As semiconductor manufacturers push to the 90nm node and beyond they will require increasingly higher resolution in the imaging techniques they use to analyse defects and develop and monitor manufacturing processes. Current imaging workhorses such as the scanning electron microscope (SEM) are inadequate for imaging structures such as gate oxides and barrier seed layers that may be only Angstroms thick. Other techniques are high cost and time consuming. Ted Tessner of FEI describes how his company aims to overcome these shortcomings through its High Throughput Atomic Resolution Analysis (HTARA) technique

As semiconductor manufacturers push to the 90nm node and beyond they will require increasingly higher resolution in the imaging techniques they use to analyse defects and develop and monitor manufacturing processes. Current imaging workhorses such as the scanning electron microscope (SEM) are inadequate for imaging structures such as gate oxides and barrier seed layers that may be only Angstroms thick. Other techniques are high cost and time consuming. Ted Tessner of FEI describes how his company aims to overcome these shortcomings through its High Throughput Atomic Resolution Analysis (HTARA) technique.

A decade ago, as semiconductor manufacturers pushed into sub-micron territory, light microscopy gave way to scanning electron microscopy (SEM) for high-resolution imaging. The industry now faces a similar transition as it crosses the 0.1µm barrier and forges ahead into the nanometre realm. SEM can achieve resolutions approaching 1nm in some specialised applications, but generally its resolution is several nanometres or more. SEMs form an image by scanning a finely focused beam of electrons over the sample surface and monitoring signals caused by the interaction of the beam electrons with sample atoms. Their resolution is limited primarily by beam spreading within the sample. While the beam itself can be focused into a very small spot, its electrons scatter as they enter the sample so that the imaging signal can originate well beyond the spot boundaries. Equally important, SEMs are limited in their ability to generate the material contrast necessary to distinguish the boundaries of critical device features. Their highest resolution signal, secondary electrons, depends primarily on sample topography to generate contrast - and topography is minimal on typical semiconductor samples. These SEM deficiencies are offset by the equipment's low cost, general ease of use and minimal requirements for sample preparation.

The most likely candidates to replace SEM for imaging semiconductor devices at higher, atomic-level resolution are high voltage field emission scanning transmission (STEM) and tunnelling (TEM) electron microscopy. In STEM, as in SEM, a finely focused electron probe is rastered over the sample surface. However, in STEM, the sample is very thin and the beam electrons pass through with a greatly reduced likelihood of scattering. Thus, the spatial resolution for STEM is primarily determined by the probe size and the brightness of the electron source (emitter). As with SEM, several signals are available for imaging in STEM. Among the most useful information carriers are beam electrons that have passed through the sample but have been strongly deflected by interactions with sample nuclei. This signal, collected by an HAADF (high angle annular dark field) detector, shows both high spatial resolution and strong material contrast, since the likelihood of electrons scattering into the detector is strongly dependent on the size of the nuclei they encounter.

Fig.1: UltraView process flow: a) prism shaped section extracted by inline dual beam from whole wafer; b) section after thinning to electron transparency in laboratory dual beam; c) sub-nanometre resolution STEM image; d) atomic resolution TEM image

By contrast with SEM and STEM, TEM forms its image by illuminating the observed area of the sample with a relatively broad beam of electrons and focusing the transmitted electrons into a real image on a charged couple device (CCD) camera or photographic plate. The process is analogous to the projection of a photographic image from film to a screen. TEM resolution is limited primarily by the wavelength (determined by the accelerating voltage) of the beam electrons and the aberrations of the electron-optical system. If the major aberrations (spherical and chromatic) are corrected, TEM can provide sub-Angstrom resolution, sufficient to distinguish individual atoms, especially the lighter elements such as Li, O and N. TEM can also provide strong material contrast due to the diffraction contrast.

The barrier to acceptance of TEM and STEM in semiconductor applications has been their exacting requirements for sample preparation. Both require samples thin enough to transmit electrons: generally less than 100nm, but thinner is almost always better. Conventional sample preparation techniques are difficult, time consuming, and unreliable, involving painstaking manual cutting and polishing operations and requiring highly trained (and highly paid) technicians. In a typical semiconductor manufacturing environment, it is not unusual for routine TEM analysis to take up to one week. With conventional techniques there is little opportunity for parallel processing, resulting in overall throughput that is simply a multiple of time to first data. Moreover, conventional techniques require the removal of the entire wafer from the production flow, thus losing potential further splits and experimental data, process testing and analysis, and potential revenue from scrapped product.

