I. INTRODUCTION

At the mass-production line of LSI, particles generated in plasma etching process decrease the production yields and the overall equipment efficiency (OEE) ^[1-7]. This is because LSI circuit is shorted and the process must be stopped for periodical maintenance with cleaning of process chambers. The particles are mainly generated by corrosion of ceramic chamber parts and flaking off of the deposited film on inner walls of chamber. The generation of particles depends on the number of processed wafers ^[2]. In general, for a short while after the periodic chamber cleaning using chemicals, the particles are mainly caused by corrosion of chamber parts due to residual chemicals and moisture ^[1]. The number of particles temporarily decreases with the decrease in the residuals. After that, the etching reaction products attach and gradually deposit on the inner walls as more wafers are processed, with the result that the products form films and thicken. Then the particles are generated from flaking off of the films mainly due to electric field stress ^[8-10].

In mass-production lines, when the number of particles, which falls on the surface of a Si wafer and is counted by the wafer surface inspection system at sampling inspection, or the total etching time increases more than the management values, the process chambers are cleaned. The management by the inspection and the etching time, however, are indirect method, so that the cycle of such the periodic maintenance is not necessarily optimized. Optimizing the maintenance schedule can reduce maintenance and manufacturing costs, and therefore improve the OEE. In other words, a continuous and in-situ monitoring method for the particles is highly required to decide timing of the maintenance. Developing such a practical method will also allow a predictive maintenance regime to be used in mass-production lines.

In this study, we have worked on to demonstrate that the plasma impedance monitoring method is valid to monitor the generation of particles. By the in-situ and simultaneous monitoring of the plasma impedance and the particle, the relationship between the time variation of them has been investigated. Here, in the previous study, we have developed a shower-head-type electrode made from MgO-based ceramics, a high-corrosion-resistance ceramic material, and investigated the number of particles due to the corrosion by the reactive plasma ^[11]. To suppress that the flaked particles are generated, a substrate made from alumina is employed instead of a Si wafer. On the other hand, in this study, a Si wafer is used and therefore the etching reaction products easily generate. Under this condition, the particles caused by the corrosion of the electrode can be relatively negligible and therefore the main cause is the flaking off of the deposited film on the chamber inner walls which is composed of the etching reaction products. For comparison, the monitoring results which were measured in the previous study are also discussed although the results were not reported in the Ref. ^[11].

II. EXPERIMENTAL METHODS

The experimental apparatus is the reactive ion etching (RIE) equipment used at the mass-production line ^[8-10]. Fig. 1(a) and (b) show schematic top and side view of the experimental setup. The upper electrode is a shower-head type and made from the conductive MgO-based ceramics ^[12]. The composition is MgO-21% MgAl$_{2}$O$_{4}$-15% YAlO$_{3}$-3% C. The purity is higher than 99.9 % and the resistivity is the order of 10$^{0}$ ${\Omega}$·cm. The electrode was the same one used in the previous study ^[11] and has been carefully cleaned and installed at the chamber again.

A Si wafer is set on the powered electrode to which the rf power of 1000 W is supplied. The electrode gap for the powered and the ground electrodes is 60 mm. As the process gas, SF$_{6}$ is supplied from the upper electrode. The flow rate is 100 sccm at a pressure of about 10 Pa. Etching with process time of 15 s is repeatedly processed, and the total etching time is 600 min.

The laser light scattering method is used for the detection of particles ^[10]. A sheet-like laser beam is introduced into the process chamber approximately 4 mm below the ground electrode. Scattered light by particle is detected by charge-coupled device (CCD) camera. The images are captured approximately every 0.6 s. The setup in this study, a particle whose diameter is more than approximately 250 nm can be detected.

