SEERs-based process control and plasma etching
06/01/1999
Electron density, electron collision rate, bulk power, and peak voltage can be used to characterize plasmas used in semiconductor production, particularly in plasma etching. Practical implementation is via self-excited electron resonance spectroscopy, which has
been used to measure the dependence of critical internal plasma parameters on variations of external parameters and changes in wafer structure. Work reported here has enabled the optimization of cleaning procedures and determination of process windows for etch processes.
The application of self-excited electron resonance spectroscopy (SEERS) to plasma processing, which is often responsible for low yields in wafer fabrication, shows the potential of integrated process diagnostics. This approach could prove particularly valuable for bringing new equipment on line and reducing production costs [1, 2].
 |
SEERS uses a passive noninvasive sensor that is insensitive to film deposition and negative ions [3, 4]. It is based on the nonlinearity of the space charge sheath at the rf driven wafer electrode across Lam`s Transformer Coupled Plasma or TCP source; it provides harmonics with the modulated sheath width and high-frequency oscillations in the plasma bulk. (An important advantage of the TCP source is the ability to control the energy of ions incident to the substrate being etched. In plasma etching, this is important for simultaneous control of etch rate and etch profile and for minimizing surface damage caused by energetic ion bombardment.)
In brief, by using a hydrodynamic approach for the plasma, the inert mass of electrons can be treated as an inductance, and collisions with neutrals, including power dissipation in the expanding sheath, as a resistance. Accounting for the capacitive behavior of the space charge sheath, the plasma behaves like a damped oscillation circuit. The nonlinear sheath capacitance excites the plasma by causing damped oscillations close to the geometric resonance frequency that lies well below the plasma or Langmuir frequency [4]. The net result is that SEERS allows simultaneous determination of an etching plasma`s average electron density, electron collision rate, and bulk power dissipated in the plasma body.
As implemented in the Hercules plasma monitoring system developed at the Adolf-Slaby-Institute, automatic SEERS measurements are made in <0.5 sec. Hercules attaches through the chamber wall of a plasma etch system via a special sensor and a 50-Omega coaxial cable (Fig. 1). Using a nonlinear model, the pitch ratio of the measured current and the real discharge current can be determined. Calibration for sensor position is not necessary. In addition, an insulating layer up to 100-µm thick on the sensor can be ignored because its capacitance adds only a very small series impedance (<< 50 Omega). The measured discharge current signal passes through an analog-digital converter with a bandwidth of 500 MHz and a 2 billion samples/sec sampling rate. The configuration also includes a capacitive voltage divider with a low-pass filter to measure the peak voltage of the substrate being etched.
We have used vapor phase deposition measurements to verify that the SEERS passive sensor does not affect plasma conditions in any way and has no influence on a metal etch process.
Parameter correlation
Electron density and power dissipated in the plasma body depend on TCP power (Fig. 2). Electron density shows a plateau for intermediate TCP power values. It is tempting to conclude from the plots for 100- and 200-W bottom power measurements that the plateau width simply increases with increasing bottom power, but 150-W measurements do not support this simple explanation.
 |
Although interesting, understanding the underlying physics and chemistry of these phenomena was not our evaluation goal. We simply tried to see whether Hercules` signals give reproducible fingerprints of etch processes and systems [5, 6].
Keeping in mind that Hercules signals directly characterize fundamental plasma parameters, it is evident that simple measurements, such as those in Fig. 2, provide a unique method for system matching. Obviously, additional variations (i.e., to TCP power, gas flows, pressure, etc.) are needed to accomplish complete etch system matching; but, instead of labor-intensive system matching by processing lots of relatively expensive specific test wafers, less expensive bare wafers can be used to match the respective plasma conditions directly. Such savings could play a significant role in any transition to 300-mm wafers [2].
 |
Conventionally, measuring the power dissipated into the plasma bulk (i.e., the power coupled into the plasma from the bottom power) can be tricky. Just measuring the generated power will not suffice because it does not give information about what is going on inside the plasma reactor. Using a dummy wafer load helps, but still its reliability is doubtful. On the other hand, Hercules allows such measurement on-line, efficiently and accurately, and provides more benefit than with a dummy load.
To understand the dependence of electron density on TCP power in a first order approach, we have to analyze the power balance (e.g., the contribution of ionic power to total bulk power). In a case with no collisions, ionic power is the product of ion energy and ionic current. Hence, ionic power depends directly on bottom power and via peak voltage on TCP power as well. If TCP power exceeds bottom power, this dependence becomes weak.
 |
Total bulk power is the sum of bulk power from the bottom electrode, which can be determined by Hercules, and TCP power dissipated inductively in the plasma. Changing electron density and collision rate, TCP power also affects impedance and power dissipation of the bottom power. Mainly dependent on electron collision rate, bottom-electrode bulk power decreases with increasing TCP power when TCP power exceeds bottom power. This is the reason for the plateaus in electron density seen in Fig. 2. So, it is not surprising that most processes with a LAM TCP are running at an inductive power higher than bottom power.
Summarizing these variations, we conclude that by using Hercules signals, we can measure power coupling into the plasma efficiently and reproducibly. They are fundamental plasma parameters and provide the potential for long-term control of system stability, matching different systems of the same type, and comparing different systems on an absolute scale. No other single diagnostic system currently provides all these features together.
Figure 3 shows electron density signatures for four etch processes that only differ in total pressure. These process signatures exhibit significant structural differences and cannot be matched onto a single process curve. The most prominent feature is the broad maximum, due to the Al-etch step and seen in the 1.3 Pa curve. For higher total pressures this maximum vanishes; for 1.7 Pa and 2.0 Pa traces, electron densities during Al etch steps vary smoothly, the absolute value obviously decreasing with total pressure.
