The usual problem with Bio Miracle is to do with $4085.
A lot of emulators that seem to get it right are treating $4085 as a modulator phase reset, which does not
happen, but it works around a different problem. Bio Miracle is the only game I know of that will try writing $4085 when the modulator is not halted (via $4087). It writes it very soon after $4087 resumes the modulator, so in accurate emulation the modulator will not have time to tick and get out of tune before $4085 is supposed to reset it.
The code for NSFPlay's FDS emulation should be an accurate reference for this, but sadly its CPU synchronization is not accurate enough for it, so it has an option for the $4085 phase reset hack on by default. ;P I plan to improve CPU synch in a future version. (Currently it ticks the audio units for 1 sample, then ticks the CPU for an equivalent number of cycles, and unless high oversampling is used it's not good enough to meet Bio Miracle's tight timing constraints). So... the FDS part of the code
is fine, I think?
I don't know in which ways the existing FPGA emulations are failing, but it might be that they're correctly implementing $4085 as not resetting the phase, but don't synchronize the timing? Not sure. Might also be that $4087 halt needs to reset the time until next modulator tick, which is important for giving Bio Miracle enough time to slip in the $4085 write before the tick.
Also worth noting that the wrapping behaviour of the modulator is different than used to be believed, mostly that it goes from -64 to 192, rather than -128 to 128. The details of this are on the wiki page.