• JSTS:Journal of Semiconductor Technology and Science
• /
• v.14 no.2
• /
• pp.184-188
• /
• 2014

There is a trade-off between read stability and writability under a full-/half-select condition in static random access memory (SRAM). Another trade-off in the minimum operating voltage between the read and write operation also exists. A new peripheral circuit for SRAM arrays, called a variation sensor, is demonstrated here to balance the read/write margins (i.e., to optimize the read/write trade-off) as well as to lower the minimum operation voltage for both read and write operations. A test chip is fabricated using an industrial 45-nm bulk complementary metal oxide semiconductor (CMOS) process to demonstrate the operation of the variation sensor. With the variation sensor, the word-line voltage is optimized to minimize the trade-off between read stability and writability ($V_{WL,OPT}=1.055V$) as well as to lower the minimum operating voltage for the read and write operations simultaneously ($V_{MIN,READ}=0.58V$, $V_{MIN,WRITE}=0.82V$ for supply voltage $(V_{DD})=1.1V$).

• Journal of the Korea Society of Computer and Information
• /
• v.25 no.3
• /
• pp.1-9
• /
• 2020

In this paper, we propose an efficient adaptive write scheme that improves the performance of write operation in MLC STT-MRAM caches. The key idea of the proposed scheme is to perform the write operation fast if the target MLC STT-MRAM cells contain a dead block. Even if the fast write operation on the MLC STT-MRAM evicts a cache block from the MLC STT-MRAM cells, its performance impact is low if the evicted block is a dead block which is not used in the future. Through experimental evaluation with a memory simulator, we show that the proposed adaptive write scheme improves the performance of the MLC STT-MRAM caches by 17% on average.

• JSTS:Journal of Semiconductor Technology and Science
• /
• v.5 no.4
• /
• pp.262-269
• /
• 2005

Semiconductor scientists and engineers ideally desire the faster but the cheaper non-volatile memory devices. In practice, no single device satisfies this desire because a faster device is expensive and a cheaper is slow. Therefore, in this paper, we use heterogeneous non-volatile memories and construct an efficient hierarchy for them. First, a small RAM device (e.g., MRAM, FRAM, and PRAM) is used as a write buffer of flash memory devices. Since the buffer is faster and does not have an erase operation, write can be done quickly in the buffer, making the write latency short. Also, if a write is requested to a data stored in the buffer, the write is directly processed in the buffer, reducing one write operation to flash storages. Second, we use many types of flash memories (e.g., SLC and MLC flash memories) in order to reduce the overall storage cost. Specifically, write requests are classified into two types, hot and cold, where hot data is vulnerable to be modified in the near future. Only hot data is stored in the faster SLC flash, while the cold is kept in slower MLC flash or NOR flash. The evaluation results show that the proposed hierarchy is effective at improving the access time of flash memory storages in a cost-effective manner thanks to the locality in memory accesses.

• Journal of the Korea Society of Computer and Information
• /
• v.24 no.6
• /
• pp.1-9
• /
• 2019

Hard disks, which have long been used as secondary storage in computing systems, are increasingly being replaced by solid state drives (SSDs), due to their relatively fast data input / output speeds and small, light weight. SSDs that use NAND flash memory as a storage medium are significantly different from hard disks in terms of physical operation and internal operation. In particular, there is a feature that data overwrite can not be performed, which causes erase operation before writing. In order to solve this problem, a hot data for frequently updating a data for a specific page is distinguished from a cold data for a relatively non-hot data. Hot data identification helps to improve overall performance by identifying and managing hot data separately. Among the various hot data identification methods known so far, there is a technique of recording consecutive write requests by using a Bloom filter and judging the values by hot data. However, the Bloom filter technique has a problem that a new bit array must be generated every time a set of items is changed. In addition, since it is judged based on a continuous write request, it is possible to make a wrong judgment. In this paper, we propose a method using a counting filter for accurate hot data verification. The proposed method examines consecutive write requests. It also records the number of times consecutive write requests occur. The proposed method enables more accurate hot data verification.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.13 no.5
• /
• pp.269-277
• /
• 2018

NAND flash memory has widely been used because of non-volatility, low power consumption and fast access time. However, it suffers from inability to provide update-in-place and the erase cycle is limited. The unit of read/write operation is a page and the unit of erase operation is a block. Moreover erase operation is slower than other operations. We proposed the Adaptive Garbage Collection (called "AGC") policy which focuses on not only reducing garbage collection process time for real-time guarantee but also wear-leveling for a flash memory lifetime. The AGC performs better than Cost-benefit policy and Greedy policy. But the AGC does not consider the operation characteristics. So we proposed the Advanced Adaptive Garbage Collection (called "A-AGC") policy which considers the page write operation count and block erase operation count. The A-AGC reduces the write operations by considering the data update frequency and update data size. Also, it reduces the erase operations by considering the file fragmentation. We implemented the A-AGC policy and measured the performance compared with the AGC policy. Simulation results show that the A-AGC policy performs better than AGC, specially for append operation.

