Specsmanship

Every Joe at the lumberyard understands that a 2×4 does not measure 2 in. by 4 in. Those "nominal dimensions" may apply at the rough-cut stage, but the actual finished dimensions are approximately 1.5×3.5 in. after air drying, planing, and rounding the corners so you don't get splinters. If you need an accurate, straight piece of wood, then after kiln drying, you must pass your stick through a second joiner/planer phase. This operation shaves off even more wood, leaving perfectly square, true surfaces on all four sides with no bow or twist.

Knots in wood grain.What about knots? Good lumber carries a grading stamp related to the severity and location of knots and other imperfections. Grading is tricky. For example, a beam held horizontally takes most of its stress near the top edge (compression) and bottom edge (tension), leaving the middle region relatively stress-free. A big knot in the middle of a board, therefore, may not degrade the rating. If you cut that same beam in half lengthwise so the big knot appears on the side of the new structure, where stress is highest, the rating plummets.

Architects and builders understand these issues. They call out the specific type and grade of lumber for every part of a new building, such as "green Douglas fir, grade standard or better." They also design in a substantial margin of safety to account for the reality of manufacturing in their industry—sloppy assembly, inability to read or follow directions, and materials a grade or two below the called-for specifications.

Similar considerations affect electronics. A 5-V capacitor doesn't last long when you use it at 5 V. A 16-bit ADC doesn't really provide 216 equally spaced quantization levels. A 150-Gbyte disk doesn't really hold 150 Gbytes. Engineers must take these issues into account. In that way, the electronics industry bears a certain similarity to the lumber business.

However, one important difference exists. Every year, electronic-component manufacturers spew forth nominal specifications, typical operating parameters, and models "indicative of nominal performance but not guaranteed." Engineers then use these models, verbatim, in simulators all over the world, to verify the correctness of their designs. Where is the safety factor?

In other words, suppose an engineer creates a mechanical model of a 2×4 that actually measures 2 by 4 inches. He then proceeds to design things using this model, assuming further that every board runs straight and true, with no knots. He makes no allowance for error. Are you laughing? A carpenter would!

Seriously, there are no rating agencies for semiconductor models, no federal regulations, and no safety inspectors. Little prevents a manufacturer from supplying a nominal model claiming one level of performance but delivering something quite different.

For this reason, designers of military, medical, and high-reliability equipment always derate component specifications. They account for extremes of power-supply fluctuation, ambient temperature, loss of active cooling systems, material variation, memory errors due to high-energy particle radiation, and the expected degree of specsmanship on the part of their component suppliers. Then, if they really want the product to work, designers test all components before putting them into the system.

Not every project deserves such scrutiny, but, if you do not allocate a healthy margin of safety in your simulations, you deserve a good knock on the head with a nominally dimensioned 1.5×3.5-in. piece of green Douglas fir, standard and better grade.