A critical analysis of techniques and future CD metrology needs
07/01/2003
Determining the limits of various CD-metrology methods is extremely important for future IC manufact
Control of feature dimension is so crucial to producing ICs that the metrology step is routinely referred to as a critical dimension (CD) measurement. Manufacturing ICs with faster clock speeds requires controlled transistor gate length variation as well as interconnect structure control. So, a CD measurement can be either a linewidth or a contact-via diameter or area.
Although different technologies are used for CD measurement (Fig. 1), each method has a different role or application. For example, scanning electron microscopes dedicated to CD measurement (CD-SEMs) and optical scattering-based CD measurement (scatterometry) are both used to control CD during manufacturing. Atomic force microscopes (AFMs) are used to measure CDs off-line during process development.
All methods are challenged by the requirement for measuring CDs and shapes of 50nm lines found in manufacturing today. The industry must carefully consider the limitations of these methods along with potential new technologies [1–3]. Although we provide an overview below, our emphasis here is on electron-beam methods.
The CD metrology challenge
With CD-SEM, a finely focused electron beam is scanned over the analysis area. Secondary electrons are emitted from the sample, and the intensity of the secondary electrons versus scan position forms an image of the surface. Scatterometry uses an ellipsometer or reflectometer to measure the wavelength dependence of light scattered from a test structure. The most important recent development is the availability of software that determines linewidth and shape based on the data instead of comparing to a library. In CD-AFM, a probe tip is scanned across the sample surface to form an image of the height of features on the sample surface.
The difficulty of CD measurements is often illustrated by comparing the linewidth with the dimensions of a silicon crystal lattice [2]. Although the gate electrode is doped polycrystalline silicon, for the purposes of argument it is useful to consider the gate electrode to be a silicon single crystal over its width. A 50nm silicon CMOS gate would be <100 unit cells of the silicon crystal lattice or <400 atomic layers wide. At the end of the 15 year horizon of the International Technology Roadmap for Semiconductors, the transistor gate length is expected to be <10 nm or 80 atomic layers. At those dimensions, even transmission electron microscopy must be carefully applied if it is going to provide a meaningful calibration.
Another aspect of CD measurement that is often under-appreciated is the nature of the quantity that each measurement controls. In comparing scatterometry to CD-SEM, several issues require consideration. First, CD-SEM determines the average linewidth over some section or length of the line. Scatterometry determines the average linewidth of many lines contained in a test structure over the width of the test structure. For the sake of discussion, we assume that the die level and wafer level distributions of linewidth are Gaussian. Figure 1 illustrates the concept that CD-SEM is measuring one linewidth from the local distribution and scatterometry is measuring the average (midpoint of a symmetric Gaussian) distribution. A key need is to determine how many lines CD-SEM needs to measure to determine average local linewidth. This brings us to a recurring theme in metrology: Process control through amplification and averaging microscopic changes.
The high throughput of CD-SEM and scatterometry separate them from CD-AFM. The speed of scatterometry makes it a candidate for use as a sensor on lithography tracks for real-time process control.
The need for 3-D control of lines, trenches, and contact-via openings also deserves mention. CD measurements are done on photoresist prior to etch and subsequent measurement. The resist line sidewall's line edge roughness (LER) impacts all subsequent processing as well as device performance. The linewidth roughness of the etched transistor gate has been linked to increasing transistor leakage current [4]. The sidewall angle of line and trench structures is another key aspect of the 3D control of lineshape. All of the CD measurement technologies can measure sidewall angle, and all but one, scatterometry, can measure line edge and linewidth roughness.
CD measurements are done by averaging a linewidth over some length of the transistor gate. At a gate length of 25nm and a process range of 10% (3σ) — total process range of 20% — one calculates the precision to be 1nm for the measurement precision (6s) to process tolerance ratio (P/T), where P/T = 20% = 6σ/(0.2 x 25). The precision amounts to <2 crystal lattice constants or 8 lattice planes or layers. Local atomic variations in linewidth can only be overcome when averaged CD measurements are made.
 |
Figure 1. Basic descriptions of a) CD-SEM, where an electron beam produced by a field emission source is finely focused and scanned across the features of interest. Secondary electrons are produced in the sample and collected above the lens. This is the so-called through-lens detector. The highly stable sample stage holds the wafer horizontally. b) Scatterometry uses spectroscopic ellipsometers or reflectometers to collect data from grating test structures. In these systems, the angle of interest is fixed and the multiwavelength data is compared to calculated optical response for the grating. Other scatterometers use a fixed wavelength and get data from multiple angles. c) CD-AFM scans a probe tip across the feature of interest, forming a 3-D image. Probe tip shape influences the data.
