I agree that it, if you have not already recapped your Mac, it would be wise to use 25v caps instead of 16v due to the negligible increase in price. However, I believe that your conclusions regarding the risk of using 16v caps are overstated:
As shown in
Table 4 on page 7 of this
AVX Technical Document, the failure rate of 47uF 16V Solid Tantalum capacitors with a recommended 50% voltage derating stands at about
0.03% versus
1.1% for no voltage derating
This is
true, but why omit the
context, page 19 of the PDF?:
I don't know the series resistance of the circuit that these SE/30 capacitors are in, nor do I know whether these SE/30 capacitors are exposed to similar surges as in the test. Without this information, it seems premature to use this particular statistic to suggest that 16v capacitors are at "too high" a risk of failure - perhaps they are at
higher risk of failure than 25v capacitors would be, but it's not clear from this statistic that this difference would be of any practical significance. Remember also that this statistic is for capacitors with 50% derating and no derating at all. In the SE/30 circuit, the capacitors are at 25% derating.
Furthermore, when examining
steady-state failures, it becomes clear that the failure rate scales
exponentially with voltage, not linearly (figure 2):
As figure 2 shows, the correction factor with a 25% derating
is an more than an order of magnitude smaller than the correction factor for a capacitor with no derating. (Yes, the correction factor for a capacitor with a 50% derating is even better, but again, there are diminishing returns here - not clear that this difference is of
practical significance).
This correction factor is for the steady-state rather than dynamic situation. As you noted:
As shown in
Table 4 on page 7 of this
AVX Technical Document, the failure rate of 47uF 16V Solid Tantalum capacitors with a recommended 50% voltage derating stands at about
0.03% versus
1.1% for no voltage derating
Suppose similar behavior (failures scale exponentially with voltage) occurs in the dynamic situation. The capacitors in the SE/30 see some derating as they are at 12v/16v. In this scenario we'd expect the failure rate of 47uF 16V Solid Tantalum capacitors with a 25% derating to be much, much closer to 0.03% than 1.1% (and remember, those figures are derived from a "highly accelerated surge test machine" and failure rates by the end consumer are expected to be lower).
Moving back to the steady-state situation, as here we can obtain some quick-and-dirty failure rate estimates.
We have the following equation:
We have from Figure 2 above that F_u, the voltage correction factor for a 16v cap with a 12v applied voltage is
0.08. F_t is obtained from Figure 3:
I don't know what the ambient temperature is for these capacitors. I'm guessing it isn't 85C. Nevertheless,
let's assume it is 85C, and obtain an upper bound estimate of the failure rate. Then F_t is
1.00.
Again, I don't know what the circuit resistance in ohms/volt is for the SE/30. To form an upper bound estimate, I'll assume it's 0.1, so F_R is
1.00.
Finally, we have a standard failure rate F_B of
1%/1000hrs.
Thus, we obtain an upper bound estimate of .08 * 1.00 * 1.00 * 1%/1000HRS, or .08%/1000Hrs.
This is an extremely small failure rate, and it's
an upper bound estimate: if anything, this overstates the real-world steady state failure rate of these capacitors. Recall we assumed that these capacitors experienced an ambient temperature of 85C...
In light of this, I think there is zero reason for anyone who has already replaced these capacitors with 16V tantalum capacitors to desolder and replace them with 25V capacitors. As far as I know,
nobody has experienced damage to their computer from using 16V capacitors in this circuit. The list of people who have experienced damage to their logic boards from a lifted pad during recapping is a significantly longer one.