Negative Illumination Bias Stress Instability (NBIS)

Thin film transistors (TFTs), served as driving devices and switching devices for the active matrix liquid crystal display (AM-LCD) and the active matrix organic light emitting diodes (AM-OLED), have been universal applied. Since most of the proposed uses of the TFT will expose the TFT to a backlight or ambient light during operation, the black matrix (BM) integrated in the pixels is usually utilized, which can obstruct the formation of the photo-leakage current. The aperture ratio is the ratio between the transparent area, excluding a pixel’s wiring area and transistor area (usually hidden by black matrix), and the whole pixel area. As the open area ratio gets higher, the light penetrates more efficiently, which results in the brighter images for the display, with the same magnitude of back-light source. Several technologies of the wiring and element design improved the aperture ratio by reducing the black matrix area, thus increasing the projector’s luminosity. Hence, if the integrated black matrix can be fully removed from the pixel of flat panel display, the highest aperture ratio might be achieved. The photo-leakage current of the α-IGZO TFTs obtained from our experimental results exhibited no significant variation, while a remarkable negative shift of threshold voltage is observed. The reason of this finding has been explained by several models as mentioned in Chapter 2. However, theα-IGZO TFTs, used as switching devices, usually working at negative gate bias side for most of the time to ensure that the transistor is under the “off-state”. For application, if the device did not display at fully off-state condition, it would possibly cause the image persistence or bright dot. Hence, the instability of negative bias under illumination for theα-IGZO TFTs is a emergent issue we need to discuss in this report [4.6].

The three light sources with different wavelength, which are produced and classified by the halogen lamp passing through the color filters, are used to illuminate

53

on α-IGZO TFTs. The wavelengths are 653 nm (with a bandgap of 1.9 eV), 556 nm (2.23 eV) and 500 nm (2.48 eV), respectively, which is measured by USB2000.

Besides, the above light intensity received by theα-IGZO surface are controlled at the same magnitude.

Biases of between -30 and 20 V were applied to the gate contact and 10 V were applied to the drain contact for various periods up to 2000 sec and all the electrical measurements of current-voltage (I-V) were carried out in vacuum chamber at 303 K using an Agilent 4156C Semiconductor Parameter Analyzer. The capacitance-voltage (C-V) characteristics of α-IGZO TFTs is measured by Agilent 4284A, where the source and drain electrodes connect to Capacitance Measurement Low (CML) and the gate electrode connect to Capacitance Measurement High (CMH). The C-V characteristic is acquired at the frequency of 200 kHz by the gate voltage sweeping from -20 V to 5 V and -25V to 0V.

Figure 4-3-1 to Figure 4-3-6 show the VG-ID and C-V electrical characteristics of α -IGZO TFTs, illuminated by the above three light sources with different wavelength and measured at 0, 100, 300, 600, 1000 and 2000 sec to record the electrical characteristics. The threshold voltage exhibited almost the same values as the illumination was performed by the light sources with the wavelengths of 653 nm and 556 nm, as seen in Figure 4-3-1 and Figure 4-3-3, while a large negative threshold voltage shift was observed from Figure 4-3-5 as the α-IGZO TFT was illuminated by a light source with wavelength of 500 nm. As a wide band gap semiconductor forα-IGZO (3 eV) [4.7], corresponding to the light wavelength of 413 nm, it is transparent to most of the visible wavelengths (400~700 nm), usually called the transparent semiconductor. Thus, the lights with wavelengths longer than 413 nm should not be absorbed by α -IGZO film and not affect the α -IGZO TFT characteristics theoretically, i.e. our experimental light with wavelength of 500 nm

54

cannot penetrate throughα-IGZO film and influence the characteristics of α-IGZO TFTs. Nevertheless, the abnormal phenomenon was observed in our experimental results, in which the energy of light was absorbed by electrons, which were excited to the conduction band minimum, resulting in a negative threshold voltage shift.

