SPC process. By properly adjusting the annealing period or the peak temperature

conditions, the PRTA cycle can be tuned to be shorter than the transient time of the

background SPC, the undesirable effect of SPC can thus be completely suppressed during

MILC process [46].

3.2 Activation Energy Extraction

In order to gain a deeper insight into the kinetics of MILC by rapid thermal and furnace

anneal, the activation energy for a-Si to become poly-Si is obtained by thermodynamic

Arrhenius plot [49], as defined by

0 exp( ) kT A E

R= ⋅ − a (3-1),

where R is the recrystallization growth rate, A0 is the coefficient constant of growth rate, Ea

is the activation energy, k is the Boltzman constant, and T is the absolute temperature. The

equation can also be expressed as

ln( ) ln( 0) ( ) kT A E

R = + − a (3-2).

The extracted parameters, R and Ea, are related to the different definitions used (some

authors include the incubation time to be part of activation energy calculation), the

apparatuses of annealing, and the preparation of specimens [50]. In this work, the measured

initial rates of MILC crystallization, as a function of annealing temperature are shown in

Fig. 3-4. The activation energy of two dissimilar methods, deduced from the slopes, was

found to be 1.55 eV for PRTA and 1.76 eV for furnace annealing. The lower activation

energy for RTA is believed to be due to photon assisted or free energy released by sudden

heating during RTA [51]. Moreover, the method of PRTA crystallization proceeds much

faster than conventional furnace annealing during MILC process. For instance, the sample

of furnace annealing was prepared at 525 °C for 21 hr to form MILC length of about 32 μm,

whereas the PRTA sample was prepared at 650 °C with 60 cycles of pulses (total heating

time of 10 minutes) to form MILC length of 12.9 μm. Despites the fact that MILC length of

PRTA is shorter than that of furnace annealing, the recrystallized growth rate per unit time

is faster than that of the conventional furnace anneal. This improvement of growth rate for

PRTA MILC is also exploited to reduce the long process time in recrystallization.

3.3 Fundamental Characteristics of PRTA NW-TFTs

The operational principles of the novel NW-TFTs are similar to those of the conventional

TFTs. The n+ poly-Si gate is used to modulate the channel potential, consequently

controlling the switching behavior of the devices. Based on one-sided or two-sided seeding

windows for MILC, the NW-TFTs are categorized as metal-induced unilateral

crystallization (MIUC) or metal-induced bilateral crystallization (MIBC). The schematic

device structures of two kinds of MILC seeding windows arrangement are shown in Fig.

3-5, including (a) asymmetric seeding window (ASW), and (b) symmetric seeding window

(SSW). Figure 3-6 presents the growth of MILC region with increasing pulsed cycles, as

observed by optical microscope.

Figure 3-7 shows the transfer characteristics of n-channel NW-TFT devices with ASW

and SSW configurations. The samples were prepared by PRTA anneal at 650 °C peak

temperature with 60 cycles of pulses (total heating time of 10 min) to form MILC

crystallized channel. The performance of ASW devices with different channel lengths are

presented in Fig. 3-7 (a). According to previous material analysis in Section 3.1, MILC

process should ideally proceed continuously in the channel during the 60 cycles of PRTA

annealing. However, SPC seems to take place earlier than expected. We may exploit the

electrical characteristics to gain insights into the crystalline quality of polysilicon channel.

The result illustrates that the NW crystallinity is dependent on the channel length, i.e., the

NW located near the Ni seeding window depicts superior crystallinity to that away from the

window. It is believed that the heterogeneous nucleation of SPC process, rather than the

Ni-induced lateral recrystallization, dominates the channel recrystallization process in the

area away from the seeding window. Considerable grain boundaries and microstructural

defects in the NW would inevitably degrade the device performance. In addition, along

