Research And Implementation Of The Embedded Memory System Based On High Performance VIM Architecture

The performance gap between processor and memory becomes more and more serious. PIM(Processor-In-Memory) technology, which merges processor and memory into a single chip, provides a new approach to solve the well known "memory wall" problem. The super-scalar processor and the VLIW processor have some performance limitation, while compared with them vector processor can exploit data-level parallelism in application and develop higher performance. With the rapid improvement of VLSI technology and semiconductor technology, the large amounts of hardware which is needed by vector processing can be integrated in a single chip. The high performance architecture VIM(Vector-ln-Memory) can take advantage of the high memory bandwidth which is brought by embedded DRAM, and make the vector processing more efficient. VIM architecture has higher performance/cost ratio than other PIMs, and it is excellent solution to "memory wall".Based on the study of PIM and vector processing technology, we have studied the VIM architecture and implemented vector processor in memory, KD-VIM-1. This thesis focus on the key technologies of embedded memory system of VIM architecture. It analyses the current embedded memory architecture over the world, and presents the embedded memory architecture .of KD-VIM-1. We have researched some crossbar models and related scheduling algorithm, and implemented the memory crossbar of KD-VIM-1 based on analysis of memory access characteristic of VIM. After have discussing the low-bit interleaving and skewed interleaving address technologies we presented address scheme of KD-VIM-1.We have implemented embedded memory system of KD-VIM-1 in VHDL. After having simulated and verified the functionality of our VHDL code, we finished synthesized into Altera Stratix FPGA chips. At last we verified embedded memory system of KD-VIM-1 based on ProcSuperStar developing board.