Ultra-low power consumption and high performance SSD storage

Abstract

The present application discloses an ultra-low power consumption and high-performance SSD memory. The ultra-low power consumption and high-performance SSD memory includes: an intelligent power management module, which is used to predict future workloads in combination with machine learning algorithms, and dynamically adjust the power consumption state of the SSD memory according to the workload, application type, and user behavior; a high-performance optimization system, which is used to optimize performance by using block-level mapping mechanism, cache preheating mechanism, protocol optimization mechanism, hybrid data management mechanism, write acceleration mechanism, write merging optimization mechanism, garbage collection mechanism, data compression and decompression mechanism; a reliability assurance module, which is used to use real-time temperature monitoring mechanism, write balancing mechanism, burst protection mechanism, backup recovery mechanism, health detection and prediction mechanism for reliability assurance; a security module, which is used to encrypt and decrypt the stored data in the SSD memory in real time. Through the above method, fine-grained power control, temperature adaptive adjustment and dynamic adjustment of performance are achieved.

Description

Technical Field

[0001] The present application relates to the field of memory technology, and in particular to an ultra-low power consumption and high performance SSD memory. Background Art

[0002] Modern mobile and wearable devices pose severe challenges to the power consumption and performance of storage systems. Although traditional SSDs excel in performance, speed, and reliability, power consumption and thermal management remain key issues in mobile devices. Users' demands for fast startup, random access performance, write speed, and high capacity density continue to increase, making the limitations of related technologies increasingly apparent.

[0003] Challenges facing storage systems:

[0004] Power consumption challenge: The battery capacity of mobile devices is limited, the storage system accounts for a large proportion of power consumption, it is difficult to balance performance and power consumption, and the temperature control requirements are increased.

[0005] Performance requirements: fast startup requirements, high random access performance, fast write speed, and high capacity density. Summary of the invention

[0006] The ultra-low power consumption and high performance SSD memory provided in this application can realize fine-grained power control, intelligent sleep strategy, temperature adaptive regulation and dynamic adjustment of performance.

[0007] In the first aspect, the present application provides an ultra-low power consumption and high-performance SSD memory, including: an intelligent power management module, which is used to combine machine learning algorithms to predict future workloads, and dynamically adjust the power consumption state of the SSD memory according to the workload, application type, and user behavior; a high-performance optimization system, which is used to optimize performance by adopting block-level mapping mechanism, cache preheating mechanism, protocol optimization mechanism, hybrid data management mechanism, write acceleration mechanism, write merging optimization mechanism, garbage collection mechanism, data compression and decompression mechanism; a reliability assurance module, which is used to adopt real-time temperature monitoring mechanism, write balancing mechanism, burst protection mechanism, backup recovery mechanism, health detection and prediction mechanism for reliability assurance; a security module, which is used to encrypt and decrypt the stored data in the SSD memory in real time.

[0008] Among them, the intelligent power management module includes: a multi-level power consumption state refinement and dynamic adjustment unit, which is used to dynamically adjust the power consumption state of the SSD memory according to the characteristics of different I / O operations; a regional power supply control and partition namespace unit, which is used to combine the regional power supply control at the Die / Plane level with the partition namespace, and perform independent power management according to the access frequency and priority of different partition namespaces; a deep sleep and fast wake-up unit, which is used to use low-power SRAM or non-volatile memory as a wake-up controller; a dynamic frequency adjustment unit, which is used to use machine learning algorithms to predict future workloads and adjust the frequency and voltage of the storage controller of the SSD memory in advance; an intelligent clock gating unit, which is used to perform clock gating according to the data flow path and the active state of the module.

[0009] Among them, the multi-level power consumption state refinement and dynamic adjustment unit is also used to reduce the controller frequency, reduce circuit activation, optimize power supply, utilize the characteristics of the high-performance user-mode storage access framework, and optimize partition management during sequential write / append operations.

[0010] Among them, the multi-level power consumption state refinement and dynamic adjustment unit is also used to increase the controller frequency, enhance circuit activation, optimize the read channel, adopt a high-performance storage stack and optimize partition management during random read operations.

[0011] The intelligent power management module is also used to predict the characteristics of the write request and adopt a corresponding write strategy according to the characteristics of the predicted write request, wherein the characteristics of the write request include compressibility, sequentiality, and randomness.

[0012] Among them, the high-performance optimization system includes: a block-level mapping module, which is used to map logical blocks to physical blocks when the corresponding scenario of sequential writing is identified; a cache preheating module, which is used to use machine learning models to analyze user behavior and application access patterns for data preloading; a protocol optimization module, which is used to use NVMe's Multiple Queues and Submission / Completion Queues features to achieve concurrent operations; a hybrid data management module, which is used to optimize write performance using partition namespace SSD append operations; a write acceleration module, which is used to use the log structure merge tree write feature to merge small write requests; a write merge optimization module, which is used to merge writes according to data type, size and life cycle; a garbage collection module, which is used to use a predictive garbage collection algorithm to perform garbage collection when the system is idle or under load; a data compression and decompression module, which is used to dynamically compress and decompress stored data.

[0013] Among them, the garbage collection module is also used to combine the wear leveling algorithm to give priority to recycling blocks whose wear levels meet the preset requirements, and to use block-level mapping ideas to perform garbage collection when the system is idle.

