eee; photos on Flickr

I'm talking about the kind of speed that otherwise sane engineers build into semiconductors at the expense of reliability, economy, safety and everything else that should be holy. They do it because it's fashionable to be fast, and the rest of the world eggs them on by over-specifying speed, by demanding nanoseconds where milliseconds would do.

It's not hard to understand the pressures for speed. After all, it's easy enough to build an adder that adds, so once you've done that, you start looking for some other measure of how much you've pushed yourself. An obvious test is how fast you can add. Another is how small you can make the adder. Another is how reliable or inexpensive you can make it, but somehow that doesn't sound as exciting to most people.

Now, I am prefectly aware that there is an honest, legitimate requirement for speed sometimes, and if you have to go fast, you have to go fast. My point is that most of us don't have to go as fast as fashion tells us to go. And when you lose sight of that fact you pay through the nose, because speed - in anything from automobiles to pep pills to transistors - is always a bad bargain.

Here we really have a question of engineering sanity in setting design priorities. If speed is at the top of the list, everything else suffers. If, instead, an engineer backs off on speed in his design criteria and looks at total economic justification, reliability, etc., letting speed fall where it may, he will design a better system.

The urge to make things small, incidentally, is a blood brother of the urge to make things fast. Miniaturization can kill, too. Interestingly enough, these two yahoo concepts - speed and size - are in mortal conflict. Fast things are usually bigger than slow things, because there is more power to be dissipated, and more crosstalk and shielding to worry about.

Let's look at cases. Suppose an engineer designing a small computer rules out "slow" TTL logic in favor of ECI, thinking that in so doing he has not only shaved his delay times but also solved his glitch problems. But he soon finds that instead of eliminating glitches he has just moved the problem up to X-band, and now he has magnetic-coupling problems he never dreamed of before. If he's clever, he winds up no worse off than when he started, except that he may find his 1/4-inch-square device burning up 600 milliwatts, or about enough to melt the markings.

But that's only part of the story. The speed freak also throws away noise rejection, which is really the name of the game in industrial design. A high-speed device is always going to be more sensitive to high-speed digital noise than a slow-speed device. And what happens when a stray 5-nanosecond glitch puts a wrong bit into memory? That wrong bit sends the computer to the wrong address, which then gives the whole computer a nervous breakdown. A computer can be operationally destroyed by the funniest little noise that occurs once every six hours.

The extra power needed to go fast, also means extra heat, and this often means a cute little fan with its dirty little filter to carry away that extra heat. Everyone knows that the filters never get cleaned, and common sense tells you that the very presence of a fan means that the equipment needs free air, so where does that leave you? With a piece of equipment that was born to destroy itself.

Extra power also means extra money to supply the power. In the speed world of TTL, every dollar's worth of ICs need a dollar's worth of power supply. At that rate, an eventual billion-dollar IC market means a billion-dollar power-supply market. Zowie.

People who design and build industrial test equipment are as subject to speed pressures as anyone. But I believe that we should put these in their proper priority, which is after the pressures to test reliably, safely, and economically. And when we look for reliability, safety, and economy, we quickly learn that slow is good.

For openers, "slow" rejects noise.

A "slow" DTL circuit in an instrument will generate 1/100 the noise and be only 1/10 as sensitive to noise as conventional TTL. Of course, that means going four or five times slower than you could with TTL, but in many

36

part of an archival project, featuring the work of nick dewolf

Requests for use are welcome via flickrmail or nickdewolfphotoarchive [at] gmail [dot] com

eee by Jordan P

3

EEE by °ˆº•·-=+=-·•ºˆ°

5 1

eee by czlowiek z polnocy

2

EEE_4205 by Ryan Yensid