作者: Joonseop Sim , Saransh Gupta , Mohsen Imani , Yeseong Kim , Tajana Rosing
关键词:
摘要: Internet of Things (IoT) has built a network with billions connected devices which generate massive volumes data. Processing large data on existing systems requires significant costs for movements between processors and memory due to limited cache capacity bandwidth. Processing-In-Memory (PIM) is promising solution address the issue. Prior techniques that enable computation in non-volatile (NVM) are designed bipolar switching mode, suffers from high sneak current crossbar array (CBA) structure. In this paper, we propose unipolar-switching logic high-density PIM applications, called UPIM. Our design exploits mode memristor can be operated 1D1R structure hence suppresses exists prior technologies. Moreover, UPIM takes advantages 3D vertical increase utilization per unit area applications. evaluation wide range applications shows achieves up 31.3× energy saving 113.8× energy-delay product (EDP) improvement as compared recent GPGPU architecture. As state-of-the-art based our 3.1× lower consumption.