3.1. Phenomenon of TEM Etching in Anodic Bonding
Compared with the cross-section of non-etching anodic bonding wafer shown in Figure 1
a, a uniform gap appears along the Silicon/Glass joint after HF etching in Figure 1
b. It can be inferred that a special layer of modified SiO2
once existed in the gap, which etches faster than the bulk Pyrex 7740 glass. It is confirmed in Figure 1
c that the modified SiO2
belongs to the Pyrex 7740 glass at the Silicon/Glass joint. Moreover, another interesting phenomenon is demonstrated in Figure 1
d—the modified SiO2
can be cured by post annealing, so the annealed bonding samples are without gap after HF etching.
To evaluate the influencing factors of the SiO2
modification, parameter studies involving voltage and temperature were conducted. Figure 2
a shows that the etched gap increased with increasing applied voltage from 400 V to 1200 V under “current control mode”. However, keeping the charge constant (108 mA·s) while increasing the applied voltage, the width of etched gaps remained at almost 600 nm. The results reveal that the voltage is not directly related to the width variation, but it is actually the transfer charge that bridges the gap width and the voltage. Higher voltage results in more transfer charges, leading to the increment of gap width. The coefficient of the fitted curve for charge density (Qd
) and gap width (W
) is 3317 mA/mm2
/nm in Figure 2
Secondly, the temperature influence on the gap width is plotted in Figure 3
a. It seems that the temperature shares the same mechanism with the voltage by increasing the transfer charge. As shown in Figure 3
b, however, the corresponding coefficient between charge density and gap width was 3937 mA·mm−2
/nm, which is larger than that in Figure 2
b. In addition, under “charge control mode” (red squares in Figure 3
a, constant transfer charge of 72 mA·s), the width still increased with the bonding temperature. Therefore, it can be inferred that the temperature also affects the gap width in other ways besides the transfer charge.
3.2. A Novel Method for Micro/Nano Structure Micromachining by TEM Etching
Based on the above results, charge transfer and temperature effect are the two main reasons for acceleration of SiO2 etching rate. Hence, they meet the second standard for structure etching: etching faster. Comparing the two acceleration methods, the charge transfer is more feasible and effective in patterning than the temperature implantation. Thus, TEM by charge transfer has the potential to be a new approach to structure micromachining. The main idea of the new etching method is to drive the charge transfer in glass, and by the modified effect of charge transfer at specific region, the opposite pattern can be etched on the glass with the difference of etching rate.
When using TEM etching to acquire micro/nano structure, it is the primary consideration to avoid anode material anodic bonding with the glass. Therefore, a conductive passivation layer is coated on the surface of the anode materials before the TEM, such as Pt or Au [18
]. Then, silicon is selected for the mould substrate due to its similar thermal expansion coefficients with 7740 glass, high surface quality, and mature process. Figure 4
illustrates the designed flowchart of TEM etching for glass structures. In the first step, the mould structure is micromachined by deep reactive ion etching (DRIE) on the 15 mm × 15 mm silicon substrate. By sputtering with the Ti/Pt/Au or Ti/Pt layer, both sides of the silicon mould are passivated. Then, the mould is contacted to the Pyrex 7740 surface with an applied pressure. As the temperature reaches the set point, the TEM process—which is similar to anodic bonding—starts up. Finally, the TEM glass is etched in diluted HF solution for several minutes.
Following the designed flowchart, the micro/nano structures were achieved by TEM etching, as shown in Figure 5
. The AFM data proves that the silicon mould patterns could be transferred onto Pyrex 7740 as expected, and it was easy to separate the mould from glass without any residues. The results verify the feasibility of this new method. It is also concluded that the TEM process shares a similar “mask patterning” procedure with glass imprint. So, TEM has the advantages of being fast and highly accurate. Moreover, since the principle of TEM is charge transfer instead of squeezing glass, the required temperature could be far below the glass transition point, which dramatically simplifies the demoulding process and prolongs the mould life.
In order to investigate TEM accelerating etching behavior, the etching depth of the Pyrex 7740 and TEM glass were tested according to the flowchart shown in Figure 6
a. A general etching trend is presented in Figure 6
b: the TEM glass etching rate was initially faster than that of the Pyrex 7740, then decreased gradually, and eventually equaled the etching rate of Pyrex 7740. Based on the fitted curves, the max etching rate of TEM glass accelerated to 59 nm/min, while the etching rate of Pyrex 7740 was about 35 nm/min. After ~14 min etching, the modified SiO2
along depth direction was completely removed, so that the height difference remained at 300 nm from then on.
Due to the isotropic property of 7740 chemical etching, the constant dimension of etched structure occurs in not only the depth direction but also 2D size. As shown in Figure 7
, even though the glass thickness dramatically decreases for 8 h etching, the 2D size of the TEM-etched structure is almost the same as samples etched for 25 min (the modified SiO2
has been completely removed). Therefore, structure micromachining with TEM etching is proven to be an inherently three-dimensional self-controlled process.
According to the etched morphologies in Figure 5
and the data of initial height difference shown in the inset of Figure 6
, glass structures of 10–20 nm depth were already patterned after TEM. Moreover, convex shape was detected at the top edge of structure boundary, while a concavity existed at the bottom edge after HF etching. The development of structure morphology in the TEM etching is characterized as shown in Figure 8
. The results show that the convex and concave occur in TEM process and wet etching process, respectively.
