Semiconductor manufacturers need higher resolution imaging and better analytical capabilities to determine what causes failures in smaller and smaller devices. As these needs approach the fundamental limits of scanning electron microscopes (SEMs), semiconductor manufacturers turn increasingly to transmission electron microscopes (TEMs), which offer resolutions down to 0.8 to
1.0 Å. Despite significant advantages in resolution and analysis capabilities, TEMs have not found great acceptance in semiconductor-manufacturing applications in large part because of the difficulty of preparing samples for them.
TEM analysis finds use at many stages of an integrated circuit’s life. During process development, a TEM can provide high-resolution metrology and chemical analyses. Once a semiconductor process is running, a TEM can operate as a metrology tool for monitoring the process and analyzing defects implicated in yield problems. Finally, a TEM can inspect returned ICs to help determine the root cause of a failure. (See “What Is a TEM?’’ below.)
To properly use a TEM, you must understand how to prepare samples and the tradeoffs in the sample-preparation techniques. TEM samples must be thin enough—typically 100–200 nm—to transmit an electron beam. For the highest resolution imaging and analysis applications, thinner is generally better. The thinner the sample, the less the beam spreads, and the better the image and analytical resolution. Optimal performance for these applications often requires a sample thickness of less than 50 nm.
A cross section of a one-of-a-kind defect on a wafer is one of the most difficult TEM samples to prepare, so we’ll use it as our example. Typically an automated inspection tool detects and classifies the defect. The tool may identify the defect only as a set of coordinates, and the defect may prove difficult or impossible to see with an optical microscope. Once you locate the defective area, you must cut it from the wafer. The sample must be small enough to fit into a TEM sample holder, which measures a few millimeters across. (Removing a defective section with conventional techniques destroys an entire wafer.)
The sample then undergoes several grinding and polishing steps until both surfaces, or sides, approach the defect. Finally, broad-beam ion milling thins the sample until it is transparent to electrons. The only way to monitor the thinning processes is to frequently move the sample to an optical microscope and inspect it. Many defects are nearly invisible until revealed by the polishing process, and the slightest misjudgment in polishing or ion milling can destroy the sample.
FIBs Prepare Samples
You can use a focused ion beam (FIB) instrument to prepare a sample.(1) An FIB instrument uses an ion beam to image the sample and to remove material from both sides of the desired section. The FIB can also “polish’’ the sample to make it transparent to electrons in a TEM (Fig. 1). By using an FIB, you can monitor material removal as it happens. Using an FIB lets you take two approaches to FIB sample preparation—prethinning or lift out. 
| | Figure 1. A photograph of a lift-out sample shows the thin cross section ready to be cut free and moved to a TEM sample grid. | Prethinning uses traditional mechanical methods to cut a small sample from a wafer and to then thin it to about 10 to 50 mm. After the initial thinning, you glue the sample onto a TEM sample grid and place it in an FIB instrument. The ion beam removes, or mills, material from both sides of the desired section, leaving a thin membrane for TEM examination. You use successively lower ion-beam currents to remove material from the defect section. The FIB’s diameter increases with beam current, so high currents can cause the beam to erode the top of the final section if you try to mill too close to a defect. The final milling uses a beam current of 50 to 300 pA, in contrast to 5 to 20 nA for the initial rough milling.
The balance between the amount of material removed during mechanical thinning and during FIB processing provides an important tradeoff. The more material removed by the faster mechanical process, the less remains for relatively slow FIB milling. But the mechanical process is more difficult to monitor, and excessive mechanical thinning may destroy the sample.
To use the lift-out approach you place the intact wafer or device into the FIB instrument, which cuts trenches on both sides of the target, using a series of reduced beam currents as in the prethinning technique. When the target reaches a thickness of about 0.5 mm, you tilt the wafer and use the ion beam to cut the target free along its bottom and one side. After further thinning, you cut the remaining side (Fig. 2). The sample, now 10 to 20 mm in length and width, and less than 100 nm thick, sits freely between the trenches, as shown in Figure 3. Using a fine glass probe, a micromanipulator, and a long-working-distance optical microscope, you lift the sample and deposit it on a formvar or carbon film supported by a TEM sample grid.
| 
| | Figure 2. The image of a lift-out sample shows the sample cut away from the sides of the bulk material but still attached at the top. The FIB instrument uses the X marks to locate the side cuts. | Figure 3. A top view of the lift-out sample shows the deep trenches milled by the FIB and the sample after final thinning. The lift-out sample is ready to be cut free and moved to the TEM. | Techniques Offer Tradeoffs
Each of the two approaches has advantages and disadvantages. The lift-out technique leaves the remainder of the wafer intact and available for additional testing, or further processing. A lift-out sample is not surrounded by any of the bulk sample, which might block the detection of characteristic x-rays, or which might limit the tilt at which you can view a sample in the TEM. Also, a lift-out sample requires less material removal, and thus less FIB-processing time.
The prethinning technique offers an advantage, though. You can return a sample to the FIB instrument for further thinning if needed. In general, you can’t return a sample on a film-coated TEM sample grid to the instrument. On the other hand, the lift-out technique lets you use defect-detection and CAD-navigation tools to locate a defect you want to test. This capability finds great use when you must locate a one-of-a-kind defect.
Both the lift-out and prethinning techniques rely on an FIB instrument, but using the instrument for TEM sample preparation can be tedious. You must precisely align, and continuously monitor, each cut. One misplaced milling operation can destroy the sample. An automated FIB system can position the beam more accurately than an operator, and can continuously monitor beam position relative to the specimen throughout the milling process.
