Search

With the development of computer architecture, even for embedded systems, GPU devices can be integrated, providing outstanding performance and energy efficiency to meet the requirements of different industries, applications, and deployment environments. Data analytics is an important application scenario for embedded systems. Unfortunately, due to the limitation of the capacity of the embedded device, the scale of problems handled by the embedded system is limited. In this paper, we propose a novel data analytics method, called G-TADOC, for efficient text analytics directly on compression on embedded GPU systems. A large amount of data can be compressed and stored in embedded systems, and can be processed directly in the compressed state, which greatly enhances the processing capabilities of the systems. Particularly, G-TADOC has three innovations. First, a novel fine-grained thread-level workload scheduling strategy for GPU threads has been developed, which partitions heavily-dependent loads adaptively in a fine-grained manner. Second, a GPU thread-safe memory pool has been developed to handle inconsistency with low synchronization overheads. Third, a sequence-support strategy is provided to maintain high GPU parallelism while ensuring sequence information for lossless compression. Moreover, G-TADOC involves special optimizations for embedded GPUs, such as utilizing the CPU-GPU shared unified memory. Experiments show that G-TADOC provides 13.2× average speedup compared to the state-of-the-art TADOC. G-TADOC also improves performance-per-cost by 2.6× and energy efficiency by 32.5× over TADOC.

According to the manufacturer’s I-V data, we need to obtain the best parameters for assessing the photovoltaic systems. Although much work has been done in this area, it is still challenging to extract model parameters accurately. An efficient solver called SFLBS is developed to deal with this problem, in which an inheritance mechanism based on crossover and mutation is introduced. Specifically, the memory pool for storing historical population information is designed. During the sub-population evolution, the historical population will cross and mutate with the contemporary population with a certain probability, ultimately inheriting information about the dimensions that perform well. This mechanism ensures the population’s quality during the evolution process and effectively improves the local search ability of traditional SFLA. The proposed SFLBS is applied to extract unknown parameters from the single diode model, double diode model, three diode model, and photovoltaic module model. Based on the experimental results, we found that SFLBS has considerable accuracy in extracting the unknown parameters of the PV system problem, and its convergence speed is satisfactory. Moreover, SFLBS is used to evaluate three commercial PV modules under different irradiance and temperature conditions. The experimental results demonstrate that the performance of SFLBS is outstanding compared to some state-of-the-art competing algorithms. Moreover, SFLBS is still a reliable optimization tool despite the complex external environment. This research is supported by an online service for any question or needs to supplementary materials at https://aliasgharheidari.com .

Chimeric Antigen Receptor (CAR)-T cell immunotherapy revolutionized cancer treatment and consists of the genetic modification of T lymphocytes with a CAR gene, aiming to increase their ability to recognize and kill antigen-specific tumor cells. The dynamics of CAR-T cell responses in patients presents a multiphasic kinetics with distribution, expansion, contraction, and persistence phases. The characteristics and duration of each phase depend on the tumor type, the infused product, and on patient-specific characteristics. We present a mathematical model which describes the multiphasic CAR-T cell dynamics resulting from the interplay between CAR-T and tumor cells, considering patient and product heterogeneities. The CAR-T cell population is divided into functional (distributed and effector), memory, and exhausted CAR-T cell phenotypes. The model is able to describe the diversity of CAR-T cell dynamic behaviors in different patients and hematological cancers as well as their therapy outcomes. Our results indicate that the joint assessment of the area under the concentration-time curve in the first 28 days and the corresponding fraction of non-exhausted CAR-T cells may be considered as potential markers to classify therapy responses. Overall, the analysis of different CAR-T cell phenotypes can be a key aspect for a better understanding of the whole CAR-T cell dynamics.

Disaggregated memory leverages recent technology advances in high-density, byte-addressable non-volatile memory and high-performance interconnects to provide a large memory pool shared across multiple compute nodes. Due to higher memory density, memory errors may become more frequent. Unfortunately, tolerating memory errors through existing memory-error protection techniques becomes impractical due to increasing storage cost. This letter proposes replication-aware memory-error protection to improve storage efficiency of protection in data-centric applications that already rely on memory replication for performance and availability. It lets such applications lower protection storage cost by weakening the protection of each individual replica, but still realize a strong protection target by relying on the collective protection conferred by multiple replicas.