The phenomenon of initial height can be explained with Reference [19
]. It mentions that with the joule heating during the anodic bonding, the electrostatic force induced by the residual charge will drive the glass transition flow at the contact interface. Similarly, the depletion layer with O2–
and the passivation layer with Pt+
attract each other due to the electrostatic force during the TEM process. Then, the mould patterns are transferred onto the glass surface, forming initial structures. In addition, due to flow behavior [20
], the glass beneath the contact interface is accumulated towards the pattern boundary, which eventually forms two peaks.
Since TEM is the only anisotropy source during the whole process, the concave feature should be related to the enhancement of charge transfer. Finite element modeling (FEM) simulations were conducted as shown in Figure 9
. The results show that “point effect” caused by mould edge increases electric field intensity to 9.7 MV/m, almost three times larger than that in the middle of the mould. Thus, high electric field intensity leads to a further charge migration in the glass, and etching depth at the corresponding region will be deeper. Based on the detail of above analysis, the whole process can be summarized in Figure 10
Additionally, some parameters in TEM etching are investigated in detail. Firstly, samples, with TEM transferred charge of 6 to 10 mA·s under 350 °C and 1000 V were etched for 15 min. As shown in Figure 11
a, the etching depth followed a linear relationship with the charge density. Secondly, the influence of pressure in TEM on etching depth was studied, and the result in Figure 11
b shows that applied force has no effect on the etching depth.
Root-mean-square (RMS) data (as listed in Table 2
) shows that there is no difference between the two kinds of etched surface. Both magnitudes were almost the same as the original. Combining roughness data with the results plotted Figure 6
, Figure 7
and Figure 11
reveals that the new etching method for structure micromachining is a surface non-damage and three-dimensional self-control process, which eliminates the requirements of time control in conventional methods such as dry etching and wet etching.
After the TEM process, the integrity of the silicon mould and the passivation layer is shown in Figure 12
. Comparing the mould with Ti/Pt, Ti/Pt/Au samples are much more vulnerable to thermal and electrical shock, and easier to be peeled off from the silicon after several times TEM process in atmosphere. Even worse, the surface of the Au is spread by the silicon gold eutectic particles [21
], which will seriously affect the following TEM process.
One of the possible reasons for the vulnerability of Ti/Pt/Au coatings could be the fact that the difference of thermal expansion coefficient (CTE) between Au and silicon (CTEAu
= 14.2 ppm/°C, CTEsi
= 2.6 ppm/°C) is much more than that between Pt and silicon (CTEpt
= 9.0 ppm/°C). Besides that, EDX results plotted in Figure 13
show that the oxygen content in deteriorated Ti/Pt/Au is higher than normal region. The relevant results have been reported by Reference [22
]. Those works reveals that Si is prone to be anodically oxidated with Au, Mo, and Cr metal in a strong electric field at atmosphere. In this way, silicon expands after being oxidated [23
], while the Ti/Pt/Au remains constant. So, both effects will lead to a greater stress due to volume expansion difference during TEM process, which makes Ti/Pt/Au vulnerable to thermal and electrical shock. Comparatively, Ti/Pt passivation layer is preferable for TEM below the 450 °C in atmosphere.
TEM etching can be used to form structures with low process temperature, three-dimensional self-control ability, and low equipment requirement. It significantly simplifies micro/nano glass structure machining. Figure 14
demonstrates the grating diffraction property of a TEM glass with square arrays.
3.3. TEM Etching Mechanism Discussion
According to the phenomena listed in Section 3.1
, TEM SiO2
possesses three characteristics: (a) the modified depth is proportional to charge transfer; (b) the modified depth is proportional to bonding temperature; and (c) the modified SiO2
can be cured by thermal annealing.
Since different compositions in SiO2
seriously impacts the wet etching rate [24
], and the charge transfer in TEM changes alkali (Na) distribution, it is reasonable to infer that the Na concentration may determine the SiO2
etching rate. However, by comparing TEM samples with and without post annealing shown in Figure 15
, there is no significant re-doping phenomenon in the annealing samples, and both maintain a gradient of Na concentration. So, it is deduced that the etching rate acceleration in TEM does not depend on the Na compositions.
It is known that component change and thermal treatment might induce crystallization, and the crystallization leads to anisotropic etching. Therefore, 7740 glass, TEM glass, and TEM glass with post annealing were checked for crystallization using XRD. The results plotted in Figure 16
reveal that all of the samples are amorphous, which means that TEM and post annealing does not result in crystallization. Thus, the assumption that crystallization accelerates etching rate also does not hold.
Another supposed reason for TEM acceleration may originate from the stress. Stress-corrosion for glass has been verified, revealing that stress concentrators are responsible for a local corrosive acceleration [25
]. For anodic bonding wafer, residual stress caused by differences of thermal expansion coefficient of Si/glass has been sufficiently studied. Both experimental results and FEM simulation show that the stress concentrates on the bonding interface [26
]. Charge transfer is also proven to be a stress implantation process due to Na gradient [27
] and ions replaced [28
]. The stress induced by charge transfer locates in the sodium depletion layer near the bonding interface. Moreover, residual stress in anodic bonding can be cured by >500 °C [30
], which is consistent with aforementioned “characteristic c”. Based on the above analysis, it is reasonable to assume that stress-corrosion may lead to TEM acceleration. Relevant experiment about stress assumption and more research will be carried out in future work.