[0014] Among them, the reliability assurance module includes: a real-time temperature monitoring unit, which is used to monitor the temperature of each area inside the SSD memory; a write leveling unit, which is used to adopt a dynamic wear leveling strategy and allocate writes according to the number of erase and write times and the health status of the block; a burst protection unit, which is used to complete data refresh and metadata protection in combination with a fast power-off detection mechanism; a backup and recovery unit, which is used to back up and restore data of the SSD memory; a health detection and prediction unit, which is used to monitor the various health indicators of the SSD memory in real time, and use machine learning algorithms to predict the remaining life and potential failures of the SSD memory.

[0015] The burst protection unit is also used to utilize the data compression and metadata management method in the predicted write request to refresh the data and metadata into the non-volatile memory when the power is off.

[0016] The security module is also used to utilize the hardware encryption engine to perform real-time encryption and decryption on the storage data in the SSD memory.

[0017] The beneficial effects of the present application are as follows: Different from the prior art, the ultra-low power consumption and high-performance SSD memory provided by the present application realizes fine-grained power control, intelligent sleep strategy, temperature adaptive adjustment and dynamic adjustment of performance through dynamic power consumption management. And realizes fast wake-up mechanism, intelligent data cache, pre-reading optimization and write acceleration through performance optimization system. And realizes temperature protection mechanism, prolongs write life, protects data integrity and provides power-off protection through reliability assurance module. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0019] Figure 1 It is a structural schematic diagram of an embodiment of an ultra-low power consumption and high performance SSD memory provided by the present application;

[0020] Figure 2 It is a structural diagram of an embodiment of an intelligent power management module provided by the present application;

[0021] Figure 3 It is a structural schematic diagram of an embodiment of a high-performance optimization system provided by the present application;

[0022] Figure 4 It is a structural diagram of an embodiment of a reliability assurance module provided in this application. DETAILED DESCRIPTION

[0023] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It will be understood that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should also be noted that, for ease of description, only some but not all structures related to the present application are shown in the drawings. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0024] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0025] Modern mobile and wearable devices pose severe challenges to the power consumption and performance of storage systems. Although traditional SSDs excel in performance, speed, and reliability, power consumption and thermal management remain key issues in mobile devices. Users' demands for fast startup, random access performance, write speed, and high capacity density continue to increase, making the limitations of related technologies increasingly apparent.

[0026] Challenges facing storage systems:

[0027] Power consumption challenge: The battery capacity of mobile devices is limited, the storage system accounts for a large proportion of power consumption, it is difficult to balance performance and power consumption, and the temperature control requirements are increased.

[0028] Performance requirements: fast startup requirements, high random access performance, fast write speed, and high capacity density.

[0029] Based on this, the present application proposes to realize fine-grained power control, intelligent sleep strategy, temperature adaptive regulation and dynamic adjustment of performance through dynamic power management. And realize fast wake-up mechanism, intelligent data cache, pre-reading optimization and write acceleration through performance optimization system. And realize temperature protection mechanism, extend write life, protect data integrity and provide power-off protection through reliability assurance module. Please refer to any of the following embodiments or any combination of embodiments for details.

[0030] See also Figure 1 , Figure 1The ultra-low power consumption and high performance SSD memory 100 comprises: an intelligent power management module 10, a high performance optimization system 20, a reliability guarantee module 30 and a security module 40.

[0031] The intelligent power management module 10 is used to predict future workloads in combination with machine learning algorithms, and dynamically adjust the power consumption state of the SSD memory according to the workload, application type, and user behavior.

[0032] In some embodiments, a more sophisticated multi-level power consumption state can be designed in the SSD memory, such as high-performance mode, balanced mode, low-power mode, deep sleep mode, perception mode, etc. In addition, the machine learning algorithm can be combined to predict future workloads, and dynamic adjustments can be made based on multiple factors such as workload, system load, application type, user behavior, etc., to accurately match the power consumption requirements of the SSD memory.

[0033] Furthermore, the power consumption state of the SSD can be dynamically adjusted according to the characteristics of different I / O operations to minimize power consumption while ensuring performance. This method fully utilizes the characteristics of the partitioned namespace SSD and the different requirements of different I / O operations for resources and performance.

[0034] The high-performance optimization system 20 is used to optimize performance by using a block-level mapping mechanism, a cache warm-up mechanism, a protocol optimization mechanism, a hybrid data management mechanism, a write acceleration mechanism, a write merge optimization mechanism, a garbage collection mechanism, and a data compression and decompression mechanism.

[0035] In some embodiments, corresponding block-level mapping mechanism, cache preheating mechanism, protocol optimization mechanism, hybrid data management mechanism, write acceleration mechanism, write merge optimization mechanism, garbage collection mechanism, data compression and decompression mechanism can be set in the high-performance optimization system 20 to obtain the optimal performance during the use of SSD memory.

[0036] The reliability assurance module 30 is used to ensure reliability by adopting a real-time temperature monitoring mechanism, a write balancing mechanism, a burst protection mechanism, a backup recovery mechanism, and a health detection and prediction mechanism.

[0037] The security module 40 is used to perform real-time encryption and decryption on the data stored in the SSD memory.

[0038] See also Figure 2 , Figure 2The intelligent power management module 10 includes: a multi-level power consumption state refinement and dynamic adjustment unit 11, a regional power supply control and partition namespace unit 12, a deep sleep and fast wake-up unit 13, a dynamic frequency adjustment unit 14 and an intelligent clock gating unit 15.

[0039] The multi-level power consumption state refinement and dynamic adjustment unit 11 is used to dynamically adjust the power consumption state of the SSD memory according to the characteristics of different I / O operations.

