Working Up

by Dwayne Phillips

I have seen many “critical” systems built with commercial products that are available at the mall, Best Buy, Circuit City, you name them. The result can be disastrous. Please be careful.

About ten years ago I worked on a small project that built a data link. We were sending short, simple text messages through a satellite. This was an odd situation that used a low data rate – a couple hundred bytes per second – on an almost-forgotten side channel of the satellite. The engineers built a circuit card that had a serial data port on it. The trick was that it used a precise data rate – one that you could not buy off-the-shelf.

The idea was simple: build this serial port on a circuit card ($20,000 in precision parts), pop it in a basic PC, connect that to a satellite communications system, and go. It didn’t work. The PC would not recognize the circuit card. Days of wrangling and experimenting were in vain. As a last gasp, the engineers put the circuit card in another PC – voila’ everything worked. They repeated the experiment a dozen times. Some PCs recognized the circuit card while others did not.

PCs were not standard devices. They had many variables in them – variables that we could not predict.

Move forward a few years. An experienced and smart engineer built a special-purpose data recording device for us. It used SD cards for storage. A key parameter was that the device had to run X hours on a single AA battery. The engineer researched, experimented, designed, and built a dozen prototypes. In testing, he discovered a wide variance in the amount of time his devices worked on a single AA battery. The variance in the batteries had to be the culprit, so he moved testing to a high-quality bench power supply.

He measured the current each SD storage card used. The variance was huge. These storage cards were all from the same manufacturer, all had the same storage capacity, all had everything the same except for the variance in the critical area of current draw.

I can repeat many other similar stories. The moral is always the same:

Take care with things that you can buy at the mall.

Consider the commercial SD memory card. A manufacturer designs an integrated circuit to hold 8GBytes of data. The manufacturer pumps them out of the foundry and then tests them. Some cards hold 8GBytes as designed. Some cards have a bad section that kills half the storage, so the manufacturer stamps it as a 4GByte card. Some cards have more bad sections, so the manufacturer stamps them as 2GByte or 1GByte or whatever works. How much current does each device draw? Who knows.

Commercial-off-the-shelf (COTS) has long been hailed as a savior to those of us who build low-volume specialty devices. Just buy the parts, plug them in, and go. Save huge amounts of money in design cost.

Yes, there is much good in COTS parts, but there are also traps that are hidden, unless you test the parts. You test and test and test and use the parts that perform to your specifications. You find yourself trading design cost for testing and selection costs. The test and selection is usually cheaper, but it is not zero.

Go ahead and use parts and pieces that you can buy at the mall. Don’t, however, do this blindly. Plan to buy two or three times the volume of parts you need and also plan for the time and labor needed to test and select parts that meet your specs. A 2GByte USB thumb drive may cost $20 at Best Buy. It will, however, cost thousands of dollars by the time you put it in a critical system.