Modeling and Analysis of a Membrane-Based Randomized-Contact Decoder

Decoders are employed to address individual NWs in crossbars (xbars) used for storage and computation. We report on the simulation and analysis of a membranebased nanowire (NW) decoder proposed by Pribat and Savage [1] that implements the randomized-contact decoder (RCD) introduced by Williams and Kuekes [2] and analyzed by Hogg et al [3] and Rachlin and Savage [4]. An RCD has two sets of parallel wires, NWs and mesoscale wires (MWs), that are orthogonal to one another. “Contacts” are made at random at NW/MW junctions. These contacts, which allow a MW to control a NW, assign codewords to NWs, some of which are unusable. In the membrane-based decoder a contact is made by filling a pore with a metallic pin. As the probability of filling pores increases, the probability of creating a MW/NW contact increases monotonically. However, the average number of usable NW addresses achieves a maximum when a small fraction of the pores is filled. The goal is to maximize the average number of usuable NW addresses. We report on the results of our simulations of the membrane-based RCD as well as corroborate the experimental results with analysis of probabilistic models.