[0040] In some embodiments, the core idea of ​​dynamically adjusting the power consumption state of the SSD memory according to the characteristics of different I / O operations is to dynamically adjust the power consumption state of the SSD according to the characteristics of different I / O operations to achieve the purpose of minimizing power consumption while ensuring performance. This method makes full use of the characteristics of the zoned namespaces SSD and the different requirements of different I / O operations for resources and performance. I / O operations include sequential write / append operations, random read operations, and garbage collection operations.

[0041] Sequential write / append operation characteristics: The characteristic of the partition namespace SSD is forced sequential write. This means that data must be written in increasing order of logical block address (LBA). This characteristic makes the write operation more physically predictable and reduces additional operations inside the Flash storage chip, such as random addressing, which can reduce power consumption. In addition, the append operation is also a sequential write, but it continues to write data in the existing zone, which is also sequential.

[0042] Characteristics of random read operations: Random reads require frequent addressing and reading of data at different physical locations, and require more frequent activation and control of circuits within the memory chip, so higher power consumption is required to meet the requirements of read performance.

[0043] Characteristics of garbage collection operations: In traditional SSDs, garbage collection (GC) operations run in the background, affecting read and write performance, while GC operations in partitioned namespace SSDs are explicitly controlled by the host, which makes it possible for this application to plan and control GC according to workloads, thereby reducing power consumption.

[0044] Performance differences between different storage stacks: Using the high-performance user-state storage access framework (the high-performance user-state storage access framework is a high-performance, low-latency storage stack that is particularly suitable for applications and scenarios that require direct access to hardware and have high requirements for I / O performance. It allows developers to bypass the traditional operating system I / O stack, achieve more refined I / O control, and make full use of the performance of storage devices, especially the characteristics of partitioned namespace SSDs) has lower latency under a single I / O request, which shows that the high-performance user-state storage access framework can provide more efficient I / O processing and can reduce power consumption time in high-performance mode.

[0045] Among them, the dynamic power consumption mode switching based on I / O operation can be divided into the following ways:

[0046] A lower power mode is used during sequential write / append operations. This low power mode involves reducing controller frequency, reducing circuit activation, optimizing power supply, utilizing features of the high-performance user-mode storage access framework, partition management optimization, and garbage collection planning.

[0047] The main reason for reducing the controller frequency is that the physical addressing of sequential write operations is predictable, so the operating frequency of the SSD controller can be appropriately reduced, thereby reducing dynamic power consumption.

[0048] Reducing circuit activation is mainly due to the characteristics of sequential writing, which can reduce unnecessary circuit activation inside the Flash chip and further reduce power consumption. For example, the number of pre-charge and discharge times of the storage unit can be reduced.

[0049] Optimizing power supply mainly provides power to the area in the storage array that is being written sequentially, reducing the power supply to other areas.

[0050] The main feature of the high-performance user-mode storage access framework is to use the low latency feature of the high-performance user-mode storage access framework during sequential writes, which can reduce the time for data transmission and processing and further reduce power consumption. The log structure merge tree write feature is used to merge small write requests, reduce write amplification, and reduce write power consumption. Random small writes are converted to sequential writes to improve write efficiency. Data is first written to the memory table in memory, and then refreshed in batches to the sorted string table file on disk, which itself has the effect of merging writes and reduces the additional overhead caused by random small writes.

[0051] Partition management optimization is mainly based on the characteristics of partition namespace, limiting sequential writes to a small number of partitions, reducing the partitions that need to be in working state at the same time, and reducing overall power consumption.

[0052] Garbage collection planning mainly utilizes the host-controlled GC feature of partitioned namespace SSDs to schedule GC operations when the system load is low, avoiding GC execution during write or read peaks, thereby reducing energy consumption.

[0053] Switch to high-performance mode during random read operations, which involves increasing controller frequency, enhancing circuit activation, optimizing read channels, using high-performance storage stacks, and optimizing partition management.

[0054] Increasing the controller frequency is mainly to meet the high performance requirements of random reads, and it is necessary to increase the operating frequency of the SSD controller.

[0055] Enhanced circuit activation is mainly for fast access to random physical addresses, and the circuits inside the Flash chip need to be activated more frequently to ensure low latency in data reading.

[0056] Optimizing the read channel mainly involves using a more efficient read channel to reduce the latency of random read operations.

[0057] The use of a high-performance storage stack mainly utilizes the low latency characteristics of the SPDK storage stack to ensure fast execution of random read operations. Use faster memory or new non-volatile storage to cache read data and improve random read performance.

[0058] Partition management optimization is mainly based on the partition namespace characteristics. During read operations, the partition containing the read data is quickly activated, and unnecessary partition activation is reduced to reduce power consumption.

[0059] Among them, the realization of dynamic adjustment is mainly reflected in state monitoring, prediction mechanism and feedback control.

[0060] Status monitoring mainly monitors the type of I / O operations (sequential write / append, random read) in real time and dynamically adjusts the power consumption mode according to the characteristics of the I / O requests.

[0061] The prediction mechanism mainly uses machine learning models to predict the type of I / O operations in the future and switch the power consumption mode in advance. For example, if it is predicted that there will be a large number of random reads, it can switch to high-performance mode in advance to reduce latency.

[0062] Feedback control mainly uses the feedback control mechanism to further fine-tune the power consumption mode according to the actual performance. For example, if the delay is detected to be too high, the power input can be increased. Conversely, if the load is detected to be low, the power consumption can be reduced.

