This invention pertains generally to the field of architectures for non-volatile memory systems and, more particularly, to the arrangement of interfaces between the system controller and its memory arrays.
Non-volatile memory systems for the storage of user data, such flash memory cards, typically are made up of a set of arrays of non-volatile memory cells, on which the user data is stored, and a controller that controls the transfer of the user data between the arrays and a host and also manages the storage of the user data on the system. Much of this system management data, or meta-data, needs also to be retained and is therefore also stored in the non-volatile arrays. An example of such meta-data is the logical to physical translation information by which the physical location of data on an array is associated with the logical address by which the host identifies the data. Other examples can include meta-block linking data and defect re-mapping data.
Historically, flash memory controller manufacturers have used static logical to physical mapping algorithms to map user data in non-volatile memory cells. These translation tables were relatively small (on the order of 64 KB for 128 GB storage capacity) and of fixed size, allowing them easily to be cached on the memory controller. This caching results in good system performance, but has an inherent issue with these static mapping algorithms is their high write amplification ratio. For older generations of flash memory types, such high write amplification ratio was acceptable for a given usage model, since the raw endurance of memory cells was of the order of 50K or more program-erase cycles. However, as the geometry of flash memory chips has continued to shrink, the raw reliability of memory cells is often reduced significantly. Because of this, many flash memory controller manufacturers have adopted dynamic mapping algorithms which provide fairly low write amplification ratios, thereby extending the life of the storage product; however, an inherent issue with these dynamic mapping algorithms is that the translation tables, used to map logical to physical addresses, can be very large, growing, for example, to the order of 500 MB. To optimize performance, the controller would preferably cache the entire translation table in the internal memory of the controller; however, it is extremely expensive to have such large pool of SRAM memory in the controller. Consequently, there is room for improvement in the handling of large and complicated metadata in such memory systems.
According to a general aspect of the invention, a memory system includes multiple banks of non-volatile memory cells and a controller to manage the transfer of user data between a host and the memory system and also to manage the storage of user data from the host in the memory banks. The controller circuit includes a host interface, by which user data is transferred between the host and the controller circuit, an internal bus connected to the host interface, and multiple memory bank interfaces connected to the internal bus, whereby user data and management data is respectively transferred between the controller circuitry and a corresponding one of the memory banks. The controller circuitry also includes a volatile memory connected to the internal bus, in which the controller stores management data for use in managing the storage of user data, and processing circuitry connected to the internal bus, where the processing circuitry dedicates one of the memory banks for the storage of management data and stores user data on the other N memory banks.
Various aspects, advantages, features and embodiments of the present invention are included in the following description of exemplary examples thereof, whose description should be taken in conjunction with the accompanying drawings. All patents, patent applications, articles, other publications, documents and things referenced herein are hereby incorporated herein by this reference in their entirety for all purposes. To the extent of any inconsistency or conflict in the definition or use of terms between any of the incorporated publications, documents or things and the present application, those of the present application shall prevail.
The techniques presented here can provide a cost effective non-volatile memory system that also provides high performance. This is done by providing the controller with an auxiliary memory interface for use with an additionally non-volatile memory bank, where the additional memory bank and interface are used for metadata, such as logical to physical translation data, with the user data being kept on the other memory banks. To give a specific example, a non-volatile memory could include a controller and (N+1) NAND flash memories, where N of these memories would store user data, but the remaining memory with its own controller interface would be dedicated to the storage of metadata. This allows for the metadata to be kept in non-volatile memory, but still quite readily accessible relative to the typical paging/overlay arrangement for metadata that is typically used in many non-volatile memory system. Before describing an exemplary embodiment, discussed below with respect to
One approach is to store the translation tables and other control data structures (or “meta-data”) in the same memory banks as user data and then load it when the device is run into a DRAM memory that is external to the controller. This is illustrated in
Controller 121 is a block diagram of an exemplary controller with some of the relevant elements shown. The controller communicates with the host by way of host interface module 133 and the non-volatile memory banks through the corresponding memory interface module 123-0, . . . , 123-n. including a Flash Control RISC Processor (FR) and a Flash Protocol Sequencer (FPS). These interfaces are then connected along the controller's internal bus structure 131. The controller also has processing circuitry, with an exemplary embodiment including a main processor 151 and a data path manager 153, which is an independent processor integrated in the same controller. The data path manager manages the flow of data once it enters the controller through host interface, directing DMA (Direct Memory Access) engines to transfer data between the various internal memories and managing data overflows and data under-runs. There could be one or more instances of such processors, each dedicated to manage transfer of data between different controller memory modules. The controller usually will also have some amount of volatile SRAM memory to use for various purposes, here including a portion for storing code 141 it uses to operate, a data cache buffer portion 143 where the controller can buffer data begin transferred between the host 101 and the non-volatile memory banks 105-0, . . . , 105-n, and a portion 145 for caching management data or metadata, such as logical to physical address translation tables. The multiple instances of SRAM memory blocks in the controller core optimized for performance, power consumption and to better manage real estate (layout) of the controller core logic, although, for the purposes of this discussion, the SRAM memory can be considered as one logical entity. To address some of the problems described in the Background, the memory system of
Under the arrangement of
The arrangement of
An alternative arrangement, that can be used for storage applications that do not require high system performance, is illustrated with respect to
As shown in
The exemplary embodiment of
Relative to the DRAM approach of
In the exemplary embodiment, the memory banks are all taken be of the same type, but with one dedicated to metadata. The device used for metadata can be pre-selected, or determined by the controller when the system is first used, and in the most basic arrangement will be kept fixed, although it could also be rotated for wear leveling or data consolidation reasons. As system data often tends to be rewritten more frequently than user data, this could lead to more frequent writes on the dedicated metadata bank; but as these writes are often relatively small (as opposed the large writes of logically contiguous data write common in user data), and bank capacity can be quite large, this should not result in the lifetime of the metadata bank being a limiting factor for the system. If there are such lifetime concerns for the metadata bank, the controller could manage it differently than the user data banks based on the different data types (small, non-contiguous writes as opposed to large, logically contiguous writes) or even operate the metadata bank in a binary format, while using multi-state (MLC) storage for user data. (More detail on memory systems using both binary and MLC storage is found in U.S. patent application Ser. No. 12/640,820 filed on Dec. 17, 2009, and U.S. patent application Ser. Nos. 12/642,584, 12/642,740, 12/642,611 and 12/642,649, all filed on Dec. 18, 2009.) In other embodiments, the dedicated bank could, for example, be a Flash memory having a NOR architecture, allows the data to be accessed on a cell by cell basis, while the other banks use a NAND architecture.
Even though the arrangement of
The foregoing detailed description of the invention has been presented for purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise form disclosed. Many modifications and variations are possible in light of the above teaching. The described embodiments were chosen in order to best explain the principles of the invention and its practical application, to thereby enable others skilled in the art to best utilize the invention in various embodiments and with various modifications as are suited to the particular use contemplated. It is intended that the scope of the invention be defined by the claims appended hereto.
This application is a continuation of U.S. application Ser. No. 12/969,167, filed Dec. 15, 2010, which is incorporated herein in its entirety by this reference.
Number | Date | Country | |
---|---|---|---|
Parent | 12969167 | Dec 2010 | US |
Child | 14845089 | US |