SiO2 film steps preferential nucleation area during annealing was observed [52]. Since NW

channels are just located at the steps of gate oxide, controlling RTA heating temperature is

supposed to be more significant during MILC treatment. However, for PRTA MILC

technique, NW-TFT devices with SSW has demonstrated that better device performance

can be achieved as shown in Fig. 3-7 (b) and Fig. 3-8, especially in terms of higher drain

current and steeper subthreshold slope, compared with the asymmetric counterparts. It can

be attributed that SSW cases may have more opportunities to develop better NW channel

crystallinity because seeding windows on both the source and drain terminals would

proceed to convert a-Si into polycrystalline silicon simultaneously. The probability of

undesired, fine grains by SPC can be reduced, and the needle-like grain with crystallization

direction along the channel is formed. As a result, the number of barrier at GBs where

conduction electrons must overcome is also reduced as well. The major device parameters

of different channel lengths are extracted and summarized in Table 3-1. Therefore, the

following discussion will focus on NW-TFT devices with SSW configuration. Also noted

that the off-state current of SSW case is rather high due to more Ni residual giving rise to

metal contamination. Further details regarding off-state leakage current mechanisms will be

discussed later.

3.4 Effects of Pulsed RTA Annealing Conditions

3.4.1 Effects of Pulsed RTA Annealing Temperature

In order to obtain ideal crystallinity of NW channels, selecting a precise heating

temperature plays an important role on PRTA MILC fabrication. The MILC annealing

conditions affect not only the recrystallization growth rate but also polycrystalline silicon

grain size, grain boundaries, and even metallic species distribution inside the film. Figure

3-9 compares several MILC annealing splits of n-channel NW-TFTs with an additional

nitride capping layer on the channel. The samples were prepared with different peak PRTA

annealing temperature for 60 cycles. The results show that the three MILC splits (i.e.,

625°C, 650°C, and 675°C) almost have identical transfer characteristics when the length of

overall channel crystallinity when the channel length is short. When the channel length

increases, the split with polycrystalline silicon formed at 650 °C has better characteristics

in term of steeper subthreshold slope, compared with the other splits of 625 °C and 675 °C.

At higher annealing temperature such as 675 °C, the SPC mechanism of the a-Si layer takes

place quickly and clogs MILC grain growth. Thus, the resultant polycrystalline silicon NW

channel is believed to contain microstructural defects with random crystalline gain

orientation. Besides, the large number of grain boundaries existing in the channel would

degrade the device performance. Another reason of device degradation may relate to Ni

concentration. Higher annealing temperature leads to higher Ni concentration in the MILC

region, grain size decreases due to many NiSi2 crystallites formation during MILC

treatment. This phenomenon, albeit sensitive to annealing temperature, is insensitive to the

annealing time [53]. On the other hand, a relatively low temperature is desirable to form

good-quality MILC polycrystalline channel with larger grain size and less grain boundaries.

However, the MILC of a-Si becomes slower with lower temperature. It will take more time

to achieve the same MILC growth region at lower temperature. The trade-off between the

MILC and SPC should therefore be carefully taken into consideration. Nevertheless,

NW-TFTs without nitride capping layer fabricated by similar MILC annealing conditions

depict similar trends, as shown in Fig. 3-10.

3.4.2 Effects of Pulsed RTA Annealing Time

In the above discussions, the device performance of different PRTA annealing

temperature depends on the channel length of NW-TFTs. For longer channel, the annealing

time is prolonged beyond the incubation time of SPC. Arranging proper annealing period

during each cycle is another approach to suppress SPC mechanism. The samples were

prepared by PRTA anneal periodically for 60 times with 5 seconds per period (i.e., total

heating time of 5 min). The peak temperature was set at 675 °C to make sure that the MILC

recrystallization region is longer than the channel length. The transfer characteristics are

shown in Fig. 3-11, together with those of the sample with the same processing conditions,

except the total heating time of 10 minutes. It can be seen that the device performance is

not improved by reducing the annealing time. It is assumed that MILC lateral-growth

crystallinity does not dominate the entire NW channel. It has been reported by R.B. Iverson

et al. that the incubation time of SPC is roughly 40 minutes in furnace anneal at 675 °C

[54]. While from our material analysis in Section 3.1, the incubation time in RTA anneal at

675°C in our case is about 11 minutes. This suggests that SPC mechanism takes place

earlier in NW channel during PRTA MILC treatment. Thus, PRTA MILC with higher

temperature annealing appears to speed up SPC mechanism and is unfavorable for

achieving good-quality grains.