[0063] The above technical effects mainly lie in power consumption reduction, performance guarantee and flexible adaptability.

[0064] The power consumption is reduced mainly through this dynamic switching mode, which can significantly reduce the overall power consumption of the SSD. During sequential write or append operations, the SSD works in a lower power consumption mode, reducing unnecessary energy consumption. During random reads, although the power consumption increases, the overall power consumption can still be kept at a low level because the read operation is usually short-lived.

[0065] The performance guarantee mainly means that when performing operations that require high performance, such as random reading, the SSD can quickly switch to high-performance mode to ensure low latency and high throughput of the operation.

[0066] Flexible adaptability mainly means that the strategy can be dynamically adjusted according to different workloads, minimizing power consumption while ensuring performance, and adapting to different application scenarios more flexibly.

[0067] Furthermore, the above technology can be coordinated with other technologies, such as data compression, intelligent caching, and garbage collection. The combined data compression technology can reduce the amount of data actually written, thereby further reducing the power consumption of write operations. Combined with intelligent caching to prefetch data, reduce the power consumption of reads. Combined with garbage collection, the explicit GC control of the partition namespace SSD can be used, combined with garbage collection scheduling, to perform garbage collection when the system is idle, avoiding performance and power consumption during peak hours.

[0068] The regional power supply control and partition namespace unit 12 is used to combine the regional power supply control at the Die / Plane level with the partition namespace, and perform independent power management according to the access frequency and priority of different partition namespaces. That is, the regional power supply control at the Die / Plane level is combined with the partition namespace, and independent power management is performed according to the access frequency and priority of different partition namespaces to reduce unnecessary power consumption. Inactive partition namespaces can be placed in a low power state.

[0069] The deep sleep and fast wake-up unit 13 is used to use low-power SRAM or non-volatile memory as a wake-up controller. In some embodiments, an ultra-low power deep sleep mode can be designed and the wake-up path can be optimized. The low latency advantage of the high-performance user-mode storage access framework is utilized to reduce the wake-up latency and reduce power consumption. Using low-power SRAM or a new type of non-volatile memory as a wake-up controller further optimizes the wake-up path.

[0070] The dynamic frequency adjustment unit 14 is used to predict future workloads using a machine learning algorithm and adjust the frequency and voltage of the storage controller of the SSD memory in advance, thereby enabling more detailed power management on the partition namespace SSD.

[0071] The intelligent clock gating unit 15 is used to perform clock gating according to the data flow path and the active state of the module. That is, more refined clock gating can be achieved according to the data flow path and the active state of the module. The relevant clock signal is activated only when a specific type of operation occurs. The specific type of operation can be set according to the above-mentioned I / O operation.

[0072] Among them, the multi-level power consumption state refinement and dynamic adjustment unit 11 is also used to reduce the controller frequency, reduce circuit activation, optimize power supply, utilize the characteristics of the high-performance user-mode storage access framework and optimize partition management during sequential write / append operations.

[0073] The multi-level power consumption state refinement and dynamic adjustment unit 11 is also used to increase the controller frequency, enhance circuit activation, optimize the read channel, adopt a high-performance storage stack, and optimize partition management during random read operations.

[0074] The intelligent power management module 10 is also used to predict the characteristics of the write request and adopt a corresponding write strategy according to the predicted characteristics of the write request, wherein the characteristics of the write request include compressibility, sequentiality, and randomness.

[0075] For write operations in solid-state drives (SSDs), power consumption is optimized by predicting the characteristics of write requests (including compressibility, sequentiality, and randomness) and adopting different write strategies accordingly. The following are the specific implementation methods: prediction of write request types, power consumption optimization strategies based on prediction results, optimization of mapping tables, and power consumption optimization of garbage collection and merging operations.

[0076] The prediction of the write request type may be a compressibility prediction or a sequentiality / randomness prediction.

[0077] The compressibility prediction first predicts whether the data of the write request is compressible. It does not try to compress all logical pages, but makes a prejudgment through the "incompressible request prediction module". It mainly analyzes the data histogram of the write request and compares it with the pre-established representative histogram data set. If the similarity exceeds the predefined threshold, the request is judged to be incompressible.

[0078] Sequential / random prediction is reflected in the fact that for incompressible requests, it will further distinguish whether it is sequential write or random write.

[0079] Sequential writing is usually generated when writing large files or streaming data, and has the characteristics of address continuity. Random writing is usually generated by small files or scattered data, and the addresses are not continuous. That is, it can be distinguished whether it is sequential writing or random writing according to the characteristics of sequential writing and random writing.

[0080] Furthermore, a “locality detection module” is used to identify small incompressible sequential subrequests. This module mainly exploits the spatial locality of write requests and aggregates consecutive small incompressible subrequests together.

[0081] Among them, the power consumption optimization strategy based on the prediction results is mainly reflected in the processing of compressible requests, the processing of incompressible sequential requests, the processing of incompressible small sequential requests, and the processing of incompressible random requests.

[0082] The processing of compressible requests is mainly reflected as follows:

[0083] For compressible write requests, the data is compressed, which reduces the amount of data written, directly reducing the power consumption of the write operation. The compressed data is written to a "container buffer".

[0084] Write multiple compressible logical pages into one physical page, and use the internal fragmentation space of the physical page to store metadata. This further improves storage efficiency, reduces the actual number of physical writes, and thus reduces power consumption.

[0085] Depending on whether there is enough space in the block to write data, there are two situations: sufficient space and insufficient space.

[0086] If there is enough space, the data is written to the last few pages of the data block.

