An Approach to Reduce Greenhouse Gases in the Semiconductor Industry Using F 2 Dissociated in Plasma for CVD Chamber Cleaning

Featured Application: CVD Cleaning process for Semiconductor industry. Abstract: The gases used in industrial cleaning processes are considered greenhouse gases with a high global warming potential (GWP). It is important to provide a viable alternative chemical vapor deposition (CVD) cleaning gas that is capable of removing efﬁciently deposited layers on the CVD chamber inner wall and other parts of the apparatus. The cleaning gas has to be environmental friendly in order to avoid accentuation of the global warming phenomena. Besides that, the alternative cleaning gas should be compatible with the existing gas delivery system and the CVD equipment that is already used by industrial companies. Only by fulﬁlling the three requirements mentioned above is it possible to replace the well-established cleaning gases. In this project, an F 2 –gas mixture for the in-situ cleaning of CVD chambers has been studied and compared with conventional cleaning methods. The conventional cleaning process is deﬁned as a cleaning procedure using either C 2 F 6 in RF plasma discharge or NF 3 in remote plasma discharge. production procedure, where 25-wafer lots were automatically processed through the AltaCVD tool. A total of 500 wafers were processed, depositing 0.8 µ m of SiO 2 on to the Si substrates followed by the F 2 -based cleaning process. The optical inspection showed no unusual wear of the process kit parts, no degradation of O-rings made of perﬂuoroelastomer, no tool downtime, and no noticeable chamber damage. The measured particles density was compared with the NF 3 clean. These results are a good indication for a possibility of substituting the NF 3 via the F 2 –gas mixture clean. The substitution does not necessitate any modiﬁcation of the existing hardware used by the semiconductor industrial companies.


Introduction
Semiconductor chip manufacturers use in their in-situ processes a variety of gases to pattern silicon wafers and to reach a rapid cleaning of chemical vapor deposition (CVD) tool chambers. To maintain a high throughput during the thin film deposition processes, it is essential to optimize the cleaning step in a way to remove perfectly the residual thin films deposited on the CVD chamber inner wall without generating any particles. The presence of particles inside the CVD chamber will lead to a degradation of the device performance.
Under industrial process conditions, non-negligible amounts of unreacted fluorinated gases ranging between 10-80% are released into the atmosphere. Unfortunately, they are considered as greenhouse gases with a high global warming potential [1]. Greenhouse gases in the upper atmosphere avoid releasing the infrared radiation to the universe by reflecting it backwards to the earth. The global warming potential (GWP) of a gas is generally measured in a 100-year time horizon and expressed in CO 2 equivalents. The GWP value depends on the lifetime of the gas in the atmosphere and its ability to absorb radiation.
In the last climate conference in Paris, which took place on 2015, a binding and universal climate target was set [2]. An agreement to limit the temperature increase to a maximum of 2 • C by 2100 was made.
Most of the gases employed by the semiconductor industry are CF 4 , C 2 F 6 , and C 3 F 8 , which are named perfluorocarbons, nitrogen trifluoride NF 3 , and sulfur hexafluoride SF6. According to the World Semiconductor Council, in 2013 the semiconductor industry used 7512 t of NF 3 , 1133 t of CF 4 , and 708 t of C 2 F 6 on a global basis [3]. Approximately 72% of all semiconductor industry gaseous emissions are caused from these three gases. Although the semiconductor industry is only a small contributor to global warming, several fluorinated gases with a high global warming potential (GWP) are intensively used, and large quantities of Perfluorinated compounds (PFCs) are emitted. NF 3 is the mostly used gas with a GWP of 17,200, followed by C 2 F 6 and CF 4 with GWPs of 1200 and 7390, respectively [4]. The main applications of these gases are to remove the residual films remaining inside deposition chambers after a plasma-enhanced chemical vapor deposition process (PECVD). The most important materials to remove are dielectric layers such as silicon oxides (SiO 2 ), carbon-containing silicon oxides, silicon nitrides (Si3N4), and, to a certain extent, conducting films such as doped polysilicon and silicide layers.
Both Fraunhofer EMFT (Research Institution for Microsystems and Solid State Technologies) and Solvay have joined their efforts in a research program to find safe alternative gases to be used in the manufacturing processes for ensuring climate safety. Our goal is focused on finding an adequate gas for the semiconductor industry, which could be easily integrated to the existing CVD equipment so that buying any new systems or including important modifications to the onsite hardware are avoided. This means that the proposed approach has to be, on the one hand, a cost-effective solution for industrial companies, and on the other hand, reduce the amount of the used greenhouse gases. Here, we propose a fluorine gas mixture as an environmental-friendly gas for the CVD cleaning processes to replace C 2 F 6 and NF 3 gases in the near future.
In this paper, we report a comparison study between cleaning processes using C 2 F 6 , NF 3 , and a fluorine gas mixture (F 2 -gas mixture) performed in a 200-mm wafer-size CVD tool for SiO 2 thin-layer deposition [5,6]. The process chamber ran with the new cleaning chemistry and was inspected on a regular basis during the period of this study in order to detect any equipment degradation related to using the new gas mixture.

