Experimental Study on Seismic Vibration Control of Stockers in Wafer and LCD Panel Fabs

Automated stocker system is widely used in semiconductor and liquid crystal display (LCD) industries for handling and storage of valuable wafers or glass panels. Massive front opening unified pods (foups) containing wafers, or cassettes storing glass panels, are placed in shelf stockers during manufacturing. Although several preventative measures have been taken, during the past earthquakes, substantial financial loss from the industries were reported, and one of the main causes was attributed to collision of the foups or cassettes and shake off from the shelfs. This paper proposes a methodology of incorporating viscous fluid dampers into the stokers to mitigate their seismic response. Unlike conventionally been done in buildings where dampers are placed between adjacent stories, it is proposed to install dampers in between the ceiling and top of the stocker. Such configuration utilizes the large velocity at the stocker top under vibration, resulting in smaller damper size, and enables a leverage mechanism that requires smaller damper force to resist the stocker’s vibration. Both shake table tests and simulation of a full-scale stocker under realistic earthquakes have been conducted. Results indicate that both displacement and acceleration responses of the stocker can be significantly reduced, and dynamic response of the ∗Corresponding Author Email address: Yung-Tsang.Chen@nottingham.edu.cn (Yung-Tsang Chen) Preprint submitted to Journal of LTEX Templates March 16, 2021


Introduction
Locating at the boundary between the Eurasian and the Philippine Sea tectonic plates, Taiwan, an island in East Asia, has seen many earthquakes resulting from movement of the tectonic plates in its history. As a result, implementing seismic design in building codes is compulsory in Taiwan, in order to construct 5 earthquake resistant buildings and infrastructures. Serving as an important base in world supply chain of electronics and computers, Taiwan has set up three major industrial zones, namely Hsinchu, Taichung, and Tainan Science Parks, to accommodate the companies involved in design, testing, and production of these products. Taiwan's economy relies heavily on the high technology industries. 10 In 2019, the three Science Parks posted a combined revenue of NT $2.63 trillion (US $87.8 billion) [1]. Prevent earthquake damage to the buildings and facilities in the Science Parks so as to keep economy growth, has become one of the main priority for the Science Parks' administration bureau. While building codes are designed mainly to provide protection of the lives and property but not con-15 tent of the buildings, damage to vibration-sensitive manufacturing equipment and consequently shut down of production during earthquakes, which might be inevitable, could impair the country's overall economy significantly.
On February 6, 2016, an earthquake with a Richter magnitude of 6.6 struck Tainan Science Park in the southern part of Taiwan. The recorded peak ground reported loss from the industries was mainly due to damage of manufacturing facilities. Substantial financial loss estimated to be tens of billions of U.S. dollars from the industries have been revealed, in particular, semiconductor and LCD industries reported significant loss due to the earthquake, which, after field 30 investigations, was attributed to the damage of the valuable wafers and glass panels.
In semiconductor and LCD industries, wafers made of silicon are used for the fabrication of integrated circuits chips, while glass panels consisting of a layer of liquid crystal material supported by two glass plates are used in the pro-35 duction of LCD monitors, both of which are crucial and valuable components.
Automated stocker system is widely used in these two industries to handle and store the wafers and glass panels during production. Massive front opening unified pods containing wafers, or cassettes storing glass panels, are placed in shelf stockers during manufacturing. The stockers in the automated stocker system shelf of the stockers are expected to be larger than the floor acceleration during earthquakes. This is escalated by the fact that maximum floor acceleration subjected to major earthquake loading is often amplified over the peak ground acceleration. In the February 6 earthquake, although the PGA was merely 300 gal, the recorded peak floor acceleration (PFA) from various companies ranged 50 from 400 gal to 600 gal, depending on the floor elevation and structural types.
As a result, foups or cassettes on the shelf stockers are prone to collide and even shake off during large stocker's vibration, which is evident as has been reported in this earthquake. Remedial measurements to tackle this problem therefore has become a top priority for company leaders in the industries.

55
In the literature, very limited research was found in mitigating vibration of factory facilities such as automated stocker system due to earthquakes. In a 3 pioneer work by Wang et al. [3], a series of shake table tests were conducted on a stocker that was 4 m wide, 3.05 m long, and 7.68 m high, with a total mass of 5650 kg. Polyvinyl chloride (PVC) panels were used to simulate glass panels in 60 the cassettes. The proposed remedial measurements included (i) installation of stoppers at the edge of the shelf, (ii) installation of braces and (iii) installation of viscous dampers. Results indicated that when the peak ground acceleration of the input excitation was larger than 300 gal, the cassette started to slide and collide with the stopper. The large inertial force from the self-weight of the 65 cassette (about 1000 kg) caused the PVC panels to eject from the cassette under impact loading. Results from the stocker with braces installed from bottom to top in all shelfs of the stocker revealed that, although the drift of the stocker decreased due to brace stiffening, the increased lateral stiffness also resulted in large shelf acceleration, which contradicted the purpose of installing the braces.