[0087] If there is insufficient space, a new data block needs to be allocated.

[0088] The processing of incompressible sequential requests is mainly reflected as follows:

[0089] For incompressible sequential write requests, they are directly written to a new physical block and the mapping table is updated. Block mapping is used to avoid compression operations on write data, thereby reducing the CPU's computational burden and power consumption.

[0090] This approach can reduce the overhead of garbage collection because sequentially written data usually becomes invalid at the same time, avoiding the cleanup operation of scattered data and further reducing power consumption.

[0091] Among them, the incompressible small sequential request processing is mainly reflected as follows:

[0092] For incompressible small sequential sub-requests with spatial locality, they are aggregated into sequential log blocks. These log blocks use block mapping, and when the log block is full, the mapping table is updated at one time, thereby reducing frequent mapping table update operations and reducing power consumption.

[0093] When the log block is full, it will be merged with the original data block. You can choose to merge completely or partially.

[0094] Among them, the incompressible random request processing is mainly reflected as follows:

[0095] For incompressible random write requests, they are written to the data block or shared log block and the mapping table is updated. These requests are handled similarly to the traditional flash translation layer and cannot be written to reduce power consumption, but can reduce overall power consumption by compressing other write requests.

[0096] Among them, the optimization of the mapping table is mainly reflected in the use of a two-level mapping table structure for address translation. Compared with traditional page mapping, this design significantly reduces the memory occupied by the mapping table, thereby reducing the energy consumption of the solid-state drive, because memory access is also an important source of power consumption. The mapping table contains the data physical block number and the log physical block number, as well as the offset of the logical page in each block, and uses a single bit to mark whether the log physical offset belongs to a shared log block.

[0097] Among them, the power consumption optimization of garbage collection and merging operations is mainly reflected in the use of the intra-block update mechanism, which allows data to be updated within the physical block, reducing the frequency of garbage collection operations to a certain extent. And merging operations, such as "switch merge" and "partial merge", both achieve power consumption optimization by reducing data movement. And switch merge only updates the mapping table and does not involve data movement. Partial merge only moves valid data and replaces the old data block with a new log block.

[0098] In summary, further optimization of power consumption is achieved by means of data compression, distinguishing between sequential and random writes, utilizing spatial locality, reducing mapping table memory usage, optimizing garbage collection and merging operations, and combining prediction of write request types.

[0099] Data compression is mainly used for compressible data. It reduces the amount of written data and reduces power consumption through compression.

[0100] The distinction between sequential and random writes is mainly for incompressible sequential writes, using more efficient block mapping and writing methods to reduce garbage collection overhead.

[0101] Spatial locality is mainly used for small incompressible sequential sub-requests. The locality detection module aggregates writes and uses log blocks to reduce the frequency of mapping table updates.

[0102] The memory usage of the mapping table is reduced mainly through a two-level mapping table structure, which reduces the power consumption of memory access.

[0103] Optimizing garbage collection and merging operations mainly reduces the power consumption of garbage collection through intra-block update mechanisms and merging operations that reduce data movement.

[0104] Through these strategies, the write operations of the SSD can be managed more finely, and different processing methods can be adopted according to different write request characteristics, thereby effectively reducing power consumption while improving write performance.

[0105] See also Figure 3 , Figure 3 1 is a schematic diagram of the structure of an embodiment of a high-performance optimization system provided by the present application. The high-performance optimization system 20 includes: a block-level mapping module 21, a cache preheating module 22, a protocol optimization module 23, a hybrid data management module 24, a write acceleration module 25, a write merging optimization module 26, a garbage collection module 27 and a data compression and decompression module 28.

[0106] The block-level mapping module 21 is used to map the logical blocks to the physical blocks when a corresponding scenario of sequential writing is identified.

[0107] In some embodiments, the traditional flash translation layer solution usually uses page-level mapping, that is, a mapping to a physical page is maintained for each logical page. The advantage of this method is high flexibility, and any physical page can be written arbitrarily, but the disadvantage is that the mapping table is very large and requires a lot of memory. For example, a 1TB SSD requires 1GB of RAM to store the page-level mapping table. In view of the fact that there are more sequential writes in personal computer (PC) application scenarios, this application introduces the concept of block-level mapping. For sequentially written incompressible data, it is directly mapped to the physical block, reducing the number of entries in the mapping table.

[0108] At the same time, in order to support the writing of compressible data, multiple logical pages are compressed and written into one physical page. In order to manage these compressed data, metadata is stored in the fragment space of the physical page to record the number, offset and size of compressed pages.

[0109] This approach reduces the number of physical page writes and allows data to be updated within blocks, thereby reducing garbage collection overhead.

[0110] Furthermore, the present application also uses a two-layer address mapping table: the first layer performs mapping from logical blocks to physical blocks, and the second layer records the physical offset of pages within the logical blocks to support intra-block updates.

[0111] In addition, the present application further optimizes the size and read / write efficiency of the mapping table by distinguishing between sequential and random requests, and compressible and incompressible requests, and selectively applying different mapping strategies.

[0112] The specific steps are as follows:

[0113] Identify sequential write scenarios: Analyze the I / O patterns in the application scenarios and identify scenarios where sequential writes dominate. For example, data backup, log writing, multimedia data writing, etc. usually have strong sequentiality.

[0114] Implement block-level mapping: For sequentially written data, directly map logical blocks to physical blocks without maintaining a separate mapping entry for each logical page.

