• 沒有找到結果。

Chapter 1: Introduction

1.3 Organization of this Thesis

Besides this chapter, this thesis is divided into the following chapters.

In Chap. 2, we describe the process flow of fabricating the n-chanel poly-silicon

TFTs. Also we present the measurement setup.

In Chap. 3, we show the response of poly-silicon TFT gas sensors to the

environments with various pressure and/or various gas compositions. The change of

electrical characteristics to different environments and the effect of structural

parameters are recorded and discussed. Results of nano-wire gas sensors are also

presented and addressed.

Finally, we conclude this thesis in Chap. 4 by summarizing the experimental results

and major findings.

Chapter 2:

Device Fabrication and Measurement Setup

2.1 Fabrication of Planar Poly-Si TFTs

The poly-silicon TFT gas sensors were fabricated on 6-inch p-type (100) Si wafers

with resistivity of 15~25Ω-cm and the thickness of 655~695μm. Note that the

wafers will serve as the back-gate of the completed devices. A wet oxide layer with

thickness of 500 Å serving as gate dielectric first grew on poly-silicon wafer. And

then a poly-Si layer with thickness of 1000 Å was deposited on the wet oxide. The

poly-Si was heavily doped with P+ at energy of 15 keV and dose of 5x1015 /cm2.

Afterwards, source/drain (S/D) photoresist patterns were formed on the poly-Si layer

by a standard lithography step, followed by a plasma etch step to form S/D (Figure

2-1). Second poly-Si layer was subsequently deposited on the wafer to serve as the

channel layer. In the fabrication, the poly-Si layers were deposited twice to form the

S/D and channel separately, so the thickness of the channel can be modulated flexibly.

In this regard, 4 different channel thicknesses, namely, 70 Å, 300 Å, 500 Å, 1000 Å,

were explored in this work. Channel pattern was then formed with another standard

lithographic and etch steps (Figure 2-2).

2.2 Nano-wire Gas Sensor Device Structures and Fabrication [16]

The nano-wire field effect transistors (NWFETs) were also fabricated on 6-inch

p-type wafers capped with a nitride layer. Figure 2-3(a) and (b) show the top (layout)

and cross-sectional views (along the A – B direction in Figure 2-3(a)), respectively, of

the device, dubbed as device A.

As shown in Figure 2-3, the Si wafers were first capped with a 100 nm-thick nitride

layer. Then, a 100nm-thick oxide layer was deposited by LPCVD. After deposition of

the oxide layer, standard photolithographic and etch steps were employed to form the

oxide dummy structures. Subsequently, a 100 nm-thick amorphous-Si layer was

deposited and then annealed at 600°C for 24 hours in N2 ambient to transform it into

poly-Si. Afterwards, the source/drain (S/D) was doped with phosphorus ion

implantation with a dose of 5E15 cm-2. After the generation of S/D photoresist

patterns with a lithographic step, a reactive plasma etching step was performed to

form the S/D regions. Using the anisotropic sidewall-spacer etching process, poly-Si

NW channels were formed at the sidewalls of the oxide dummy structures during the

S/D etching step. By precisely controlling the etching time, the cross-sectional

dimensions of poly-Si NW channels could be scaled down to sub-100 nm scale.

Subsequently, all devices were then covered with a 200-nm-thick TEOS oxide

passivation layer. Finally, the poly-Si NW was exposed by a 2-step dry/wet etching

process. Note that in the etching steps the wet etching rate of TEOS oxide should be

carefully adjusted so that an oxide layer would remain on source/drain regions while

the channels were exposed. This would prevent the conduction from source to drain

via the test solution.

2.3 Gas Sensing Measurements

The electrical characteristics of devices were measured in different environmental

conditions (pressure and composition), including vacuum (2x10-1 torr), nitrogen,

ammonia, steam, mixture of ammonia and nitrogen, mixture of nitrogen and steam,

and so on. For this purpose, a measurement system schematically shown in Figure 2-4

was constructed. In the system a closed chamber is used to house the test devices. A

pump connected to the chamber is used to pump out the air in the chamber and the

pressure as low as 2x10-1 torr can be achieved. After pumping, various gases (nitrogen,

ammonia, and steam) can be injected into the chamber to vary the chamber gas

composition. Note that nitrogen and ammonia are stored in the steel cylinders, while

the steam (vaporized H2O) is stored in sampling bags. Flow of the gases is controlled

by valves. The electrical characteristics are measured by Keithley4200 with setup

measurement software, Interactive Characterization Software (ICS). This

measurement system is owned by Laboratory of Enzyme and Protein Engineering,

Department of Biological Science and Technology, National Chiao Tung University.