Orphan transactions are those whose parental income sources are missing at the time that they are processed. These transactions typically languish in a local buffer until they are evicted or all their parents are discovered, at which point they may be propagated further. To date, there has been little work in the literature on characterizing the nature and impact of such orphans, and yet it is intuitive that they should affect the performance of the Bitcoin network. This work thus seeks to methodically research such effects through a measurement campaign on live Bitcoin nodes. Our data show that about 45% of orphan transactions end up being included in the blockchain. Surprisingly, orphan transactions tend to have fewer parents on average than non-orphan transactions, and their missing parents have a lower fee, larger size, and lower transaction fee per byte than all other received transactions. Moreover, the network overhead incurred by these orphan transactions can be significant, exceeding 17% when using the default orphan memory pool size (i.e., 100 transactions), although this overhead can be made negligible, without significant computational or memory demands, if the pool size is simply increased to 1000 transactions. Finally, we show that when a node with an empty mempool first joins the network, 25% of the transactions that it receives become orphan, whereas in steady-state this quantity drops to about 1%.

We aim to tackle existing problems about deep learning serving on GPUs in the view of the system. GPUs have been widely adopted to serve online deep learning-based services that have stringent QoS(Quality-of-Service) requirements. However, emerging deep learning serving systems often result in poor responsiveness and low throughput of the inferences that damage user experience and increase the number of GPUs required to host an online service. Our investigation shows that the poor batching operation and the lack of data transfer-computation overlap are the root causes of the poor responsiveness and low throughput. To this end, we propose E $^2$ 2 bird, a deep learning serving system that is comprised of a GPU-resident memory pool, a multi-granularity inference engine, and an elastic batch scheduler. The memory pool eliminates the unnecessary waiting of the batching operation and enables data transfer-computation overlap. The inference engine enables concurrent execution of different batches, improving the GPU resource utilization. The batch scheduler organizes inferences elastically to guarantee the QoS. Our experimental results on an Nvidia Titan RTX GPU show that E $^2$ 2 bird reduces the response latency of inferences by up to 82.4 percent and improves the throughput by up to 62.8 percent while guaranteeing the QoS target compared with TensorFlow Serving.

We study the distribution of confirmation times of Bitcoin transactions, conditional on the size of the current memory pool. We argue that the time until a Bitcoin transaction is confirmed resembles the time to ruin in a corresponding Cramer-Lundberg process. This well-studied model gives mathematical insights in the mempool behaviour over time. Specifically, for situations where one chooses a fee, such that the total size of incoming transactions with higher fee is close to the total size of transactions leaving the mempool (heavy traffic), a diffusion approximation leads to an inverse Gaussian distribution for the confirmation times. The results of this paper are particularly interesting for users that want to make a Bitcoin transaction during heavy-traffic situations, as evaluation of the well-known inverse Gaussian distribution is computationally straightforward.

In this paper Percheron optimization algorithm (POA) has been designed for voltage stability enhancement and power loss reduction. Percherons are used in agriculture segment and carrying of weighty goods. In each group a stallion will be there and it takes control of the group activities. Based on the Pecking-order food and others things are accessed. Ranking of the Percheron has been done with reference to the fitness value. Search space has been modeled by the arrays of zeros and ones. Memory updating of the Percheron is done by Percheron Memory Pool (PMP). In the process some Percheron will be left out if it has poor fitness value. By means of considering L (voltage stability)—index POA verified in IEEE 30- bus system. Then without L-index POA appraised in 30 bus test systems. POA condensed the power loss proficiently with augmentation of voltage stability and minimization of voltage deviation.