Hardware
Before describing the PECVD systems used to realize this work, we would like to first give some clarifications about the safety needed for using an F 2 -gas mixture with the existing hardware.
The delivery lines of an F 2 -gas mixture (F 2 /Ar/N 2 ) from the gas cylinder to the CVD chambers were modified; however, most parts of the already existing single-wall stainless steel lines could be reused. A dry N 2 gas flow insured the line purge for an easy change of the F 2 cylinder. In some papers [7], it is relevant to use double-wall stainless tubes for flowing pure F 2 gas, or a gas mixture containing more than 40% of F 2 . For our study, as the F 2 -gas mixture has lower F 2 content, it was possible to flow it using the existing hardware facilities. Based on the long-term experiences of the Solvay Fluor GmbH in the field of F 2 and F 2 -gas mixtures, a passivation of the newly installed gas facilitation was performed [8]. It was important to slowly flow the F 2 -gas mixture in the gas delivery system. The formation of metal fluorides inside the stainless steel surface of the gas line (Fe, Cr, Ni) due to reactions of the stainless steel elements with fluorine atoms permits achieving a durable passivation. Thus, the gas lines are protected from any corrosion. Further safety precautions were taken into account by placing the gas cabinet, the gas delivery system, and the F 2 cylinder inside a closed cabinet with a safety exhaust. F 2 sensors installed in the safety exhaust of the gas cabinets near the system were connected to the safety valve at the F 2 -gas cylinder and automatized to cut off gas flow in case of F 2 detection.
In this study, the test runs were realized in two different PECVD systems that were located in the clean room of the Fraunhofer EMFT. 200-mm Si substrates were used. Both PECVD chambers were evacuated down to approximately 40 mTorr using a dry pumping system. One of the PECVD systems was a lamp-heated parallel-plate PECVD reactor from Applied Materials. TEOS (Tetraethyl orthosilicate)-based SiO 2 films were deposited by applying 13.56 MHz RF power (radio frequency). The standard cleaning chemistry was C 2 F 6 /O 2 with a small amount of NF 3 addition in the second cleaning step. An additional F 2 /Ar/N 2 gas line was installed, connecting a new gas cabinet located outside the cleanroom with the PECVD reactor.
The other PECVD chamber was made by the company Kobus, which is situated in France. The Aluminium Nitride (AlN) susceptor, which is a chuck for silicon wafers, has a resistive heating element filament to heat it to a desired chuck temperature. The standard cleaning process uses a mixture of C 2 F 6 and O 2 for chamber cleaning after oxide deposition. After polysilicon deposition, a NF 3 gas using a remote plasma source (RPS) unit was used as the chamber clean. The RPS running at 400 kHz (MKS; "Paragon", 6 kW RF power) was mounted on the top of the Alta-CVD reactor by Fraunhofer EMFT to allow chamber cleaning using the F 2 -gas mixture as well as the widely used NF 3 process. Therefore, the original gas mixing block was modified. The RPS sits directly above the top gas inlet of the chamber lid, whose hardware was adapted to shorten the distance between the RPS outlet and the chamber inlet in order to reduce the path of the generated F-radicals by some cm to reach the CVD chamber.
It is worth mentioning that the same type of the gas mass flow controller (MFC) was used in both chambers for direct comparison and control of the total amount of F 2 -gas mixture flow. The gas mixture used for the RPS cleaning technique was 30% F 2 /Ar/N 2 . Since the gas cabinet for this chamber was directly located near the PECVD chamber in the cleanroom, only 10-l cylinders with a maximum pressure of 100 bar were allowed to be placed in the grey room, which is the service area of the clean room, to ensure safe operating conditions. For safety reasons, it is strongly advisable to avoid adding burnable gases to F 2 .

