1999 looks like a very busy year

11/01/1998

1999 looks like a very busy year

Robert HaavindEditor in Chief, [email protected]

While semiconductor fabrication has undergone some dramatic changes throughout its history, in recent times the industry has settled on digital CMOS circuits on silicon. By far, this represents the bulk of today`s processing, with a little BiCMOS and some gallium arsenide sprinkled here and there for special applications. Silicon dioxide has proven to be a very handy dielectric, with aluminum, tantalum and titanium for conductors and vias.

In the earliest times, silicon and germanium transistors competed for dominance, but silicon won out. Major developments such as planar junctions, epitaxial growth, and ion implantation had dramatic impacts on wafer fabrication methods, just as chemical mechanical polishing has more recently.

On the device side, bipolars had a big speed edge on MOS for a long time, but shrinking geometries helped cut the difference, and the shift to complementary MOS, which dramatically reduced quiescent power, clinched the top spot for digital CMOS. Increasing density allowed digital circuits to replace many analog functions, so it became less important to put analog onto the same chip for most applications. While fabs certainly play with recipes, especially where analog or memory circuits are combined with processors and logic, the convergence on CMOS is virtually universal.

For higher frequency applications, gallium arsenide has found a niche, but even here silicon CMOS may eventually play a role. CMOS processors now run at hundreds of megahertz, and there are experiments with synthesizing digital rf circuitry, even into the GHz range, on CMOS chips.

With CMOS processes everywhere, even camera chip developers started to explore how theymight challenge charge-controlled devices withnew designs based entirely on conventional CMOS. Early versions required a tough trade-off between the space taken by rather feeble photosensors and the circuitry to compensate for their shortcomings. Steady progress now makes the CMOS camera chip a promising contender.

Just as the industry has started to get comfortable with a fairly stable set of processes, changes are creeping in, with much more variety on the way. As we plan future articles in Solid State Technology, it is clear that we will be delving into a diversity of new process developments during 1999 and thereafter.

Deep submicron interconnects will move to copper (until recently a very dirty word in cleanrooms), which, along with low-k dielectrics, will cut RC time delays. The dual damascene process developed for copper may also now be applied to aluminum to cut process steps. Actually, there are variations on dual damascene processing, and this, too, may add to the influx of new processes into the fabs. The quest for faster processors is also bringing silicon-on-insulator wafers into the mainstream mix.

Memory chipmakers are exploring a whole range of new high-k dielectrics and compatible metals, so their material sets are changing, too. The giant magnetoresistive effect now being applied to heads for high density storage, calling for multiple sputtered films only a few angstroms thick, may become a candidate for nonvolatile semiconductor memories, competing with flash.

Even germanium is re-emerging as a companion to silicon for high-frequency circuits, but great care must be taken to introduce Ge into Si processing.

While a number of these examples may prove viable only for niche markets, the proliferation of new materials and processing approaches for deep-submicron devices makes it likely that at least some of them will be showing up in most future fabs. If you are just settling down to a well-established manufacturing routine, and feeling thankful for it, don`t get too comfortable. 1999 is likely to be a very busy year no matter what happens in the marketplace.