iMARS：一种存内计算结构的推荐系统
iMARS: An In-Memory-Computing Architecture for Recommendation Systems
摘要：
推荐系统（RecSys）通过基于历史数据预测用户的偏好来向用户推荐项目。典型的RecSys处理大型嵌入表和许多嵌入表相关操作。传统计算机体系结构的内存大小和带宽限制了RecSys的性能。这项工作提出了一种存内计算（IMC）架构（iMARS），用于加速基于深度神经网络的RecSys的过滤和排名阶段。iMARS利用在基于铁电FET的IMC结构内实现的IMC友好嵌入表。电路级和系统级评估表明，与MovieLens数据集的GPU对应物相比，\fw实现了16.8倍（713倍）的端到端延迟（能量）改进。
英文摘要：
Recommendation systems (RecSys) suggest items to users by predicting their preferences based on historical data. Typical RecSys handle large embedding tables and many embedding table related operations. The memory size and bandwidth of the conventional computer architecture restrict the performance of RecSys. This work proposes an in-memory-computing (IMC) architecture (iMARS) for accelerating the filtering and ranking stages of deep neural network-based RecSys. iMARS leverages IMC-friendly embedding tables implemented inside a ferroelectric FET based IMC fabric. Circuit-level and system-level evaluation show that \fw achieves 16.8x (713x) end-to-end latency (energy) improvement compared to the GPU counterpart for the MovieLens dataset.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
