Abstract
A novel WIDE I/O DRAM architecture called 3-D WiRED is proposed, with an enhanced DRAM core to enable low latency and energy-efficient memory access. Through detailed time-energy analysis of a WIDE I/O DRAM prototype, we have identified the need to reduce the capacitance of bitlines (BLs), memory bus (M_bus), and global data path to reduce random access latency, read/write energy, and activation-precharge (ActPre) energy. Trace-driven simulation analysis was performed to compare 3-DWiRED with other state-of-the-art DRAM architectures. Memory access traces for the PARSEC benchmark suite were extracted from detailed cycle-accurate simulations using gem5. rank-based round-robin scheduling scheme and a closed page policy were used for all simulations. Energy-per-bit, average-latency, and energy-delay product values for the memory subsystem were obtained from DRAMSim2. The results indicate that the performance and energy efficiency of contemporary wide I/O DRAM core can be greatly improved by aggressively using TSVs at subarray-level granularity. However, several challenges still need to be overcome to support such a fine-grained 3-D integration.
Original language | English |
---|---|
Article number | 7116520 |
Pages (from-to) | 71-80 |
Number of pages | 10 |
Journal | IEEE Design and Test |
Volume | 32 |
Issue number | 4 |
DOIs | |
State | Published - 2015 |
ASJC Scopus subject areas
- Software
- Hardware and Architecture
- Electrical and Electronic Engineering