Limitations of CD-SEM
CD-SEM has been the main method of CD measurement for many years. Since its introduction, CD-SEM has been continuously improved. Some time ago, the beam voltage was reduced to ~1keV or less in an attempt to minimize charging and damage to photoresist and device structures. Ten years ago, field emission electron sources improved CD-SEM by increasing the brightness 100¥ (i.e., the number of electrons/unit area) and reducing the beam size.
3-D control of lineshape has been approached in several different ways. Commercially available innovations, such as tilt-beam CD-SEM, allow for a 2-D type determination of line shape. Software for determining line shape from top down CD-SEM has been developed, but is not commercially available. The concept associated with this has been incorporated into one supplier's software that compares the line scan over the edge of a line to the corresponding profile from a line having optimum shape. Thus, the path to the future CD-SEM improvements can be found by attacking the same problems that were solved in the past.
The electron source can be further reduced in size and increased in brightness. The nanotip field emitter is already being investigated. The nanotip is an electron source of atomic size that has a brightness 50x or more higher than that of a conventional field emitter. Vladar and Postek at NIST have demonstrated that such sources can be installed in existing CD-SEMs and that they do provide a significant improvement in beam current, scan speed, and resolution [5]. The challenges facing nanotips are tip lifetime and stability. Tips need to last approximately as long as present field emitter tips (i.e., 12 to 18 months) and must provide comparably stable probe currents. Further improvements in electron optics can reduce lens aberrations that effect image quality, which is a necessary part of extending CD-SEM. These improvements have been demonstrated in a transmission electron microscope lens [6] and in a prototype form in a laboratory SEM built by JEOL.
 Figure 2. The bloom of high secondary electron intensity at line edges is used to locate the line edge and thus determine the linewidth. |
null
Traditionally, the CD-SEM is viewed as having a large depth of field (DOF). However, improvements in resolution have resulted in a significant loss of DOF [7]. The three parameters that characterize the SEM — resolution, beam energy, and DOF — are not independent, but interrelated. Consequently, only two of the three can be freely chosen at one time. Although better resolution improves precision, this gain is obtained at the expense of DOF, and this penalty is especially severe at low beam energies [7]. If low-energy beams are to be used, no really satisfactory solution yet exists other than measuring each line two or three times with different lens focus settings. A composite image can then be synthesized, although at the cost of extra exposure time and higher doses. Higher voltage CD-SEMs (30–200keV) would overcome the DOF issue while still maintaining adequate resolution. High-voltage electron beam columns have been used in transmission electron microscopes for years, and recent advances have resulted in ultra-small electron beams.
Before discussing higher-voltage CD-SEM, an analysis of the ultimate limits of the present approach to CD-SEM seems appropriate. As noted above, CD-SEM forms images from the low energy secondary electrons that are produced as the electron beam scans across the sample. At the edge of a rectangular line, secondary electron intensity blooms (Fig. 2). CD-SEM determines linewidth based on the distance between the line edges of a section of the feature even though the true location of the edge is not easily known due to the width of the bloom.
For many years, line edge was determined from different algorithms that attempted to deduce true edge position. The width of the bloom is a function of the range of secondary electrons in a material. Thus, when the line is very narrow, the blooms from the two edges overlap and determination of linewidth is no longer possible (Fig. 3).
As a result, the ultimate limit of CD-SEM seems to be ~5nm for silicon lines, and the consequences of this bloom effect already make precision linewidth determination difficult. 50nm wide photoresist lines used to fabricate advanced transistors have rounded tops with no sharp edges. Blooms overlap and new line algorithms are already being used in manufacturing (Fig. 4). Thus, linewidth measurement can be improved using lineshape sensitive algorithms. For example, algorithms could be made sensitive to sidewall angle effect [8] and roughness. A combination of aberration correction, nanotips, and shape-sensitive algorithms may extend low voltage CD-SEM for one or two technology nodes.
Other improvements are also required: The level of automation for alignment, focusing, and stigmation needs to be greatly increased. It is perhaps not realized how accurately the focus must be set if the optical transfer function (i.e., the Fourier filter through which the signal must be passed) is not to be compromised. Without some machine-aided ability to focus accurately and rapidly, the loss in performance is unacceptably high. The minimum focus step on most SEMs is so coarse (30–50nm) that optimum focus can only be achieved by chance and a human operator has no objective basis on which to decide when the best condition is achieved.