Scientific applications often require high-bandwidth shared storage to perform joint simulations and collaborative data analytics. Shared memory pools provide a chance to satisfy such needs. Recently, a high-speed network such as Gen-Z utilizing persistent memory (PM) offers an opportunity to create a shared memory pool connected to compute nodes. However, there are several challenges to use scientific applications on the shared memory pool directly such as scalability, failure-atomicity, and lack of scientific metadata-based search and query. In this paper, we propose MOSIQS, a persistent memory object storage framework with metadata indexing and querying for scientific computing. We design MOSIQS based on the key idea that memory objects on PM pool can live beyond the application lifetime and can become the sharing currency for applications and scientists. MOSIQS provides an aggregate memory pool atop an array of persistent memory devices to store and access memory objects to accelerate scientific computing. MOSIQS uses a lightweight persistent memory key-value store to manage the metadata of memory objects, which enables memory object sharing. To facilitate metadata search and query over millions of memory objects resident on memory pool, we introduce Group Split and Merge (GSM), a novel persistent index data structure designed primarily for scientific datasets. GSM splits and merges dynamically to minimize the query search space and maintains low query processing time while overcoming the index storage overhead. MOSIQS is implemented on top of PMDK. We evaluate the proposed approach on many-core server with an array of real PM devices. Experimental results show that MOSIQS gains a 100% write performance improvement and executes multi-attribute queries efficiently with $2.7\times $ less index storage overhead offering significant potential to speed up scientific computing applications.

Once exotic, computational accelerators are now commonly available in many computing systems. Graphics processing units (GPUs) are perhaps the most frequently encountered computational accelerators. Recent work has shown that GPUs are beneficial when analyzing massive data sets. Specifically related to this study, it has been demonstrated that GPUs can significantly reduce the query processing time of database bitmap index queries. Bitmap indices are typically used for large, read-only data sets and are often compressed using some form of hybrid run-length compression. In this paper, we present three GPU algorithm enhancement strategies for executing queries of bitmap indices compressed using word aligned hybrid compression: (1) data structure reuse (2) metadata creation with various type alignment and (3) a preallocated memory pool. The data structure reuse greatly reduces the number of costly memory system calls. The use of metadata exploits the immutable nature of bitmaps to pre-calculate and store necessary intermediate processing results. This metadata reduces the number of required query-time processing steps. Preallocating a memory pool can reduce or entirely remove the overhead of memory operations during query processing. Our empirical study showed that performing a combination of these strategies can achieve 32.4$$\times$$ × to 98.7$$\times$$ × speedup over the current state-of-the-art implementation. Our study also showed that by using our enhancements, a common gaming GPU can achieve a $$15.0\times$$ 15.0 × speedup over a more expensive high-end CPU.

Deep-neural-networks-based online visual tracking methods have achieved state-of-the-art results. One of the core components of these methods is the memory pool, in which a number of samples consisting of image patches and the corresponding labels are stored to update the online tracking network. Hence, the mechanism of updating the stored samples determines the performance of the tracking method. In this paper, a novel memory mechanism is proposed to control the writing and reading accesses of the memory pool using credit assignment network ${H}$ , which learns features of the target object. This memory mechanism comprises the writing and reading mechanisms. In the writing mechanism, network ${H}$ produces credits for the current tracked object and the samples in the memory pool. This ensures that the reliable samples are written into the memory pool and the unreliable samples are replaced if the memory pool is full. In the reading mechanism, network ${H}$ assigns an importance score to each sample selected to update the online tracking network. The state-of-the-art tracking methods with and without the proposed memory mechanism are evaluated on the CVPR2013 and OTB100 benchmarks. The experimental results demonstrated that the proposed memory mechanism improves tracking performance significantly.

Autonomous navigation systems consist of multiple software modules, such as sensing, object detection, and planning, to achieve traffic perception and fast decision making. Such a system generates a large amount of data and requires data processing and communication in real-time. Although accelerators, such as GPUs, have been exploited to speed up data processing, communicating GPU messages between modules is still lacking support, leading to high communication latency and resource contention. For such a latency-sensitive and resource-limited autonomous navigation system, high performance and lightweight message communication are crucial and demanding. To obtain both high performance and low resource usages, we first propose a novel pub-centric memory pool and an on-the-fly offset conversion algorithm to avoid unnecessary data movement. Secondly, we combine these two techniques and propose an efficient message communication on a single GPU. Finally, we extend this approach to multi-GPU and design a framework that natively supports GPU message communication for Inter-Process Communication. With comprehensive evaluation, results show our approach is able to reduce communication latency by 53.7% for PointCloud and Image messages compared to the state-of-the-art approach. Moreover, in the real autonomous navigation scenario, our approach reduces the end-to-end latency by 29.2% and decreases resource usage up to 58.9%.

