If the manufacturer has not informed the change to a die-shrink part, I guess the chip specifications (mostly the timing specs) published in datasheet has not changed. In that case the only change was in the impedance (especially the inductance) of the package pins. Would this make such a huge impact on the board performance unless there was inherent design issue on board, which was on the border line?
I've seen the same problem with a LM324 op amp die shrink. Reduced yield problems were directly correlated with a die shrink (and no, we weren't informed of the change). In this case, the die shrink was physically obvious (even if the electrical performance change wasn't immediately so) as the IC was bought in slice form and cut, glued and wire bonded into the module.
In this case, poor layout practice was a factor (a non-inverting input track ran parallel to an output track for perhaps a couple of centimetres), but modules using the larger die didn't break into oscillation.
Obviously the smaller transistors in the shrunk die had a higher cut-off frequency and the internal compensation probably didn't work with as much margin as in the larger, "electrically equivalent" IC.
I can understand the op amp manufacturer's point of view in not informing us of the change given the part was electrically equivalent and the die shrink part would probably have been approved on the basis of a desktop review anyway...
When everything else checks out right, then you look for noise, and often that noise is RFI. I wonder if his employers purchasing people approved the substitution of the "die-shrunk" parts as part of a cost reduction, or if the supplier just made the change to improve their profits. IN many cases it happens that "equivalent parts" are not close enough to function as required, and so problems arise when they are used.
It was good detective work indeed.
Blog Doing Math in FPGAs Tom Burke 20 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...