When Second Sources aren’t

Editor’s note : James Bryant (retired), Analog Devices, is back with this informative and intriguing blog.

[Q] : We have been manufacturing a system using an Analog Devices' component for many years. Recently our purchasing department bought a consignment of cheaper replacement devices from another company and the whole batch of systems using them failed to work. Every specification on their data sheet is the same as ADI's, and when we test them they are well within those specifications. What is happening?

[A] The operation of your system relies on some feature of our product which is not featured on the data sheet and is different in the second source.

This can apply to any component, not just ICs or active devices. In an entirely different context I keep a bottle of hot sauce in my kitchen to adjust the heat of curries and other hot dishes. I recently bought a new bottle – my usual brand was unavailable so I bought another with the same Scoville rating (i.e. the data sheet specification was the same). This second sauce was horrible – the heat was indeed the same, but it had much slower onset (first taste seems bland, but the tongue burns several seconds later, which may cause the cook to judge too soon and add too much) and, while my usual sauce just contributes heat, the second sauce has a bitter acrid taste in addition to its heat which spoils the carefully planned flavor of the dish. Obviously the flavor of electronic components is irrelevant, but other unspecified parameters may be critical.

While I should like to tell you that all will be well if you remain loyal to Analog Devices – and that is probably true – it is quite important that you ensure that there is sufficiently close liaison between your purchasing department and system designers that when a change of component is proposed, for cost or any other reason, the new component is evaluated to ensure that it is truly compatible. Such evaluation must include hardware tests as well as any paper and/or simulation exercises since no model can simulate every feature of a device, and to improve their run-time macro-models often deliberately simplify structures which the modeller considers unimportant. It is impossible to discuss every possible issue in a large book, let alone a short article, but the principle involved in anticipating trouble is simple: ask yourself what non-ideal component behaviour may cause a circuit to malfunction – and check it out.

I shall list a few that I have encountered over the years, but I am trying to teach a mind-set here, rather than present a (far from exhaustive) list of potential problems.

  1. A recent RAQ considered unused device pins – I have seen second-sources where an unconnected pin in the original had an internal connection in the second source.
  2. Some op-amp inputs are high impedance even with large differential voltages, others have high impedance when used with negative feedback (and hence small differential Vin ) but their protection circuitry reduces Zin dramatically when the differential Vin >600mV.
  3. A modern (“improved”) second source of an old circuit may use a faster IC process and have unexpectedly wider bandwidth, potential instability, and broadband noise.
  4. Or a faster logic process may be vulnerable to ns glitches (which its predecessors ignored).
  5. A 10 nF 50 V ceramic capacitor may have much lower inductance, and hence HF impedance, than a cheaper foil capacitor of the same value.
  6. And even two manufacturers' standard (Cat 5) Ethernet cables may have sufficient differences of loss and crosstalk that only one works in a particular system.

Think, test and retest your hardware, and always remember Murphy's Law.

1 Unused Pins

2 “If it can go wrong, it will go wrong”

9 comments on “When Second Sources aren’t

  1. MWagner_MA
    February 10, 2016

    This is precisely the senario I instruct my younger colleagues to be concerned with.  The lesson is:  If you are depending on a parameter or characteristic of a part that ISN”T specified on a datasheet, you simple DON”T USE THAT part.  This is regardless of the fact that your prototype may work at the moment!  Eventually, some change may be made, and that parameter you depended on changes as well, and whalla! Your worst nightmare.  Now there are always exceptions to the rule, but if you have choices – chose another part.

  2. JNW
    February 10, 2016

    Years ago, in a different life, I had to deal with a company standard comunications interface that was in several products.  We had 5 sources.  

    Product 1 could use source A and B

    Product 2 could use source A and E

    Product 3 could use source A,B,C, and E

    Product 4 could only use source D


    It was all board layout issues as there were no differences in the schematic.


  3. MWagner_MA
    February 10, 2016

    That can happen, then it becomes a learning exercise for the layout designer/engineer.  In those cases you would likely find excessive noise or ripple on a power rail that some devices were sensitive to.  I have also seen flagrant disregard for capacitive loading on opamps.  Some phase margins vary from one mfg to another (analog).  If designs are doing that – the designs really aren't done, and you need a root cause analysis.