In this study, Id-Vg and Id-time measurements were constantly performed. For the

Id-Vg measurements, Vd was fixed at 0.5 V while Vg was kept at low values in order

not to cause stability and reliability issues on the test devices. For the Id-time

measurements, first we probed the basic electrical characteristics of the device and

chose an appropriate gate voltage in the subthreshold region to ensure obvious change

if the varied environment indeed affects the device. In the process the drain voltage

was also kept at 0.5 volt.

Some major parameters of the transistors, including threshold voltage (Vth) and

subthreshold swing (SS), are defined below [17].

The threshold voltage:

Vth = VFB – (Qd+qD)Cox + 2ΦF (2-1).

In the expression, Qd (Coul/cm2) is the effective charge stored in the depletion region

of channel when MOSEFT is turned on, Cox is the capacitance of gate oxide per unit

area, D is the doping concentration of the channel, and VFB is the flat-band voltage

with the following relation:

VFBMS – Qi/Cox (2-2),

ΦMSMS, (2-2a),

where ΦM is the work function of gate material ,ΦS is the work function of the

channel material, and Qi is the charge in oxide that includes mobile ions, trapped

charge, fixed charge, and charge in interface of oxide and silicon. ΦF is the silicon

bulk potential:

ΦF = ln(p/ni) x (kT /q) = ln(ni/n) x (kT/q),

where kT/q is thermal energy, k is Boltzmann constant, T is absolute temperature, q is

the electron charge, p is the concentration of holes at thermal equilibrium, n is the

concentration of electrons at thermal equilibrium, and ni is the intrinsic concentration

of electrons and holes.

The on-current in the linear region (Vg > Vth, Vg-Vth > Vd ) can be expressed as

Id = μCox x (W/L) x [(Vg - Vth) x Vd – Vd2/2] (2-3),

where μ is the carrier mobility, W is the channel width, and L is the channel length.

The on-current in the saturation region (Vg > Vth, Vg-Vth < Vd ) can be expressed

as :

Id = μCox x (W/L) x (Vg - Vth)2 x 1/2 (2-4).

Definition of subthreshold swing (SS) is :

(2-5).

These electrical parameters can be obtained from the Id-Vg curves at Vd of 0.5V.

Chapter 3 Electrical Characteristics of Back-Gate TFT Gas Sensor

3.1 Basic Electrical Characteristics of Devices with Different Channel Structure under Atmosphere

The electrical characteristics of NWFET and back-gated planar TFTs with channel

thickness: 70Å, 300Å, 500Å and 1000Å are shown in Figure 3-1. The measurements

were done under atmosphere and at room temperature. For the planar devices the

drain current increases with increasing channel thickness in general. The device with

70Å channel thickness exhibits the lowest on and off current. This phenomenon is

attributed to the high resistance of the ultra-thin channel. However, the device with

70Å thickness channel also depicts slightly better subthreshold swing (SS) over the

remaining planar devices. This implies that device with thinner channel has better

gate controllability.

By comparing the transfer characteristics of NWTFT with those of the planar

devices, as shown in Figure 3-1, clearly the NWFET exhibits much improved

performance in terms of higher on-current, larger on/off current ratio, and steeper SS.

Part of the improvement can be attributed to nanowire’s high surface/volume ratio.

However, some hydrogen species are suspected to have been contained in the poly-Si.

These H species come from the underlying SiN (see Figure 2-3) which was deposited

with H-related reaction gas (e.g., SiH4 and NH3), thus abundant of hydrogen atoms

are incorporated in the SiN film [18]. Portion of the H species diffuse into the poly-Si

nanowire and passivate the defects existing therein in the subsequent process steps,

thus the device shows improved performance.

We can speculate reasonably that there is a conducting path from the source to

drain near the film surface (Figure 3-2) in addition to the nominal conduction channel

manipulated by the back gate near the oxide/film interface. The surface conduction

path is significantly affected by the environment as there is no dielectric passivation,

and less affected by the back gate unless the channel film is sufficiently thin. Below

are the results of a series of experiments carried out to examine the above postulation.

3.2 The Effects of Measurement Environment on Poly-Si TFTs

3.2.1 Device Characteristics in Vacuum and in Nitrogen Ambient

Typical characteristics of the planar TFTs with various film thickness measured

under different test environments are shown in Figures 3-3(a)~(d). In these

measurements, the device was first tested in normal ambient which had a moisture

level of 42%, and then in the closed test chamber (see the description in Chap. 2)

which was pumped down to 0.2 Torr. After the second measurement performed in

vacuum, the vacuum chamber was injected with high purity N2 (99.999%) to increase

the pressure to around 1 atm, and then the third measurement was performed. Finally

the chamber was open to expose the device to the air again. We then repeated the

measurements to check if the transfer characteristics of the device recovered to the

original characteristics.

