Active spectral-control techniques for improving OPC
12/01/2006
Improved optical proximity correction is needed to meet the tighter CD control budgets for advanced lithography processes. Laser bandwidth variation is one of the factors that contributes to the optical proximity effect (OPE). Advanced active control technologies for bandwidth stabilization and improved optical proximity correction (OPC) models incorporating laser spectral characteristics can be used to reduce this contribution to OPE error.
The variation of critical dimension (CD) with pitch in a microlithography process, also known as OPE, is characteristic of the exposure and process conditions and is sensitive to variations in those conditions. Maintaining a stable process is important in realizing the full effectiveness of mask OPC. As CD budgets shrink, it becomes important to reduce OPC errors due to variations in both the process conditions and tool operating parameters. In the exposure tool, particular attention must be given to those parameters that have the greatest effect on image contrast, and hence OPE, such as focus, dose, partial coherence, flare, and laser bandwidth.
Reducing OPC error
There are two aspects to reducing the OPC error. First, the accuracy of the OPC software needs to effectively model the contributions from these sources of variation. Second, once the OPC model has been created at a well-characterized set of process conditions, those conditions must be stabilized in order to minimize deviations in process performance. This article describes the particular contribution of laser bandwidth to OPE and discusses recent improvements in both simulation techniques and bandwidth control technology aimed at minimizing this contribution to OPE error.
 Figure 1. Simulation of the variation in normalized CD with bandwidth for various NA lenses. |
The effect of laser bandwidth on CD has been studied for a number of years [1-3]. It is well known that chromatic aberration in the scanner projection lens causes the multiple wavelengths within the finite width of the laser spectrum to focus at slightly different positions at the wafer and introduce a focus blur that causes a loss of image contrast. Typically, the contrast of isolated features is more sensitive to focus blur than dense features, which leads to a characteristic change in iso-dense bias with bandwidth, as shown in the ArF simulation results in Fig. 1. Recent work [4] has compared simulation and experimental results over a range of pitches and has shown that variations in the spectral properties of the laser light are starting to have an effect on OPE, which is similar in magnitude to other sources of variation, such as shifts in focus, dose, and partial coherence.
The shape of the curve shown in Fig. 1 is quadratic, suggesting that the CD variation due to bandwidth can be minimized by decreasing both bandwidth variation and the mean bandwidth. Both approaches have been an important part of laser light source development for new process generations because the effects of bandwidth variation increase both at shorter wavelength and higher lens numerical aperture (NA). The increased sensitivity of CD variation with bandwidth at higher NA is shown in Fig. 1. It can be seen that the effect of the same range of bandwidth stability is quite similar at two different NAs, primarily due to the lower mean bandwidth used for the higher NA lens. However, the tighter CD tolerances of low k1 and immersion ArF processes are driving requirements for additional enhancements to bandwidth stability. The required level of control strongly depends on the specific design and process conditions and can be guided by simulations of CD sensitivity to bandwidth.
As OPC models are improved to take into account the effect of laser bandwidth and other imaging system variations, it is important to provide an accurate description of the equipment performance and to continually increase the accuracy of the simulation techniques used to predict these effects. The simulation results described here use recently developed techniques that decompose full laser spectral shapes within the lithography simulator to produce high accuracy calculations of the laser bandwidth effects [5]. Figure 2 shows the importance of using the actual spectral shape to provide an acceptable level of accuracy, since there is a significant deviation between this result compared to the fitting functions commonly used to describe the laser spectrum. The use of full spectral data typically results in significantly increased computation time, but new methods have recently been proposed that reduce the simulation time by as much as 100×, while maintaining accuracy to within ~0.1nm of conventional methods [6].
Although the characteristics of the full laser spectrum affect the OPE response, fast on-board real-time measurement of the complete spectrum is very difficult, and it is more practical to use a bandwidth metric to approximate the spectrum. The two most commonly used metrics are full width at half-maximum (FWHM) and E95, defined as the width of the spectrum that contains 95% of the integrated spectral intensity. Previous work has shown that the E95 bandwidth is better correlated to OPE variation [1]. Although the accuracy of past E95 measurement techniques has been limited by sensitivity to various systematic errors, recent developments in bandwidth metrology technology [7] have provided the enhanced level of accuracy and precision needed to support new bandwidth control strategies. Other bandwidth metrics have also been suggested and are under investigation [8].
