Tuesday, July 19, 2005

Electronic Advances

I began my engineering career in 1959, and it is not only amazing how technology has advanced since then, but I feel privileged to have been able to see much of it first-hand.

The junction transistor was invented in 1948, but it needed development before it was practical to use it. Computers had used vacuum tubes until the first fully transistorized computer in 1954, just five years before I started my career. I remember trying different circuit configurations that could acheive the same function, then counting the transistors so that I could select the lowest-cost configuration. That was important, because each transistor cost about $5 to $10. (Other kinds of transistors cost $45 or more.) Now, about 10,000 transistors cost one cent.

You could see the transitors back then. Each one looked like a little tin can, 1/4-inch wide, with three wires sticking out. Each transistor in the computer, with the help of a resistor and a capacitor, served as a switch that could turn electrical current on and off. It took a configuration of 12 such switches just to add two 'bits' (binary digits), including the 'carry' from a nearby digit position. That would occupy about a 4-inch by 10-inch area on a circuit board. Now that transistors are so much cheaper, twice as many are used for the same function, and it's all smaller than the period at the end of this sentence.

Back then, a computer was a room full of refrigerator-sized cabinets. And because everything was so costly, the computers were as simple as possible. Today's computers, although much smaller physically, are bigger in terms of the numbers of equivalent parts. I remember one of those refrigerator-sized cabinets was a memory storing just 256 words. Just a few days ago, I bought a memory card for a digital camera with one million times larger memory, and it is the size of a penny.

Some kinds of circuits need a 'matched pair' of transitors, for accurate balance. That was hard to acheive when transistors were made as individual devices. The solution was to make them as a pair. It was something like two cookies side-by-side on the same baking sheet, baked at the same time, would come out nearly identical. These were sold in the same little tin cans, but with six wires sticking out of each can. Later, when someone figured out how to make the resistors and capacitors on the same piece of silicon as the transitors, the 'integrated circuit' was born. At first, the integrated circuits where packaged in the same little six-wire cans as the matched transistors. With just two of those integrated circuits, we could make one 'flipflop' (a circuit for storing one bit), which previously occupied about a 4-inch by 6-inch area. Today, thousands of 'flipflops', or millions of transitors, fit on one integrated circuit. I remember that as companies like Motorola and Texas Instruments made more and more dense integrated circuits, they would brag about how many transistors were in each circuit. Now, nobody bothers to count.

The steady increase in integrated circuit density was described by Moore's Law (see here and here), which observed that the number of transistors per square inch on integrated circuits had doubled every year since the integrated circuit was invented. As the rate slowed, this was later revised to 'double every 18 months' and then to 'double every 24 months'. A few years before I retired, I was asked to design a circuit that could not be built -- yet. I was asked to design the most powerful digital correlator that could be built on one 'chip' in 2010. So I had to use Moore's Law to estimate -- to predict -- how much addition logic could be put into one integrated circuit in 2010. It was tricky, because Moore's Law was sometimes also stated as a doubling of speed rather than density, and addition can not only be done twice as fast by using circuits that are twice as fast, but also by using twice as much circuitry.

This wasn't the first time that I have designed for the future. Most electronic designs are rushed into production, to try to beat the competition. But when designing for GPS (the Global Positioning System) satellites, the design cycle is much slower-paced. The main reason is that if the circuits in a satellite fail, it is VERY expensive to send a repairman (a.k.a. space-walking astronaut) up to fix it. It is also very expensive to launch the satellite to begin with, and often a launch fails, and millions of dollars are suddenly lost. So the design cycle is deliberately slow and very careful, with lots of checking and testing. Then when a GPS satellite is built, it is not launched right away -- it is put into storage, and launched only when an older satellite fails so badly that it needs to be replaced. So, for example, my Phase Meter, which was invented in 1995 and patented in 2002, and destined for new GPS designs, is still not in space yet.

No comments: