An ultrasonic laser sonar technique for copper damascene CMP metrology
02/01/2001
Part One of a Series
Michael Colgan, Chris Morath,
Guray Tas, Rudolph Technologies, Flanders, New Jersey
Malcolm Grief, SpeedFam-IPEC, Chandler, Arizona
 |
overview
The picosecond sonar technique has proven to be a powerful tool for process development and production process monitoring for copper dual damascene fabrication. Along with demonstrated capability to measure seed, barrier, and electroplated copper, the technique can be used to monitor CMP process control parameters.
Copper interconnects have revolutionized the semiconductor industry. Processes for traditional aluminum interconnects have been perfected by more than 30 years of experience, but they are no longer adequate for the demands of copper dual damascene patterning. The chemical-mechanical planarization (CMP) step is critical for both aluminum and copper, but the increased CMP challenges with copper have led to significant new technological improvements. Further refinements are still in development.
To form a level of aluminum interconnect wiring, the aluminum is plasma etched; an oxide dielectric layer is deposited; interlevel contact vias are etched; and tungsten interlevel interconnect plugs are deposited. The tungsten and oxide are then polished by CMP to form a planar surface for the next aluminum interconnect level. When aluminum is detected in the polishing slurry mix, the CMP step is complete.
 Figure 1. A cross-section of a copper interconnect structure showing three basic failure modes after CMP (orange = copper, green = TaN, purple = ILD). |
In the copper dual damascene patterning process, vias and trenches are etched into the oxide interlevel dielectric (ILD). Thin adhesion/barrier layers and a copper "seed" layer are vacuum-deposited. This multilayer metal film stack is then overlaid by a thick electrodeposited copper layer that fills the etched vias and trenches. CMP is used to polish down the thick copper layer to a flat surface, exposing the oxide and separating the copper surface into discrete wires in the trenches. This inlay patterning method requires simultaneous polishing of the soft copper and the complete removal of the thin, hard barrier layer, while minimizing erosion of the underlying oxide.
Accurate, reliable, and robust endpoint detection is critical with copper CMP, since only a few seconds separate under-polished and over-polished structures. Optimizing the copper CMP process requires fine-tuning of the slurry chemistry and pad pressure [1]. Several new approaches are also being explored, including multistep, multislurry recipes [2] and noncontact, nonslurry processes [3]. Metrology plays a major role in CMP development and will be critical in production process monitoring.
Process control requirements
CMP has three basic failure modes: residual barrier, dishing, and ILD erosion. Figure 1 illustrates each of these modes.
 |
Figure 2. Picosecond sonar echo from a silicon dioxide film on silicon, curve fitted for a 7596Å oxide film.
When the wafer is under-polished, residual barrier material or copper can remain on oxide areas. Such residue causes electrical leakage that can degrade device performance or lower product yield.
When the wafer has been polished to the point where no residual barrier is left, the differences in the polishing rate of copper and oxide can cause dishing and erosion. Dishing of large copper features occurs because copper is generally polished more quickly than the surrounding oxide. Erosion occurs on dense arrays of narrow lines. With erosion, both the copper lines and the oxide between them are over-polished, and the thickness of the entire structure is affected. Dishing and erosion both result in thinner lines and higher resistance. In addition, dishing or erosion of one metal level can cause a dip in the oxide deposited for the next level. The copper pool that forms in this dip may not be completely cleared during the next CMP step, causing short circuits in that layer.
Process development and optimization can be monitored using test wafers featuring special test structures that are designed to be measured with various metrology tools. For production process monitoring, these test patterns can be incorporated in scribe line areas on product wafers. This requires the use of a metrology system that can measure in test sites as small as a 50µm square. Properly designed test structures can be used to detect and measure barrier residue, dishing, and erosion. These test structures typically contain oxide and copper pads of various sizes and arrays of metal lines with various linewidths and pitch.
Metrology requirements
The general metrology requirements for CMP process control include a small measurement spot size, repeatability of ±1% or better, and the ability to make measurements on both very thin and thick films on all interconnect levels of product wafers. At a wafer diameter of 300mm, dedicated monitor wafers are not cost-effective. Additionally, usable metrology techniques should be capable of measuring on solid areas of transparent and opaque films, as well as on arrays of metal lines, without significant interference from underlying materials or structures.