70
For the case with four viscous dampers installed in the first shelf, results showed also a reduced drift response of the stocker but in a less extent as compared to the braced counterpart; however, the measured shelf acceleration still increased slightly as compared to the original un-reinforced stocker, which failed to meet the design objective.

75
Another work done by Wang et al. [4] focused on strengthening the joints between the base of the stocker and the floor, as the poor seismic performance of the stocker was deemed to be attributed to the poor detailing of the connectors, leading to rocking of the stocker during earthquakes. In this work, a smaller stocker that was 1.69 m wide, 3.95 m long, and 4.18 m in height with a total 80 mass of 1700 kg, was used as the test frame. Similar to previous work by Wang et al. [3], lateral braces and stoppers at the end of the shelf edge were installed to stiffen and to stop the cassettes from shaking off, respectively. To prevent the stocker from sliding, more foot mounts were also installed. While several improvement was made on the connection details, results from the shake table 85 tests nonetheless indicated that the overall sliding and drift of the frame were reduced, but shelf acceleration was compromised, in particular, the stocker's peak shelf acceleration was amplified by nearly 2.5 times as compared to the 4 un-reinforced counterpart. Even with the mounting of the stopper to keep the cassettes from shaking off, the large impact force due to collision of the cassette 90 and stopper caused damage to the stopper, and eventually leading to ejection of the PVC panels.
Viscous fluid dampers are often applied in bridges and buildings to mitigate structural sway, by providing additional damping to the main structures.
They have been proven to adequately protect structures against earthquakes 95 [5,6,7,8]. While most of the applications of viscous dampers in civil engineering are on the mitigation of structures due to seismic and wind-induced vibration, in the literature some research works are focused on controlling the vibration of specific objects or building content in the structure. For example, Asfar and Akour [9] presented a numerical study for the suppression of self-excited oscilla-100 tor using an impact viscous damper. Lin et al. [10] proposed a micro vibration mitigation system using viscous dampers to reduce the vibration in a high-tech building. Hong et. al. [11] presented a three-dimensional analytical study of a hybrid platform on which high-tech equipments are mounted for their vibration mitigation. The literatures mentioned above have proven that small, customized 105 viscous dampers are effective means of reducing the unwanted vibration of objects or building content. However, in the application of adopting viscous fluid dampers in stockers by Wang et al. [3], where four dampers were installed diagonally at first shelf of the stocker, the dampers did not perform well. Although not explicitly shown, the main cause may be attributed to the small inter-shelf 110 drift under given excitations, which restricted the types of dampers that can be used, and this affected the dampers' performance significantly.
In this paper, a practical approach of incorporating viscous fluid dampers into the stokers to mitigate their seismic response is proposed. To tackle the issues with simple yet feasible solution, unlike conventionally been done by Wang

Experimental program
A test specimen representative of typical stockers used in semiconductor industry is selected at the outset. The test stocker is provided by a semiconductor company, with the same structural details as those used in its factories.
Although the stockers in different companies may vary, the aim of the experi-130 mental program is to prove that the proposed methodology will certainly work on the chosen stocker, it can also be easily adapted to suit different types of stockers. Design and details of the test specimen, instrumentation, and test setup are described as follows.

Details of test stocker, steel frame, and viscous dampers 135
The test specimen is of frame type with shelfs installed at upper half of the stocker, as can be seen in Fig. 1 For optimal performance of the dampers on vibration control of the stocker, a numerical study by Chen [12] was first conducted. From the parametric study of the dampers, a linear damper with a damping coefficient C of 4.9 N·sec/mm 160 is suggested, and customized dampers are first manufactured as recommended, followed by component tests of the damper by inputting sinusoidal excitation at various frequencies and amplitudes. Table 1 summarizes the test results. It can be seen from Table 1 that the maximum deviation of the damping coefficients obtained from various tests is 6.4%. 165

Test setup and instrumentation
The test specimen has four rows of shelfs at upper half of the stocker, capable of storing wafer boxes during manufacturing, as can be seen in the test setup shown in Fig. 3. The stocker was bolted using 8 M10 bolts to a horizontal frame with dimensions 1.5 m by 1.35 m and the frame was fixed to the shake 170 table with 4 M10 bolts. The stocker was also laterally supported at its base by stainless steel brackets, which were bolted to the shake table. The one story steel frame simulating the seismic response of a semiconductor fabrication plan was supported by four base piers of 0.43 m in height, as shown in Fig. 3.
Both connection (frame to piers and piers to shake table) are fixed with bolts.