  4. araasch
    February 10, 2016

    I would investigate why the second source does not work vs the ADI part, then confirm that the ADI part is being used correctly as per the manufacturer's recommendations in the design.  For instance, there can be power supply sequencing issues where the existing design with ADI just barely works while the alternative does not. However, if ADI makes a process improvement that does not impact its specified parameters your design may fail with the new ADI parts as well.  This is why it is important to find out why the design does not work with the replacement, it may lead to the understanding of a hidden issue with your existing design and give you time to correct it before your design is impossible to build with any vendor's product.

    Working with the FAE for the part in question is something to consider as well.  They have access to information that may not be in the public domain, or may suggest avenues of investigation to help you find the problem at hand.


    February 10, 2016

    Years ago, I was working for a company that designed complex test equipment.  The entire box-full of circuit boards was driven by a RESET circuit on one of the PCBs.

    We used one section of a Texas Instruments 7438 to drive the RESET line, but when The TI part became unavailable, we switched to Motorola.  While the specs of the devices were virtually identical,the Motorola parts sometimes worked, but usually did not.  Initially, the technicians simply swapped out Motorola parts until they found one that worked. But as you can imagine, we had a huge number of field failures.  When we could get the TI parts again, we switched back and the failure rates dropped to near zero.

    The engineers in the company passed the word around “NEVER BUY MOTOROLA – they aren't as good as TI (or even other manufacturer's parts).  Purchasing was told to NEVER buy Motorola!

    Then I looked at the issue.  I found that the RESET line went to 8 circuit boards, and each one had a 220 ohm pullup resistor to 5V.  And while the IOL of both The TI and Motorola parts were 48mA, in our circuit, the device actually would actually have tosink 167mA in order to pull the line to 0.4V  (5V – 0.4V/(220/8)!

    Somehow, the TI part was sinking that amount of current, while the Motorola part could 'only' sink 88mA (measured).  Yes, the TI part was 'stronger', but the Motorola part still exceeded the datasheet value  by a factor of almost 2.

    I added two transistors to the RESET circuit (it was easier to change 1 PCB than to change the pullups on the 8 other boards).  Now the circuit worked with ANY manufacturer's 7438 part. 

    So, in this case, the second source didin't work, but the problem was a bad design. The answer was really on the datasheet all along, not some 'unlisted' parameter.

  6. xcvnb
    February 10, 2016

    About 35 years ago I worked for Datapoint Corporation which afforded itself the luxury of a components engineering group to write specs and qualify vendors of electronic parts to company part numbers.  Three parts come to mind that were problematical for  manufacturing anyway: uA723 voltage regulator, 74S161 4 bit loadable counter, and second sources of Mostek MK4164 dynamic RAM ics.

    The 723 received from the parts' inventor had production fallout so high our manufacturing engineers got a new company part number that excluded the problem child allowing only qualified second sources.  They then ECOed the PCB BOM to replace the old, problematical part number with the new more failure resistent part number parts.

    The problematical 161 counter was a TI second source qualified by datasheet and by test but one of the CPU boards we manufactured with them (four 161's constituted the memory address register) consistently failed test.  The remedy for that was to disqualify that vendor's part for our corresponding company part number.

    The MK4164 problem was more subtle.  The Mostek datasheet showed waveforms and specified strengent strobe pulse timing for address row and column address access (RAS and CAS) which Mostek's competetors carefully implemented.  Mostek's actual parts, however, performed perfectly with much looser RAS and CAS timing.  This allowed if not encouraged such carelessness in RAM RAS and CAS timing that second source memory devices would fail in the same boards that had been tested and would function perfectly with only Mostek parts.  Although the problem was a little hard to identify it was easily fixed since RAS and CAS pulse timing was usually set by tapped delay line connections.  All that became necessary was to change the PC boards' delay line taps to create the exact, conforming datasheet timing needed.

    So an IC inventors' parts may turn out to be unusable in manufacture, carefully qualified second source parts may not consistently work in particular circuits, and the component's datasheet may not tell all or even mean what it says.

    Bottom line: caveat emptor!

    Caveat emptor!

  7. Palavi
    February 12, 2016

    well i appreciates your comment

  8. saiopen
    February 15, 2016

    nice post

  9. raijakson
    February 15, 2016

    good post

Leave a Reply

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