As the world is progressing faster, to compete with the demand, the need for proficient computing technology has increased, resulting in huge volumes of data. Consequently, the extraction of relevant information from such a massive volume of data in a short time becomes challenging. Hence, automatic text summarization (TS) has emerged as an efficient solution to this problem. In the current study, the automatic TS problem is formulated as a multi-objective optimization problem, and to mitigate this problem, the modified cat swarm optimization (MCSO) strategy is employed. In this work, the population is represented as a collection of feasible individuals where the summary length limit is considered as a constraint that determines the feasibility of an individual. Here, each individual is shaped by randomly selecting some of the sentences encoded in the binary form. Furthermore, two objective functions, namely “coverage and informativeness” and “anti-redundancy,” are used to evaluate each individual’s fitness. Also, to update the position of an individual, genetic and bit manipulating operators and the best cat memory pool have been incorporated into the system. Finally, from the generated non-dominated optimal solutions, the best solution is selected based on the ROUGE score for the summary generation process. The system’s performance is evaluated using ROUGE-1 and ROUGE-2 measures on two standard summarization datasets, namely DUC-2001 and DUC-2002, which revealed that the proposed approach achieved a noticeable improvement in ROUGE scores compared to many state-of-the-art methods mentioned in this paper. The system is also evaluated using the generational distance, CPU processing time, and cohesion, reflecting that the obtained summaries are readable, concise, and relevant being fast converging.

Byte-addressable Non-volatile Memory (NVM) technologies promise higher density and lower cost than DRAM. They have been increasingly employed for data center applications. Despite many previous studies on using NVM in a single machine, there remain challenges to best utilize it in a distributed data center environment. This paper presents Gengar, an RDMA-enabled Distributed Shared Hybrid Memory (DSHM) pool with simple programming APIs on viewing remote NVM and DRAM in a global memory space. We propose to exploit semantics of RDMA primitives to identify frequently-accessed data in the hybrid memory pool, and cache it in distributed DRAM buffers. We redesign RDMA communication protocols to reduce the bottleneck of RDMA write latency by leveraging a proxy mechanism. Gengar also supports memory sharing among multiple users with data consistency guarantee. We evaluate Gengar in a real testbed equipped with Intel Optane DC Persistent DIMMs. Experimental results show that Gengar significantly improves the performance of public benchmarks such as MapReduce and YCSB by up to 70 % compared with state-of-the-art DSHM systems.

In the memory forensics, the Pool Tag Scanning based on the memory pool tag requires a detailed search of the physical memory when scanning the kernel driver object, which is very inefficient. The object scanning of Windows kernel driver by using the pool tag quick scanning is proposed. The method uses the quick pool tag scanning to reduce the memory range of the scan, and then scan the driver object according to the characteristics of the kernel driver object quickly, to help investigator to determine whether the driver is normal. Experimental results shows that the scanning efficiency for object scanning of kernel driver is improved greatly by using the quick pool tag scanning technology and the time spent in the scanning step is reduced while ensuring the false alarm rate is same.

One of the most commonly known vulnerabilities that can affect a binary executable is the stack-based buffer overflow. The buffer overflow occurs when a program, while writing data to a buffer, overruns the buffer’s boundary and overwrites adjacent memory locations. Nowadays, due to multiple protection mechanisms enforced by the operating system and on the executable level, the buffer overflow has become harder to exploit. Multiple bypassing techniques are often required to be used in order to successfully exploit the vulnerability and control the execution flow of the studied executable. One of the security features designed as protection mechanisms is Data Execution Prevention (DEP) which helps prevent code execution from the stack, heap or memory pool pages by marking all memory locations in a process as non-executable unless the location explicitly contains executable code. Another protection mechanism targeted is the Address Space Layout Randomization (ASLR), which is often used in conjunction with DEP. This security feature randomizes the location where the system executables are loaded into memory. By default, modern day operating systems have these security features implemented. However, on the executable level, they have to be explicitly enabled. Most of the protection mechanisms, like the ones mentioned above, require certain techniques in order to bypass them and many of these techniques are using some form of address memory leakage in order to leverage an exploit. By successfully exploiting a buffer overflow, the adversary can potentially obtain code execution on the affected operating system which runs the vulnerable executable. The level of privilege granted to the adversary is highly depended on the level of privilege that the binary is executed with. As such, an adversary may gain elevated privileges inside the system. Most of the times, this type of vulnerability is used for privilege escalation attacks or for gaining remote code execution on the system.