Sample Preparation Adds Artifacts
All methods of TEM sample preparation cause changes in the sample that can introduce artifacts into images and analyses. (Understanding the artifacts will help you better interpret images and analysis data, but actual analysis of the images goes beyond this article’s scope.) Bombarding the sample with 5- to 50-keV gallium ions causes the most concern. The expected damage caused by the ion beam includes implantation of gallium ions, amorphization of crystalline structures, mixing of components, and loss of fine structural detail.
Experiments confirm the existence of these damage types in a thin layer at the sample surface. As you might expect, the thickness of the layer strongly depends on the angle of incidence and the ion energy. Studies show the layer ranges from a little over 3 nm at 10 keV and a 1º tilt to nearly 40 nm at 50 keV and normal incidence (perpendicular).(2)
To prevent top-surface damage, you may deposit a layer of platinum or other metal using an FIB instrument prior to the start of any milling. The metal smooths sample topography that might cause uneven milling, and it forms a sacrificial layer. Top-surface damage normally occurs due to implantation of the platinum ions early on in the deposition step. During the final stages of thinning, the ion beam will remove some of the sacrificial metal instead of affecting the sample below it. Without the sacrificial layer, the gallium ions used for milling would cause greater damage to the surface.
Damage at the top of the sample may or may not be important, however, depending on the depth of the features you want to see. Normally, the areas of interest exist well away from the 30-to-60-nm layer on the top surface. But if areas of interest exist near the surface, even the deposition of the protective platinum layer can introduce unacceptable artifacts. Low-energy deposition of platinum (5 keV) may help reduce the damage. Or if an electron beam is available, you can use it to first deposit a thin layer of platinum to protect the sample during subsequent deposition of a thicker metal layer using the ion beam at full energy. Alternatively, you could use conventional means to deposit a thin layer of gold or carbon over the sample surface prior to loading it in the FIB system. This layer would protect the sample during subsequent ion-beam-induced deposition.
Damage can also occur on the sidewalls of samples. Because the beam strikes at a low grazing angle, less damage occurs on the sides than on the top, but damage to the sidewalls affects the entire sample. You can reduce sidewall damage by not inspecting the face with direct ion imaging after you perform the final thinning and cleaning operations. If you avoid ion-beam imaging, the amorphous layer induced by ion milling represents only a few percent of the sample volume. If an electron beam is available, you can use it to inspect the final sample without causing damage.
Polishing Reduces Damage
You can also reduce damage to the face of a sample by using a final low-energy “polishing’’ step. Although users report no direct measurements of beam damage, a final cleaning with 5- to 10- keV gallium ions at a 5º to 10º angle improves image contrast in lattice imaging in the TEM.
FIB sample preparation offers several advantages over conventional techniques. It avoids the need to scarifice the whole wafer, permits the use of automated defect detection data, increases the yield of good samples, and reduces the difficulty of preparing samples. T&MW
Richard Young is a product manager who specializes in FIB systems for product design. He holds a Ph.D. from the Cavendish Laboratory (Cambridge, UK), and he has been working with FIB systems and their applications for more than 12 years.
Eric Van Cappellen is IC-TEM product manager at FEI. He has a Ph.D. in physics and has done research in quantitative x-ray microanalysis in scanning and conventional-transmission electron microscopy.
Peter Carleson works in the areas of development, demonstration, and training at FEI. Prior to joining FEI, he worked at the Oregon Graduate Institute of Science and Technology where he used FIBs to create micro-optical structures. He received his B.S. degree in physics from Linfield College.
FOOTNOTES
1. Lee, Randall G., and William C. Monigle, “FIBs Probe and Fix Semiconductor Problems,’’ Test & Measurement World, Newton, MA. May 1998, pp. 53–60.
2. Young, Richard, et al. “High-Yield and High-Throughput TEM Sample Preparation Using Focused Ion Beam Automation,” Proceedings of the 24th International Symposium for Testing and Failure Analysis, ASM International, Materials Park, OH. 1998. What Is a TEM? | | A transmission electron microscope (TEM) uses a beam of high-energy electrons to project a magnified image of a sample onto a fluorescent screen or other viewing device. Its optical configuration resembles a 35-mm slide projector with the sample taking the place of the photographic slide. A TEM uses electrons instead of light, and the sample removes energy from the beam due to electron scattering rather than light absorption (figure).
A TEM illuminates an entire sample and uses electromagnetic lenses to focus the transmitted electrons into a highly magnified image. In contrast, a scanning electron microscope (SEM) scans a finely focused probe of electrons over the imaged area, simultaneously monitoring an induced signal. It maps the signal variations to form a 2-D representation of the sample surface. TEMs require thin samples, typically 100 nm or less, that can transmit most of the incident electrons. TEMs offer a resolution of 0.8 to 1Å, an order of magnitude better than the resolution of SEMs. Because TEMs can use the atomic lattice of the silicon substrate as an internal calibration standard, they can make very accurate dimensional measurements.
The interactions between electrons and sample atoms generate a number of secondary signals that find use in both TEMs and SEMs. Transmitted electrons are the primary imaging signal in TEMs. Secondary electrons (sample electrons ejected by beam electrons) and backscattered electrons (beam electrons scattered backward by collisions with sample nuclei) provide the primary imaging signals in SEMs. The TEM can analyze transmitted electrons to determine the energy they lost when scattered in the sample. The energy loss indirectly characterizes the state and type of the scattering atom. Electron diffraction analysis can provide information about the crystalline structure of the sample.--Richard Young, Eric Van Cappellen, and Peter Carleson
| |