The advent of focused ion beam (FIB)-based sample preparation has allowed the simplification and automation of much of the sample preparation process. The ion beam can cut sections from a bulk sample and thin them to electron transparency under precise, automatic, electronic control. Of great importance in semiconductor operations is FIB's ability to remove a sample from a wafer while allowing the rest of the wafer and its unsampled dies to continue in the production process. FEI's High Throughput Atomic Resolution Analysis (HTARA) simply integrates FIB-based sample preparation with SEM, STEM, and TEM imaging into a complete and seamless data acquisition process.

Integrated process

FEI's UltraView system is an integrated sample preparation and imaging process that provides easy, fast, cost effective generation of critical atomic resolution data for semiconductor process development, failure analysis, process monitoring, and other applications. Though flexible in configuration, the full process integrates an in-line dual beam (SEM/FIB) system (FEI Expida) for sample extraction, a lab based small chamber dual beam system (FEI Strata) for sample thinning and SEM/STEM imaging, and a TEM (FEI Tecnai) for atomic resolution imaging. The process begins with automatic navigation to one or more sites on the wafer in the in-line dual beam system. At each site the FIB beam cuts a prism shaped section (Figure 1a) from the wafer and, in conjunction with the nanoprobe/manipulator, attaches the section to a TEM grid. Up to six samples can be attached to the same grid. The grid is placed in a sealed transfer capsule (Figure 2) that can be extracted from the dual beam system without breaking chamber vacuum via the auto-unloader. The capsule and samples can then be transported to the lab for further processing while the wafer continues in the manufacturing process. The sample is then transferred in the lab to a small-chamber dual beam system, where it is milled to final thickness (Figure 1b) and can be imaged with sub-nanometre resolution by STEM (Figure 1c). If atomic resolution is required the grid is subsequently transferred to a TEM for final imaging (Figure 1d).

Fig.2: Sealed capsule used to transport sample between systems

Although the process is integrated through both software and hardware interfaces, it is also designed to be flexible and to permit a building block approach in its configuration. For instance, if the lab-based dual beam is omitted, final thinning may be performed in the in-line dual beam and the sample sent directly to the TEM for imaging. Alternatively, the TEM may be omitted and final imaging performed in the STEM of the lab-based dual beam. Each tool can continue to function in stand-alone applications as well as be integrated into the HTARA process. Likewise, the components may be procured individually, making significant stand-alone contributions pending ultimate integration into the full HTARA process.

Quantifying benefit

Time to first data, analytical throughput, analytical yield and cost of ownership are all parameters that directly impact the profitability of a semiconductor manufacturing operation. In process development they control the analytical feedback loop and affect the rate at which a new process ramps to production yields. In high volume production they similarly control the duration of yield excursions and thereby the magnitude of each excursion's contribution to overall process losses. In an effort to quantify the potential benefit of UltraView relative to conventional alternatives, we monitored the progress of a typical 300mm analysis through the HTARA process. Tables 1-3 present a high level view of the data. Table 1 demonstrates an 85-minute time to first data in an unoptimised process (exclusive of the time required to physically walk from one system to the next). Figure 3 illustrates the opportunity for parallel processing in the various component systems. This enables output of the second sample less than one hour after the first and an average throughput of about one sample per hour in the one sample per grid case (these numbers include time for moving samples between tools). Figure 4 repeats the experiment analysing six samples per grid. Although the time to first data increases to six hours in this case, the average throughput doubles to two samples per hour. In addition, the automated "hands off" nature of the UltraView process also promises very high reliability over extended use.

Table 1: Time to first data for a single sample on the sample grid

Fig.3: Parallel processing and average sustained throughput for a single sample per grid. Includes time to transfer sample from tool to tool

Fig.4: Parallel processing and average sustained throughput for six samples per grid

Whether performed in-house or commercially, conventional TEM has proven to be a relatively expensive analytical technique with typical cost per analysis in the range of $1500-3000. This compares very unfavourably with SEM at $150-200 per analysis. A simple UltraView cost of ownership model based on a throughput of two samples per hour on a 24/7 schedule and including tool costs, personnel costs, overhead, and tool down time, gives a average cost of less than $400 per sample analysis. At this rate, the cost of analysis breaks even with conventional methods at about 0.2 samples/hour (5/day) in a 24/7 operation or 1 samples/hour (8/day) in a single shift 5 day-week operation along with the benefit of 1.5 hours to first data. Alternatively, at an aggregate expenditure of $15,000 per day, UltraView will produce nearly 50 analyses while conventional techniques will produce five.

Additional cost savings not included in this analysis are likely available from a reduction in salaries required by less highly trained operators as well significant savings in unscrapped wafers. While offering a direct reduction in analytical costs, UltraView's most significant economic benefits will accrue from shortened process ramps and faster recovery from yield excursions.

Author:

Ted Tessner PhD, Director UltraView, FEI.