Etching: Closed-loop bias voltage control for plasma etching
02/01/2000
Roger Patrick, Scott Baldwin, Norman Williams Lam Research Corp., Fremont, California
The stated goal for high-density etchers of completely decoupling plasma generation from ion energy control cannot be achieved with conventional methods of monitoring RF power delivery. By using a closed-loop control system based on a voltage sensor to control the RF voltage at the chuck, the interaction of plasma density and bias voltage seen in conventional systems can be eliminated.
The RF source power that sustains plasma discharge and the RF bias power that determines ion energy at the wafer surface are important plasma etch parameters to control in order to reduce variability and improve process results. High-density plasma sources, such as electron cyclotron resonance, inductively coupled, and helicon wave, decouple plasma density from ion energy and potentially allow each to be set independently. They are still interactive, however, so changes in the source power do produce variations in the bias [1, 2]. Moreover, other process parameters, such as pressure, chemistry, and film composition in complex stacks, can produce changes in the bias.
In the pursuit of tighter requirements for CD control and selectivity, especially on larger wafer sizes, it is necessary to have tighter control and complete separation between the key RF power parameters for plasma etching systems. Unfortunately, in current plasma etch systems, the power parameters are still interactive and poorly controlled. This article describes a method of isolating and controlling bias on the wafer and thereby the ion energy in high-density reactors. Benefits include a greater degree of accuracy and increased repeatability, both in individual etchers and across multiple systems.
Conventional bias control
The RF setpoint in a process recipe is typically the generator output power. This power is fed to the process chamber via a matching network that transforms the complex load of plasma and chamber so they appear to the generator as a purely resistive 50omega load. This control maximizes the power delivered to the load, but it does not guarantee that the efficiency of power transfer is 100%. It has been shown [1, 2] that the efficiency of power transfer, h, to the load is given by:
 |
null
where Rl is the resistance of the load, and Rm is the resistance of the match. This equation implies that the efficiency is 50% when the resistances of the load and match are equal and only approaches 100% when Rl > Rm. As has been shown [2], in practice the efficiency is usually considerably less than 100%. In addition, since the efficiency depends on the load resistance, which in turn depends on the process conditions, the efficiency depends on process conditions.
The equation shows that any variation in match resistance from unit to unit will lead directly to differences in power delivered to the load for the same generator setpoint. Variations in etching results from machines in the field running the same process recipes are commonly encountered as a result.
Further variation in machine-to-machine etching performance can be attributed to the lack of reproducibility in stray capacitance. If the stray capacitance between the powered electrode and ground is large (on the order of several hundred pF), then increased circulating currents may be set up, reducing the overall apparent efficiency of RF delivery by burning more power in the matching network. Thus, changes in process conditions, source power, matching resistance, and stray capacitance can all contribute to unpredictable results in a single etch system or across multiple systems in one or more fabs.
Alternative bias control
Rather than settle for the limitations of monitoring generator output, monitoring and controlling certain RF parameters at the wafer chuck has been attempted with varying degrees of success.
For example, controlling the delivered power requires a sensor capable of measuring power accurately in a non-50omega environment. The delivered power, P, is given by
 |
null
 |
Figure 1. Correlation of sheath potential and monitored peak RF voltage for various power and pressure conditions, including an aluminum etch cycle.
 |
Figure 2. Schematic layout of a TCP plasma etch system.
 |
Figure 3. Peak RF voltage at the bias chuck as a function of TCP power and chuck power settings.
where I is the current, V is the voltage, and f is the phase angle. For a plasma load that is largely capacitive, the phase angle is close to 90°, and the power factor, cos(phi), is close to zero. Small errors in the measurement of phi produce large errors in both cos(phi) and the measured value for the delivered power. This difficulty in making accurate, repeatable measurements of delivered power makes it an unsuitable variable for process control.
 |
Figure 4. Voltage bias control maintains a setpoint of 100V when the source power is varied.
 |
Figure 5. Dependence of photoresist etch rate on a) TCP and bottom power settings; and b) TCP power and bias voltage.
To control ion energy directly would require measuring the sheath voltage (Vsh). One parameter closely related to Vsh is the dc bias (Vdc). To measure Vdc, it would be necessary to have a conductive pickup on the wafer chuck. Most commercial systems do not have this pickup; in any case, readings from it could be subject to drift because of etching or the deposition of polymer and etch products.