Previous study [4.8] indicated that there are high density of state (about 10²⁰ cm^-3) in the α-IGZO bandgap, which is 2.3 eV below the conduction band minimum, and shallow defects of 10¹⁷ cm^-3 often related to electrical characteristic, which is about 0.2 eV below the conduction band minimum as shown in Figure 4-3-7. The origin of the high density of state near the valance band maximum is that the oxygen vacancy state (VO) keeps a large space, which traps two electrons and forms a deep fully-occupied state. On the contrary, the shallow defect states are due to the oxygen deficiency, which does not form a large space, and therefore electrons cannot be trapped at V_O site, which results in the doping of free electrons to conduction band minimum.

In order to investigate these abnormal findings about the light absorption of α -IGZO film, we proposed simple band diagrams to explain the reasons. As seen in Figure 4-3-8 (a), as the light with wavelength of 653 nm is introduced, the electrons existing in the deep level cannot be excited to the conduction band minimum due to the lack of energy. Figure 4-3-8 (b) showed that the light energy with wavelength of 556 nm can excite the electrons through the successive two steps: one excitation occurs from deep levels to tail states; the other excitation appears from tail states to conduction bands. Due to the limitation of density of states for tail states is about 10¹⁷ cm^-3, the largest number of electron excited from deep levels to tail states is about 10¹⁷ cm^-3, which is not large enough to change theα-IGZO conductivity and cause negative threshold voltage shift. Because the electron which is generated by light is slow and the amount could not be able to change the conductivity. As a result of the

55

light with wavelength of 500 nm is able to excite electrons from the deep level to conduction band minimum, as seen in Figure 4-3-8 (c), it would cause a huge negative threshold voltage shift. Accordingly, the light with wavelength of 500 nm can induce larger amount of electrons and holes inα-IGZO than the others.

For the sake of electrical operation, electrical reliability ofα-IGZO TFT is also required to investigate, in which the illumination with the three lights of different below the flat band voltage with the increasing stress time, as shown in Figure 4-3-10.

The characteristics variation of α-IGZO TFT experienced a negative bias illumination with wavelength of 556 nm is larger than the variation with wavelength of 653 nm. Similarly, a negative threshold voltage shift is observed in Figure 4-3-11, while the C-V stretched-out phenomenon is not obvious in the bias stress under light illumination of 556 nm wavelength as shown in Figure 4-3-12. Furthermore, as theα -IGZO TFT is illuminated by the light with wavelength of 500 nm and experienced a negative gate bias, the threshold voltage shift remains the same direction and is larger than theα-IGZO TFT illuminated by the light with wavelength of 500 nm only, as seen in Figure 4-3-13. Interestingly, the C-V stretched-out phenomenon disappeared, shown in Figure 4-3-14, as the device underwent a negative bias with illumination of 500 nm wavelength.

Oh at al. [4.9] claimed that the origin of the C-V curve stretch-out and negative threshold voltage shift under negative bias illumination stress is due to the light,

56 stress with visible light illumination, we considered that the photo-excited electrons stimulated from the VO state to the conduction band can be survived due to the lack of the direct recombination between the excess holes and photo-excited electrons. The above two mechanisms result in a larger negative shift in the transfer curves of α -IGZO TFT. Furthermore, the another role of V_O⁺² states is the donor-like states, which is above the high density of states and 2.3 eV below the conduction band minimum. Consequently, the V_O⁺² states are positive without the occupying electrons and neutral as the electrons filled these states. Hence, the occupancy of these electrons at the ionized oxygen vacancy states would result in the stretch-out of C-V curve.

Such model is not appropriate to explain our experimental data, since the serious stretch-out of C-V curve dose not accompany a negative threshold voltage shift under negative bias illumination stress with light wavelength of 653 nm. However, the C-V stretch-out phenomenon gradually disappears and the negative threshold voltage shift is obvious as the negative bias illumination stress with light wavelengths of 556 and 500 nm is applied.

