混合式的記憶體系統（Hybrid Memory System）是一個新穎的研究，其原理 是是藉由將不同的記憶體混合應用，並互取各個記憶體系統的優點，以達到最佳 化目標的記憶體系統。目前的一般記憶體系統架構，通常以傳統的DRAM （Dynamic Random Access Memory）做為主記憶體並儲存大量資料。雖然DRAM 的存取速度以及耗能相較於其他記憶體是較平均且較好，但DRAM記憶體是屬於 揮發性的記憶體（Volatile memory），除了存取動作之外，每過一段時間必須進行 資料的維護保存，如此便造成更多額外的功率消耗。因此藉由整合不同的記憶體 晶片，可以降低記憶體的整體功率消耗，有效改善傳統架構的耗能問題。然而對 於如何整合不同的記憶體晶片上工作頁的配置問題，便成為了混合式記憶體架構 下一個關鍵的因素。本論文中，我們將提出在混合式記憶體架構下進行Task Division與Page Allocation演算法。在整合不同記憶體晶片架構上，並同時考慮到 記憶體晶片間不同的存取耗能、存取延遲等特性，我們可以由實驗結果說明，我 們的方法可以有效的提升混合式記憶體的表現。

Due to the technology limit on DRAM and SRAM, non-volatile memory (NVM) such as PCRAM and STT-RAM attracts more and more attention in recent years. Non-volatile memory has the advantages of non-volatility, circuit density and ignorable leakage power. On the other hand, it suffers from the bad write latency and poor write endurance problems. Therefore, a hybrid memory consists of traditional memory and non-volatile memory becomes an important direction of future memory architecture design. In this paper, we allocate data to memory by dividing tasks into page segments, and propose a mixed ILP formulation for pages allocation on the hybrid DRAM/PCRAM memory architecture. Because a divided page segment has more allocation flexibility than a single task, we can further taking the advantages of good leakage consumption of PCRAM and short write latency of DRAM. Experimental results show that our page-based approach performs better than the task-based approach for all objective functions, including optimization of latency, power consumption, and product of both terms.