In the figures we can find that there is a big change in the Id-Vg curves as the test

environment is switched from atmosphere to vacuum. When measured in vacuum, the

drain current became low and insensitive to the variation of gate voltage. As N2 was

injected into the chamber, the Id-Vg curve remained the same as that measured in

vacuum.

When the device was put back to the normal air, the I-V characteristics would

recover, but the extent depended strongly on the film thickness. As can be seen in

Figure 3-3(a), the TFT with 70Å-thick channel almost return to the original

characteristics while the other devices with thicker channel do not t recover

completely (Figures 3-3(b) ~3-3(d)).

Note that the high purity N2 is very dry and contains less moisture, unlike the

normal air which has a relative humidity of 42 %. Thus we postulate that the above

difference in I-V characteristics is due to the existence of moisture and the interaction

between moisture and the poly-Si films. Note the moisture contains H-related species

(such as H or OH) which can help passivate the defects contained in the poly-Si [19].

The schematic illustration about the passivation of defects in the grain boundaries

with H from the moisture in the ambient is shown in Figure 3-4 (a). In the figure and

following discussion, for simplicity, we assume H is the main species responsible for

the passivation of the defects. Note that the fresh devices should contain specific

amount of H throughout the film since the last step in device fabrication was a

de-ionized water rinse after the removal of photoresist with H2SO4/H2O2. According

to the water passivation effect [14], some H species should remain in the films even

when the device was dried. At vacuum, the amount of active defects in the poly-Si

increases because the vacuum tends to drive out the H from the grain boundaries and

destruct the bonding, such as Si-H, leaving the defects unpassivated (see Figure 3-4).

The amount of dangling bonds increases massively and affects the electrical

characteristics considerably. As the high-purity N2 is injected into the chamber such

characteristics are retained due to the dry ambient. After returning to the air, the Id-Vg

curve is restored due to the diffusion of H from the moisture contained in the normal

air. However, the concentration of H decreases with increasing depth due to the

diffusion process. The device with70 Å-thick channel restores soon due to its shallow

channel (Figure 3-5(a)). However, for devices with thicker channel it needs longer

time to recover completely. The gated channel (bottom path shown in Figure 3-2) thus

retains a high amount of un-passivated defects, as shown in Figure 3-5(b). This

explains why the characteristics cannot recover completely in Figures 3-3(b)~(d) as

the test environment returns to the normal air.

3.2.2 Id-time Measurements

In this Id-time measurement, the setup of measurement environment is the same

as that described in last sub-section. The measurement scheme is stated in Sec. 2.3,

and the Vg is set at 0.5V. In Figure 3-6, the Id of a TFT with 300Å channel thickness

initially increased as measured under normal atmosphere. To check this phenomenon

we compared the transfer curves of the device before and 60 sec after the above

Id-time measurement. The results are shown in Figure 3-7. We can see that the

threshold voltage becomes smaller while the SS remains unchanged after the Id-time

measurement. To confirm and figure out the mechanism, we made a series of

measurements performed on one another device with the results are shown in Figure

3-8. First, we measured the transfer curve of the device with 300Å channel thickness

under normal air (curve (1) in Figure 3-8). And then we made the Id-Time

measurement (Vg = 3.7V > 0V ) (curve (1) in Figure 3-9) on the device for 2100 sec,

and then the transfer curve was measured again (curve (2) in Figure 3-8). As the trend

shown in Figure 3-7, the threshold voltage of the curve (2) in Figure 3-8 shifts

leftward. And then we made Id-Vg measurement with gate voltage sweeping from 0

to -5V, as shown in Figure 3-10. After that we measured the curve 3) in Figure 3-8

and found the threshold voltage shifts slightly rightward with respect to the curve (2)

shown in the same figure. Before measuring the curve (4) in Figure 3-8, we made

another Id-Time measurement with Vg=-3.7V (< 0V) (curve (2) in Figure 3-9). The

curve (4) shifts leftward and becomes close to the original curve (1) in Figure 3-8.

From the above experimental results we postulate the instability in the transfer

curves of the test devices as measured in the normal ambient is due to the action of

mobile ions, like sodium or potassium ions, presenting in the oxide. This is a well

known issue for old MOSFET technology [20]. Since there is no passivation dielectric

covering the devices, these contamination species are likely to appear in the test

samples. As a positive gate voltage is applied for a sufficient long time, these ions

tend to accumulate near the oxide/channel interface and the threshold voltage is thus

decreased (see Figure 3-11(a)). In opposite situation as the gate bias is negative, these

ions would be attracted toward and accumulate near the oxide/gate interface, leading

to an increase in threshold voltage.

Now let’s return to discuss the results shown in Figure 3-6. While the air was

pumped out to vacuum the text chamber, the Id dropped drastically which is

consistent with the results shown in Figure 3-3(b). After the measurement

environment turned back to atmosphere, the current increased again. The air has

obvious passivation to the device.