Method to Estimate Cleaning Efficiency of F 2 -Gas Mixture
The etching rates of the SiO 2 layer were gained by measuring SiO 2 film thicknesses pre-etched and post-etched under the desired etching conditions. The obtained etch rate of SiO 2 reflects the efficiency of the CVD chamber clean. Firstly, the 200-mm blank silicon wafers were thermally oxidized to obtain a 530-nm oxide layer. This protective layer was necessary to prevent the silicon test wafers from unwanted edge etching of the silicon substrate during the cleaning step, measure the etched SiO 2 during the cleaning process, and allow an easy wafer substrate recycling. Secondly, a 2.4-µm thick layer of PECVD-based SiO 2 film was deposited in two cycles. The SiO 2 layer thickness was measured using a reflectometry and/or ellipsometry (tencor UV1280 SE, KLA-Tencor, California, USA) method. A silicon wafer with a defined layer thickness of SiO 2 was introduced inside the chamber on the susceptor, and the chamber clean was run with the desired gas for a fixed cleaning time. By partially etching the SiO 2 , the wafer surface is specular.

Comparative Study of RF Plasma Cleaning Using C 2 F 6 and F 2 -Gas Mixture
The purpose of this experiment is to compare the cleaning effect of F 2 -gas mixture to the standard used cleaning process based on C 2 F 6 /O 2 chemistry. A Taguchi Type L9 design of experiments (DOE) was adapted for conceptualizing the experiment. Etch rate data generated with the conventional 13.56-MHz PECVD system (FOGALE Nanotech , Nimes, France) were stable and well repeatable within a ±3% range [9].
Susceptor temperature was set to 400 • C. The temperature of the chamber wall was set constant at 55 • C by continuous water cooling. Figure 1a-d shows the SiO 2 cleaning rates, under RF plasma cleaning, dependent on varying four parameters: chamber pressure, total flow amount of gas, RF power, and spacing between showerhead and the susceptor. The etching uniformity of SiO 2 films was below ±5%. The etch rate is an average value of at least four identical wafers with an SiO 2 layer. These results are in good agreement to previously reported data [9,10]. The addition of Ar gas in the F 2 -gas mixture is needed, on one side, for a safe and stable plasma ignition, and on the other side, to keep the etch non-uniformity of SiO 2 below ±8%. In the absence of an argon element, the etch non-uniformity of SiO 2 rose up to ±40%.

142
These results are in good agreement to previously reported data [9,10]. The addition of Ar gas in the 143 F2-gas mixture is needed, on one side, for a safe and stable plasma ignition, and on the other side, to bars are ±8% and ±3% for plasma cleaning using F2/Ar/N2 mixture and C2F6/O2, respectively.

153
As can be seen in Figure 1, for all four varied parameters, the curves of the  [12]. Therefore, F atoms form more easily from the F2-gas mixture at comparably low plasma energy.