Stabilizing 95 bandwidth
Stabilizing E95 bandwidth has been a focus of lithography laser design for some time now, and considerable effort has been invested in passive improvements to bandwidth stability, for example using technologies to reduce acoustic disturbances in the discharge region and reduce sensitivity to optical power loading. Although passive bandwidth stabilization techniques have been effective in meeting OPE control requirements, future, tighter OPE specifications will require the more precise performance provided by active control techniques. These methods use either feedback or feedforward control to regulate the bandwidth of the light source while maintaining other key performance specifications.
 Figure 2. Comparison of simulated CD variation with bandwidth using real laser spectra and analytical approximations. |
The performance of such active spectral control (ASC) methods depends heavily on both the statistical accuracy of the state-of-the-art on-board metrology [7], used to measure E95 bandwidth and also on the method of changing the bandwidth. Active control enables not only improved stability of E95 bandwidth, but also the ability to regulate E95 bandwidth to a desired set point. This E95 set point may be chosen so as to minimize OPE or to provide tool-to-tool matching.
 Figure 3. Variation in energy output and bandwidth with MO-PA timing offset on a Cymer XLA laser. |
Variations in laser bandwidth may be categorized by the time scale and magnitude of their effect on E95. Examples include changes in laser energy set point, which has a low magnitude effect on E95 but occurs on a very fast timescale, typically msec to sec. Changes in duty cycle (the ratio of firing time to nonfiring time) and gas depletion affect E95 bandwidth in the seconds-to-hours timescale with larger magnitude. Effects of optical component aging are experienced in the days-to-weeks timescale, and are the largest magnitude variations. For optimum active control of laser bandwidth, at least a dual stage design is essential to compensate for variations over the wide range of magnitudes and timescales seen by the laser. The two stages work together to control bandwidth over the full range of operating conditions. A coarse controller targets large magnitude changes that occur at low frequency, such as large E95 set point changes, gas depletion effects, and the long timescale component of duty cycle changes. A fine controller targets the smaller magnitude but higher frequency disturbances, such as laser energy and the fast component of duty cycle changes. The coarse controller also serves to continually center the fine controller within its control range.
Coarse bandwidth control
An example of a choice for the coarse controller is F2 gas injection, which adjusts the F2 concentration in the laser chamber. Increasing the F2 concentration increases the laser gain and speeds the build-up of energy in the cavity. The stored energy is depleted more rapidly because there is more stimulated emission, so the laser is above the oscillation threshold for less time. This results in fewer round trips, which decreases the line narrowing of the light, and thus the E95 bandwidth increases. Decreasing the chamber F2 concentration has the opposite effect.
Adjustment of F2 concentration provides a large enough range of bandwidth control to correct for sources of variation due to long-term duty-cycle variations, gas aging and component aging. The advantage of using this technique as a coarse controller for E95 bandwidth is that it has a slow effect on other chamber performance parameters and the fast controllers (energy, wavelength, and timing) are effectively decoupled and are able to track without error.
Fine bandwidth control
An example of a choice for the fine controller is adjustment of the relative time delay, denoted DtMOPA, between the firing of the master-oscillator (MO) and power-amplifier (PA) chambers in a dual chamber laser. Since the MO pulse becomes more line-narrowed over its duration, as explained earlier, if the PA chamber is fired later relative to the MO chamber, it selects a more line-narrowed portion of the MO pulse, and the effective E95 bandwidth of the laser decreases. There are two principal advantages of using differential firing time as a fine controller for E95 bandwidth. First, the measurement of E95 and the change of DtMOPA both occur on a tens-of-pulses time scale, allowing for very fast bandwidth control. Second, the extremely tight timing control capability of the pulse power design of the laser allows a range of adjustment which is large enough to correct for sources of bandwidth deviation such as laser energy and duty cycle variations.
During normal laser operation, DtMOPA is continually monitored and adjusted to optimize the efficiency of the laser. As shown in Fig. 3, at the peak of the efficiency curve, pulse energy stability is least sensitive to any timing jitter, and the system can operate with relatively poor timing control. However, other laser performance parameters, including bandwidth, vary monotonically with timing offset, also shown in Fig. 3, and require highly accurate timing control in order to maintain stability. The single power supply design in Cymer XLA lasers provides sub-nsec control of timing, which allows all performance specifications to be met with excellent stability over a range of DtMOPA values and not only at the optimum timing offset. This design concept enables the use of fast timing control feedback to provide superior bandwidth stabilization.
