"Set Cff to 220nF, not 100nF. The datasheet lies. Also, the ground plane matters more than your marriage. Run the transient analysis for 500ms, not 50. The instability shows up late."