ConciseSignal
Following

Researchers propose row-parallel computing inside DRAM

A joint academic team from Syracuse University, FAU, and TU Dresden has introduced a DRAM design called RAPID that performs arithmetic operations directly within memory rows. Unlike existing methods requiring costly data rearrangement for compatibility with processors, RAPID enables computations in the same row-parallel format used by CPUs. It does this by adding migration and inversion cells to DRAM subarrays, reducing data reorganization overhead.

Why it mattersDRAM that can compute in a CPU-compatible format could greatly speed up data processing and reduce power use in memory-intensive tasks. Integrating computation into memory hardware may drive advances in AI and high-performance computing.

Sources covering this

Semiconductor EngineeringRow-Parallel DRAM Computing Cuts Data-Reorganization Overhead (Syracuse, FAU, TU Dresden)4:16 PM →
Concise Signal DailyEnterprise AI, security & business tech.Weekdays, 7am Eastern · Sample issue

More in Chips