We also used the ISE-TCAD to simulate the C-V transfer curve of amorphous silicon (α-Si) TFT. In this simulation, the donor-like interface defects of 10¹² cm^-2 below the intrinsic Fermi energy (Ei) and presenting a uniform distribution within 0.2 eV (show in the inset of Figure 4-3-15) for α-Si near the drain side is assumed. The Figure 4-3-15 shows that the stretch-out phenomenon of C-V curve appeared as the assumption of donor-like interface defects is introduced. Accordingly, such stretch-out phenomenon of C-V curve induced by negative bias illumination stress

57

can be ascribed to the donor-like defects generated by the gate bias stress. As seen in Figure 4-3-10, the two dash lines defined the C-V stretch-out region between A and B points, which is corresponding to a donor-like defects in the deep energy band gap of α -IGZO, presented in Figure 4-3-16 (a). In the methodology of the C-V measurement, the C-V transfer curve we measured is the total capacitance of source/drain overlap and channel for α-IGZO TFT. When the applied gate voltage is larger than the flat band voltage (V_FB) of α-IGZO TFT, the electrons accumulate gradually in the channel area. However, only the source/drain overlap capacitance is obtained with the gate voltage below the flat band voltage (VFB) ofα-IGZO TFT.

Hence, the additional capacitance value appearing in the C-V stretch-out region can be related to the donor-like interface defects. We would assume the donor-like defects are among the whole channel but only the position where the donor-like defects near source and drain side could contribute to the C-V transfer curve stretch-out because of carrier from electrode have chance to react with those defects. On the other hand, due to the C-V transfer curve stretch-out phenomenon in the gate voltage below flat band voltage region, the carrier density in the channel is not high enough to react with defect that contribute to C-V transfer curve stretch-out as show in Figure 4-3-16 (b).

However, the stretched-out of C-V curve, which accompanied a larger negative shift of threshold voltage in the I-V curve, disappeared gradually when the negative bias illumination stress with the shorter light wavelengths (556 and 500 nm) were utilized. The different results between the lights with long wavelength (653 nm) and the shorter wavelengths (556 and 500 nm) is ascribed to the photon energy with wavelength of 653 nm cannot excite sufficient electron-hole pairs. These light-generated holes would attack the interface between the gate insulator and α -IGZO bulk to form the donor-like interface traps with the application of the negative bias illumination stress. However, as the light with wavelength of 653 nm is applied,

58

there have no sufficient holes to be accumulated around the donor-like interface traps.

Hence, the C-V transfer curve shows the stretch-out phenomenon without threshold voltage shift, as shown in Figure 4-3-17 (a). On the contrary, negative bias illumination stress with wavelength of 500 nm could generate large amount electron-hole pairs, which have been already explained in the experiment of illumination only forα-IGZO TFT as shown in Figure 4-3-8. These large numbers of holes generated by the negative bias illumination stress could generate interface donor-like defects, and then these holes could also accumulate at the dielectric/channel interface causing a huge negative threshold voltage (V_TH) shift as shown in Figure 4-3-17 (b). Besides, the donor-like trap is positive charged because of it above the Fermi energy level when device suffer negative gate bias stress. As we measure C-V transfer curve from -25V to 0V, the donor-like trap turn from above the Fermi level to below Fermi level. Hence, the donor-like trap tends to capture an electron to become neutral, this change cause an additional interface capacitance as shown in Figure 4-3-18 (a). On the contrary, we could not see the C-V transfer curve stretch-out in Figure 4-3-14, because donor-like trap is screened by holes accumulated at dielectric/channel interface. The donor-like trap could not capture an electron to become neutral, so no additional capacitance detected as shown in Figure 4-3-18 (b).

