Some Integrated Circuit Scenarios

• Science fiction uses scenarios as the setting for drama.
• Others (eg, GBN) use scenarios as a basis for planning:
  • Build those wild-eyed scenarios, each at the extreme of plausibility but in a different direction.
    • For each scenario, write up a little (fake) history, explaining why in retrospect this outcome was inevitable. In particular, this would include a list of symptoms and anti-symptoms.
    • Taken together, list responses common to different scenarios, with a major goal of maintaining flexibility.
  • Track relevant events as time passes
    • Update existing scenarios accordingly
    • Create new scenarios accordingly

In this talk I'll discuss several IC scenarios, mainly dealing with threats and extreme promises.

But first I'd like to turn the game around and try to explain our bizarre past:

One might simplify the explanation to a virtuous circle of development:

• Lithography in service of electronics
• Electronics in service of computers
• Computers in service of software
• Software running on computers in service of as a substitute for more and more of the technological arts.

Some might say this list is too parochial, even as pop history. On the other hand, this progress can transform all of technology.
Different aspects of the semiconductor business have their ups and downs, but overall the industry is in the blessed position of being the economic key to success in almost every human endeavor.

Integrated circuits are not just a better mousetrap, they are close to being a universal better mousetrap.

So what could cripple or block further progress of this sort? Here are some possibilities, of varying plausibility, speed of onset, and unpleasantness:

Insurmountable physical limitation FOOBAR (replace FOOBAR with your current favorite) brings Moore's progress to an end.

This form of "crying wolf" has been so common over the years that there's a tendency to shrug it off. The underlying cosmic limitations on computation appear to lie very far in the future. Nearer term (but still beyond many folks' favorite personal retirement dates), there seem to be limits on lithography. If economic demand for improvement remains, I'd bet on human creativity to keep pushing hardware power forward for some time to come.

Economic demand for improved chip technology diminishes because we can't figure out how to write create software to exploit more powerful hardware. There are intriguing variants on this scenario, including:

• Massive multicore systems ease our thermal problems but leave us with chips that general purpose software can't exploit.
  This scenario would diminish demand in certain areas, with interesting tradeoffs.
  (My own interest in the multi-core programming issue lies in the fact that Evolution has long ago solved the problem of general-purpose massive parallelism. Now the software folks may have to become very focused on discovering just what those solutions are! If Kilo-core and Mega-core desktops are not just going to be sales hype, the applications running on them should be remarkable!)
• Software projects that endeavor to exploit increasing hardware power fail in more and more spectacular ways. (Human pilots occasionally run one aircraft into another; a bug in an automatic fleet manager could bring a dozen aircraft to the same point in space and time.)

Economic demand for chip technology collapses because of wide-area hardware failures of embedded microcontrollers.

In 2007, such a failure in the Bay Area would probably be as serious as a major earthquake. As we approach ubiquity, embedded microprocessors become a single-failure point for all our technology, and a wide-area failure would be one of the worst categories of physical disaster. This single failure point threat is so deadly that even if we didn't have any particular cause imagined for it (such as nuclear EMP), this scenario should still demand serious attention. Unfortunately, the economic win represented by embedded micros is so great that we tend toward wilful dismissal of the danger.

Given the probable ubiquity of embedded microcontrollers, what can be done to mitigate the danger of wide-area failures?

• Design systems with this threat in mind. Systems should include "dumb" fallback modes, even if low-performing.
• Encourage the development of a variety of semiconductor technologies with different failure characteristics -- that is, avoid a monoculture of hardware substrates. (This might even have happy side effects, encouraging low-performance physics, that while relatively slow, might be very easy to fab and design for.)
• Encourage a new hobby: the manufacture and ownership of appliances and vehicles that (gasp!) contain no semiconductor logic.

Note that this doesn't directly affect progress in chip technology. It simply diverts that progress into more and more elaborate countermeasures (think 100000 transistors to implement a "secure flipflop").

The most likely outcome would be an extended "Chip Enforcement" war. Ultimately, if the notion of the Secure Hardware Environment were successful, we might actually enter an era where reality could be legislated (see Karl Schroeder's novel Lady of Mazes).

Suppose our progress is not blocked. How ubiquitous can ubiquitous computing come to be?

Nowadays, most biologists recognize three great Domains of Life: Bacteria, Archaea, Eukaryota [illustration from Wikipedia].

Perhaps we should consider that we are seeing the beginning of a new Domain of Life: ubiquitous IC logic. (True, the IC Domain is not descended from our hypothetical biological common ancestor -- but it is the creation of a twig on that great Tree of Life.)

One might criticize this fancy because ICs cannot reproduce without the aid of existing life forms and their products -- but this is also true of the animals, including humans!

Possible characteristics the IC Domain:

• Probably very fast evolution
• At the "center", the "server farms" of such an era should be clearly superhuman (okay, so this is another path to the Technological Singularity)
• At the "edges", the material world could awaken.

And a piece of good news (at least reasoning by analogy with our current "tree of life"): We humans would probably continue as the fallback rescue system for the new critters who think they are running the show.