In-memory hamming similarity computation in resistive arrays

This paper develops a framework to calculate Hamming similarity between vectors stored in resistive memory. A single-parameter model is proposed for the resistive measurement channel, which is then used to analytically reveal an interesting tradeoff between this parameter and succeeding in the calculation task. We suggest coding techniques that can improve this tradeoff under natural usage assumptions. The proposed constructions work to change the Hamming weight of the stored vectors without corrupting the Hamming distance between pairs of vectors.