The impedance monitoring system measures load impedance from a 50 ${\Omega}$ transmission line ^[13,14]. The attenuated forward and reflected rf powers with a phase difference are measured by using the directional coupler, which is installed between the rf power supply and the matching circuit, and the measurement instrument, Cross Domain Analyzer$^{\mathrm{TM}}$ (CDA; Advantest). The load impedance, Z$_{L}$＝R$_{L}$＋jX$_{L}$, is the impedance of the load side from the output port of the matching circuit, where R$_{L}$ and X$_{L}$ are the real and imaginary parts of the load impedance, respectively, and j is the imaginary unit. The characteristic impedance, Z$_{CIM}$, calculated by the measured forward and reflected rf powers involves both the Z$_{L}$ and the impedance of the auto-matching circuit. The matching circuit contains the variable capacitors hence the system simultaneously measures the capacitances of them to monitor the Z$_{L}$. That is to say, by separating the impedance of the auto-matching circuit from the Z$_{CIM}$ based on the circuit equation, the Z$_{L}$ is monitored. During each etching process, the forward and reflected rf powers are recorded, and the values of R$_{L}$ and X$_{L}$ are calculated at a rate of 10 mega samples per second (MS/s) for 100 ${\mu}$s with the time intervals of 200 ms for the data acquisition. One of the biggest advantages of this method is that the system can be applied to the equipment in mass-production line without remodeling of them. Whereas the particle monitoring method is not simple that the system can be applied to such equipment although this is valid to in-situ monitoring of the particle.

Fig. 1. (a) Schematic top; (b) side view of the experimental setup.

III. RESULTS AND DISCUSSION

Fig. 2 shows the number of particles detected per one hour in the case that alumina substrate ^[11] or a Si wafer is set on the powered electrode. When the alumina substrate is used, the number of particles detected is few over the total etching time, whereas when a Si wafer is used, a lot of particles are detected. Under a Si wafer condition, after the etching time of 5 h the more the etching time increases, the more the number of the particles increases. More than one handled of particles per hour are observed at the etching time from nine to ten hours.

As shown the result of alumina substrate, the etching reaction of MgO-based ceramics hardly progresses, so that the particle caused by the corrosion is very low. On the other hand, the result of a Si wafer shows the typical tendency of particle generation, which is empirically known at mass-production line, caused by the flaking off of the deposited film on the chamber inner walls ^[1,2]. Under the mass-production condition, the wafers are processed one after another hence the etching reaction products gradually deposit on the inner walls of process chamber. The particles are generated by the flaking off of the deposited film and the number of particles increases with the number of the processed wafers, that is the total etching time.

Fig. 3(a) and (b) show the time variation of the load impedances R$_{L}$ and X$_{L}$ measured under the Si wafer condition. The averaged values measured during each etching process are plotted. The R$_{L}$ shown in Fig. 3(a) is approximately constant for 10 h, whereas the X$_{L}$ shown in Fig. 3(b) gradually decreases with the increase in the etching time, i.e., with the increase in the number of particles as shown in Fig. 2. Here, the R$_{L}$ mainly depends on plasma parameters such as electron density and electron temperature ^[15], so that the result shown in Fig. 3(a) indicates that plasma discharge condition does not change during this experiment. The X$_{L}$ shown in Fig. 3(b) decreases by approximately 1 ${\Omega}$ during 10 h. In the first half of the total etching time of 10h, the decrease in the X$_{L}$ is relatively gradual and the variation is large. Compared to these changes, after approximately 5 h of the etching time, the decrease in the X$_{L}$ becomes relatively large and the variation becomes low. The change in the X$_{L}$ would reflect the process of the deposition of the film on the chamber inner walls caused by the etching reaction product. In the first half of the etching time, the film begins to deposit sparsely on the inner walls, and by approximately 5 h, the film would deposit on the entire surface of the inner wall. And then the film stably thickens with the etching time, therefore the X$_{L}$ proportionally decreases because the increase in the film thickness leads to the increase in the capacitance composed of inner wall materials ^[15,16]. The extending of the sparse deposition area of the film would relate to the decrease in the variation in the X$_{L}$. Here, it is noteworthy that this tendency corresponds to the change in the number of particles shown in Fig. 2: before 5 h, the films form locally on the inner walls, so that the number of particles is relatively few, whereas after 5 h, the films generally deposit on the entire inner wall and thicken with the etching time hence the number of the particles increases.