175
Preliminary system identification test of the frame was conducted, and result indicated that the fundamental frequency and damping ratio of the frame were 3.06 Hz and 0.3%, respectively. Fig. 4 shows photographs of the final test setup.

7
After the assembly of the stocker and frame on the shake table, to measure the acceleration responses of both, accelerometers were first installed, two on 180 top of the steel frame (AS1 and AS2), two on top of the stocker (ASTK1 and ASTK2), and two on the highest shelf of the stocker (ASTK3 and ASTK4), as shown in Fig. 5. To record the actual inputted ground acceleration, an accelerometer (AG) was also installed on the shake table. Ground (table) displacement relative to the strong floor was measured by a linear potentiometer 185 (DG). In addition, two laser displacement sensors (LD1 and LD2) were placed between the stocker top and the steel frame to measure the stroke of the two viscous dampers, while one laser displacement sensor (LD3) was placed at the reference frame to measure the movement of the steel frame relative to the strong floor. The recorded stroke histories of the two dampers also represent 190 the relative displacement between the stocker and the ceiling as represented by the steel frame. A data acquisition system with 16 channels was in place to record data for dynamic signals at a sampling rate of 100 Hz. The shake table is an uni-axial earthquake simulator with a payload of 100 kN and a maximum traveling distance of ±125 mm.

195
To assess seismic performance of the stocker, one representative historical earthquake, namely the 1995 Kobe earthquake, was considered. The time history of the Kobe earthquake with PGA scaled to 150 gal and its amplitude spectrum are shown in Fig. 6. The performance of the stocker with and without added viscous dampers was evaluated using the selected earthquakes at floor 200 level. The input earthquakes were first scaled linearly based on the desired peak floor acceleration to peak ground acceleration ratio, and were subsequently used as the input floor excitations for the stocker. Due to the fact that the recorded peak floor acceleration from various companies in Tainan Science Park ranged from 400-600 gal during the 2016 February 6 earthquake, for the shake table 205 test, the maximum PFA was scaled from 150 to 700 gal (when implemented with the seismic dampers), to accommodate the possible scenarios in an earthquake event. It should be noted, however, that for the original stocker without dampers, Kobe earthquake with a PFA scaled to 150 gal was used, to avoid 8 possible damage to the specimen. Seismic response of the original stocker with 210 larger PFA, e.g. 400 gal to 700 gal, was obtained by scaling linearly the response of the stocker subjected to the same earthquake but with a PFA of 150 gal, assuming linear elastic response of the stocker.
To test the effectiveness and efficiency of the added reaction beam-viscous dampers assembly, the 1995 Kobe earthquake with target PFA levels of 400, 600, 215 and 700 gals were pre-selected. In order to drive the earthquake simulator, which is displacement control, the input acceleration history needs to be integrated twice to derive displacement history, followed by a baseline correction to give final displacement input. The base line correction of the earthquake record is needed as double integration of an earthquake acceleration history may be 220 different from the corresponding displacement history of the same earthquake.
The technique proposed by Chiu [13] is adopted in this paper to process the acceleration data to derive the displacement history of the Kobe earthquake for the earthquake simulator. The resulting PFA for Kobe earthquake from experiments were 415, 614, and 717 gals.

Test Results
Performance of the stocker with the added reaction beam-viscous dampers system is mainly assessed by the response reduction in acceleration measured by the accelerometers at top of the stocker (ASTK1 and ASTK2) and at the highest shelf (ASTK3 and ASTK4), as well as the response reduction in overall 230 displacement of the stocker. Experimental results from shake table tests will be discussed in detail as follows.