A preferred alternative control variable is the peak-to-peak RF voltage (Vpp) or the amplitude (Vpk, equal to Vpp/2). Figure 1 shows experimental measurements confirming that there is a good correlation between Vsh and Vpk, which is independent of pressure and power and does not change when gas composition varies during etching. In the figure, process pressure is expressed in millitorr (mT), and source power is expressed in watts (W). The aluminum etch from which data were taken used a standard BCl3/Cl2 chemistry.
To determine Vsh, the plasma potential (Vp) was measured with a Langmuir probe, and Vdc was measured by a voltage tap on a conductive wafer. The relation between these quantities is:
 |
null
As a control variable, Vpk is relatively easy to measure, and the sensor can be located outside the chamber immediately below the wafer chuck, downstream from the matching network.
Closed-loop bias voltage control
To test the effectiveness of using Vpk for a closed-loop, bias power control system, we conducted tests on the Lam Research TCP 9400PTX silicon etch system and the TCP 9600PTX metal etch system [3]. The basic configuration of these systems is the same, and it is shown schematically in the block diagram in Fig. 2.
The plasma source consists of a flat spiral transformer coupled plasma (TCP) coil that couples power inductively into the chamber. Power is fed to the coil via a capacitive "T" match, which is highly efficient because it lacks any inductive components. The wafer bias is provided by a second power supply, which drives the wafer chuck. Since the combination of plasma and chamber load seen by the second supply is capacitive, a match with inductive components that make it inherently "lossy" is required to make it look purely resistive. The voltage sensor is attached immediately below the chuck to monitor Vpk downstream from the match network. Under recipe control, a feedback loop adjusts the generator output to maintain the desired voltage setpoint. By operating in this mode, "point of use" control of the delivered RF is maintained irrespective of any losses in the match.
Decoupling ion energy from plasma density
One goal for a high-density etcher is to have direct, independent control of both the plasma density and the ion energy at the wafer surface. It has been shown elsewhere [4] that the ion density in this kind of inductively coupled etcher is a linear function of TCP power and independent of bias power. Thus, ion density can be set independently simply by setting the source power. However, if the output of the bottom generator is used as the bias control variable, there are confounding interactions with other process parameters, such as top power, pressure, and gas composition, which make independent control of ion energy difficult to achieve. Controlling the voltage at the wafer chuck provides a more complete orthogonal separation of the process variables.
Figure 3 shows the measured peak voltage at the wafer chuck as a function of bottom generator power while the top power is varied. This clearly shows coupling between top and bottom power, since the voltage is a function of both of them. By contrast, as shown in Fig. 4, if the bias voltage is held constant as the top power is varied, the control loop adjusts the power output of the bottom generator to compensate.
Deconvolution of apparent process parameter interactions
Interactions between process parameters, such as top and bottom power, may give misleading results during Design of Experiment (DOE) studies. The real nature of the relationship between etch results and machine settings is thus obscured. For instance, the effect of top and bottom power on photoresist etch rate might be needed to optimize a metal etch process.
Figure 5a shows the dependence of etch rate on TCP power for various bottom powers. The data show etch rate increasing or decreasing, and top power dependent on the bottom power setting. Figure 5b, however, shows data plotted with RF voltage as the control variable rather than bottom power. It is apparent that there are orthogonal linear relationships between resist etch rate and top power, and between etch rate and voltage. The apparently complex behavior in Fig. 5a arises because voltage is really the fundamental control parameter rather than bottom power. Interactions between top and bottom power and voltage, shown in Fig. 3, obscure the simple underlying trends in photoresist etch rate.
Similar effects can be shown for the interaction of bottom power and pressure on the measured RF voltage. Again, the effects are separated when the voltage at the chuck is controlled directly. There is an interactive effect of varying gas chemistry and bottom power on bias voltage, which is eliminated by directly controlling the voltage. Experiments with source power, pressure, and chemistry have all shown that these process parameters affect the RF bias voltage at the chuck in a conventional power delivery system. Only by directly controlling voltage can an unambiguous determination of the impact of process parameters on etching results be made.
Equation 1 illustrates that the power delivery efficiency is a function of the plasma load. As a consequence, changes in plasma chemistry associated with etching of different materials lead to changes in power delivery to the chuck. Etching wafers with different films — e.g., oxide, polysilicon, or metal — with the same recipe will lead to different power deliveries and different bias voltages. Clearing one film and etching another will also lead to changes in bias voltage and delivered power at endpoint.
