In this letter, we propose a data randomization scheme for endurance and interference mitigation of deeply-scaled multilevel flash memory. We address the relationships between data patterns and the raw bit error rate. An on-chip pseudorandom generator composed of an address-based seed location decoder is developed and evaluated with respect to uniformity. Experiments performed with 2x-nm and 4x-nm NAND flash memory devices illustrate the effectiveness of our scheme. The results show that the error rate is reduced up to 86% compared to that of a conventional cycling scheme. Accordingly, the endurance phenomenon can be mitigated through analysis of interference that causes tech shrinkage.