3.5 Leakage Mechanisms

The major off-state currents of NW-TFT fabricated by MILC process are related to

several possible paths as depicted in Fig. 3-12. One is through the drain/channel junction

(path 1) and the other is through the gate-to-drain overlap region (path 2). The conduction

mechanisms in the two regions are strongly dependent on the strength of local field which

is determined by the gate and drain biases, referred to as the gate-induced drain leakage

(GIDL). The field-dependent conduction mechanisms are shown in Fig. 3-13. When |VGD|

is high, strong electric field would lead to trap-assisted (Fig.3-13(b)) or band-to-band

tunneling (Fig. 3-13(c)). Major conduction path can be identified by investigating the

dependence of leakage of devices with same channel on the gate-to-drain overlap area [55]:

It is mainly through path 2 as the dependence is linearly proportional to the gate-to-drain

overlap area, and through path 1 as the leakage is independence of the gate-to-drain overlap

area. Typical poly-Si NW devices fabricated in our work follow the path 2 [56]. This is

mainly attributed to the low dopant concentration for the poly-Si layer near the oxide

interface and the relatively low density-of-state (DOS) of the poly-Si film so that the band

bending (related to the electric field) at interface of the poly-Si layer shown in Figs.3-13(b)

and (c) is easy to trigger. To address the issue, we’ve also proposed and demonstrated two

ways to suppress the leakage: (1) Use of a hardmask dielectric (e.g., nitride) on top of the

side-gate to reduce the field strength at the channel interface [57]. (2) Promotion of the

dopant concentration in the region of the poly-Si layer near the interface to hinder the

inversion of the drain region (Figs.3-13(b) and (c)) [58]. In one of our previous work, we

also found that, as a poly-SiGe layer is used to replace the poly-SiGe as the NW channel

and the S/D material, the dependence of leakage of devices with same channel is

independent of the gate-to-drain overlap area. This is ascribed to the DOS of the poly-Si

which is much larger than the poly-Si [59]. The high DOS in poly-SiGe would pin the

Fermi level and thus the band-bending is less likely to occur.

For MILC samples, one additional leakage mechanism is attributable to the

contamination of residual Ni. Ni-related species accumulated either at the inter- or

intra-grains may provide more trap states in the band gap of Si, resulting in trap-assisted

tunneling leakage current. Besides, the off-state current of NW-TFT with SSW devices is

more severe. Except the fronts of seeding Ni-silicide from the opposite sides may confront

each other causing metallic species left inside the channel, main reason is one of the

seeding windows located at the gate-to-drain overlap region. Many residual Ni-related

species trap in grain boundaries between the MIC and MILC region. Hence, either forward

or reverse operation modes were used; the leakage currents of SSW devices would be

identical and transfer characteristics are illustrated in Fig. 3-14. Here, the reverse mode of

operation represents the exchange of source and drain electrodes. An increase in leakage

can be seen if ASW devices operated under reverse mode, owing to the fact that the

seeding window is now located in the drain terminal instead of source terminal.

Chapter 4

NW-TFTs Fabricated Using Different Crystallization Strategy

4.1 The Effect of Stress on PRTA NW-TFTs

In this thesis, the effects of capping an additional 100 nm-thick PECVD Nitride film

over the 100 nm LTO layer prior to PRTA MILC treatment on the device characteristics are

studied. Figure 4-1 compares the transfer characteristics of NW-TFT devices with and

without the Nitride-capping layer. We can see that better subthreshold slope and higher

on-current are obtained with the Nitride capping. To study the effect of stress, a 100 nm

on-current are obtained with the Nitride capping. To study the effect of stress, a 100 nm

相關文件