Figure 6a shows the change in voltage signal at endpoint when etching an aluminum film and the exposing underlying oxide layer with a conventional power delivery system. There is roughly a 30V change as the film clears. This change, which goes unobserved and uncontrolled in machines with conventional power control schemes, can be undesirable. Without the user's being aware of it, bias voltage may increase after poly endpoint when clearing to a thin gate oxide film. Figure 6b shows that the Vpk is a strong function of how much aluminum is exposed on photoresist patterned wafers during an aluminum etch process when run in conventional power delivery mode. Since photoresist etch rate has been shown in Fig. 5b to be a strong function of Vpk, if bottom power is used as the control variable for a recipe, then large variations in resist etch rate with photoresist coverage on a wafer would be expected.
 |
Figure 6. Variation in RF peak voltage a) at endpoint when an Al film is cleared, exposing underlying oxide; and b) with exposed Al area.
 |
Figure 7. Photoresist etch rate a) vs. generator bias power setpoint for different matching networks; and b) vs. peak RF voltage for the same matching units.
 |
Figure 8. Oxide etch rate vs. reflected power for a deliberately mismatched RF power delivery circuit operated using either bottom generator power or peak RF voltage as the setpoint control variable.
Reducing machine-to-machine variation
When voltage is controlled directly at the chuck, variability due to variations in match resistance is eliminated. Figure 7a shows oxide etch rate plotted as a function of bottom generator power for two different matching networks. There is an approximate 5% increase in etch rate for one match compared to the other. If the data are replotted with voltage as the control variable, however, the differences in the matches have no effect.
Similarly, the effect of variation in the stray capacitance between the chuck and ground can be compensated. When stray capacitance was varied between approximately 600 and 900pF, the oxide etch rate was lower for the case with highest capacitance because of the lower efficiency of the delivery circuit. The effect of variations in stray capacitance is eliminated by controlling the voltage at the chuck.
Eliminating the effect of RF mismatch
For conventional methods of power delivery, if the RF circuit is not perfectly matched, power is reflected back to the generator, and the power delivered to the plasma load is reduced. As shown in Fig. 8, reflected power increases as the RF circuit is progressively detuned, and, as a consequence, etch rate falls off.
However, if the voltage at the chuck is maintained constant, the process is no longer sensitive to poor tuning of the matching network. The oxide etch rate remains stable even as the reflected power increases.
Conclusion
In this article, it has been shown that the stated goal for high-density etchers of completely decoupling plasma generation from ion energy control cannot be achieved with conventional methods of monitoring RF power delivery. By using a closed-loop control system based on a voltage sensor to control the RF voltage at the chuck, the interaction of plasma density and bias voltage seen in conventional systems can be eliminated. This method of power control will also lead to better reproducibility in process performance by giving "point of use" control of ion energy at the wafer, eliminating the possibility of variation due to component variation in the RF delivery circuit between the generator and the wafer chuck.
This patent-pending [5] voltage control technology is now available for high-density plasma etching systems.
Acknowledgments
The authors acknowledge contributions to this work from J. Jafarian, N. Benjamin, and S. Siu.
References
- R. Patrick, N. Williams, C.G. Lee, "Application of RF sensors for real time control of inductively coupled plasma etching equipment," Proc. SPIE, Vol. 3213, p. 64, 1997.
- N. Williams, C.G. Lee, J. Jafarian, R. Patrick, "Characterization of radio frequency power control using a RF sensor in an inductively coupled plasma etcher," Electrochemical Soc. Proc., Vol. 97-99, p.197, 1997.
- J.S. Ogle, US Patent No. 4,948,458, Aug. 14, 1990.
- R. Patrick, P. Schoenborn, H. Toda, F. Bose, "Application of a high density inductively coupled plasma reactor to polysilicon etching," J. Vac. Sci. Technol. A, 11 (4), p. 1296, 1993.
- R. Patrick, N. Williams, patent pending.
Roger Patrick received his MA and D Phil in physical chemistry from the University of Oxford. He has been with Lam Research for five years and is currently the director of conductor technology. Previously, he was Etch Section manager for LSI Logic in Santa Clara. Lam Research, 4650 Cushing Parkway, Fremont, CA 94538; ph 510/572-8300, fax 510/572-6349, e-mail [email protected].
Scott Baldwin received his doctorate in mechanical engineering from Stanford University. He has been with Lam Research Corp. since April 1996, and he is currently the manager of technology development for conductor etch.
Norman Williams received his BSc and PhD in physics from Manchester University. He has worked for more than 30 years in the semiconductor industry. He is currently working on a number of problems related to plasma etch control and reproducibility.