Figure 4-3-19 shows the recovery in the dark environment after negative gate bias stress with 500 nm wavelength illumination. We could found that the C-V transfer curve show stretch-out phenomenon form gradually as the recovery time getting longer. Because some part of holes which accumulated at dielectric/channel interface by negative gate bias stress with 500 nm wavelength illumination recombine with electron in the dark environment. The donor-like trap no longer screen by those holes, it could become neutral as below Fermi level and attribute to additional capacitance. Figure 4-3-20 which shows the recovery in the dark environment after

59

negative gate bias stress with 653 nm wavelength illumination. Because no sufficient holes accumulate at dielectric/channel interface the donor-like trap would not be screen. The C-V stretch-out remains the same in the recovery period.

Here, we proposed another model to explain the stretch-out phenomenon of C-V transfer curve disappear under negative bias stress with 500 nm wavelength of light due to the hole trapping in the gate insulator. Once the hole trapping in the gate insulator, the electrical field of insulator will enhance. Thus, the electric field in the channel will decrease which lead to the Fermi level cannot “go through” donor-like defects as shown in Figure 4-3-21. Hence, we could not measure the C-V transfer curve stretch-out. Both models are possible occur simultaneously and not contradictory.

The light wavelength (556 nm) is between the two other lights; both the C-V stretch-out and negative threshold voltage shift can be observed as the negative bias illumination stress with light wave length of 556 nm was introduced. It indicates that only part of the generated-hole accumulated at the interface causing partial donor-like trap be screened or some holes trapped in the gate insulator. As the experimental results Oh et al. have reported.

60

Figure 4-3-1 The I-V transfer curve under 653 nm wavelength of light illumination.

Figure 4-3-2 The C-V transfer curve under 653 nm wavelength of light illumination.

61

Figure 4-3-3 The I-V transfer curve under 556 nm wavelength of light illumination.

Figure 4-3-4 The C-V transfer curve under 556 nm wavelength of light illumination.

62

Figure 4-3-5 The I-V transfer curve under 500 nm wavelength of light illumination.

Figure 4-3-6 The I-V transfer curve under 500 nm wavelength of light illumination.

63

Figure 4-3-7 The subgap density of state distribution forα-IGZO. [4.8]

Figure 4-3-8 The schematic of electron in the deep trap under three different wavelength illumination condition.

64

Figure 4-3-9 The I-V transfer curve under 653 nm wavelength of light illumination and VG-VTH = -30V.

Figure 4-3-10 The C-V transfer curve under 653 nm wavelength of light illumination and VG-VTH = -30V.

65

Figure 4-3-11 The I-V transfer curve under 556 nm wavelength of light illumination and VG-VTH = -30V.

Figure 4-3-12 The C-V transfer curve under 556 nm wavelength of light illumination and VG-VTH = -30V.

66

Figure 4-3-13 The I-V transfer curve under 500 nm wavelength of light illumination and VG-VTH = -30V.

Figure 4-3-14 The IC-V transfer curve under 500 nm wavelength of light illumination and VG-VTH = -30V.

67

Figure 4-3-15 The ISE-TCAD C-V simulation for α-Si TFT with donor-like state below Ei.

Figure 4-3-16 (a)The schematic of donor-like trap generate region within the bandgap of a-IGZO under NBIS. (b) Only the donor-like trap near source and drain could contribute to C-V transfer curve stretch-out.

68

Figure 4-3-17 The schematic is (a) hole could generate donor-like interface trap (b) too much holes accumulate at the interface.

Figure 4-3-18 (a) the donor-like trap could become neutral as below Fermi level. (b) the donor-like trap could not become neutral as below Fermi level because the accumulated holes screen it.

69

Figure 4-3-19 The C-V stretch-out appears in the recovery under dark environment after negative gate bias stress with 500nm wavelength of light.

Figure 4-3-20 The C-V stretch-out appears in the recovery under dark environment after negative gate bias stress with 653nm wavelength of light.

70

Figure 4-3-21 The schematic of α-IGZO band diagrame with donor-like defects (a) without hole trapping (b) with hole trapping, which measure from neagetive gate bias.

71