RED: Energy Optimization Framework for eDRAM-based PIM with Reconfigurable Voltage Swing and Retention-aware Scheduling

TL;DR

RED is an energy optimization framework for eDRAM-based processing-in-memory that uses reconfigurable voltage swing and retention-aware scheduling to significantly reduce energy consumption and improve efficiency in AI applications.

Contribution

The paper introduces a novel reconfigurable eDRAM design and scheduling method that optimizes energy use in PIM, addressing memory power consumption which was previously overlooked.

Findings

Achieves up to 3.05x higher energy efficiency than SRAM-based PIM.

Reduces eDRAM macro energy consumption by up to 74.88%.

Requires only 3.5% area and 0.77% energy overhead for scheduling.

Abstract

In the era of artificial intelligence (AI), Transformer demonstrates its performance across various applications. The excessive amount of parameters incurs high latency and energy overhead when processed in the von Neumann architecture. Processing-in-memory (PIM) has shown the potential in accelerating data-intensive applications by reducing data movement. While previous works mainly optimize the computational part of PIM to enhance energy efficiency, the importance of memory design, which consumes the most power in PIM, has been rather neglected. In this work, we present RED, an energy optimization framework for eDRAM-based PIM. We first analyze the PIM operations in eDRAM, obtaining two key observations: 1) memory access energy consumption is predominant in PIM, and 2) read bitline (RBL) voltage swing, sense amplifier power, and retention time are in trade-off relations. Leveraging them, we propose a novel reconfigurable eDRAM and retention-aware scheduling that minimizes the runtime energy consumption of the eDRAM macro. The framework pinpoints the optimal operating point by pre-estimating energy consumption across all possible tiling schemes and memory operations. Then, the reconfigurable eDRAM controls the RBL voltage swing at runtime according to the scheduling, optimizing the memory access power. Moreover, RED employs refresh skipping and sense amplifier power gating to mitigate the energy consumption overhead coming from the trade-off relation. Finally, the RED framework achieves up to 3.05x higher energy efficiency than the prior SRAM-based PIM, reducing the energy consumption of eDRAM macro up to 74.88% with reconfigurable eDRAM and optimization schemes, requiring only 3.5% area and 0.77% energy overhead for scheduling.