Capacity of Resistive Random-Access Memory Channel: Upper Bound and Achievable Rate Under Suboptimal Decodings

The achievable rate of code over resistive random-access memory (ReRAM) channel with finite selector failures was published in our recent work. The rate was derived under the assumption of independent and identically distributed (i.i.d.) input. In this work, focusing on the ReRAM channel with a single selector failure in the memory array, we derive an upper bound on achievable rate under arbitrary input distribution. This upper bound is within 0.02 bits from the achievable rate of i.i.d. input, indicating that i.i.d. is very close to optimal for large memory arrays. Moreover, we analyze the achievable rate of random code over ReRAM channel with suboptimal decodings where the decoder ignores the channel correlation. Our result indicates that in this case the achievable rate is limited by the capacity of a memoryless channel. We reveal both weak and strong asymptotic properties of ReRAM channel to prove this. The proof can be directly extended to the case of ReRAM with an arbitrary number of selector failures in the memory array.