Figure 4. shows the overall performance of the DtMOPA (fine) and F2 inject (coarse) dual-stage controller. The data show max and min E95 bandwidth measurements taken every 30 sec during a seven hour test. The lower set of curves show the behavior when only the coarse controller is active; the upper ones show performance when the fine controller is also used to both stabilize E95 bandwidth and shift it up to a set-point value of 0.35pm. The improvement is significant and E95 bandwidth has been stabilized to well within the limits of the bandwidth metrology itself. This result uses feedback control only and further improvement is expected with added feedforward control.
The DtMOPA controller described is not an option for single-chamber systems so an alternative choice for a fine controller is manipulating the curvature of the linewidth-narrowing module (LNM) grating surface. Altering the optical wavefront within the laser’s line-narrowing element by changing the grating surface is an effective optical method of regulating E95 bandwidth with fine resolution relatively quickly.
Conclusion
Tighter CD control and OPE requirements are placing stringent new demands on the precision of modeling and control of all process parameters. The contribution of laser bandwidth to CD variation is one factor being investigated. We showed that faster, more accurate simulation tools that incorporate laser bandwidth are needed to improve the accuracy of OPC models and identify key future bandwidth control needs.
 Figure 4. E95 bandwidth control comparison on a Cymer XLA laser using coarse controller only (lower plot), and coarse and fine controllers (upper plot). |
Although passive bandwidth control schemes were sufficient to meet CD control targets in the past, active control methods are under development to provide the next level of E95 bandwidth stabilization and control. We described a concept for a dual-stage control technique for active adjustment of bandwidth over a wide range of amplitudes and timescales. F2 injection, DtMOPA and LNM grating curvature were shown as suitable control techniques for either the coarse or fine controllers in this scheme. Combining these control technologies with the latest advances in E95 bandwidth metrology allowed high performance E95 stabilization and regulation. Further work on three-stage controllers promises to offer even tighter levels of bandwidth control.
References
- A. Kroyan, I. Lalovic, N.R. Farrar, “Effects of 95% Integral vs. FWHM Bandwidth Specifications on Lithographic Imaging,” Optical Microlithography XIV, Proc. SPIE, Vol. 4346, pp. 1244-1253, March 2001.
- A. Kroyan, I. Lalovic, N.R. Farrar, “Contribution of Polychromatic Illumination to Optical Proximity Effects in the Context of Deep-UV Lithography,” Proc. 21st Annual BACUS Symposium, Proc. SPIE, eds. G.T. Dao and B.J. Grenon, Vol. 4562, pp. 1112-1120, 2002.
- K. Lai, I. Lalovic, R. Fair, A. Kroyan, C. Progler, N.R. Farrar, “Understanding Chromatic Aberration Impacts on Lithographic Imaging,” J. Microlithography, Microfabrication and Microsystems, Vol. 2, Issue 2, pp. 105-111, 2003.
- K. Huggins, T. Tsuyoshi, M. Ong, R. Rafac, C. Treadway, D. Choudhary, et al., “Effects of Laser Bandwidth on OPE in a Modern Lithography Tool,” Optical Microlithography XVIII, Proc. of SPIE, ed. Donis G. Flagello, Vol. 6154, 2006.
- M. Smith, J. Bendik, I. Lalovic, N. Farrar, W. Howard, C. Sallee, “Modeling and Performance Metrics for Longitudinal Chromatic Aberrations, Focus-drilling, and Z-noise; Exploring Excimer Laser Pulse-Spectra,” to be published in Optical Lithography XIX, Proc. of SPIE, 2007.
- J. Bendik, I. Lalovic, to be published.
- R.J. Rafac, “Overcoming Limitations of Etalon Spectrometers Used for Spectral Metrology of DUV Excimer Light Sources,” Optical Microlithography XVII,Proc. of SPIE, ed. Bruce W. Smith, Vol. 5377, pp. 846-858, 2004.
- T. Brunner, D. Corliss, S. Butt, T. Wiltshire, C.P. Ausschnitt, M. Smith, “Laser Bandwidth and Other Sources of Focus Blur in Lithography,” Optical Microlithography XVIII, Proc. of SPIE, ed. Donis G. Flagello, Vol. 6154, 2006.
Nigel Farrar is VP of lithography applications at Cymer Inc., 17075 Thornmint Ct., San Diego, CA 92127; ph 858/385-5527, e-mail [email protected].
Wayne Dunstan is senior control system scientist at Cymer Inc.
Robert Jacques is a senior fellow in control systems at Cymer Inc.