Too much of a good thing can constrain designers to a single path.
Since its pronouncement in 1965, Moore's Law has been the defining paradigm of the electronics industry. ("Law" is a misnomer, it's a "conjecture"; F=ma is true law).
Under its guidance, we relentlessly drive towards ever smaller features, higher densities, larger wafers and more chips per wafer. Yet there's a problem with living under this law and its corresponding roadmap.
True, the implications of Moore's Law have enabled the spectacular growth of the industry, but it has come at an enormous cost, literally and figuratively. Before you start flinging silicon wafers at me, let's step back and take a different perspective.
The R&D needed to advance process technology to the next node is extraordinarily costly, and matched by the billion dollar fabs needed to take it to full production. In turn, vendors need extremely high volumes to pay for all this, while the market opportunities at these very high volumes are fading. How many products exists that will need the tens or even hundreds of millions of a given IC?
What we have is a circular situation: High up-front costs are paid for by large volumes, but at decreasing per-chip pricing. All this while the low pricing drives volume, which will pay for everything. Then the cycle starts again, except the cost of developing that next-generation of process is ten times higher than the previous one.
Since capital costs are so high, the industry loses a fortune when fab utilization and yield aren't close to 100 percent.
At some point, the merry-go-round stops, for three reasons: costs simply get too high to support; the market opportunities shrink; and most inviolate, the laws of physics dictate that this can't go on forever.
That's the point I am afraid we are reaching. The vaunted, celebrated roadmap has been both our guide and our straitjacket
The undeniable success of process technology advances based on Moore's Law have squeezed out funding for other approaches, such as biological-based components, analog computing or integrated electro-optics, to name just a few. The money and opportunities available for significantly different approaches, which may offer a way around or beyond Moore's Law, are relatively miniscule compared to what is spent on our conventional process technology.
Our relentless pursuit of semiconductor process improvements has squelched technological diversity, which history shows is a very risky way to grow and survive, offering no pathway out at its endpoint. As an industry, we may now be paying the price.
We may not be able to spend our way past the "Moore's Law ends here" sign and the corresponding implications. After all, there is a law even stronger than Moore's Law, namely, the law of unintended consequences.