radical species then of interest in the electronic materials area. The fact that most of the 144 references in the paper are to sources of these molecular parameters reinforces the above remarks on the lack of centralized data sources. When molecular parameters were unknown, they were estimated. More than 10 years later, a number of these parameters have been measured, but no systematic review has been performed to assess the reliability of such estimation attempts. The minimum detectability estimates for absorption include a 100 cm path length. In a practical reactor, this path length would require a multipass geometry, a potential barrier to implementation. An alternative approach to achieving the same predicted sensitivity would be to assume smaller minimum detectable absorbances than the 10-3 assumed in the paper. Lower values are routinely achieved in the laboratory but would have to be demonstrated for a particular reactor.

The SPIE paper discussed above was intended to promote work in two areas: diagnostics of semiconductor processing systems and laboratory measurements of supporting data. It is of interest to examine a few of the experiments of both types carried out following publication of the paper, the idea being to exemplify some general thoughts about the relationship between diagnostics and their supporting databases. The general points are as follows:

1. Quantitative diagnostic measurements, and comparison with the predictions of quantitative models, are necessary for the quantitative understanding of any phenomenon.
2. The level of quantitative accuracy required for a measurement to be useful is determined by the uncertainty limits expected from current or projected predictive models. This in turn determines the uncertainty levels that are acceptable in the supporting database.
3. Data that are enabling, in the sense that diagnostic experiments will not be attempted in their absence, are often different from the accurately quantified molecular parameters that allow derivation of absolute molecular concentrations and other system quantities.
4. It is important to realize that although at least one molecular parameter must be known accurately for a diagnostic experiment to yield quantitative results, typically at least one system parameter must also be known, and its uncertainties may determine the overall accuracy.
5. To summarize, database needs should be evaluated using at least three criteria: (a) What unknowns prevent first diagnostic experiments from being attempted? (b) What unknowns prevent meaningful comparisons with theoretical predictions? (c) Would reduction in database uncertainties for either reason have no effect because of continuing difficulties in quantifying system parameters?

Examples of the interaction between the fundamental database and the work of diagnostics developers can be obtained from an extensive series of studies using tunable infrared diode laser spectroscopy.³⁰

Another method that has great promise in studies of processing plasmas is the two-photon allowed laser-induced fluorescence (TALIF) method. This method measures atomic species' densities using the high peak power of a commercially available 10 ns pulsed laser. Since the atomic species are very reactive, it is relatively easy to devise titration reactions that allow absolute calibration of the atom densities. TALIF has been applied mostly to hydrogen dissociation studies. A listing of other candidates for the TALIF approach and their titration reactions is given in Table 3.2. The work required to ensure linearity of the detector response over a wide dynamic range is quite demanding.

The point to be made here in the context of database needs is that the titration calibration, necessary to quantify the fluorescence collection efficiency, also has the effect of removing the TALIF determination of absolute concentrations from any dependence on atomic or molecular parameters. This independence of the spectroscopic and kinetic database would not apply, however, in cases in which the quenching environment of the experiment was very different from that in which the titration was done. Some data useful in plasma