If the 24MB 1-T SRAM is really integrated on the Hollywood die, that would explain everything for me. Flipper was ~51M transistors on a 0.18 process. About half was logic and half was 1-T. For the sake of argument, lets say that a 0.09 process allows 4x the number of transistors with the same die size (I think Flipper was 10mmx10mm). 24MB of 1-T takes up ~192M transistors. Throw in an extra 24M transistors of logic and you arrive at 216M transistors. Even if you through in a few million more, this still borders on feasible. They could even increase the die size a bit without hurting yields too much. I'd imagine that die sizes have been increasing on average anyway due to the move to 12-inch wafers (although I don't know what the wafer size was for NEC's 0.18 process, or what the wafer size is for the Hollywood 0.09 process).