SLOW DOWN AND GROW UP

Dick Pountain/09:29/01 February 1996/Idealog 18

I'm sure there's a rule somewhere that says columnists shouldn't use  their columns to praise rival magazines, but I'm equally sure our esteemed publisher won't object if that magazine is Scientific American, 150 years old and in no sense a competitor to PC Pro. I've been reading SciAm, with a few gaps, since I was 16 years old and it's still the only magazine whose appearance I greet without even the slightest twinge of "Oh no, something else to read!". I bought my first ever copy to read on the train, travelling down to London to go to university in the early '60s, so it became inextricably fused in my mind with growing up and independence. 

Though it became thin and ill for a while in the tough early '90s, it survived and is returning to form. Its covers are still a joy - a single strong (usually painted) image - compared to the supermarket windows of our computer mags. Its prose style is still cool, jargon free and authoritative. Its diagrams are still the best in the business, small works of art, models of clarity and concision. Its truly great book reviewer Philip Morrison has now retired but thankfully still writes a short column. And in case you're thinking SciAm lacks sex appeal, I bet even Loaded doesn't carry a review of 'Copulation, Masturbation and Infidelity' by R. Baker and M. Bellis.

However there is actually a point to this gushing paean. A few times each year Scientific American will run a feature about the computer or semiconductor business, and often these will spot trends that those of us who are immersed in the fervid business itself have missed. Sure enough, the January 96 issue bears the arresting cover line "Microchip progress: end in sight?". The article in question is by Jerry D. and G. Dan Hutcheson (father and son), who advance the argument that at last we may be seeing a limit to the shrinking of chip fabrication technology, and hence to the flow of ever cheaper chips. Now, hacks have been flogging this old nag around the track once a year for a long time, but the reason we should perhaps listen this time is that Jerry is a top physicist/engineer who founded VLSI Research Inc., and more to the point he invented the return-on-investment model that all the big semicon companies use to guide their investment plans. In short, he knows.

J and D's argument is based on economics. Lots of people have argued from the physics, for example that as the metal tracks on a chip keep shrinking, eventually quantum effects will stop them conducting predictably, or they won't be able to dissipate heat, or whatever. J and D will have none of this. Yankee optimists of the best sort, they argue that smart engineers will always find a way around problems of mere physics. The trouble is those ways cost more and more to find, and more again to implement. Nowadays everyone, even the Sunday papers, knows about Moore's 'Law' which states that the number of transistors you can cram onto a chip will double every 18 months. The problem is that Moore's Law doesn't say how much it will cost to build a fabrication plant to do the cramming. Intel's next two fabs in Oregon and Arizona are costing $1.1 billion and $1.3 billion, while Motorola's next may cost $2.4 billion. That is not pocket change; we are talking Moon Shot money, GDP-of-a-Third-World-Country money.

In fact the costs of equipping a semiconductor plant exhibit a Moore's Law of their own, doubling in cost every three years, or at half the rate of chip shrinkage. The Hutchesons argue from this not that the chip industry will come shuddering to a halt, which would be absurd, but something far more horrible - it may have to grow up. They argue from the history of other technology-based industries like the railway, aviation and car industries, and very convincingly the argument is.

All these industries went through an early phase of technical innovation and exponential growth, during which the product was constantly getting cheaper, just as transistors do today. Take the aviation industry; it started fast and furious (just 40 years from the Wright brothers to the Flying Fortress); it began by serving military markets before moving into civilian ones; and in the early years there was a continual reduction of the cost per passenger-mile and journey time simultaneously; anything sound familiar here?

So what happened? The industry topped out with two products, the Boeing 747 as the high-water mark of capacity, and Concorde as the (uneconomical) high mark of speed. But the industry didn't die. Instead research emphasis has moved away from speed onto efficiency, passenger comfort and low noise. Products have become diversified with many smaller planes tailored for specific markets. Fares no longer fall exponentially but people keep on flying.

So what would happen to the computer industry if the same were to happen with chips? Try to imagine for a moment that next year's PC WON'T be twice as fast as this year's, WON'T have twice as much memory, and - God Forbid - might even be more expensive. You might actually keep a PC for several years. Software writers might not be permitted to add yet another layer of bloat to their operating system/word processor/mail exchange. Stability might concentrate people's minds on the horrendous inefficiency of current software production, and advanced programming techniques might be taken more seriously. It doesn't sound too bad to me.

Are there any great milestones that we wouldn't be able to reach? Well, we're topping out just as 24-bit full color, real-time animation, and 3D capabilities finally coincide with the capabilities of the human eye and brain. A coincidence? Maybe not.