3.3 Effects of Ammonia

In this experiment, the device was first tested in normal ambient. Then the

device was put into the chamber and the chamber was pumped down to 0.2 Torr.

After the second measurement performed in vacuum, the nitrogen was injected into

the chamber to increase the pressure to 1 atm, and then the third measurement was

performed. Afterwards, 1.5 ppm ammonia (NH3) was injected into the chamber.

Finally, the fourth measurement was performed.

The results are shown in Figures 3-12(a) to (d) for devices with different channel

thickness. Also the results of Id-time measurement performed on a TFT with 300 Å

channel thickness are shown in Figure 3-13. Basically the results shown in Figures

3-12(a)~(d) are the same as those shown in Figures 3-3(a)~(d) at the first three stages

measured in normal air, vacuum, and N2 ambient. As the ammonia is introduced, no

obvious effect on the Id-Vg curve is observed for all the test devices regardless of the

channel thickness. Figure 3-13 also shows similar trend. The observation implies the

1.5 ppm ammonia does not result in significant passivation effect as the moisture does

to recover the device characteristics.

In [21] it was mentioned that ammonia (NH3) might dissociatively absorb at 300

K to form SiH and SiNH2 on the surface of Si, as shown in Figure 3-14. Though slow

thermal decomposition of NH2 to N and H2 occurs even at 320 K, most Si-NH2 is

stable up to 630 K [22]. However, owing to the trace amount of NH3 presenting in the

ambient, these H-related species cannot exhibit significant passivation effect on the

device characteristics.

3.4 Effects of Moisture and Moisture/Ammonia Mixture

Here we first examined the response of device characteristics to the injection of

moisture and ammonia. In this test the former three stages of measurements were the

same as those described in former section. Then the humidity was injected into the

chamber until the moisture content in the chamber reached 42.6 % and the fourth

measurement was performed. Finally, 1.5 ppm ammonia was injected into the

chamber before the fifth measurement was performed.

The results are shown in Figures 3-15(a)~(d) for devices with different channel

thickness. Since the effects of vacuum and N2 ambient have been addressed in former

sections, here we just discuss the effect of moisture and moisture/NH3 mixture. As

can be seen in the figures, the input of moisture can indeed improve the device

characteristics. In [23], it was mentioned that in a H2O molecule the oxygen end tends

to bond with a Si-Si dimmer on Si surface, whereas the hydrogen ends tend to attach

to dangling bonds within the neighboring row of dimmers. The bindings, Si-H and

Si-OH, can passivate the defects in poly-Si (Figure 3-16). However, as mentioned

above, such passivation is limited to the region closest to the surface of poly-Si,

therefore the characteristics cannot fully recover.

One interesting phenomenon observed in Figures 3-15(a)~(d) is the dramatic

increase in the drain current as the ammonia is mixed with the steam in the

environment. In [24], the extent of the hydrolytic reaction in wet ammonia: H2O +

NH3 -> NH4+ + OH-, was mentioned (Figure 3-17). Formation of NH4+ can get more

OH- dissociated from H2O and thus the passivation effect can be greatly enhanced as

compared with ammonia or moisture alone.

However, some remarks on the effect should be given. First, the dependence

between the extent of drain current enhancement and the channel thickness is not the

same as that illustrated in Figures 3-3(a)~(d). From Figure 3-15(d), the characteristics

measured under moisture/ammonia mixture have higher off-current than the original

curve. Moreover, in the figures the SS for the condition is larger than the original

transfer curves. It is obvious that most of the conduction current pass through near the

top surface of channel (Figure 3-2) and thus weaken the controllability of back-gate.

From this observation, we conclude that the passivation mechanism of

moisture/ammonia mixture mainly occurs in the top surface region of the poly-Si

channels.

Next we investigate the interaction of ammonia with the moisture in the normal

air and the resultant impact on device performance. In the testing of Figure 3-18, we

made first device measurement under atmosphere. After closing the test chamber we

injected ammonia into the chamber directly without vacuum process and then

performed the second measurement. Finally we opened the chamber and made the

third measurement after 5 min the ammonia dispersed away from the chamber (5

minutes later from opening the chamber).

In Figures 3-18(a)~(d), as NH3 is injected, not only the Vth but also the SS

reduces dramatically, unlike the situation shown in Figures 3-15(a)~(d). As mentioned

in Sec. 3.1, the fresh devices contain an amount of H-related species. It is speculated

that those H-related species could accelerate the penetration of ammonia and thus the

passivation is not restricted to the regions near the surface. In other words, the region

close to the oxide/channel interface is also passivated by the ammonia even when the

film is thick. From the above results and discussion, we infer that the moisture in the

air reacts with ammonia and results in the enhancement of defect passivation. For the

air reacts with ammonia and results in the enhancement of defect passivation. For the

相關文件