During the development of copper CMP processes, resistance measurements and profilometry can be used extensively to measure blanket or structure films on monitor wafers. While resistance measurements can be correlated with the thickness of copper lines, however, they give no information on dishing and erosion-related topography. Profilometry, on the other hand, provides surface profiles only and furnishes no information on underlying layer thicknesses. For example, profilometry cannot measure the thickness of a copper line in a dielectric trench. Profilometry is often combined with an optical determination of oxide thickness for calibration. It is not clear, though, whether either of these techniques will be suitable for production monitoring, and neither technique can describe both the surface profile and the line thickness simultaneously.
Ultrasonic metrology for CMP process control
Picosecond ultrasonic laser sonar metrology (PULSE) [4], as implemented in the MetaPULSE instrument from Rudolph Technologies Inc., is well suited to CMP applications. In this technique, a laser flash of 10-13 sec in duration generated by a pump laser induces an ultrahigh-frequency sound wave that travels from the top surface of the structure through the film stack. The sound reflects from each interface back to the top surface as an echo. A film stack produces a train of echoes, one from each layer, with the thickness of each layer indicated by the time it takes the sound wave to transit through that layer. The echoes are detected by a probe laser that senses changes in reflectivity of the film surface that occur when the echo returns to the surface.
Films with thicknesses ranging from 40Å to 3µm can be measured in this way. This capability has been demonstrated on a variety of line array patterns. Test sites as small as 40x40µm can be measured using a spot size of 15µm with 1s repeatability of 1% or better for typical applications. The time delay between echoes from the interfaces of the multiple layers allows simultaneous measurements of underlying layers.
Picosecond sonar technology is best known for measuring metal films [5]. Measurement of transparent films is also possible, although commercial applications have only recently been pursued. In the case of a transparent film, the underlying opaque layer (e.g., silicon or copper) absorbs the light from the pump laser. The probe laser beam then detects the pump beam-generated strain (sound) wave as it moves upward through the transparent film. The result is an oscillatory signal rather than an isolated echo. Two characteristic features of signals from a picosecond sonar system for a transparent silicon dioxide film on silicon are shown in Fig. 2. The first, a phase shift and a level shift, is due to reflection of the strain pulse at the top surface of the film. The second is due to the reflection and absorption of the echo at the bottom of the film. These features indicate the thickness of the transparent film via one-way and round-trip transit times. The best fit and thickness measurements are calculated by modeling software within the instrument.
Detection of residual barrier films
The picosecond sonar method can detect when a residue as thin as 15-30Å is present on the oxide. When the residue thickness exceeds 40Å, the precise thickness of the film can be measured. For very thin films, such as residual Ta barrier films after CMP, an echo pattern results. The echoes are not distinct when the absorption length of the pulse is comparable to, or greater than, the film thickness. Instead, the echo signal is a damped oscillation or ringing due to multiple reflections of the strain pulse.
Picosecond sonar metrology can also be used to determine the thickness distribution of any residual barrier layer. This data can be used to optimize the polish rate of the CMP tool. Figure 3 shows a line scan profile of a copper trench and the adjacent oxide having residual TaN barrier. The scans show that the TaN barrier is thinnest near the copper line.
Dishing
Dishing is defined as the difference in surface height between a copper feature and the surrounding oxide. Limiting dishing is critical since the thickness of the copper determines the line resistance. Profilometry typically is used to measure dishing, but this technique does not indicate the thickness of the copper line and it is not generally performed on product wafers. Picosecond sonar simultaneously measures both the actual surface profile and layer thicknesses and can measure the percent of dishing on polished copper over structures as small as 40µm wide. Figure 4 shows a picosecond sonar profile measurement after CMP of a test pattern array of 50µm lines and spaces. The points in Fig. 4 represent actual measurements of copper and oxide thicknesses, not just a determination of surface topography.
Plotting the total thickness of the copper and barrier along with the thickness of the oxide makes the dishing effect obvious. For production process monitoring, a faster sample of one copper and one oxide point would be sufficient to predict dishing effects within the product die. Such a scan is shown in Fig. 5 and can be used to indicate polishing uniformity. In this example, the oxide and copper are both thinner near the edges of the wafer, but the copper thickness has a more complex dependence on position. This indicates both the amount of copper dishing and the difference in copper/oxide polishing rates across the wafer.
Erosion
Erosion refers to the over-polishing that can occur on dense copper interconnect patterns. Both the copper lines and the oxide regions between the lines are polished more than less densely patterned regions. Figure 6 shows both picosecond sonar results and a profilometer scan generated after CMP of an array of 100µm metal lines and spaces that shows both erosion and dishing of the copper lines. The surface profile results are in good agreement, showing that the densely patterned region is nearly 500Å thinner at the left side of the array compared to the field region. The picosecond sonar oxide thickness measurements show the amount of erosion on each piece of the pattern, and the copper line measurements show a similar variation in thickness as well as dishing. The feature at the far right of both scans is a very narrow line. The surface profile shows the ability of picosecond sonar measurements to simultaneously measure topography and physical copper line thicknesses.
 |
Figure 7. a) Picosecond sonar profile of an array of copper lines with a 0.4µm linewidth, 0.65µm pitch, and a 0.35µm thick layer of SiO2 on silicon showing strong oscillation; b) Fourier transform of the same signal reveals the dominant frequency.
