Today, discrete transistors seem almost like anachronisms, although they're still widely used in many demanding applications. Costs range from nearly nothing to tens of dollars or more for certain specialized parts. An IC the size of that venerable CK-722 might have hundreds of millions of transistors, each of which costs the buyer a few microcents.

Ironically, some of the problems that plagued vacuum tubes and lead to their near-demise now haunt transistorized products. In 1946, all of the computer capability in the world consumed a few hundred kilowatts. Today a single server farm sucks many megawatts. According to http://blogs.business2.com/greenwombat/2007/02/photo_originall.html, in 2005 server farms worldwide needed the equivalent of 14 one-gigawatt power plants. Google's data center in The Dalles, Oregon reputedly has cooling towers four-stories tall.

Transistors come in many varieties, the field-effect transistor (FET) being the most important. Invented in 1960 (drawing on Shockley's work) by John Atalla, it was at first a novelty. RCA introduced a series of logic chips using FETs, but they were used only in specialty, low-power applications due to their low speed. Everyone knew the technology would never replace the much more useful junction transistor.

Now, of course, FETs are the basis of the digital revolution. The speed problems were solved, and their extremely low power requirements made it possible to pack millions on to a single IC.

A three tube radio didn't generate all that much heat, but group 18,000 into a computer and the air conditioning system becomes a significant problem. The same holds true for all kinds of transistors: a single IC with hundreds of millions low-power FETs will thermally self-destruct. So, ironically once again, vendors are grappling with different technologies like multicore to get better MIPs per milliwatt ratios.

At the same time Morse was perfecting the telegraph, the first real electrical system, Rudolf Clausius codified the basic idea of the second law of thermodynamics, which has haunted the entire history of electronics. Multicore may or may not be a solution to MIPs/mW today, but put huge numbers of low-power CPUs on a single core and Clasius's law will surface yet again. I suspect that long before the transistor's 100th birthday entirely novel, low-entropy technologies will be invented. And those, too, will fall to inexorable thermal scaling problems.

Jack Ganssle (jack@ganssle.com) is a lecturer and consultant specializing in embedded systems' development issues. For more information about Jack click here .