Kobe Earthquake with PFA=415 gal
As mentioned previously, for the earthquake on February 6, 2016 in Tainan, Taiwan, although the PGA was merely 300 gal, the recorded peak floor accel-   Table 2 summarizes the test results. It can be seen from Table 2 that maximum acceleration occurs at left side of stocker top (ASTK2), more specifically, the peak acceleration is reduced 255 from 2422 to 916 gal, equivalent to a peak acceleration reduction of 62%. Similar peak acceleration reduction is also observed at right side (ASTK1) of the stocker (53%), left side (ASTK4)of top shelf (58%), and right side (ASTK3) of top shelf (47%). The reduction in root-mean-square acceleration response for all sensor locations is also significant. It is worth noting that the rotation of the stocker is 260 well controlled by the added dampers, as the difference in acceleration response at two sides of the stocker, e.g. the difference between ASTK1 and ASTK2, seems have been minimized. Fig. 7(b) shows the measured displacement response of the stocker top and the stroke history of one of the dampers. It can be seen from Fig. 7(b) that 265 the roof displacement is reduced significantly. A peak response of the original stocker is observed to be 101.2 mm; with the implementation of the damper system, the peak is reduced to 73 mm, equivalent to 28% response reduction. It can also be seen from Fig. 7(b) that the maximum damper stroke is measured as 5.7 mm, which is well within the acceptable damper's stroke of 55 mm. For Since shake table test of the stocker with input Kobe Earthquake at a PFA of 416 gal has shown rather promising results, it is of interest to know whether the added dampers can provide similar protection for earthquakes with higher intensity. To this end, the same Kobe Earthquake but with a higher PFA of 614 gal is used, as up to 600 gal PFA was observed from the onsite measurement.  Table   2 summarizes the test results. As can be seen from Fig. 8 and Table 2, for the original un-controlled stocker, when it is compared with previous results (Kobe earthquake with 416 gal PFA as input), the stocker shows overall higher 285 acceleration response at its top and top shelf for all four sensors (ASTK1-4).
The peak accelerations at ASTK1,2,3, and 4 are 2745, 3587, 2365, and 3163 gal, respectively. When the dampers are added, the peak accelerations drop to 1097, 1105, 1095, and 1076 gal at ASTK1,2,3, and 4, respectively, corresponding to reductions of peak acceleration of 60%, 69%, 54%, and 66%. Similar response 290 reduction can be observed from the root-mean-square acceleration response. In this scenario, the stocker with the added dampers show overall higher response reduction as compared to the previous test with a PFA of 416 gal. Torsion of the stocker is also well controlled by the dampers, as the maximum acceleration on two ends of the stocker are fairly close. Moreover, the measured displacement 295 response of the stocker top, as can be seen from Fig. 8(b), is reduced significantly (up to 69% R.M.S. reduction as shown in Table 2). The maximum stroke of the damper (8.1 mm), although increases slightly as compared to the case for PFA=416 gal (5.7 mm), is still well within the acceptable limit of 55 mm. In this earthquake scenario, the overall response acceleration and displacement of 300 the stocker are both well controlled.

Kobe Earthquake with PFA=717 gal
Shake table tests of the stocker subjected to Kobe Earthquake with PFA=416 and 617 gal have shown very promising results in reducing both the acceleration and displacement responses; however, to accommodate possible scenarios 305 in future earthquakes, an earthquake event with a PFA larger than 617 gal may worth exploring. To this end, the Kobe earthquake with a PFA of 717 gal, to represent an ideal case of a 700 gal earthquake, is adopted as the seismic input at floor level.  Table   2 show that, if the stocker is equipped with the seismic dampers, the peak The proposed scheme also integrates the stocker into the ceiling, which enables 330 an efficient leverage mechanism for seismic control of the stocker, as the long moment arm measured from the base to the stocker top reduces the required damper force to resist the stocker's vibration. Table 2 summarizes the results of the stocker to Kobe earthquake with PFAs of 415, 614, and 717 gal. As can be seen from Table 2

Simulation of the Frame-Stocker-Damper System
In addition to the experimental program of the stocker which confirms the feasibility of the proposed reaction beam-viscous damper system, in this paper, a finite element analysis aimed at simulating the seismic response of the stocker is also conducted. The finite element software ETABS is used to create the 345 structural model for the stocker and to simulate the response of the stocker under input earthquakes. Results from finite element modeling will be compared with those from shake table testes, to verify the accuracy of the output and to prove whether the modeling provides a reliable and efficient means to support the design of stockers with the proposed reaction beam-viscous dampers system.

System Identification
As shake table tests of the stocker have been conducted, it will be beneficial for later analysis if the dynamic characteristic of the test stocker, including natural frequencies and damping ratios, can be identified via system identification using experimental data. In this study, system identification using ARX 355 (Auto-Regressive with eXogenous) [14], a linear regression model, is conducted and described briefly below.

