**5.2 Custom macro cells for C3S designs (ongoing work)**

In our follow-on work, we plan to examine decomposition of the extended C3S microarchitecture model into its fundamental building blocks, and implement them as custom macro cells. These new macro cells can be added to our earlier *TNN7* library to create a new extended library with custom macros for supporting C3S designs. We envision distinct specialized macros for "storage", "predict", "update" and "voting" functionalities that are generic enough to build *Where*, *What* and *Output* minicolumns at arbitrary scales and modalities. A key goal here is to have a rich set of macros that support diverse sensory modalities to enable implementation of potentially new types of mini-columns in the future. Currently, content-addressable memories (CAMs) seem to be a promising candidate as building blocks for implementing mini-columns for Reference Frames. In order to effectively predict relevant information based on stored knowledge in RFs, we believe these CAMs would need to support fuzzy matching; not strict exact matching as done conventionally. Implementing fuzziness in an RF can help recover similar features that can aid rapid learning of new patterns and features. This is largely ongoing and future work.