If charging cannot be eliminated, something could be done to help the situation by changing the nature of the signal collected. Efficient backscattered detectors for 1keV operation are now available, or a simple energy filter can be used to select secondary electrons with higher than average energies. Either of these entities is much less susceptible to charging artifacts
Why, after 50 years, are we still collecting the SEM signal with a crude analog system? It is easy to demonstrate that this leads to many of the uglier artifacts in line profiles. Moving to a digital system would eliminate overload, slew rate limits, and bandwidth as sources of trouble.
In summary, the limits of the CD-SEM are the results of the limited brightness and finite size of the electron source, restricted DOF, and the physics of imaging with secondary electrons. One possible scenario suggests that these limitations can be removed by using much higher energy electrons, and choosing to image with electrons that are scattered from the sample with little or no loss of energy.
High-energy CD-SEM
A totally new approach to CD measurement might overcome many of the problems facing present generation CD-SEM. One challenge facing higher voltage CD-SEM is that of perception. The initial view of many in the semiconductor field is often that the higher voltage beam will cause considerably more damage, but the physics of 100–200keV electrons suggests otherwise.
null
null
The energy deposited in the sample by the beam (i.e., the energy available to damage resists and oxides) reaches its maximum value for incident energies around 100eV. At higher energies, the rate of energy deposition (E) falls ~1/E, while the interaction volume of the beam increases ~E5. Consequently, the energy deposited/unit volume falls very rapidly and so does the corresponding rate of damage. This is confirmed by measurements of the swelling or shrinkage of polymers as a function of energy. Similarly, detailed models of the charge implanted in gate oxide layers show that threshold shifts at 200keV would only be a few millivolts compared to 0.5–1 V values experienced for equivalent exposure doses at 1keV beam energy. Mizuno and others are evaluating device damage to verify these predictions [9].
The situation for charging is similarly benign. Low energy electrons can result in large amounts of negative charging because all of the beam energy is deposited within a thin polymer or oxide film. 200keV electrons, by comparison, have a range of nearly 0.25mm in silicon and so deposit their charge deep in the substrate, leaving little or none in the surface resist layers. Under these conditions, the charging that does occur is positive in sign and self-limiting in magnitude. Because the high-energy incident beam is much stiffer than a 1keV beam, image artifacts and distortion due to charging are unlikely. The next step should be assessing the effect of higher voltage beams on typical resist and etched features.
A second issue that a new approach can overcome is the limitations of imaging with secondary electrons. These secondary electrons are still generated by the sample at 200keV, although at a reduced rate compared to lower energies, but using these to perform metrology would limit spatial resolution to the value determined by the avalanche diffusion discussed earlier.
To take advantage of the much smaller beam spot size at high energies, it is necessary to collect a different kind of signal. The most appropriate choice would be "low-loss" electrons [10]. These are electrons scattered by the sample that leave with energies lying within a few percent of the energy of incident electrons. Those electrons that satisfy this criterion can only have traveled a very small distance within the specimen before being scattered out again by a single high angle collision (Fig. 3). These two facts ensure that the signal produced in this way has a resolution that can approach atomic levels, and so is much superior to the secondary electron alternative.
Another advantage of this higher voltage CD-SEM is that most of the technology is well known to SEM suppliers, and high voltage electron beam columns have been made for transmission electron microscopes for years. While the effective use of low-loss imaging will require some innovations in probe forming lenses and in detector design, the advances required are straightforward and require no new technology. Aberration correction lenses for SEMs have already been designed [11, 12]. The biggest hurdle is providing test data that convinces the semiconductor industry that higher voltage CD-SEM will not damage patterned wafers. If the industry accepts this new approach, 100 or 200keV CD-SEMs could be available in time for the 65 or 45nm technology nodes.
Electron holography
For many years, electron holography, exemplified by the point projection microscope (PPM), has been proposed as a means of extending e-beam metrology to the end of the roadmap. The beauty of electron holography is that it accomplishes the goal of process control through amplification and averaging microscopic changes.
In the PPM, a coherent electron beam from an atomic-sized field emission source reflects from the sample at roughly normal incidence, and the hologram is formed above the beam source. The PPM is intended to be capable of imaging a large area that could result in high throughput. Using the hologram, which gives information in Fourier space, to give linewidth, you obtain average linewidth over the test area. The hologram is a result of electron waves interfering after scattering from the line structures in the test area.
Because this Fourier space image is the result of the interference of scattered waves, it averages over individual lines. If instead one chooses to use the reconstructed real-space image, then the linewidths of individual features can be measured and analyzed. For example, a 3-D form of a line could be reconstructed from the hologram, giving information about the sidewall.
