LM4950: LM4950 Bias Problem

We have a speaker amplifier design using an LM4950 amplifier that has been in production for several years.  Just recently we've been experiencing a problem with the amplifier where it will only operate for less than two seconds after power-up.  It appears to enter a shutdown-like state and will produce no audio output.

Our application is very much like the Typical Bridge-Tied-Load Audio Amplifier Application Circuit shown in Figure 1 of the data sheet.  Our circuit has a voltage gain Av of 28 -- RFa is 301k ohms, and the other three gain-setting resistors are 21.5k ohms.  Cbypass is 10 microfarads tantalum, and Cs (supply bypass) is 1 microfarad tantalum.  CINa is .1 microfarad.  The SHUTDOWN pin is fed by a pair of 100k ohm resistors configured as a voltage divider such that the SHUTDOWN pin receives VDD/2.  VDD in our application is 12 to 13.5 volts.

What we're seeing when the amplifier fails is that at power-up the BYPASS voltage will climb from zero volts to VDD over a period of about 1.8 seconds -- it does _NOT_ stop at VDD/2 as it should.  The amplifier will produce audio output during the time BYPASS is moving between 0 volts and VDD, but goes silent once VDD is reached.  The amp does not appear to be under any thermal stress, it is cool to the touch even without a heatsink.

We have determined that once the amplifier is in this anomalous bias state, we can restore it to working operation by momentarily grounding the SHUTDOWN pin.  When we do this, the BYPASS voltage will slew from VDD to 0 volts (take about 4 seconds), then climb back up to VDD/2 (about 550 miliseconds).  The amp then operates normally until the next power cycle.

I have at least one unit exhibiting the odd bias behavior that I can "fix" by placing a capacitor of 1 microfarad or larger between the SHUTDOWN pin and ground. This appears to hold the SHUTDOWN pin active-low momentarily at power-up and causes it to bias normally.  I am not proposing this as a production fix as the voltage presented to the SHUTDOWN pin could exceed VDD/2 when power is removed from the amplifier, but it may provide a hint as to what's going on.

As mentioned, we've been using this design for years without issue and are just now seeing this problem.  We suspect that it may be related to a particular batch or batches of parts -- the first line of device marking on failing devices that we've seen is VM27AB, VM28AB, and VM28AC.  Devices that we have with device marking VM54RA seem to be fine.

Do we just have some bad parts?  Is this a known issue that we can fix somehow (we hope without a major rework)?  Or is there maybe an issue with our particular circuit?

Any thoughts appreciated.  Thanks.

 - Pat