The abstract of this arXiv preprint argues that DRAM continues to limit the performance, energy efficiency, and robustness of modern systems. It notes that many prior works have proposed DRAM techniques, including support for in-DRAM computation and improvements to memory access latency and parallelism.
To make such techniques practical, the paper introduces Terracotta, a flexible DRAM interface and memory controller. The idea is to lower the barriers that prevent new DRAM features from being adopted in real systems.
The source is an arXiv preprint (2610.06475) with a truncated abstract, so details of the implementation and evaluation are not available from the provided text. The description is limited to the motivation and the proposed approach.