• Transactions of the Society of Information Storage Systems
• /
• v.7 no.1
• /
• pp.7-12
• /
• 2011

The flash memory has been remarked as the next generation media of portable and desktop computers' storage devices. Their features include non-volatility, low power consumption, and fast access time for read operations, which are sufficient to present flash memories as major data storage components for desktop and servers. The purpose of our study is to upgrade a traditional mirroring scheme based on SSD storages due to the relatively slow or freezing characteristics of write operations, as compared to fast read operations. For this work, we propose a new storage management scheme called Memory Mirror-Switching based on traditional mirroring scheme. Our Mirror-Switching scheme improves flash operation performance by switching write-workloads from flash memory to RAM and delaying write operations to avoid freezing. Our test results show that our scheme significantly reduces the write operation delay and storage freezing.

• International journal of advanced smart convergence
• /
• v.12 no.4
• /
• pp.164-170
• /
• 2023

With the recent advances in AI (artificial intelligence) and HPC (high-performance computing) technologies, deep learning is proliferated in various domains of the 4th industrial revolution. As the workload volume of deep learning increasingly grows, analyzing the memory reference characteristics becomes important. In this article, we analyze the memory reference traces of deep learning workloads in comparison with traditional workloads specially focusing on read and write operations. Based on our analysis, we observe some unique characteristics of deep learning memory references that are quite different from traditional workloads. First, when comparing instruction and data references, instruction reference accounts for a little portion in deep learning workloads. Second, when comparing read and write, write reference accounts for a majority of memory references, which is also different from traditional workloads. Third, although write references are dominant, it exhibits low reference skewness compared to traditional workloads. Specifically, the skew factor of write references is small compared to traditional workloads. We expect that the analysis performed in this article will be helpful in efficiently designing memory management systems for deep learning workloads.

• International Journal of Internet, Broadcasting and Communication
• /
• v.10 no.4
• /
• pp.6-11
• /
• 2018

As embedded memory technology evolves, the traditional Static Random Access Memory (SRAM) technology has reached the end of development. For deepening the manufacturing process technology, the next generation memory technology is highly required because of the exponentially increasing leakage current of SRAM. Non-volatile memories such as STT-MRAM (Spin Torque Transfer Magnetic Random Access Memory), PCM (Phase Change Memory) are good candidates for replacing SRAM technology in embedded memory systems. They have many advanced characteristics in the perspective of power consumption, leakage power, size (density) and latency. Nonetheless, nonvolatile memories have two major problems that hinder their use it the next-generation memory. First, the lifetime of the nonvolatile memory cell is limited by the number of write operations. Next, the write operation consumes more latency and power than the same size of the read operation.These disadvantages can be solved using the compiler. The disadvantage of non-volatile memory is in write operations. Therefore, when the compiler decides the layout of the data, it is solved by optimizing the write operation to allocate a lot of data to the SRAM. This study provides insights into how these compiler and architectural designs can be developed.

• Proceedings of the IEEK Conference
• /
• 2002.07b
• /
• pp.932-935
• /
• 2002

In this paper, we introduce new FTL(Flash Translation Layer) driver algorithm that tolerate the power off errors. FTL driver is the software that provide the block device interface to the upper layer software such as file systems or application programs that using the flash memory as a block device interfaced storage. Usually, the flash memory is used as the storage devices of the mobile system due to its low power consumption and small form factor. In mobile system, the state of the power supplement is not stable, because it using the small sized battery that has limited capacity. So, a sudden power off failure can be occurred when we read or write the data on the flash memory. During the write operation, power off failure may introduce the incomplete write operation. Incomplete write operation denotes the inconsistency of the data in flash memory. To provide the stable storage facility with flash memory in mobile system, FTL should provide the fault tolerance against the power off failure. SSR (Simple Sector Remapper) is a fault tolerant FTL driver that provides block device interface and also provides tolerance against power off errors.

• Journal of IKEEE
• /
• v.16 no.3
• /
• pp.265-273
• /
• 2012

Data stability and leakage power dissipation have become a critical issue in scaled SRAM design. In this paper, an advanced 8T SRAM cell improving the read and write stability of data storage elements as well as reducing the leakage current in the idle mode is presented. During the read operation, the bit-cell keeps the noise-vulnerable data 'low' node voltage close to the ground level, and thus producing near-ideal voltage transfer characteristics essential for robust read functionality. In the write operation, a negative bias on the cell facilitates to change the contents of the bit. Unlike the conventional 6T cell, there is no conflicting read and write requirement on sizing the transistors. In the standby mode, the built-in stacked device in the 8T cell reduces the leakage current significantly. The 8T SRAM cell implemented in a 130 nm CMOS technology demonstrates almost 100 % higher read stability while bearing 20 % better write-ability at 1.2 V typical condition, and a reduction by 45 % in leakage power consumption compared to the standard 6T cell. The stability enhancement and leakage power reduction provided with the proposed bit-cell are confirmed under process, voltage and temperature variations.