This paper co-designs SLC Mask and MLC Weight Twin FeFET devices and Strong Lottery Ticket Hy-pothesis (SLTH)-based Neural Network (NN) algo-rithm to achieve highly error-tolerant low power Computation-in-Memory (CiM). SLC Mask FeFET masks or not weights stored in MLC Weight FeFET, and CiM power consumption is reduced. SLC Mask FeFET shows 86% inference ac-curacy even at 2,000 endurance cycles. Uniformly randomized VTH weight of MLC Weight FeFET achieves 87% inference accuracy against 10-year da-ta-retention and read-disturb. Moreover, proposed Shared-IL SLTH CiM and Common-mask SLTH CiM save CiM area by sharing select transistor or Mask FeFET.