• 한국컴퓨터정보학회논문지
• /
• 제22권5호
• /
• pp.1-9
• /
• 2017

Recently, flash memory has been in a great demand from embedded system sectors for storage devices. However, program/erase (P/E) cycles per block are limited on flash memory. For the limited number of P/E cycles, many wear leveling techniques are studied. They prolonged the life time of flash memory using information tables. As one of the techniques, block erase table (BET) method using a bit array table was studied for embedded devices. However, it has a disadvantage in that performance of wear leveling is sharply low, when the consumption of memory is reduced. To solve this problem, we propose a novel wear leveling technique using Sampling-based Block Erase Table (SBET). SBET relates one bit of the bit array table to each block by using exclusive OR operation with round robin function. Accordingly, SBET enhances accuracy of cold block information and can prevent to decrease the performance of wear leveling. In our experiment, SBET prolongs life time of flash memory by up to 88%, compared with previous techniques which use a bit array table.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제15권3호
• /
• pp.837-852
• /
• 2021

Recurrent neural network (RNN) architectures have been used for language modeling (LM) tasks that require learning long-range word or character sequences. However, the RNN architecture is still suffered from unstable gradients on long-range sequences. To address the issue of long-range sequences, an attention mechanism has been used, showing state-of-the-art (SOTA) performance in all LM tasks. A differentiable neural computer (DNC) is a deep learning architecture using an attention mechanism. The DNC architecture is a neural network augmented with a content-addressable external memory. However, in the write operation, some information unrelated to the input word remains in memory. Moreover, DNCs have been found to perform poorly with low numbers of weight parameters. Therefore, we propose a robust memory deallocation method using a limited retention vector. The limited retention vector determines whether the network increases or decreases its usage of information in external memory according to a threshold. We experimentally evaluate the robustness of a DNC implementing the proposed approach according to the size of the controller and external memory on the enwik8 LM task. When we decreased the number of weight parameters by 32.47%, the proposed DNC showed a low bits-per-character (BPC) degradation of 4.30%, demonstrating the effectiveness of our approach in language modeling tasks.

• 전기의세계
• /
• 제23권1호
• /
• pp.73-78
• /
• 1974

Digital simulation algorithms and program for multimachine dynamic stability have been developed which represent the effects of machines much more complety than have been available previously. Emphasis is given to the savings of the memory spaces required, thus making it possible to use a small computer with limited capacity of core storage (without auxiliary storage). Both d- and q- aris quantities are fully represented, and the speed-governing and voltage-regulating system available are ertensive, thus allowing a very close approximation to any physical system. Facilities for dynamic and nonlinear loads are also included. The computational algorithms and program developed have been shown to be extensive and complete, and are very desirable features minimizing memory spaces for stability calculations. The capabilities have been demonstrated by several case studies for an actual power system of 44 generators, 22 loads and 33 buses. About 13-K words of memory spaces have been required for the case studies on the basis of two words per real variable and a word per integer variable.

• 대한임베디드공학회논문지
• /
• 제14권2호
• /
• pp.79-86
• /
• 2019

For decades, in-memory data structures have been designed for DRAM-based main memory that provides symmetric read/write performances and has no limited write endurance. However, such data structures provide sub-optimal performance for NVM as it has different characteristics to DRAM. With this motivation, we rethink a conventional red-black tree in terms of its efficacy under NVM settings. The original red-black tree constantly rebalances sub-trees so as to export fast access time over dataset, but it inevitably increases the write traffic, adversely affecting the performance for NVM with a long write latency and limited endurance. To resolve this problem, we present a variant of the red-black tree called a hierarchical balanced binary search tree. The proposed structure maintains multiple keys in a single node so as to amortize the rebalancing cost. The performance study reveals that the proposed hierarchical binary search tree effectively reduces the write traffic by effectively reaping the high capacity of NVM.

• Journal of information and communication convergence engineering
• /
• 제12권3호
• /
• pp.181-185
• /
• 2014

The lifetime of a solid-state drive (SSD) is limited because of the number of program and erase cycles allowed on its NAND flash blocks. Data cannot be overwritten in an SSD, leading to an out-of-place update every time the data are modified. This result in two copies of the data: the original copy and a modified copy. This phenomenon is known as write amplification and adversely affects the endurance of the memory. In this study, we address the issue of reducing wear leveling through efficient handling of write requests. This results in even wearing of all the blocks, thereby increasing the endurance period. The focus of our work is to logically divert the write requests, which are concentrated to limited blocks, to the less-worn blocks and then measure the maximum number of write requests that the memory can handle. A memory without the proposed algorithm wears out prematurely as compared to that with the algorithm. The main feature of the proposed algorithm is to delay out-of-place updates till the threshold is reached, which results in a low overhead. Further, the algorithm increases endurance by a factor of the threshold level multiplied by the number of blocks in the memory.