[0115] Support for intra-block updates: In order to support intra-block updates of data, it is necessary to record the offset of the page within the block. The secondary mapping table in the IBU solution or other similar lightweight solutions can be used.

[0116] Dynamically switch mapping modes: For scenarios with a lot of random writes, you can dynamically switch back to page-level mapping, or use other mapping schemes that are more suitable for random writes.

[0117] Combined with data compression: Consider combining data compression technology to further reduce physical space usage and reduce garbage collection overhead. Compress multiple logical pages and write them into one physical page, and store metadata in the fragmented space.

[0118] Metadata management: In the block-level mapping scheme, metadata needs to be managed efficiently, such as using lightweight data structures or caching metadata in fast memory.

[0119] Partition Namespace SSD partitions are essentially logical blocks that can only be written sequentially, which is very similar to the block-level mapping for sequential writing in this application.

[0120] Therefore, the block-level mapping of the present application can be directly applied to the partition management of the partition namespace SSD, further simplifying the address mapping and reducing the overhead of the mapping table.

[0121] In addition, the partitioned namespace SSD provides an explicit partition reset operation, which can more effectively control garbage collection and reduce mapping table entries caused by invalid pages.

[0122] The advantage of block-level mapping is that it can reduce the size of the mapping table, save memory, improve read and write performance, and reduce power consumption.

[0123] Among them, memory saving is mainly reflected in that the mapping table is usually stored in RAM. Reducing the size of the mapping table can significantly reduce memory usage and reduce hardware costs.

[0124] Among them, improving read and write performance is mainly reflected in the fact that smaller mapping tables mean faster search speeds, which can improve the performance of read and write operations, especially in high-concurrency scenarios.

[0125] Among them, reducing power consumption is mainly reflected in reducing memory access to reduce power consumption, which is very important for mobile devices and wearable devices.

[0126] The cache preheating module 22 is used to use machine learning models to analyze user behavior and application access patterns and preload data. For example, more advanced machine learning models are used to analyze user behavior and application access patterns for more accurate preloading. In addition, the characteristics of the partition namespace SSD are combined to achieve more efficient data prefetching, such as prefetching data for the entire partition based on the access pattern. Edge computing is used to preprocess data to reduce cache occupancy.

[0127] The protocol optimization module 23 is used to utilize the Multiple Queues and Submission / Completion Queues features of NVMe to achieve concurrent operations. For example, the NVMe protocol stack is deeply optimized to reduce software overhead. NVMe features such as Multiple Queues and Submission / Completion Queues are utilized to achieve more efficient concurrent operations. A high-performance user-mode storage access framework is used to explore new storage interfaces and protocols, such as computational storage, to sink computing power to storage devices and reduce the need for data transmission.

[0128] The hybrid data management module 24 is used to optimize write performance by using the partition namespace SSD append operation. For example, the partition namespace SSD append operation is used to optimize write performance, especially in small write scenarios (see the above content for details). A hybrid data management strategy is used to obtain optimal performance.

[0129] The write acceleration module 25 is used to merge small write requests by utilizing the log structure merge tree write feature. This operation can reduce write amplification. Furthermore, data compression technology in predictive write requests can be used to increase write speed.

[0130] The write merge optimization module 26 is used to perform write merge according to data type, size and life cycle. For small and frequent writes, a more aggressive merge strategy is adopted.

[0131] The garbage collection module 27 is used to use a predictive garbage collection algorithm to perform garbage collection when the system is idle or under low load. That is, a predictive garbage collection algorithm is used to perform garbage collection when the system is idle or under low load to avoid affecting performance during peak hours. In addition, a wear leveling algorithm can be combined to give priority to recycling blocks with higher wear levels to extend the life of the SSD. In addition, a block-level mapping idea is used to perform garbage collection when the system is idle.

[0132] The data compression and decompression module 28 is used to dynamically compress and decompress the stored data. Among them, dynamically compressing and decompressing the stored data can reduce the amount of data actually written, improve the writing speed, and reduce the storage space occupied. Further, different compression algorithms can be selected according to the data type, and a hardware acceleration unit can be used for efficient compression and decompression. And a method for processing compressible and incompressible data in combination with a predicted write request.

[0133] The garbage collection module 27 is also used to combine the wear leveling algorithm to give priority to recycling blocks whose wear levels meet preset requirements, and to use block-level mapping ideas to perform garbage collection when the system is idle.

[0134] See also Figure 4 , Figure 4 The reliability guarantee module 30 includes: a real-time temperature monitoring unit 31 , a write balancing unit 32 , a burst protection unit 33 , a backup recovery unit 34 and a health detection and prediction unit 35 .

[0135] The real-time temperature monitoring unit 31 is used to monitor the temperature of each area inside the SSD storage. A more sophisticated temperature sensor network is integrated to monitor the temperature of each area inside the SSD, and more refined power consumption control and performance adjustment are performed according to the temperature gradient. The partitioned namespace SSD allows the host to more finely control storage management, including garbage collection operations. It is convenient for the host to dynamically adjust the garbage collection strategy according to the data of the temperature sensor, for example, to reduce the GC frequency when the temperature is high, or to delay high-intensity GC operations. Heat pipes or new heat dissipation materials are used to improve heat dissipation efficiency.

[0136] The write leveling unit 32 is used to adopt a dynamic wear leveling strategy to allocate writes according to the number of times a block is erased and written and the health status. A dynamic wear leveling strategy is adopted to make a more intelligent write allocation according to the number of times a block is erased and written and the health status. Combined with the characteristics of data compression in the predicted write request, the life of the SSD is extended by reducing the amount of writes. Host-Aware wear leveling is adopted to optimize according to the file system information of the host.

