Abstract
The role of the VPD collection solution and its composition has been examined before) and water have been touted as a best means for collection efficiency. The role of this solution is most important, as an improper solution will “collect” less contaminant and “dirty” wafers then judged as “clean.” Relative newer applications of [1] , most recently with the advent of copper processing in the semiconductor industry. Solutions with varying concentrations of hydrofluoric acid (HF), hydrogen peroxide (H 2O 2), nitric acid (HNO 3Pd, Pt, and Ag in process, and the importance of Au as a major player in minority carrier lifetime make these elements important for contamination control and ensuring collection solutions are properly selected.
Discussed in the 2007 ITRS Yield Enhancement section and again in the 2008 update, the importance of the edge and bevel to process yield is a critical challenge going forward [2] . From a metal contamination standpoint, the edge and bevel have been looked at and reported on specifically for copper and via TXRF and ICP-MS[3] and using an innovative mechanical system to look specifically at these areas of the wafer. We extend these initial discussions by using an automated system to look at the edge and bevel and looking at contamination for a host of metals.
Introduction
Routine and accurate monitoring of trace metal contamination is an important step in semiconductor processing. As device features shrink and dielectrics scale toward the atomic level, metal concentrations of 5E09 atoms/cm 2can affect semiconductor performance and product yields [4]. Contamination in processing can affect yield in multiple ways. For example, during high temperature processing steps, metals can diffuse into the silicon substrate and act as recombination centers adding electronic states into the band gap of silicon. As a result minority carrier lifetimes are degraded. Metal contamination can also adversely affect silicon oxidation rates and become incorporated in gate oxides where they increase leakage currents by degrading the gate oxide integrity.
Vapor phase decomposition (VPD) is a proven method of wafer preparation for subsequent measurement of trace metal contamination [5]. In this technique the entire oxide layer of the wafer is decomposed and concentrated into a single droplet, which then can be measured by inductively coupled plasma mass spectrometry (ICP-MS). Pairing the VPD wafer preparation with ICP-MS analysis gives fab engineers increased sensitivity and lowers detection limits (into the mid E07 atoms/cm 2for most elements on a 300 mm silicon wafer). VPD wafer preparation allows for good sensitivity because it concentrates a large volume of the wafer area scanned into a single measurement.
In the first step of VPD, the entire wafer is exposed in an enclosed chamber to hydrofluoric acid (HF) vapor. Through specific chemical reactions, HF acid completely dissolves the thermal or native oxide on the wafer, and changes the surface of the wafer from hydrophilic to hydrophobic. This decomposed oxide layer is collected for measurement by scanning a drop of slightly acidic solution across the surface of the wafer. After all the condensation has been collected into the acid droplet, the droplet can be placed into a vial for ICP-MS measurement.
In this paper, we will discuss two important factors in VPD ICP-MS analysis. First, we will examine the VPD preparation for its capability to properly recover contamination off the wafer surface for newer semiconductor processing materials of gold (Au), palladium (Pd), platinum (Pt), and silver (Ag). Second, we will discuss use of an automated VPD process to collect trace metal contamination specifically at the edge and bevel of wafers.
For our first study, in order to judge whether the VPD preparation is efficient, VPD collection experiments are performed. A collection experiment is completed simply by spiking a wafer with a known amount of metal contamination and then using VPD ICP-MS to determine how much of the contamination is recovered. The experimental amount of metal contamination obtained from the experiment is then divided by the known amount of contamination placed on the wafer to provide a recovery percentage. General guidelines state that recoveries between 75% and 125% are acceptable [6]. A typical illustration of recovery data is shown in Table 1 for routine metals analyzed in the semiconductor industry.
The composition and chemistry of the VPD scan fluid are critical to recovery. To improve the recovery of certain elements, specific concentrations of acids such as hydrofluoric acid, hydrogen peroxide, and nitric acid are used to dissolve metal contamination from the wafer surface into the acid droplets. Inappropriate choice of acid concentration and mixture in the scan solution can greatly affect element recovery (Table 2a), resulting in inaccurate results or gross underestimation of wafer contamination levels.
In this paper, aqua regia, which is famous for its ability to dissolve gold and platinum, is selected as the medium of the precious metal VPD scanning fluid. The experiments were repeated three times for gold, silver, platinum and palladium, and the element recoveries were calculated (Table 2b). The average recoveries of the four metals were between 80% and 90%. The spiked precious metals are simply deposited by pipetting onto the wafer surface and dried, where the background concentration of precious metals on the wafer surface is less than 10E9 atoms/cm2. Although this deposition method is different from typical semiconductor processes, aqua regia is obviously more suitable for the collection of gold, silver, platinum and palladium considering that VPD has actually scanned the possible metal contamination in the oxide layer.
