I will start with a joke:

A 747 is in route from Los Angeles to Tokyo. The Captain comes on the P.A. "We were having trouble with our number 4 engine and had to shut it down. Not to worry, everything is fine, though we will now be arriving an hour later than scheduled." An couple of hours goes by before the Captain's voice is heard again, "Sorry for the interruption, the number 2 engine has just failed. That pushes our arrival time back another hour." A bit later the P.A. crackles, "Good afternoon, this is your Captain again, no need to panic, we still have plenty of power, but we've now lost our third engine and will be arriving in Los Angeles a little more than three hours late." A distracted passenger leans into to the person in the next seat and says, "If that last engine fails, we'll be up here all week."

I have a materials scientist friend who relayed a story of his own (this one true). First some background. He spent the bulk of his career advising the science driving IC chip fabrication machinery. Then, near the end of his career his attention was redirected towards photo voltaic solar technology. This coincided with global climate change awareness and the green energy momentum that was spurring government research funding and anticipating new markets.

But solar is a very different animal than integrated circuits. The arch of the IC market has been unlike anything human kind has ever witnessed. We are all familiar with the geometric acceleration known as Moore's Law – a doubling of processor density (number of transistors per unit area) every 21 months! Computer chips have doubled in power and halved in price every 21 months for the last 40 years! Today's computer chips have nine hundred thousand times as many transistors as chips had in 1968.

How was this possible? It is because the physical structure necessary for the behavior that defines a transistor (changes from 1 to 0 or 0 to 1 each time a current is applied) works at any granularity all the way down to a scale of just several tens of atoms. And because the number of these gates you can fit side to side on a flat surface grows as the square of the reduction in gate size (if you make your transistors 10 times smaller, you can fit 100 times as many in the same area).

The first transistors were so large (trillions of trillions of atoms across) that even at a Moore's Law pace, we were guaranteed a run of at least 45 years before we would reach a fundamental limit. We are now in the midst of enjoying the last 3 or 4, 21 month cycles Moore's Law predicts.

Solar is a different game altogether. The first indicator of this difference is that the percentage scale has always been used to measure performance in photo voltaic systems. Which means we have always been within two orders of magnitude of the known limit. Obviously, the amount of energy arriving at any square meter of earth surface is known, it doesn't change, and it isn't very large. Within a few years of the discovery of the photo voltaic effect, we were building collectors in the tens of percentage points of perfect efficiency. And it has always been obvious that even if we could build a perfect collector we would never be able to collect more than the meager 1000 watts that arrive at any square meter of surface.

My friend tells a story of a meeting he had with his superiors at the major chip fab equipment firm for which he worked. They were excited by the doubling of the 3% efficiency of a new substrate. My friend could sense that their excitement reviled a natural tendency to falsely overlay the insane history of IC improvement onto this solar domain. He put it to them this way:

"It took twenty years to achieve a doubling of efficiency of amorphous silicon photo voltaic, to go from 3 to 6 percent efficiency. If you know the history of IC fabrication you might think that that the next doubling of efficiency in amorphous silicon PV would take half of 20 years – just 10 years. In 15 years, by that same projection PV would achieve 24 percent efficiency, 48% in 17.5 years, and 96% efficiency in just over 18 years. You surpass 200% efficiency just 2 months later? Obviously something about this growth projection is wrong-headed."

And then my friend further frustrated his superiors with this prescient parable:

"So let us explore a more likely scenario. Imagine a you were trying to get all of the juice out of lemon. You squeezed that lemon as hard as you could for 20 years. Along the way you learned many things about lemon squeezing. Would you therefore expect to get two times that amount of juice if you squeeze the same lemon for another 20 years?"

Both of these stories are pertinent as we reach the apex of the era of the growth of computing power and walk hesitantly into the era of the growth of computational relevance.

[to be continued]