The elegy for Moore’s Law is premature
The bottomline: Moore’s law is NOT dead.
If you’re perplexed by Moore’s law essence, it is perhaps because you have not fully perceived the fact that it’s not a physical law, it describes only an observation of the calculation power of microchips. It is certainly one of the most bizarre laws in the high tech scientific world. It really describes the advance of technological capability. But let’s be happy that it is not as unusual as some other physics laws such as:
The faster you move, the heavier you get.
Which was pointed out by Albert Einstein in his theory of relativity or retrocausality was hypothesized by many people in the past, from David Hume’s definition of the cause and Kant’s idea of synthetic a priori truth, Michael Dummett and Anthony Flew (1954) and Max Black (1956) all the way to quantom computing eras that states:
Events in the future can affect what happened in the past.
You must now feel better about Moore’s law complication 😆 The best explanation I have heard is that it sets a goal for engineers and scientists. They fear that others will match Moore’s Law if they don’t, so they work and work until they manage to reach their expectation.
Note that there is a potential competitive advantage to disingenuously pronouncing Moore’s Law dead. Do that and others may give up, and then when your advance comes out, to everyone’s surprise, you can dominate the market
Why does Moore’s Law continue to be true?
I borrowed a plot from the Wikipedia page on which Richard Muller (Prof Physics, UC Berkeley, author “Now, The Physics of Time”) has superimposed a line.
Remember, Moore’s Law predicts the MAXIMUM number of transistors, NOT THE AVERAGE.
Thus, points that fall below the line aren’t really relevant, if there are some on the line or above it. You’ll see in the plot that Moore’s Law is alive and well, at least up to the current date (August 2019).
Every two years engineers and scientists will pronounce Moore’s Law dead, only to be confounded when it continues on — Richard Muller
Every decade the reason given for the expected imminent end of Moore’s Law has changed. For the last decade, the argument involved the size of the atom. As the fundamental components of circuits, often called “transistors” for historical reasons, or more recently “nodes”, have gotten smaller and smaller, they are now approaching atom sizes. That’s not quite true; the current density has been achieved with 5 nm nodes. The companies TSMC and Samsung are reported to have plans for 3 nm nodes. Note that for a silicon crystal, the size of the lattice parameter is 0.54 nm. So the 3 nm node size is less than six times bigger than the characteristic spacing of the atoms in silicon! No wonder Moore’s Law seems to be doomed. Frankly, most knowledgeable people considered 6 nm to be impossible. Certainly you can’t go much smaller! You can’t make a transistor smaller than an atom, can you?
I think so. So why do I propose that the elegy for Moore’s Law is premature? One simple answer is that current devices are not yet fully exploiting the third dimension. There is lots of space above the silicon wafer.
The challenge is cooling; packing nodes in 3D rather than 2D creates an overheating problem. But note that I have reduced a previous physical limit (the size of the atoms) to an engineering limit (cooling). And over the past 50 years, the engineers have proven to be amazingly inventive and resourceful. It has been that very inventiveness and resourcefulness that is responsible for Moore’s Law in the first place.
An orthogonal view:
A graph of processor transistor counts versus time, and drawing a line through it to illustrate that Moore’s Law is still going strong is questionable. There are however curves that also fit the data and show Moore’s Law leveling off dramatically.
A linear curve fit to Moore’s Law is less sensical on its face. What if we eventually conquer the physical limits such as atom size and the speed of light by engineering solutions.
Another interesting graph beside this one would be the cost of a device fabrication facility. This cost seems to double along with transistor count. In the previous generation or two you saw companies like IBM and Intel going in on joint fab development, and whole nations funding a fab. Do you think this is actually sustainable? I don’t. Yet another plot that would be interesting in this discussion would be leakage current versus feature size. This graph is the reason why transistors are getting smaller, but power consumption has leveled off.
A few more turns of the process crank are possible at great cost, but Moore’s Law is not going to give us the same significant improvement in horsepower as it did 20 years ago.
Other Sources:
https://plato.stanford.edu/entries/causation-backwards/
https://www.quora.com/Why-is-Moores-law-dead
