Differentiable Content Addressable Memory with Memristors

Memristors, Flash, and related nonvolatile analog device technologies offer in-memory computing structures operating in the domain, such as accelerating linear matrix operations array structures. These take advantage of tunability large dynamic range. At other side, content addressable memories (CAM) are fast digital lookup tables which effectively perform nonlinear Boolean logic return a match/mismatch value. Recently, CAMs have been presented merging storage search with output. However, CAM blocks cannot easily be inserted within larger adaptive system due to challenges training learning binary outputs. Here, missing link between crossbar arrays CAMs, namely differentiable memory (dCAM), is presented. Utilizing that act “soft” outputs, dCAM enables fine-tuning operation performance. Four applications quantitatively evaluated highlight capabilities: improved data pattern storage, robustness noise variability, reduced energy latency performance, an application solving satisfiability optimization problems. The use envisioned core building block fully systems employing multiple types compute memories.