The Time Machine and Its Lessons
By George A. Riley
About 10 years ago, I held an aluminized silicon wafer in hand, curiously regarding a dull, non-reflective square covering most of the surface. My microscope revealed acres of closely spaced bumps, more than 200,000 of them on the wafer. These were not connections, nor a flip chip assembly, only a demonstration of an incredible bumping density. This month, I heard that the present density leader among pixel detectors is a 2,000 × 2,000 array, four million bumps on a single chip. Who would have forecasted that, back in 1993?
Since forecasting the future is at best a hazardous enterprise, I suggest that together we 'reforecast' the past. Return with me to the year 1993, when I co-authored my first flip chip paper. Stand with me on that small eminence in flip chip land. Let us together attempt to predict the size and shape of our city in the year 2004. What trends would we have anticipated? More important, which trends would we have missed? And what might we learn from our journey through time?
In 1993, much packaging research focused on flip chip. Most is aimed at new bumping materials and techniques. Flip chip then (as now) usually meant solder bump. Gold and gold-tin bumps existed for special applications, but emergent multichip modules and chip-on-board assemblies were still generally wire bonded. Low cost bumps, such as electroless nickel, were nurtured in laboratories. Anisotropic conductive film was hiding in Japan. High-density plated solder bumps, gold stud bumps? For most of us in 1993, they were laboratory curiosities, or stimulating conference papers. Some might succeed, but who knew which?
Looking forward from 1993, we might have correctly foreseen general tends: continuing growth, although slower than expected. Improved performance, decreased size. These were extrapolations, not forecasts. But at the next level of detail, (Which? How? When? Why?), we would have missed much.
Certainly our major forecast miss in 1993 would have been chip scale packaging. CSP was being pioneered then, and its prophets were still voices in the wilderness. The whole purpose of flip chip was "smaller, faster." Putting the chip into a package seemed a backward step. Most flip chip development then was in the hands of the researchers. These researchers were not yet focused on manufacturing issues such as testing, handling and fit to the installed base of production equipment. As flip chip emerged from the laboratories, manufacturability and cost became the major drivers. In 1993, they were secondary considerations.
Another major forecast miss would be the third word, omitted from the above mantra of "smaller, faster, cheaper." Flip chip aimed at performance enhancement or size reduction, for applications where performance or size justified the cost and complexity. Would we have picked flip chip as an inexpensive approach for cell phones, smart cards, RF-ID tags? Not likely in 1993.
Flip chip was created for active devices on silicon. Integrated passive flip chips were beneath our notice, and not cost-justifiable. Today, wafer-level packaging has made integrated passives a major application. Wafer-level packaging itself was beyond our imagining in 1993.
Lead-free bumping, a social and political requirement, not a technical one, was as 'unforecastable' as an earthquake, and may prove to be as costly. To predict then that our industry would be forced to abandon a proven technology in favor of spending millions of dollars taming less tractable ones, to avoid an unproven hypothetical risk, would have been lunacy. However, political and social developments are outside of the laws of logic.
Some of our 1993 forecast came true. Flip chip was expected to have strong and steady growth, and it did. New techniques were expected; they appeared. Our forecast might be compared to planting a tree. We know that with proper care the tree will grow, but we can't clearly foresee which branches will be largest, and which might even wither.
Is there a lesson in this hindsight? It might shape our evaluation of forecasts, and of forecasters, to remember that there are no facts about our future, only conjectures. In predicting the future, forecasters will likely get some things right. Their major flaws will be sins of omission, not of commission.
More important, our hindsight should temper our views and expectations of our changing technology. I freely predict that flip chip will continue its strong growth, including techniques and applications not foreseen in 2004; that it will drive new technologies, yet unknown; and that some developments will be looked back on as surprises in 2014.
GEORGE A. RILEY, founder and owner, may be contacted at FlipChips Dot Com, 210 Park Avenue #300, Worcester, MA 01609; (508) 753-3572; e-mail: firstname.lastname@example.org.