View from the Valley - Nobody Goes to Jail

Moore's Law is probably the most widely known characteristic of the semiconductor industry. Many are amazed at how uncannily accurate the law has proven to be. Since the law has for years been used to set long term development goals, there is an alternate view that the law has become a self-fulfilling prophesy. Either way, it's had an amazing run.

Gordon Moore laid down the basics of his Law in his 1965 paper. It has since been tweaked by Dr. Moore, himself, and interpreted by many. He observed, "The number of transistors that can be placed inexpensively on an integrated circuit has doubled approximately every two years."

To those of us in the industry, we don't give it a second thought. I would guess, however, that each of us may have a little different interpretation if we dig deeper. First, it applies only to the IC chip itself. It doesn't involve packaging and it doesn't involve speed or other IC performance. Second, it describes an "inexpensively" cost factor. It is not what a scientist can achieve in a laboratory, but rather what can be done in a commercial environment. Third, there is the dimension of time that is not precise. Fortunately, this evens out over a period of many years.
Historical progress has always caused me to focus on two main drivers for Moore's Law: wafer size and minimum feature size. In the back of my mind, I assumed you needed both to stay on target. When I started in the industry, wafer sizes were 1" and are now 12" (300 mm). Each diameter step approximately doubled the wafer area, and thus the number of chips, for around same processing cost. Over the past 35 years, geometries have gone from 8 micron (8000 nm) to 45 nm, roughly doubling the number of transistors per square inch every two years.

For the last couple of years, we have been facing a practical decision: to go to 450 mm wafers or not. There is no doubt that our industry has the technical wherewithal to design, build and operate 450 mm capacity. The pesky fly in the ointment is the word "inexpensively" in Moore's Law. I have broadened this to include "total cost," which includes not only the cost of manufacturing the chips, but the cost of designing and building the equipment and developing the processes. There is significant data to indicate that the semi fab tooling industry has still not fully recouped the investment in going to 300 mm.

For our industry to remain healthy, a move to 450 mm must allow the fab tooling vendors to recoup their investment and make a profit. Remember, these same vendors must also continue development to allow jumps to new nodes, which have been the main driver in the past. There is certainly a case that can be made that a move to 450 mm, with all its intended effort and cost, will dilute the focus and funding on node jump development to the point that no net gain will be realized.

The industry as a whole may well be better off by staying at 300 mm and driving node development in the x, y and even z dimensions. No matter what, the semiconductor industry will continue to make progress so that new generations of transistors will be less expensive than their predecessors. Whether we are able to maintain adherence to Moore's Law remains to be seen. But if we don't quite follow the Law . . . at least nobody goes to jail.