Electron density is much more sensitive to etch-system status (i.e., cold or warm) than is peak voltage [5]. Like the cold effect on process signatures seen in Fig. 3, first-wafer effects can be detected using the Hercules sensor [6]. For example, electron density measured for the first three wafers of a lot being Al-etched shows that the first wafer effect is most prominent in the main etch step of the first wafer, resulting also in a longer etch time (Fig. 4).
Cleaning characterization
Because the Hercules sensor measures absolute parameters, it can be valuable for characterizing and comparing etch-system status and stability, particular for etch-chamber cleaning (Fig. 5), which is still a challenge in many fabs. We obtained the data in Fig. 5 after a wet chamber clean and leak testing, by sequentially processing one bare-silicon wafer, two product wafers, and five resist-coated wafers (the latter for chamber conditioning using a specific recipe). We repeated this sequence of eight wafers, four times (for a total of 20 resist-coated wafers). Finally, we processed a final bare-silicon wafer and two product wafers. Each data point in Fig. 5 represents the mean value of the electron collision rate (i.e., its process average) for a single wafer (for the product wafers, the data point represents the second product wafer).
 |
Without sophisticated analysis, this data shows the capability of Hercules-generated data to characterize the cleaning procedure:
 |
We think that other etch-chamber cleaning procedures can be similarly characterized. In general, data from Hercules can be used to compare any cleaning on an absolute scale; today`s etch system status can be compared with that of six months ago or with other etch systems. Also, cleaning procedure intervals and procedures (e.g., how many wafers are really needed) can be optimized and controlled.
Long-term control
Our work also showed differences in the behavior of collision rates when we monitored 77 lots going through an aluminum etch step and a barrier (TiN/Ti) etch step (Fig. 6); we correlated our data to quick and main cleaning steps, and mass flow controller drift [6]:
We believe that parameters obtained from Hercules are more closely related to the plasma-wafer "system" than any other parameters that can be monitored today. In turn, these parameters are sensitive to changes in product, process, and etch-system status. Therefore, with assurance that wafers have not changed, for example, resist thickness, or aluminum and barrier-layer thickness and quality, then changes in Hercules data clearly indicate a change in system conditioning or in the etch system itself.
For such monitoring, it is crucial to consider properly the difference between relative data (i.e., optical endpoint signal, which needs frequent adjustments) and absolute data (i.e., SEERS). Using relative data, it is practically impossible to monitor process performance or etch system stability unambiguously for a selected system over an extended period, match etch systems, or compare process performance for the same product on different etch systems. Absolute data, on the other hand, provide direct relevance to what is going on inside a reactor chamber.
Perhaps the greatest advantage of this technology and its application is that it can be added to a fab-wide sensor actuator network, thereby contributing to production implementation and cost controls. Imagine the possibility of using a SEERS-based plasma-process "fingerprint" to transfer technology from research to a production line or from 200-mm to 300-mm wafers. In addition, reliable plasma monitoring and better real-time and run-by-run process control can improve the mean time between failure and mean time between assist, resulting in a lower cost/wafer. Ultimately, adaptive process control methods will allow systems to self-tune, compensating for effects such as chamber aging and incoming product variations.
Conclusion
Wafer effects and the influence of cleaning procedures in Cl2-based metal etchers clearly show the ability to monitor etch system and process stability by measuring internal plasma parameters with SEERS, as implemented in a Hercules system. Collection of this data enables optimization of etch-system cleaning steps and determination of process windows for etch processes in capacitively coupled plasmas. As such, this technology becomes an on-line diagnostic tool that is able to identify faulty wafer processing without a significant loss in semiconductor manufacturing productivity. The ultimate benefit is a reduction in cost/wafer and increase in system up-time.
Acknowledgments
The authors thank Ch. Kölbl at SIEMENS AG and H. Steinmetz at Lam Research GmbH for their support and helpful discussions.
References
1. D.J. Austin, Solid State Technology, p. 184, Sept. 1997.
2. Y. Ra, C.-H. Chen, J. Vac. Sci. Technol., A11, 2911, 1993.
3. M. Klick, W. Rehak, M. Kammeyer, Jpn. J. Appl. Phys., 36, 4625, 1997.
4. M. Klick, Frontiers in Low Temp. Plasma Diag. II, Bad Honnef, Germany, 1997.
5. S. Wurm, Lam Research Technical Symposium, Genf, 1998.
6. S. Wurm, et al., Proc. Of the Plasma Etch User Group, Fourth Intern. Workshop on Ad. Plasma Tools and Process Eng., pp. 2-7, Millbrae, CA, May 1998.
Stefan Wurm received his PhD in physics from the Technische University M?nchen, Germany. He joined Siemens Semiconductor Group in 1996, and is presently a Siemens assignee at International Sematech, 2706 Montopolis Dr., Austin, TX 78741; ph 512/356-7419, fax 512/356-7848, e-mail [email protected].
Walter Preis received his degree in precision mechanics from the Fachhochschule M?nchen, Germany. He joined Siemens Semiconductor Group in 1984 and is a senior engineer in plasma etch at SIEMENS AG Semiconductors, PO Box 100944, 93009 Regensburg, Germany; e-mail [email protected].
Michael Klick received his PhD in physics from the Ernst-Moritz-Arndt University of Greifswald, Germany. In 1992 he was one of the founders, and is now R&D manager, of the Adolf-Slaby-Institute GmbH, Rudower Chaussee 6a, 12489 Berlin, Germany; e-mail [email protected].