With the rapid development of deep convolutional neural networks, more and more computer vision tasks have been well resolved. These convolutional neural network solutions rely heavily on the performance of the hardware. However, due to privacy issues or the network instability, we need to run convolutional neural networks on embedded platforms. Critical challenges will be raised by limited hardware resources on the embedded platform. In this paper, we design and implement an embedded inference framework to accelerate the inference of the convolutional neural network on the embedded platform. For this, we first analyzed the time-consuming layers in the inference process of the network, and then we design optimization methods for these layers. Also, we design a memory pool specifically for neural networks. Our experimental results show that our embedded inference framework can run a classification model MobileNet in 80ms and a detection model MobileNet-SSD in 155ms on Firefly-RK3399 development board.

To promote the work capacity and efficiency in a weakly structured logistics sorting scene, a pick-and-throw method based on reinforcement learning is proposed. First, a D N-based learning algorithm is used to obtain a picking feasibility distribution map for guiding an optimal picking action. Second, Deep Deterministic Policy Gradient (DDPG) algorithm is utilized to train a throwing policy. The throwing policy outputs a throwing velocity of the end-effector of a robot to throw the picked object to a target area. In addition, a memory pool optimization algorithm is also proposed to enhance the convergence of the throwing policy. Experiments are conducted in both simulation and real scenarios and results demonstrate that the proposed method can improve the sorting efficiency and expand the sorting space significantly.

As one of the classic fields of computer vision, image classification has been booming with the improvement of chip performance and algorithm efficiency. With the rapid progress of deep learning in recent decades, remote sensing land cover and land use image classification has ushered in a golden period of development. This paper presents a new deep learning classifier to classify remote sensing land cover and land use images. The approach first uses multi-layer convolutional neural networks to extract the image features, attached through a fully-connected neural network to generate the sample loss. Then, a hard sample memory pool is created to collect the samples with large losses during the training. A batch of hard samples is randomly extracted from the memory pool to participate in the training of the convolutional fully connected model so that the model becomes more robust. Our method is validated by testing the classic remote sensing land cover and land use dataset. Compared with the previous popular classification algorithm, our algorithm can classify images more accurately with a shorter training iteration.

Cognitive radio (CR) and energy harvesting (EH) have been deemed two promising technologies in the spectrum-scarce and energy-limited wireless networks. In this paper, the cooperative cognitive radio network (CRN) with EH is considered, where a secondary user (SU) close to the secondary base station (SBS) employs power splitting for EH and assists to relay the data for another SU far away from the SBS. A SU sum-rate maximization problem is formulated under the constraints of the power budget at the SBS, the interference threshold of the primary network, and SU QoS. To tackle this problem, a resource allocation algorithm based on an improved deep neural network (DNN) is proposed. In order to accelerate the convergence of the DNN loss function, transfer learning is exploited to initialize the DNN weights. The loss between the DNN output and the optimal transmit power obtained by the conventional solution is stored in the memory pool, where the samples with large losses are used to train the DNN. Simulation results show the efficiency of our proposed DNN-based resource allocation scheme, which outperforms the normal DNN-based resource allocation and conventional resource allocation scheme in terms of the computation time.

Text analytics directly on compression (TADOC) has proven to be a promising technology for big data analytics. GPUs are extremely popular accelerators for data analytics systems. Unfortunately, no work so far shows how to utilize GPUs to accelerate TADOC. We describe G-TADOC, the first framework that provides GPU-based text analytics directly on compression, effectively enabling efficient text analytics on GPUs without decompressing the input data. G-TADOC solves three major challenges. First, TADOC involves a large amount of dependencies, which makes it difficult to exploit massive parallelism on a GPU. We develop a novel fine-grained thread-level workload scheduling strategy for GPU threads, which partitions heavily-dependent loads adaptively in a fine-grained manner. Second, in developing G-TADOC, thousands of GPU threads writing to the same result buffer leads to inconsistency while directly using locks and atomic operations lead to large synchronization overheads. We develop a memory pool with thread-safe data structures on GPUs to handle such difficulties. Third, maintaining the sequence information among words is essential for lossless compression. We design a sequence-support strategy, which maintains high GPU parallelism while ensuring sequence information. Our experimental evaluations show that G-TADOC provides 31.1x average speedup compared to state-of-the-art TADOC., Comment: 37th IEEE International Conference on Data Engineering (ICDE 2021)