Contamination and defects at the wafer edge and bevel areasAnother important consideration for VPD analysis is discussion of the edge and bevel and the inclusion of these areas in the VPD scan. Contamination and defects at the wafer edge and bevel areas has become an increased concern in semiconductor processing. The 2008 International Roadmap for Semiconductors (ITRS) has identified that contamination in these areas can make a significant impact on yield loss for device processing [2]. Contamination at the wafer edge and bevel can also be an easy path for cross-contamination as these areas as they are often contact points in semiconductor processing. Wafers can pick up metallic contamination at the edge and bevel from wafer carriers, processing equipment that use edge grip handling [3], and contamination in these areas has been found on incoming silicon [7]. Because contamination at the wafer edge and bevel area can eventually affect device areas [4] and because witness wafers are used for contamination monitoring experiments, it is important to ensure these areas are monitored in a VPD ICP-MS process.
Manual scanning at the wafer edge and bevel presents some practical problems. It is difficult to maintain control of the scanning droplet in these areas of the wafer. For a wafer edge scan it is difficult to precisely maintain a specified edge exclusion (for example, a 3 mm area vs. 5 mm area from the edge) without automated control of the scanning droplet. Scanning at the wafer bevel is impossible by manual VPD without a specially developed apparatus or automated equipment.
Typical VPD ICP-MS contamination experiments at the wafer edge and bevel areas are completed with separate measurements. As more contamination is typically found at the wafer edge and bevel, it is common in the VPD preparation for the wafer scanning to be done with multiple droplets. The first scan is performed to collect the wafer surface area and subsequent scans to cover the edge exclusion, edge, and/or wafer bevel. Separating scan area measurements with multiple VPD scan droplets gives engineers more data to determine where on the wafer the contamination is located. With this fingerprint information engineers can pin-point process steps where metal cross-contamination occurred. Performing repeat experiments, and especially at the edge-exclusion area, is possible via a manual VPD scanning process. However chances of cross-contamination are certainly increased in a manual mode as well. Moreover, scanning on the bevel/edge cannot be performed in a manual mode. To limit the number of manual steps and allow bevel/edge experiments, automated VPD equipment is utilized in these experiments and described here.
Different VPD experimental procedures may lead to differences in the results of judging “good” and “bad” wafers. The data in Table 3 illustrate why it is necessary to unify the VPD pre-processing, such as scanning the entire wafer surface or not including the edge removal area. Wafers 1 and 2 (for reasons of repeated analysis) were scanned over the entire wafer surface and included edge removal areas. Common contaminant elements such as aluminum, calcium, iron, potassium and sodium exceed the 4E9 atoms/cm2 specification. However, when wafers 3 and 4 (same process as wafers 1 and 2) were analyzed by VPD, the edge removal was 2mm, and the analysis results were quite different from the previous ones, with a metal-limited pass. These data indicate that the VPD scan settings in a sense determine the “pass” or “fail” result of wafer metal impurity monitoring.
If a silicon wafer is taken directly from the supplier’s delivery box, the comparison analysis is performed before and after the cleaning process, and the wafer is scanned twice. The first was a 5mm edge removal collection on the wafer surface. The second is an edge scan, starting at 5mm of edge removal, then 2mm across the edge, and finally moving the scanning droplet across the edge of the wafer so that contamination at the chamfer is also captured (Figure 1).
Figure 2 shows the results of the contamination comparison, where the edge bevel scan fluid of the wafer before cleaning contained the highest concentration level of metal contamination. Compared with the results of scanning fluid on the wafer surface before cleaning, it has the highest content of aluminum, calcium, iron, magnesium, potassium, sodium and zinc. Nickel and zinc contamination was only present in the edge bevel scan fluid of the wafer before cleaning. Results showed that the cleaning process reduced contamination of all components, including surface and edge chamfers. The caveat here is that incoming wafers are often considered “clean” when in fact the contamination levels in the edge bevel area are relatively high. These sources of contamination are most likely caused by grabbing or touching within the wafer carrier.
Summarize
Firstly, the results of VPD scanning recovery experiments show that aqua regia medium can be used to collect gold, palladium, platinum and silver elements. But aqua regia is not a panacea. For ruthenium and iridium precious metals, aqua regia cannot provide good solubility, so it is not the best choice for scanning fluid. Further research is needed in the future as different metals are introduced into advanced manufacturing processes.
Secondly, the VPD pre-processing settings should be unified, which affects the comparison between the measurement results and the specification requirements. Experiments using an automated VPD system have shown that wafers that appear to be “clean” may have contamination in wafer edge removal areas or edge bevels. These areas should be prioritized when troubleshooting during a metal contamination event.
Finally, incoming wafers from suppliers, even if “clean,” are monitored against process or witness wafers, which may be desirable.