• 정보과학회 논문지
• /
• 제42권8호
• /
• pp.972-978
• /
• 2015

비휘발성 메모리는 바이트 단위 접근과 비휘발성을 지원한다. 이러한 특성들은 비휘발성 메모리를 캐시, 메모리, 디스크와 같은 메모리 계층 구조 가운데 하나의 영역으로 사용을 가능케 한다. 비휘발성 메모리의 흥미로운 특성은 데이터 보존 기간이 실제로는 제한적인 기간을 가지고 있다는 것이다. 게다가 데이터 보존 기간과 쓰기 지연간의 트레이드오프가 존재 한다. 본 논문에서는 이를 활용하여 비휘발성 메모리를 파일 캐시로 사용하는 새로운 관리 기법을 제안한다. 제안하는 기법은 기존의 캐시 관리 기법과는 반대로 짧은 데이터 보존 시간으로 데이터를 저장하고 쓰기 성능을 개선한다. 제안하는 기법은 LRU 대비 평균 접근 지연 시간을 최대 31%, 평균 24.4%로 감소시킴을 보인다.

• 대한임베디드공학회논문지
• /
• 제10권6호
• /
• pp.391-397
• /
• 2015

Nonvolatile memories in memory hierarchy have been investigated to reduce its energy consumption because nonvolatile memories consume zero leakage power in memory cells. One of the difficulties is, however, that the endurance of most nonvolatile memory technologies is much shorter than the conventional SRAM and DRAM technology. This has limited its usage to only the low levels of a memory hierarchy, e.g., disks, that is far from the CPU. In this paper, we study the use of a new type of nonvolatile memories - the Phase Change Memory (PCM) with a DRAM buffer system as the main memory. Our design reduced the total energy of a DRAM main memory of the same capacity by 80%. These results indicate that it is feasible to use PCM technology in place of DRAM in the main memory for better energy efficiency.

• 대한전기학회논문지:시스템및제어부문D
• /
• 제55권10호
• /
• pp.446-451
• /
• 2006

In this paper, we propose an implementation of JPEG2000 codec on the ARM9 Processor which includes independent memory management facility. The codec and memory management facility together can control the encoding and the decoding process effectively within available memory area. Embedded appliances like cellular phones have very limited internal memory which can't be expanded easily. However, they should provide various applications and services using restricted memory resources. The proposed codec with memory management can provide image quality that is identical to the original image on embedded platform. The implemented codec has no memory conflict with other applications. It shows that the proposed codec can manage memory resources efficiently.

• JSTS:Journal of Semiconductor Technology and Science
• /
• 제2권3호
• /
• pp.197-204
• /
• 2002

DRAM, SRAM, and FLASH memory are three major memory devices currently used in most electronic applications. But, they have very distinct attributes, therefore, each memory could be used only for limited applications. MRAM (Magneto-resistive Random Access Memory) is a promising candidate for a universal memory that meets all application needs with non-volatile, fast operational speed, and low power consumption. The simplest architecture of MRAM cell is a series of MTJ (Magnetic Tunnel Junction) as a data storage part and MOS transistor as a data selection part. To be a commercially competitive memory device, scalability is an important factor as well. This paper is testing the actual electrical parameters and the scaling factors to limit MRAM technology in the semiconductor based memory device by an actual integration of MRAM core cell. Electrical tuning of MOS/MTJ, and control of resistance are important factors for data sensing, and control of magnetic switching for data writing.

• 대한임베디드공학회논문지
• /
• 제13권3호
• /
• pp.151-160
• /
• 2018

NAND flash memories have the advantages of fast access speed, high density and low power consumption, thus they have increasing demand in embedded system and mobile environment. Despite the low power and fast speed gains of NAND flash memory, DRAM disk buffers were used because of the performance load and limited durability of NAND flash cell. However, DRAM disk buffers are not suitable for limited energy environments due to their high static energy consumption. In this paper, we propose WAP-LRU (Write pattern Analysis based Placement by LRU) hybrid disk buffer management policy. Our policy designates the buffer location in the hybrid memory by analyzing write pattern of the workloads to check the continuity of the page operations. In our simulation, WAP-LRU increased the lifetime of NAND flash memory by reducing the number of garbage collections by 63.1% on average. In addition, energy consumption is reduced by an average of 53.4% compared to DRAM disk buffers.