Collisions of low-energy electrons with atoms and molecules have been of both fundamental and practical interest since the early days of the quantum theory. Indeed, one of the first successes of quantum mechanics was an explanation of the curious transparency of certain gases to very slow electrons [Mott:87a]. Today, we have an excellent understanding of the physical principles involved in low-energy electron collisions in gases, and with it an ability to calculate the cross section, or probability, for various electron-atom collision processes to high accuracy [Bartschat:89a]. The case of electron collisions in molecular gases is, however, quite different. Although the same principles are involved, complications arising from the nonspherical shapes of molecules and their numerous internal degrees of freedom (vibrations and rotations) make calculating reliable cross sections for low-energy electron-molecule collisions a significant computational challenge.
At the same time, electron-molecule collision data is of growing practical importance.
Plasma-based processing of
materials [Manos:89a], [JTIS:88a]
relies on collisions between ``hot'' electrons, with kinetic energies
on the order of tens of electron-volts (
), and gas molecules at temperatures of hundreds of
to generate reactive fragments-atoms, radicals, and
ions-that could otherwise be obtained only at temperatures high
enough to damage or destroy the surface being treated. Such
low-temperature plasma processing is a key technology in the
manufacture of semiconductors [Manos:89a], and has applications in
many other areas as well [JTIS:88a], ranging from the hardening of
metals to the deposition of polymer coatings.
The properties of materials-processing plasmas are sensitive to operating conditions, which are generally optimized by trial and error. However, efforts at direct numerical modelling of plasmas are being made [Kushner:91a], which hold the potential to greatly increase the efficiency of plasma-based processing. Since electron-molecule collisions are responsible for the generation of reactive species, clearly, an essential ingredient in plasma modelling is knowledge of the electron-molecule collision cross sections.
We have been engaged in studies of electron-molecule collisions for a number of years, using a theoretical approach, the Schwinger Multichannel (SMC) method, specifically formulated to handle the complexities of electron-molecule interactions [Lima:90a], [Takatsuka:81a;84a]. Implementations of the SMC method run in production mode both on small platforms (e.g., Sun SPARCstations) and on CRAY machines, and cross sections for several diatomic and small polyatomic molecules have been reported [Brescansin:89a], [Huo:87a;87b], [Lima:89a], [Pritchard:89a], [Winstead:90a]. Recently, however, the computational demands of detailed studies, combined with the high cost of cycles on CRAY-type machines, have led us to implement the SMC method on distributed-memory parallel computers, beginning with the JPL/Caltech Mark IIIfp and currently including Intel's iPSC/860 and Touchstone Delta machines . In the following, we will describe the SMC method, our strategy and experiences in porting it to parallel architectures, and its performance on different machines. We conclude with selected results produced by the parallel SMC code and some speculation on future prospects.