(An slightly edited version of this column appeared in EE Times, November 6, 2006)
I've always been fascinated by engineers who push a design to a performance extreme in a selected dimension, such as resolution, absolute accuracy, or power dissipation. These designers have to understand and analyze every source of error and shortfall, then work out a way around each, but without excessively compromising other factors. Call it a kind of “extreme scene investigation”, though I doubt we'll see a TV show any time soon.
My first encounter with this type of engineering was over 25 years ago, in an article by Jim Williams, now staff scientist at Linear Technology Corp (Milpitas CA). It discussed the construction of a precise scale, to be used by a nutrition lab for weighing newborns and every bit of food they ate, as well as their “output”. The scale needed resolution to 0.1 ounce and excellent absolute accuracy. But those specifications were achievable, albeit with effort, until you added the next two: it had to use only standard, off-the-shelf components, and never need calibration once placed in use.
Jim's article discussed how he uncovered every potential source of error in components, circuitry, and system, as well as those due to stray currents and EM fields; he then methodically knocked each one down through very careful engineering technique. At the time, high-resolution A/D converters were expensive, idiosyncratic rarities, nor were there good, easy-to-use voltage reference ICs, among other deficiencies. As an example of how he achieved his goal, he used a standard band-gap reference, aged it, tested it to find the drive-current value where it had minimum drift, and then ran it at that value.
To me, this sort of extreme attention to detail, and a successful design, is part of the art and “existential pleasures of engineering”, to borrow the title of Samuel Florman's excellent 1976 book. Overcoming first-order error sources, and then 2nd and 3rd-order sources, is a challenge for which many designs clearly don't have the luxury, or need. But for those that do, it's engineering at its exhilarating ultimate.
For example, when engineers at Draper Laboratory, (Cambridge, MA), were developing the ultra-accurate missile-guidance gyroscopes, they used extremely precise tiny ball bearings, of course, to minimize gyro-wheel friction. But the residual friction was still too much, so they floated the entire gyro assembly in a high-density fluid, to make the gyro assembly buoyant and so reduce the effect of wheel's mass on the bearings. Good enough? Not quite. Since the density of the fluid varied with temperature, they kept the entire assembly at a specific temperature, to eliminate yet another residual error source.
Most engineering designs today do not involve such extremes of performance. Instead, there's a more carefully nuanced balance among factors such as overall performance, power, cost, size, and time to market. In fact, in many applications, there's no point in exceeding a specification—after all, if you need to achieve an industry-standard requirement such as the data rate for IEEE 802.n, exceeding that rate conveys no benefit. But sometimes it's nice to think about how you would design if you really had a focused goal on extreme performance, and what that implies for engineering discipline and execution.