Abstract: Magnetic random access memory (MRAM) uses the high and low resistance states of magnetic tunneling junction (MTJ) to store binary information, and the MTJ reference resistor network is often used as the hard decision threshold. MTJ resistance is easily affected by process deviation, voltage and temperature (PVT), which challenges the reliability of MRAM read operation in large PVT range. In this paper, a hard decision channel quantizer model based on MTJ reference resistance network scheme is proposed. The influence of key parameters such as temperature and process fluctuation on different resistance values of MTJ is considered. By comparing the hard decision thresholds provided by seven MTJ reference resistor combinations, it is found that the decision thresholds provided by 2(RP//RAP) reference network can achieve the average read decision error rate of the approximate maximum mutual information quantization scheme in the range of MTJ process deviation of 6%~20% and temperature of 233K~400K. Moreover, it shows good following performance under large temperature and process fluctuation range. The results show that the proposed 2(RP//RAP) hard decision threshold approximates the theoretical optimal scheme of maximum mutual information quantization under different error correction algorithms within the above temperature and process deviation range. Moreover, the decoding performance of BCH (127, 92, 5) codes is better than that of LDPC and Polar codes.