The timing that the film entirely deposits on the inner walls and subsequent the increase in the film thickness and the number of particles can be indirectly monitored by the plasma impedance measurement. The results demonstrate that this non-invasive and in-situ monitoring system is valid to monitor the tendency of particle generation. Here, the deposited film is formed from the etching reaction product, so that the permittivity of the film and its flaked particles differs if other etching process recipe is used. In the equivalent circuit model of CCP discharge, the X$_{L}$ can be described as the series-connected capacitance components composed of the plasma sheath and the inner wall materials containing the deposited film ^[15]. Hence, theoretically, the X$_{L}$ depends on the permittivity of the deposited film ${\varepsilon}$$_{\mathrm{d}}$, as X$_{L}$ ${\propto}$ ${-}$1/${\varepsilon}$$_{\mathrm{d}}$.

Fig. 4(a) and (b) show the time variation of the load impedances R$_{L}$ and X$_{L}$ when the alumina substrate is used. The results show that the R$_{L}$ is approximately constant and the tendency of decreasing in the X$_{L}$ is not observed for 10 h. The results shown in Fig. 2 and 4(b) indicate that when the particles are not generated, the X$_{L}$ does not decrease. Note that compared with the variation in the X$_{L}$ in Fig. 3(b), that in Fig. 4(b) does not lessen as the etching time increases. As mentioned above, the etching reaction product hardly generates and the deposited film forms slightly and only sparsely under this condition hence the variation in the X$_{L}$ does not decrease. The X$_{L}$ depends on the capacitances of the plasma sheath and the inner wall materials ^[15,16]. The deposited film is dielectric hence the deposition on the surface of the inner walls decreases the ground-potential part. This would stabilize the plasma sheath conditions and decrease the variation in the X$_{L}$.

To consider the relationship between the time variation of the plasma impedance and that of the etching thickness of Si wafer, the thickness is estimated by the etching rate. Si bare wafer is processed under the etching condition same as that is used in the experiment of Fig. 3(a) and (b). The etching of 15 s is repeatedly processed and the wafer is totally etched for 10 min. From the etching depth, the etching rate is estimated to approximately 0.5 ${\mu}$m/min. The etching thickness for Si bare wafer can be considered to increase in proportion to the etching time hence the total thickness for 10 h is estimated to 300 ${\mu}$m. As shown in Fig. 3(b), the X$_{L}$ changes approximately 1 ${\Omega}$ for 10 h. Hence the sensitivity of the impedance value according to the etching thickness can be calculated as 0.0033 ${\Omega}$/${\mu}$m. In mass-production line, the period of the chamber maintenance must be set for the etching time longer than 10 h, therefore the present sensitivity that can detect the change in the impedance of 1 ${\Omega}$ for 10 h would be effective. In near future work, we will try to upgrade the system and improve the sensitivity.

Fig. 2. Number of particles detected per one hour of the etching time.

Fig. 3. Time variation of (a) the real part of load impedance, R$_{L}$; (b) the imaginary part of the load impedance, X$_{L}$ under Si wafer condition. The averaged values measured during each etching process are plotted.

Fig. 4. Time variation of (a) the real part of load impedance, R$_{L}$; (b) the imaginary part of the load impedance, X$_{L}$ under alumina substrate condition. The averaged values measured during each etching process are plotted.

IV. CONCLUSIONS

The number of particles which originates from etching reaction product is investigated under the condition that the gas shower type electrode made from the MgO-based ceramics and a Si wafer are used. The measurement results of plasma impedance clarify that the imaginary part of plasma impedance decreases with the increase in the number of particles. This method is valid to monitor the tendency of the increase in particles. In this feasibility study, the results also suggest that the plasma impedance monitoring system is effective for the monitoring of the detailed process of the film deposition on the inner chamber wall. In near future work, we will upgrade the monitoring method and investigate the relationship between the film deposition, the particle generation, and the plasma impedance in detail. This non-invasive and in-situ monitoring system can be easily applied to the process equipment at mass-production line, contributing to the development of the predictive maintenance method and improvement of the production yield and the OEE.