[0137] The burst protection unit 33 is used to complete data refresh and metadata protection in combination with a fast power-off detection mechanism. For example, a more efficient capacitor or a new energy storage device is used to provide a longer power-off protection time. In addition, the fast power-off detection mechanism is combined to complete data refresh and metadata protection in a very short time. In addition, the data compression and metadata management method in the predictive write request is used to quickly refresh the data and metadata when the power is off.

[0138] Among them, when the power is off, the SSD needs to quickly refresh the data and metadata in the cache to the non-volatile memory. This mainly faces the following challenges:

[0139] Data loss: Data in the cache (including user data and metadata) is volatile and will be lost after a power outage.

[0140] Data inconsistency: If a power outage occurs during a data write or update operation, it may cause data inconsistency (such as partial write).

[0141] Metadata corruption: If metadata such as the address mapping table is corrupted, data cannot be read correctly.

[0142] Time limit: When power is lost, the SSD has only a short time (usually a few milliseconds) to complete the data refresh operation.

[0143] Fast refresh strategies may include: using compression to reduce the amount of writes, efficient metadata storage, internal fragmentation space, streamlined metadata, incremental metadata updates, log-style writing, reserved refresh space, fast power-off detection, and the use of supercapacitors or new energy storage devices.

[0144] Using compression to reduce the amount of writes is mainly reflected in the use of data compression methods that predict write requests. When power is off and refreshed, the data and metadata to be written are compressed, thereby reducing the amount of data actually written to the flash memory and shortening the refresh time.

[0145] Internal fragmentation space is mainly reflected in the use of internal fragmentation space of physical pages to store metadata, avoiding additional metadata writing overhead.

[0146] Simplified metadata mainly involves refreshing only necessary metadata, for example, recording the status of unfinished operations instead of refreshing the entire mapping table every time.

[0147] Incremental metadata updates mainly involve refreshing only the metadata that has changed, rather than refreshing all of it every time.

[0148] Log-based writing mainly uses log-based writing to sequentially write the data and metadata to be refreshed into the flash memory, which can reduce random writing, simplify recovery, and perform atomic operations.

[0149] Reducing random writes mainly means that sequential writes are faster than random writes and can reduce the wear of flash memory. Simplifying recovery mainly means that by replaying the log, unfinished operations before power failure can be restored. Atomic operations mainly mean that a group of related operations are recorded and refreshed as an atomic unit to ensure data consistency.

[0150] The reserved refresh space mainly involves reserving a portion of the flash memory specifically for power-off refresh, thereby reducing the address lookup overhead during refresh.

[0151] The fast power-off detection is mainly reflected in the use of a fast power-off detection mechanism to trigger the refresh operation in a timely manner.

[0152] The use of supercapacitors or new energy storage devices is mainly reflected in the use of supercapacitors or new energy storage devices to provide sufficient power-off protection time to ensure that the refresh operation can be completed.

[0153] The backup and recovery unit 34 is used to back up and recover the data of the SSD storage. That is, it can provide a more flexible data backup and recovery strategy, for example, support incremental backup, snapshot backup and cloud backup. RAID technology or erasure code technology is used to improve the redundancy and reliability of data.

[0154] The health detection and prediction unit 35 is used to monitor various health indicators of the SSD memory in real time, and use machine learning algorithms to predict the remaining life and potential failures of the SSD memory. For example, an integrated intelligent monitoring module is used to monitor various health indicators of the SSD in real time (such as the number of erases, the number of bad blocks, temperature, etc.), and a machine learning algorithm is used to predict the remaining life and potential failures of the SSD, and to provide early warning.

[0155] The burst protection unit 33 is also used to utilize the data compression and metadata management method in the predictive write request to refresh the data and metadata into the non-volatile memory when the power is off.

[0156] The security module 40 is also used to use the hardware encryption engine to perform real-time encryption and decryption on the stored data in the SSD memory. That is, the hardware encryption engine is integrated to perform real-time encryption and decryption on the stored data to protect the security of user data. It also supports multiple encryption algorithms and provides key management functions. It can also refer to the data compression scheme in the predicted write request and perform encryption during the compression process.

[0157] Through the comprehensive application of the above technical solutions, it is expected that the SSD storage of this application will achieve significant improvements in the following aspects:

[0158] When SSD memory is used as storage for mobile terminals, it has the characteristics of reduced power consumption, shortened startup time, improved random read and write performance, and improved write performance.

[0159] Power consumption can be reduced by 50% or more, and power consumption can be further reduced through more sophisticated power management and predictive optimization.

[0160] The boot time can be shortened by 60% or more. The optimized boot mechanism and fast wake-up technology enable the device to respond instantly.

[0161] Random read and write performance can be improved by 50% or more. Smarter cache and page mapping as well as hardware acceleration units significantly increase data access speed.

[0162] The write performance can be improved by 60% or more. Optimized write merging, garbage collection scheduling, and data compression technologies make writes faster.

[0163] When SSD memory is used in wearable devices, it has the characteristics of extended standby time, reduced response delay, reduced heat generation, and improved write life.

[0164] The standby time can be extended by 80% or even higher. More aggressive power management and deep sleep mode significantly extend the device's standby time.

[0165] Response latency can be reduced by 40% or more. Optimized performance mechanisms and fast wake-up capabilities reduce data access latency.