13
Consider a single input and single output ARX mode, the mathematical model can be described using a linear differential equation as: (1) where y[·] and x[·] represent respectively the output and input signal of the system, a i and b i respresent the coefficients for output and input signal, respectively, n a and n b are the dimensions of the output and input signal. By taking z transform of Eq. (1), the frequency response function of the system can be written as: where where r j = p j p j , p j is the j − th root to x[z] = 0 and p j is the complex Re(pj ) ], Re(p j ) and Im(p j ) are the real part and imaginary part of p j , respectively. 360 Therefore, if the system coefficients a i and b i in Eq. (1) can be identified, the frequency response function, natural frequencies and damping ratios of the system can be obtained. In the ARX model, the measured acceleration history of the shake table is treated as the input, while the response acceleration on the frame top is treated as the output. Since two accelerometers (ASTK1 and ASTK2) are installed at two sides of the stocker, and there is obvious torsional 14 effect in the test model, the average acceleration of the two sensors is used as the output in the translational direction of the stocker, namely: The torsional response of the stocker can be extracted from subtracting ASTK2 from ASTK1 first, followed by dividing by the length between the two accelerometers (l s ) as: Since translational and rotational responses can be obtained from Eqs. (5) and (6), respectively, both translational and rotational modes of vibration can be extracted. For the one-story steel frame, from experimental results there is no obvious torsion thus only acceleration response in translational direction is identified.

365
For system identification purpose the frame and the stocker are subjected to Kobe Earthquake with different PFA at their base. Figure 10 shows the Fourier amplitude spectrum of the stocker in translational and rotational directions with different earthquake intensity, while Table 3 summarizes the natural frequencies and damping ratios from system identification of the stocker and frame. It can 370 be seen from Table 3 that, when the PFA= 129 gal, in the translational direction natural frequencies and damping ratios for modes 1 and 2 are 1.84 Hz and 4.17% and 3.00 Hz and 5.35%, respectively. From Figure 10(a) the first mode's peak is much larger than the second mode, indicating that mode 1 should be the translational model. It can also be seen from Table 3 that in the rotational 375 direction natural frequencies and damping ratios for modes 1 and 2 are 1.84 Hz and 5.72% and 3.03 Hz and 2.52%, respectively. From Figure 10(b) the peak for the second mode is much larger than the first mode, implying the second mode should be torsional mode of the stocker. Therefore, by combining the observations from the two amplitude spectra, one can summarize that the first 380 mode of the stocker is a translational mode, and its frequency and damping ratio are respectively 1.84 Hz and 4.17%. The second mode is a torsional mode, and its frequency and damping ratio are 3.03 Hz and 2.52%, respectively. Similar trend is also observed in the case for PFA=217 gal, as can be seen from Figure 10 there is no obvious torsion and the natural frequency of the first mode is 3.07 Hz. Damping ratio is 0.07 % when the PFA equals 129 gal, and it increases slightly to 0.24 % when the PFA reaches 217 gal.

Structural Modeling
Since the one-story frame is used to replicate one story of the fabs, the  Although not explicitly shown, the first mode frequency from the shake table  the smaller acceleration response in simulation. It can also be seen from Table 4 that the maximum stocker displacement at top center from experiments agrees reasonably well with the ETABS' output, with a 12.5% difference. in the original stocker is well under control. Fig. 12 (b) shows the comparison of the stocker's displacement history. A big difference is observed in the stocker's displacement, as can be seen in Fig. 12 (b). This may be attributed to the fact that in the experimental setup the stocker's displacement relative to the shake table is calculated indirectly from the recorded stocker's displacement relative to the ground (LD3), the shake table movement relative to the ground (GD), and two dampers' displacement (LD1-2) as follows:

Kobe Earthquake with PFA=415 and 614 gal
where (LD3−GD) gives the displacement of the steel frame relative to the shake In addition to the input Kobe earthquake with a PFA of 400 gal, in the ETABS simulation the target PFA is further increased to 600 gal, as it is the 460 highest floor acceleration observed in the past earthquake events in the Science Parks in Taiwan. Fig. 13 shows the comparison of test results and the output from ETABS of the stocker under Kobe earthquake with an achieved PFA of 614 gal. It can be seen from Fig. 13(a) that, the ETABS simulation shows very similar response acceleration from both accelerometers. The differences 465 in maximum acceleration in stocker top is about 2%, which shows very good agreement between the simulation and the experiment. Fig. 13 (b) shows the comparison of the stocker's displacement history. As can be seen from Fig.   13 Fig.14 that the energy been dissipated increased with increasing earthquake intensity, which is expected as larger damper force is involved in a more violent earthquake scenario.

Conclusion
A methodology is proposed in this paper for semiconductor and liquid crys-