Promise you won't laugh:)
2N6802 NMOS 2N6804 PMOS 2N2222 NPN 2N2906 PNPSPICE doesn't care about breakdown voltage on these devices. Again, the goal at this stage is simply to evaluate different architectures. The mosfets show a very dramatic difference with and without the bipolar driver, so the driver seems a necessary part of the design. The actual parts selection and detailed design is still way downstream. There's still lots of work to look at short circuit protection, etc.
For example, short circuit protection by current limiting doesn't seem to be possible in this type of follower. If the output were designed to limit at some fixed current, the output voltage would be limited to Imax
- Rload. However, the input voltage could still go anywhere, which means the problem is transferred to the input stage, which now has to handle the voltage difference between input and output.
Given that very low leakage diodes may be impossible to find, and series current limiting mosfets only go to about 500V, this creates a rather difficult issue. Perhaps limiting the input current with a large series resistor might work, but this increases the noise. And I still don't know how the op amp would cope with 1mA or so current at the input. Some of the datasheets show the input driven well past the supply rails, but I still need to test with the actual device.
As far as fixing the SPICE models, there's several ways to go. Jim proposed that Level 7 might solve the problem. Also, measurements on working hardware might show the performance is more than adequate for the requirements and no further SPICE work is needed. Or bench work might uncover different problems that SPICE can't see.
For example, is the op amp really capable of 1 ppm performance, and where do you get high value resistors with 0 ppm voltage coefficient? Will the oscillations show up, and will they be impossible to kill without adding too much resistance in series with the gates?
So there are lots of issues. If improving the SPICE model would help solve these problems, then that would be the way to go. But right now, I don't think it would make that much difference, and I can work on the other problems while you and Jim sort out the modeling issue:)
Mike Monett