[0166] Heat reduction can be reduced by 35% or even higher. Intelligent temperature control management and efficient heat dissipation design reduce device heat generation and improve user experience.

[0167] The write life can be increased by 50% or even higher. More balanced writes and more complete protection mechanisms extend the life of the SSD.

[0168] In summary, the ultra-low power consumption and high-performance SSD memory 100 provided by the present application realizes fine-grained power control, intelligent sleep strategy, temperature adaptive adjustment and dynamic adjustment of performance through dynamic power management. And realizes fast wake-up mechanism, intelligent data cache, pre-reading optimization and write acceleration through performance optimization system. And realizes temperature protection mechanism, prolongs write life, protects data integrity and provides power-off protection through reliability guarantee module 30.

[0169] In the several embodiments provided in this application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation described above is only illustrative, for example, the division of the modules or units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0170] If the integrated units in the above other embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, RandomAccess Memory), disk or optical disk and other media that can store program codes.

[0171] The above description is only an implementation method of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. An ultra-low power consumption and high performance SSD memory, characterized in that: include: An intelligent power management module, which is used to predict future workloads in combination with machine learning algorithms and dynamically adjust the power consumption state of the SSD memory according to the workload, application type, and user behavior; High-performance optimization system for performance optimization using block-level mapping mechanism, cache warm-up mechanism, protocol optimization mechanism, hybrid data management mechanism, write acceleration mechanism, write merge optimization mechanism, garbage collection mechanism, data compression and decompression mechanism; Reliability assurance module, which is used to ensure reliability by using real-time temperature monitoring mechanism, write balancing mechanism, burst protection mechanism, backup recovery mechanism, health detection and prediction mechanism; A security module, used for encrypting and decrypting the stored data in the SSD memory in real time; The intelligent power management module comprises: A multi-level power consumption state refinement and dynamic adjustment unit, which is used to dynamically adjust the power consumption state of the SSD memory according to the characteristics of different I / O operations; The regional power supply control and partition namespace unit is used to combine the regional power supply control at the Die / Plane level with the partition namespace, and perform independent power management according to the access frequency and priority of different partition namespaces; Deep sleep and fast wake-up unit for using low-power SRAM or non-volatile memory as wake-up controller; A dynamic frequency adjustment unit, used to predict future workloads using a machine learning algorithm and adjust the frequency and voltage of a storage controller of the SSD memory in advance; Intelligent clock gating unit for clock gating based on data flow path and module activity status.

2. The ultra-low power consumption and high performance SSD memory according to claim 1, characterized in that: The multi-level power consumption state refinement and dynamic adjustment unit is also used to reduce controller frequency, reduce circuit activation, optimize power supply, utilize the characteristics of high-performance user-mode storage access framework and partition management optimization during sequential write / append operations.

3. The ultra-low power consumption and high performance SSD memory according to claim 1, characterized in that: The multi-level power consumption state refinement and dynamic adjustment unit is also used to increase the controller frequency, enhance circuit activation, optimize the read channel, adopt a high-performance storage stack, and optimize partition management during random read operations.

4. The ultra-low power consumption and high performance SSD memory according to claim 1, characterized in that: The intelligent power management module is also used to predict the characteristics of the write request and adopt a corresponding write strategy according to the predicted characteristics of the write request, wherein the characteristics of the write request include compressibility, sequentiality, and randomness.

5. The ultra-low power consumption and high performance SSD memory according to claim 1, characterized in that: The high performance optimization system comprises: A block-level mapping module, for mapping logical blocks to physical blocks when a corresponding sequential write scenario is identified; Cache preheating module, which uses machine learning models to analyze user behavior and application access patterns and preload data; Protocol optimization module, used to utilize the Multiple Queues and Submission / Completion Queues features of NVMe to achieve concurrent operations; Hybrid data management module for optimizing write performance using partition namespace SSD append operations; Write acceleration module, used to merge small write requests by taking advantage of the log structure merge tree write feature; Write-merge optimization module, used to merge writes based on data type, size, and life cycle; Garbage collection module, which uses a predictive garbage collection algorithm to perform garbage collection when the system is idle or under low load; The data compression and decompression module is used to dynamically compress and decompress the stored data.

6. The ultra-low power consumption and high performance SSD memory according to claim 5, characterized in that: The garbage collection module is also used to combine the wear leveling algorithm to give priority to recycling blocks whose wear levels meet preset requirements, and to use block-level mapping ideas to perform garbage collection when the system is idle.

7. The ultra-low power consumption and high performance SSD memory according to claim 1, characterized in that: The reliability assurance module comprises: Real-time temperature monitoring unit, used to monitor the temperature of various areas inside the SSD storage; Write leveling unit, used to adopt dynamic wear leveling strategy and distribute writes according to the number of erase and write times and health status of blocks; A burst protection unit is used to complete data refresh and metadata protection in combination with a fast power-off detection mechanism; A backup and recovery unit, used for backing up and restoring data of the SSD storage; The health detection and prediction unit is used to monitor the health indicators of the SSD storage in real time and use machine learning algorithms to predict the remaining life and potential failures of the SSD storage.

8. The ultra-low power consumption and high performance SSD memory according to claim 7, characterized in that: The burst protection unit is also used to refresh data and metadata into the non-volatile memory when power is off by using the data compression and metadata management method in the predictive write request.

9. The ultra-low power consumption and high performance SSD memory according to claim 1, characterized in that: The security module is also used to utilize a hardware encryption engine to perform real-time encryption and decryption on the storage data in the SSD memory.

Images