No abstract

Atmospheric pressure ionization mass spectroscopy (APIMS) has been demonstrated as a quantitative analysis technique for gas phase contaminants such as moisture, oxygen, carbon dioxide, and methane in ultrapure bulk gases (nitrogen, argon, and hydrogen). Previous analytical techniques typically provide quantitative analysis above 10–100 ppb detection limits. The APIMS technique offers true sub-parts-per-billion capability, with detection limits as low as 1 ppt. The APIMS works by ionizing a small fraction of the matrix gas in a corona discharge plasma. These matrix gas ions then transfer their ionic charge to impurities through collisions. This transfer of charge is usually thermodynamically favored due to the relatively high ionization potential of the matrix gas. The result is preferential ionization of the impurities, leading to a sensitivity increase of as much as 106 over standard electron impact mass spectrometers. The APIMS is calibrated by mixing parts-per-billion and parts-per-trillion level standard gases using a double-dilution blending system of unique design. Nonlinear calibration relationships based on models of plasma reaction kinetics have been developed to successfully account for nonlinearities seen in the calibration data. A selective moisture removal technique has been developed to eliminate interference effects due to unwanted moisture in the plasma reactor during analysis for other contaminants.

The range of concentrations over which an atmospheric pressure ionization and mass spectroscopic (APIMS) can measure impurity levels in ultrahigh purity gases accurately has been extended from parts‐per‐trillion to parts‐per‐million levels using novel calibration methods. Nonlinear calibration relationships have been derived through kinetic models of the plasma reaction chemistry. These relationships account for background levels of impurities which cause curvature in log‐log plots of calibration data at low concentrations and correct for curvature induced by plasma saturation in high concentration regimes. Nonunity reaction orders such as have been documented for oxygen calibration in nitrogen also can be handled. These relationships have been fit using weighted nonlinear regression to calibration data that extend over ranges of concentration of more than five orders of magnitude. The resulting APIMS calibration relationships can be applied to analysis of unknown sample gases to provide quantitative analysis capability over wide ranges of concentration.

Partial support from SEMATECH/SRC through Contract No. 88-MC-508 on Multilevel metallizations is appreciated. The key donation of the Precision 5000 by Applied Materials and its maintenance support by IBM are gratefully acknowledged.

The challenges associated with controlling particulate and chemical contamination to achieve high semiconductor device yields are demonstrated with data showing the influence of cleanroom air, semiconductor processes and tools, gases, chemicals, and Dl water. Because typical film thicknesses are much smaller than pattern feature sizes, defects that are as small as one hundredth of the lithographic dimension must be controlled. Scaling device dimensions by a factor of 1/3 to 1/2 will require almost a factor of 10 reduction of particulate levels in order to maintain the prescaling yield. To achieve a 78 percent yield (0.25 defects/cm2) in a typical submicron process containing 250 process steps, each step must contribute no more than 0.001 killer defects/cm2 on average. Storage of wafers for 1 hr in a Class 1 vertical laminar flow cleanroom is sufficient to reach this level. Considerably more defects are introduced in other process steps involving automated tools or chemical/gas exposure. As device dimensions are reduced, the contamination associated with liquids will become relatively more important than that from gases and cleanroom air.