As I alluded to in my previous post, I was looking for a cheaper alternative to the BSS83 for situations where it wouldn't be used as a transmission gate. The first alternative I looked at had been the Fairchild FDV301N, which came in a SOT-23 package that is basically the same size as the BSS83, only with 3 pins instead of 4. It had really nice characteristics, including price: only 4.7 cents each in 1K+ quantities.
When I got them on the bench I was greatly disappointed to find it had one major flaw: it has a static protection diode between the source and gate. If the gate ever goes negative with respect to the source, the FDV301's protection diode conducts. This means it's useless in any configuration other than when the source is grounded. Sadly, I put it aside. The next one I experimented with was the already mentioned DMN26; it too has protection diodes, but they're back-to-back Zeners which don't affect its use in normal circumstances. The performance characteristics were good, but they're so tiny I put them aside as a last resort. My third pick was the BSS138, which looked good but, as described in the previous posting, was slow as molasses in winter.
Further searches turned up no really good candidates. Others I looked at were either expensive or had very high input and output capacitance and weren't really suitable. I even took another look at the two Micrel parts, thinking maybe I could do the project more easily in P-channel, but they were more expensive than the BSS83 and had huge input capacitances (600pF for the MIC94050 vs 27pF for the BSS138, 14pF for the DMN26, 9.5pF for the FDV301, and 1.5pF for the BSS83). I decided to leave the decision for later.
I'd been trying to come up with a reliable way to automate the identification of which transistors were being used as transmission gates and had to be 4-pin devices, and which could be more conventional 3-pin. After a few evenings toying with ideas, two obvious situations came to mind. One was where one end of the channel was tied to +5V, and the other was where one end went to ground. A simple text search through the netlist told me that 104 went to VDD (+5V), and a whopping 1041 went to ground.
That got me thinking. The problem with the FDV301N was that the source needed to go to ground, and 1041 of 1749, or 60%, did just that. Using the FDV301N for those 1041 would save $262, or about half the cost of the transistors. So I took another look at the FDV301 and couldn't find any good reason not to use it in those situations.
What about using the DMN26 to replace some more BSS83s? It turns out that replacing the 104 that go to VDD would only save another $10. That alone wasn't worth the headache of using them. I could do further analysis to differentiate transmission gates from other uses, but even if I could replace half of the remaining 603 with DMN26s the savings is only another $55 due to the shifting price breaks. Unless I could find an iron-clad way to identify transmission gates, there's an increasing risk of making a board that wouldn't work, and since each bare PCB alone is $180 that doesn't seem worthwhile.
Thus I've settled on only two types of transistors: FDV301N for everything that goes to ground, and BSS83 for everything else.