162
In contrary, to reach the same density of F atoms from C2F6 molecules, a high excitation energy is 163 required.

164
The concentration of F atoms is correlated to the etch rate for capacitively-coupled plasma 165 discharge. A kinetic study showed that SiO2 removal by spontaneous chemical reaction is dominant 166 Figure 1. a-d. SiO 2 etch rates using F 2 /Ar/N 2 mixture and C 2 F 6 /O 2 as a function of (a) pressure, (b) plasma power, (c) spacing between the showerhead and the susceptor, and (d) gas flow . The error bars are ±8% and ±3% for plasma cleaning using F 2 /Ar/N 2 mixture and C 2 F 6 /O 2 , respectively.
As can be seen in Figure 1, for all four varied parameters, the curves of the SiO 2 etch rate show the same trend for both gas chemistries under the same conditions. However, the cleaning rate with the F 2 -gas mixture is higher in comparison to the one obtained for the cleaning gas based on C 2 F 6 . It was observed that the distance between the parallel plate reactor, showerhead and susceptor is an important factor, as are the RF power and working pressure, which influence the cleaning rate of the PECVD chamber. The SiO 2 etch rate for the F 2 -gas mixture increases by a factor~1.3 due to the lower bond dissociation energy of 155 kJ/mol at 298 K needed to break the F 2 molecules [11].
In comparison to C 2 F 6 molecules, a higher dissociation energy, which is approximately 402 kJ/mol at 298 K, is needed [12]. Therefore, F atoms form more easily from the F 2 -gas mixture at comparably low plasma energy. In contrary, to reach the same density of F atoms from C 2 F 6 molecules, a high excitation energy is required.
The concentration of F atoms is correlated to the etch rate for capacitively-coupled plasma discharge. A kinetic study showed that SiO 2 removal by spontaneous chemical reaction is dominant at high temperatures of 300 • C-400 • C, while SiO 2 removal by ion-enhanced removal is dominant below 100 • C [13]. Consequently, the increase of the flow rate of the F 2 -gas mixture presumably increases the atomic F concentration present inside the CVD chamber, resulting in an increase of the oxide layer's etch rate. However, a high flow rate of the clean gas had no time to reach a complete dissociation.
The gas mixture (20% F 2 / 70% Ar + N 2 ) provides a very high efficiency of cleaning in conventional CVD reactors. The total amount of fluorine consumption can be reduced by a factor of four to 10, depending on the chosen process parameters and the PECVD chamber type. This estimation is based on the comparison of the required total mass flow of F atoms between the two chemistries for the same amount of SiO 2 etched. Figure 2 summarizes the etch rates of the SiO 2 layer as a function of the fluorine mass in the total cleaning gas used by flowing into the chamber for one minute. An RPS unit was used to activate the desired cleaning gas. In our case, an F 2 -gas mixture and NF 3 were the cleaning gases to be tested. Several F 2 -gas mixtures of F 2 , Ar, and N 2 were received from the Solvay company, having a maximum concentration of 30% of F 2 . The percentage of the N 2 and Ar in the F 2 -gas mixture were varied to get the optimal mixture giving a high etch rate of the CVD chamber. The reason was to find the best F 2 -based clean gas mixture to be tested on site by our industrial partner.

Comparative Study of RPS Plasma Cleaning Using NF 3 and F 2 -Gas Mixture
The RPS unit permits a high dissociation of NF 3 , reaching more than 95%. From the reached etch rate based on the F 2 -gas mixture, we suppose that the dissociation of the F 2 molecules reached high values comparable to the NF 3 chemistry. Additionally, a certain amount of species will recombine on their way into the process chamber, independently of the cleaning chemistry used. The RPS unit is optimized and reprogrammed specifically to the use of NF 3 gas, which is widely employed in the industry. Figure 2 shows that the SiO 2 etch rates increase linearly by increasing the F 2 content in the total gas flow, whatever the cleaning gas used. For example, for an F 2 -gas mixture containing a total mass fluorine equal to 0.25 g reaches an etch rate of 1150 nm/min. However, by using NF 3 containing the same total mass of fluorine, a lower etch rate 750 nm/min is obtained.   cleaning gas used by flowing into the chamber for one minute. An RPS unit was used to activate the 179 desired cleaning gas. In our case, an F2-gas mixture and NF3 were the cleaning gases to be tested.

180
Several F2-gas mixtures of F2, Ar, and N2 were received from the Solvay company, having a maximum 181 concentration of 30% of F2. The percentage of the N2 and Ar in the F2-gas mixture were varied to get 182 the optimal mixture giving a high etch rate of the CVD chamber. The reason was to find the best F2-183 based clean gas mixture to be tested on site by our industrial partner.