Picosecond sonar measurements on large features are fairly straightforward, but when the lines and spaces become narrower than the measurement beam diameter, the measurements become more complex. In general, the signal generated by an array of narrow metal lines is a combination of copper and oxide signals. The details of this type of signal depend, at a minimum, on line thickness (metal level), linewidth, and pitch. Fortunately, it is possible to optimize measurements for many structures of interest. Successful measurements have been made of regular line arrays with widths between 10 and 0.1µm, and metal density from 10% to 90%. This covers the full range of test structures needed to optimize a CMP process and to control the process on a long-term basis using on-product measurements.
Future work
New modeling approaches are currently being developed to extract even more information on sub-µm line arrays. When excited by the pump beam, very narrow copper lines can independently vibrate like the bars of a xylophone. These individual line vibrations can persist for thousands of picoseconds. Figure 7a shows the response of an array of copper lines having a 0.4µm linewidth and 0.65µm pitch, in a 0.35µm SiO2 film on Si. A Fourier transform of the same signal (Fig. 7b) reveals the fundamental and harmonic oscillations of the metal lines. The experimentally determined frequencies are in close agreement with a model that calculated the possible vibrational modes and expected frequencies for copper lines of that depth and width embedded in SiO2. In addition, information can be extracted that could allow estimation of the thickness of the barrier layer surrounding the copper lines and detection of voids in the barrier layer [6].
Conclusion
The picosecond sonar approach, as implemented in the MetaPULSE tool, offers several advantages when qualifying and monitoring a copper CMP process. First, the small spot-size and nondestructive nature of the measurement allows monitoring of test sites as small as 40x40µm on product wafers. Other benefits include 0.2-1% long-term repeatability of measurements on both opaque metal and transparent dielectric films. The measurement range for opaque films is from <40Å to >3µm. The time-resolved nature of picosecond sonar discriminates against interference from underlying films, permitting on-product measurements of copper and ILD polish rates, as well as residual barrier, dishing, and erosion measurements at all interconnect levels. This same capability enables simultaneous measurement of film stacks such as the thin Ta, TaN, and Cu seed layers used in dual damascene Cu patterning. n
References
- A.E. Braun, "Copper Moves CMP to Center Stage," Semiconductor International, Vol. 22, No. 14, pp. 54-62, Dec. 1999.
- L. Yang, "Modeling CMP Copper for Dual Damascene Interconnects," Solid State Technology, Vol. 43, No. 6, pp. 111-121, June 2000.
- D.S. DeBear, J.A. Levert, S.P. Mukherjee, "Spin-etch Planarization for Dual Damascene Interconnect Structures," Solid State Technology, Vol. 43, No. 3, pp. 53-60, March 2000.
- C.J. Morath, G.J. Collins, R.G. Wolf, R.J. Stoner, "Ultrasonic Multilayer Metal Film Metrology," Solid State Technology, Vol. 40, No. 6, pp. 85-92, June 1997.
- G.J. Collins, "Measuring and Characterizing Opaque Multilayer Metal Film Stacks on Product Wafers," MICRO, Vol. 18, No. 6, pp. 93-106, June 2000.
- J.M.E. Harper, et al., "Microstructural Analysis of Copper Interconnects using Picosecond Ultrasonics," MRS Proceedings, Vol. 612, Materials Research Society, Warrendale, PA, 2000.
Michael Colgan received his PhD in physics from Rutgers University. He is the metrology development group leader at Rudolph Technologies, where he is responsible for developing new metrology products, product extensions, and improvements.
Chris Morath received his PhD in physics from Brown University. He is the section manager of metrology and advanced systems development at Rudolph Technologies.
Guray Tas received his PhD in condensed matter physics at Brown University. He is currently the manager of advanced applications at Rudolph Technologies.
Malcolm Grief received his BSc Hons in materials science from Manchester University in the UK. He is currently the technical manager responsible for copper CMP process development at SpeedFam-IPEC.
For more information, contact George Collins at Rudolph Technologies, One Rudolph Road, Flanders, NJ 07836; ph 973/691-1300, fax 973/691-5480, e-mail [email protected].