Advertisement

Analog Angle Blog

When Engineers Are a Little Too Smart for Themselves

Back in the day, when microprocessors were just beginning to rule the world, I was on a design team that included a very senior, extremely detail-oriented designer. His schematics were works of art, done with technical skill and achieving near perfection.

Many times, we needed no changes between his initial hand-drawn design and the final one, except when a vendor's part did not actually perform as claimed on the data sheets, or when there were unforeseeable mechanical-fit problems in layout which required some adjustment of the design. All this was in the days before modeling, simulation, and the other tools we routinely use now, which made his work even more impressive.

As was common at the time, his designs used many 7400 TTL digital-logic ICs, available as second sources from about a dozen vendors (and exemplified at the time to most EEs by the Texas Instruments' mustard-colored data books which the really senior engineers had in a hardcover version).

The much beloved Texas Instruments TTL data books (in soft cover).

The much beloved Texas Instruments TTL data books (in soft cover).

That's how I learned an important lesson about the transition between engineering and production/manufacturing.

At one of the final design reviews, someone asked if a diode wasn't needed between the TTL IC's output and the next stage in this particular configuration. In response, this senior engineer explained how he had ensured that the output of the gate would properly interface with the subsequent circuit: he saw that one — just one — of the many sources for the 74XX IC he was using incorporated an internal diode in their implementation, thereby eliminating the need for an external diode. Pretty smart! Here is a closer look at the TTL devices, courtesy of Wikipedia.

That's where the project leader jumped up: “You mean we are counting on the fact that this vendor internally uses a certain topology? And they could change that at any time with a different die design, but still meet their output specs?”

That's a real warning sign, of course. Then the purchasing liaison pointed out that they did not like to be locked into a single vendor for parts which were otherwise widely available, and 74XX parts were pretty much in the “buy the cheapest ones available, from anyone on approved vendor list” group.

Both the project leader and the purchasing fellow were absolutely right. It's unwise to count on an internal aspect of a part in a case such as this one, even if it cleverly saves the cost of an external diode. What would have happened is that the prototypes would have worked, and maybe even some early production units. Then at some random time in the future, we'd have gotten mysterious failures in production test or the field which would consume days and weeks to figure out — if we ever did.

Sometimes you can be so smart that you outsmart yourself, or inadvertently plant a delayed-action design bomb in your product. I've seen similar situations, though not quite as drastic, with mainstream analog parts such as op amps, as there are so many of them that seem so similar, offered by a single vendor as well as multiple vendors. So when your design counts on a second- or third-tier parameter or specification in order to work “just so,” and somehow that value changes or the part's source is changed (and maybe you even OK'd the change, having forgotten why you wanted that particular part in the first place), it could also mean you'll be spending long nights firefighting in crisis-mode a few months or years later.

Have you ever seen a designer be a little too clever in hardware or software? And confess: Have you ever done it yourself, only to later regret it?

5 comments on “When Engineers Are a Little Too Smart for Themselves

  1. jkvasan
    February 8, 2013

    BS,

    Nice post.

    We were designing a load cell based measurement module and used a 24 bit ADC. We laid the PCB as per the advise of the manufacturer. Soldering was good and the code was working perfectly. We had oscillations, used all averaging techniques and still the reading was oscillating. 

    Our circuit had decoupling capacitors on the AVCC , one 4.7 MFD and 0.1 MFD. This combination was normally used in many of our digital circuits and in some of the stepper motor controller chip circuits. Our engineer faithfully reproduced this part and used it at many places  in the circuit. After discussing with one of the EMC consultants, it turned out that due to the internal composition and the values of these parallel capacitors, a parasitic voltage is created and this was disturbing the measurement. The oscillation disappeared once the 4.7 MFD was removed.

    As some of them here are EMC and analog experts, may be , this could be discussed further.

  2. eafpres
    February 8, 2013

    Great reminder, Bill.  I can recall a couple of instances where an engineering designed in a part for a function it wasn't specified for, and was counting on some related property never changing, even though it wasn't on the data sheet.  

    In one case a clever RF/Antenna engineer figured out how to make a 0402 wire-wound inductor radiate, thereby having an SMT (surface mount technology) antenna before they were commonplace.  Unfortunately the whole thing was a really high Q design, and inductors that met spec would not work in some cases.  Thankfully it never went to production, or yield would have been terrible in test.

  3. Brad Albing
    February 12, 2013

    We had a similar problem once with some N-channel JFETs we were using in a circuit that detected the presence (or lack) of a gas flame by using “flame rectification.” The circuit would work quite well, but every once-in-a-while, we'd start getting bad FETs. Eventually decided it was pro'ly due to ESD in the manufacturing area causing damage to the FETs' gate-source junction. No good specs on the data sheet….

  4. Brad Albing
    March 28, 2013

    We bought enough of them tho' to get some additional screening done by the manufacturer.

  5. zeeglen
    November 18, 2015

    When purchasing demanded a 2nd source for a capacitor used in a resonant LC tank circuit I grabbed a part number of one that looked good – NP0, low loss – but never actually ordered the part to test it in the circuit.  Of course production did use the 2nd source and it did not work.  A closer look at the small print in the data sheet revealed that the tan loss angle was spec'd at 1KHz instead of 1MHz.

    For another engineering blooper see http : / / www . edn . com/electronics-blogs/tales-from-the-cube/4327332/The-case-of-the-stolen-capacitor  (remove spaces)

    Then there was the purchasing guy who told us our choice of VCXO (voltage controlled crystal osclillator) was way too expensive and he could source one for a lot cheaper that met the same spec.  Good thing we looked at his choice – ours had voltage controlled frequency for use in a PLL, his was voltage controlled all right – HI for ON, LO for tristate.

     

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.