An Intelligent Memory for Data-Parallel Applications
Kang Yi · 1999
Major advances in Merged Logic DRAM (MLD) technology coupled with the popularization of memory-intensive applications provide fertile ground for architectures based on Intelligent Memory (IRAM). The contribution of this research is to explore one way to use the current state-of-the-art MLD technology for general-purpose computers. A novel IRAM architecture named FlexRAM is proposed here. To satisfy requirements of general-purpose and low programming cost, we embed the FlexRAM chips in the memory system and let them default to plain DRAM if the application is not enabled for intelligent memory. Inside a FlexRAM, chip we build two levels of processors. These embedded processors in FlexRAM can perform parallel processing in addition to having the advantage of a lower memory access latency and a higher memory bandwidth. Many research issues related to FlexRAM are discussed. We describe the FlexRAM architecture including the memory architecture, the processor architecture and the communication models. The rationale for our architecture design is based on general principles and constraints from the technology and existing computer systems. We address design issues arising from the FlexRAM logic and physical design. We show that it is feasible to implement a FlexRAM chip with current MLD technology. An Evaluations based on high-level simulation is present. Decent speedup numbers are shown that the FlexRAM architecture is effective for many data-parallel applications. How to program FlexRAM is also explored, and a new programming model for IRAM - Intelligent Memory Operation is suggested. We identify and analyze a range of real applications for FlexRAM, including applications in the domain of data mining, computational biology, decision support and multimedia.