Natural killer cells are critical effectors of antiviral host defense. This review summarizes the recent literature, current paradigms, and molecular mechanisms underlying the ability of these innate lymphocytes to mount adaptive responses against viral infection in humans and mice., NK cells express a limited number of germline-encoded receptors that identify infected or transformed cells, eliciting cytotoxicity, effector cytokine production, and in some circumstances clonal proliferation and memory. To maximize the functional diversity of NK cells, the array and expression level of surface receptors vary between individual NK cell “clones” in mice and humans. Cytomegalovirus infection in both species can expand a population of NK cells expressing receptors critical to the clearance of infected cells and generate a long-lived memory pool capable of targeting future infection with greater efficacy. Here, we discuss the pathways and factors that regulate the generation and maintenance of effector and memory NK cells and propose how this understanding may be harnessed therapeutically.

Clonal expansion is a core aspect of T cell immunity. However, little is known with respect to the relationship between replicative history and the formation of distinct CD8+ memory T cell subgroups. To address this issue, we developed a genetic-tracing approach, termed the DivisionRecorder, that reports the extent of past proliferation of cell pools in vivo. Using this system to genetically ‘record’ the replicative history of different CD8+ T cell populations throughout a pathogen-specific immune response, we demonstrate that the central memory T cell (TCM) pool is marked by a higher number of prior divisions than the effector memory T cell pool, due to the combination of strong proliferative activity during the acute immune response and selective proliferative activity after pathogen clearance. Furthermore, by combining DivisionRecorder analysis with single cell transcriptomics and functional experiments, we show that replicative history identifies distinct cell pools within the TCM compartment. Specifically, we demonstrate that lowly divided TCM display enriched expression of stem-cell-associated genes, and that such lowly divided cells are superior in eliciting a proliferative recall response. The latter data provide the first evidence that a stem cell like memory T cell pool that reconstitutes the CD8+ T cell effector pool upon reinfection is marked by prior quiescence.

— Blockchain technology is used by most Bitcoin systems to store all historical transaction information. Blockchain is a chain of blocks similar to the linked list structure and can be changed to a fork structure, in which there are two types of forks: useful fork or an intentional fork structure. A useful fork may appear when the rules of the Bitcoin system are updated. On the other hand, the intentional fork may appear when a miner has supercomputer properties, generates a set of blocks as a private branch, and does not publish this branch to the blockchain until its length exceeds the length of the main branch. A set of blockchain transactions will be rollbacked when the intentional fork occurs in the Bitcoin system, user waiting times will increase, and miner rewards will illegally increase. A Memory pool publisher algorithm is suggested in this paper to avoid the fork issues in the Bitcoin system, for instance: intentional fork, rollback problem, users waiting time. The proposed algorithm is to make the system a single publisher and divide the block's construction into two phases. A miner constructs a block and sends it to the memory pool as the first phase. The memory pool will send the construction block to the blockchain as the second phase. The findings indicate that the proposed algorithm has a strong potential to avoid the blockchain's intentional fork problem and thus minimize user waiting times for the rollback problem.

Recently, traditional compute-oriented system architecture is changing and diverting. The increasing demands for big data and artificial intelligence accelerate the needs for new and alternative architecture in computing system. One of the emerging area for alternative computing architecture is memory-centric or disaggregated computing. The memory-centric or disaggregated computing can solve the requirement for huge memory in system wide. In this paper, we present preliminary performance evaluation results of the fabric-attached memory system with industry standard based memory pool prototype hardware. The memory pool prototype hardware is configured as block device for benchmarking. For evaluation, well-known benchmarks – sysbench and resnet – are used. The preliminary benchmark results show that the overall access performance of the fabric-attached prototype hardware is comparable to SSD, and deep learning performance is close to NVMe.

As applications requiring a lot of memory grows, memory-centric computing systems which has a lot of memory pool are drawing attention. And the high-speed interconnection for accessing memory pools in this system is a very important part. In this paper, the bridge hardware for connecting PCI Express bus which is widely used in general systems, to Gen-Z protocol adapter which is one of the new high-speed networks, is shown, and its maximum performance is measured. This hardware has shown that it can support up to 720MB/s for memory writes and 3.5GB/s for block writes. This bridge hardware is expected to upgrade the performance of existing hardware and can expand the application field of the Gen-Z memory system.