It's a long story. But briefly it was from trying to simulate a power amplifier I designed back in 1994 and had published in Electronics World magazine Sept 1995. But the simulators I used then did not have accurate enough MOSFET and BJT models for my circuit so I created some macromodels. I developed one for power MOSFET's I called the VCG model (circuit) that used a diode and resistor for the subthreshold region and I wanted to solve the D+R gain function but couldn't because that was the unsolved problem in electrical engineering. But around 2002 I spotted Thomas Banwell's article showing how to solve it using the Lambert W-function inverse. Then I realised it just might be possible to also solve the equations for a complete power amplifier. That's how it came about.
One incentive to try and develop a circuit equation solution was the frustrations of SPICE simulators when they fail to converge; which happens (by Murphy's Law) just when you have spent days getting all the macromodel's right and arranged them into a complete amplifier ... but then it won't run. More days later and it runs but only just. Even after decades of SPICE tweaks I still find some simple circuits fall into the non-convergence 'pot-hole'. Someone fix it, please!