184
The RPS unit permits a high dissociation of NF3, reaching more than 95%.   The high etch rate obtained for the F 2 -gas mixture is explained by the presence of an optimal percentage of the N 2 gas. Another explanation is the much lower plasma energy needed to completely dissociate F 2 -gas mixtures compared to NF 3 [11]. It is important to take into account the recombination phenomena depending on the operating parameter and the hardware geometry [14].
To gain more knowledge about the role of N 2 in the chamber cleaning, we represent in Figure 3 the SiO 2 etch rate as a function of the N 2 concentration in the F 2 -gas mixture for three different total masses of fluorine. A N 2 concentration between 35-45% in the F 2 -gas mixture is believed to be optimal. However, a concentration of F 2 higher than 30% in the mixture was not possible to study due to road traffic safety rules in Germany, which forbid the transportation of gas cylinders with such a concentration of F 2 . According to the literature, N 2 takes the role of the oxygen catalyst by forming N 2 O during the SiO 2 cleaning, and thus increasing the removal rate of SiO 2 from the reactor surfaces [15,16].

209
To gain more knowledge about the role of N2 in the chamber cleaning, we represent in Figure 3 210 the SiO2 etch rate as a function of the N2 concentration in the F2-gas mixture for three different total 211 masses of fluorine. A N2 concentration between 35-45% in the F2-gas mixture is believed to be 212 optimal. However, a concentration of F2 higher than 30% in the mixture was not possible to study 213 due to road traffic safety rules in Germany, which forbid the transportation of gas cylinders with 214 such a concentration of F2. According to the literature, N2 takes the role of the oxygen catalyst by surfaces [15,16]. 226 Figure 3. SiO 2 etch rates per minute as a function of the N 2 percentage in an F 2 -gas mixture for different total mass flows of fluorine in the F 2 -gas mixture. The following parameters were kept constant: total flow, spacing between 200 mils to 650 mils, T = 400 • C, pressure between 3 Torr for inner clean and 4 Torr for outer clean, clean time = 45 s. The error bar is ±10% for plasma cleaning using an F 2 -gas mixture.

Particles Generation and Process Kit Inspection
A "mini-marathon run" was performed to simulate a typical production procedure, where 25-wafer lots were automatically processed through the AltaCVD tool. A total of 500 wafers were processed, depositing 0.8 µm of SiO 2 on to the Si substrates followed by the F 2 -based cleaning process. The optical inspection showed no unusual wear of the process kit parts, no degradation of O-rings made of perfluoroelastomer, no tool downtime, and no noticeable chamber damage. The measured particles density was compared with the NF 3 clean. These results are a good indication for a possibility of substituting the NF 3 via the F 2 -gas mixture clean. The substitution does not necessitate any modification of the existing hardware used by the semiconductor industrial companies.

Conclusions
According to the present study, we demonstrate that the F 2 -gas mixture is an adequate candidate to be widely extended for cleaning CVD apparatus in conjunction with the existing hardware equipment. Its efficiency, by looking to the etch rate parameter, is largely higher than that of C 2 F 6 /O 2 using an RF discharge frequency of 13.56 MHz. It was proven that a CVD cleaning-based remote plasma source (RPS) using an F 2 -gas mixture is comparable, and in some conditions even better than the one offered by an NF 3 clean. By realizing a 500-wafer run, the etch rate stability was confirmed using the cleaning based on an F 2 -gas mixture. No abnormal amount of generated particles inside the CVD chamber and erosion of the process kit parts were observed.
Chamber cleaning with an F 2 -gas mixture compared to conventional ones, based on C 2 F 6 discharged by RF plasma or NF 3 by remote plasma, increases equipment uptime and throughput, on the one hand. On the other hand, it decreases the usage of the amount of chemical products, and therefore the cost of materials. As the molecular fluorine has no global warming potential, the F 2 gas represents a beneficial cleaning gas alternative for the CVD equipment.
Author Contributions: Jamila Boudaden and Robert Wieland designed the experiments based on previous realized work and realized the research project. Stephan Altmannshofer and Robert Wieland contributed to the experimental work. Jamila Boudaden wrote the paper with input from all authors. Micheal Pittroff and Ignaz Eisele discussed and commented on the manuscript.