Thus, the PPM both amplifies microscopic changes and provides more statistical information over the analyzed area. The development of the PPM has taken longer than expected, and perhaps higher voltage CD-SEM is a far more likely candidate for 45nm node measurements [1].
Scatterometry-based CD measurement
Scatterometry is an excellent means of high throughput CD control. It accomplishes the goal of process control through amplification and averaging microscopic changes. As mentioned above, the average CD over a test structure or grating is determined from the change in the spectroscopic response of an ellipsometer.
Recently, real-time determination of CDs from ellipsometric data has replaced the library approach that was used previously. This approach makes scatterometry even more attractive as sensor-based metrology on lithography tracks.
Scatterometry has excellent precision values for both lines in resist and etched polysilicon. In the near term, the main limitation of scatterometry is the lack of ability to measure contacts and vias. This issue has two solutions. One is to use a line-array test structure for via-contact resist and etch steps; the results of this approach can be correlated to cross-sectional data. The second solution is to develop optical models that have 3-D capability; the present algorithms used in scatterometry are purely 2-D.
The CD limit of scatterometry is not well characterized. Wavelengths toward the UV region of the spectrum are considered to be sensitive to shape changes, while wavelengths toward the IR are considered to be more sensitive to the CD change itself [13]. This suggests that using an ellipsometer that adds wavelengths further into the UV will not necessarily extend scatterometry to small feature sizes. A study of the limits of scatteromery are clearly in order.
There are many unexplored means of extending scatterometry to very small geometries, such as using multiple angles of incidence. One key aspect is that sub-10nm silicon lines at the end of the roadmap will have different optical properties than ~50nm wide lines today. The absorption edge of silicon shifts due to quantum confinement and surface states. The effect of these shifts on scatterometry is presently unknown.
CD measurement by AFM
Atomic force microscopes and scanning tunneling microscopes have produced some of the most astounding images of the past 10 years. Based on this success, AFMs dedicated to CD measurement are commercially available. AFMs can produce 3-D images that allow full determination of sidewall shape, line edge roughness, and linewidth roughness. Some types of samples are challenging. Highly undercut lines are more difficult to measure. The dimensions of future dense features, such as lines with a 1:1 pitch and high aspect ratio trench-vias, will require new probe-tip technology. Although their low throughput has limited use to process R&D, recently introduced CD-AFMs have made great strides in automation and throughput. Even greater strides in throughput are expected in the future.
One interesting use of CD-AFM is as a reference measurement system. NIST and International Sematech have worked together to characterize the measurement performance of CD-AFM for this purpose. CD-AFM will be a great means of evaluating the line edge and linewidth roughness capabilities of different CD-SEMs and image algorithms.
Alain C. Diebold, International Sematech, Austin, Texas
David Joy, University of Tennessee and Oak Ridge National Laboratory, Oak Ridge, Tennessee
References
1. D.C. Joy, The Future of the CD-SEM, Microlithography World, August 2002, pp. 4–6.
2. A.C. Diebold, IEEE Transaction on Semiconductor Manufacturing,15, pp. 169–182, 2002.
3. The Requirements and Limits of Metrology Technology, Proceedings of the 2003 International conference on Characterization and Metrology for ULSI Technology, Austin, March 24–28, 2003, to be publish in late 2003.
4. K. Patterson, et al., Metrology, Inspection, and Process Control for Microlithography XV, SPIE Vol. 4344, 2001, pp. 809–814.
5. A. Vladar, M. Postek, report to ISMT Metrology Council, June 2001.
6. P. E. Batson, et al., Nature 418 617, 2002.
7. M. Sato, F. Mizuno, Proceedings EIPBN 2000, (Palm Springs:CA), pp. 35–40.
8. S. Villarrubia, et al., Proc. SPIE 4689, (2002).
9. F. Mizuno, S. Yamada, J. Vac. Sci. Technol. B 13, (1995), pp. 2682–2687. O.C. Wells, Scanning Electron Microscopy, (McGrawHill, NY).
10. J. Zach, M. Haider, Nucl. Instrum. and Methods, A363, pp. 319-329.
11. S. Joens, Micros. and Microanalysis, 7, Suppl. 2, pp. 875–876.
12. F. Terry, private communication and presentation at the July 2002 Metrology Technical Working Group meeting.
Alain Diebold received his PhD from Purdue University. He is senior fellow at International Sematech, 2706 Montopolis Dr., Austin, TX 78741; ph 512/345-7680, fax 512/345-7680, e-mail [email protected].
David Joy received his PhD from the University of Oxford. He is a member of the Distinguished Scientist Program at Oak Ridge National Laboratory and the U. of Tennessee.