Flash-memory system with enhanced smart-storage switch and packed meta-data cache for mitigating write amplification by delaying and merging writes until a host read

• US 8,452,912 B2
• Filed: 10/08/2009
• Issued: 05/28/2013
• Est. Priority Date: 10/11/2007
• Status: Expired due to Fees

• Current Assignee: Super Talent Technology, Corporation
• Original Assignee: Super Talent Electronics, Inc.
• Inventors: Lee, Charles C., Yu, Frank, Ma, Abraham C.
• Primary Examiner(s): Parikh, Kalpit
• Application Number:
  US12/576,216
• Publication Number:
  US 20100023682A1
• Time in Patent Office: 1,328 Days
• Field of Search: 711/103, 711/156, 711/143, 714/766, 710/54
• US Class Current: 711/103
• CPC Class Codes: G06F 12/0246 in block erasable memory, e.g. flash memory
• CPC Class Codes: G06F 12/0866 for peripheral storage systems, e.g. disk cache
• CPC Class Codes: G06F 2212/1036 Life time enhancement
• CPC Class Codes: G06F 2212/214 Solid state disk
• CPC Class Codes: G06F 2212/7203 Temporary buffering, e.g. using volatile buffer or dedicated buffer blocks
• CPC Class Codes: G06F 2212/7207 management of metadata or control data
• CPC Class Codes: G11C 13/0004 comprising amorphous/crystalline phase transition cells
• CPC Class Codes: G11C 16/102 External programming circuits, e.g. EPROM programmers; In-circuit programming or reprogramming; EPROM emulators
• CPC Class Codes: G11C 19/0808 using magnetic domain propagation
• CPC Class Codes: G11C 2211/5641 Multilevel memory having cells with different number of storage levels
• CPC Class Codes: G11C 2216/30 Reduction of number of input/output pins by using a serial interface to transmit or receive addresses or data, i.e. serial access memory

Abstract

A flash memory solid-state-drive (SSD) has a smart storage switch that reduces write acceleration that occurs when more data is written to flash memory than is received from the host. Page mapping rather than block mapping reduces write acceleration. Host commands are loaded into a Logical-Block-Address (LBA) range FIFO. Entries are sub-divided and portions invalidated when a new command overlaps an older command in the FIFO. Host data is aligned to page boundaries with pre- and post-fetched data filling in to the boundaries. Repeated data patterns are detected and encoded by compressed meta-data codes that are stored in meta-pattern entries in a meta-pattern cache of a meta-pattern flash block. The sector data is not written to flash. The meta-pattern entries are located using a meta-data mapping table. Storing host CRC'"'"'s for comparison to incoming host data can detect identical data writes that can be skipped, avoiding a write to flash.

Claims

1. A smart-storage switch solid-state-drive (SSD) comprising:

a Non-Volatile Memory (NVM) that retains data when power is lost;

a volatile memory that loses data when power is lost;

a smart storage switch which comprises;

a protocol converter, coupled to a host, for receiving host commands to access the NVM and for receiving host data and a host logical block address (LBA);

a LBA range first-in-first-out (FIFO) having entries for storing the host LBA and a length of the host data;

a volatile memory controller for interfacing to the volatile memory;

a flash controller for interfacing to the NVM;

a processor for controlling transfer of host data among the protocol converter, the LBA range FIFO, the volatile memory controller, and the flash controller;

a sector data buffer, in the volatile memory, for storing the host data before writing to the NVM; and

a lookup table, in the volatile memory, that maps pages in the host LBA to physical pages in a physical-block address (PBA) in the NVM;

wherein the lookup table is page-mapped;

wherein the processor operates to reduce writes to the NVM to reduce write acceleration;

wherein the LBA range FIFO comprises sub-divided entries that are generated by the processor when a new host command has new host data that overlaps host data for a prior entry in the LBA range FIFO, the processor dividing the prior entry into a plurality of sub-divided entries in the LBA range FIFO;

wherein sub-divided entries that are completely overlapped by new host data are invalidated and not written to the NVM;

whereby writes to the NVM are reduced by invalidating overlapping sub-divided entries in the LBA range FIFO.

2. The smart-storage switch solid-state-drive of claim 1 wherein the processor delays writing the host data for entries in the LBA range FIFO into the NVM until a host read is received by the protocol converter,

whereby host writes to NVM are delayed until the host read is received.

3. The smart-storage switch solid-state-drive of claim 2 wherein the processor pre-fetches data from the NVM and loads the pre-fetched data before a start of the host data when the start of the host data is not aligned to a boundary;

wherein the processor post-fetches data from the NVM and loads the post-fetched data after an end of the host data when the end of the host data is not aligned to a boundary,

whereby host data is aligned to boundaries by pre-fetching and post-fetching.

4. The smart-storage switch solid-state-drive of claim 3 wherein the boundary is a page boundary.

5. A smart-storage switch solid-state-drive (SSD) comprising:

a Non-Volatile Memory (NVM) that retains data when power is lost;

a volatile memory that loses data when power is lost;

a smart storage switch which comprises;

a protocol converter, coupled to a host, for receiving host commands to access the NVM and for receiving host data and a host logical block address (LBA);

a LBA range first-in-first-out (FIFO) having entries for storing the host LBA and a length of the host data;

a volatile memory controller for interfacing to the volatile memory;

a flash controller for interfacing to the NVM;

a processor for controlling transfer of host data among the protocol converter, the LBA range FIFO, the volatile memory controller, and the flash controller;

a sector data buffer, in the volatile memory, for storing the host data before writing to the NVM; and

a lookup table, in the volatile memory, that maps pages in the host LBA to physical pages in a physical-block address (PBA) in the NVM;

wherein the lookup table is page-mapped;

wherein the processor operates to reduce writes to the NVM to reduce write acceleration;

wherein the protocol converter further receives a host cyclical-redundancy-check (CRC) from the host;

further comprising;

a CRC generator that receives the host data from the protocol converter and generates a generated CRC from the host data;

wherein the protocol converter discards the host data when the host CRC mismatches the generated CRC;

wherein the host CRC or the generated CRC is stored in the volatile memory as a stored CRC;

wherein when the host LBA matches a cached LBA for a matching entry in the volatile memory, the stored CRC is read from the volatile memory and compared to the host CRC or generated CRC to signal a CRC match;

when the CRC match is signaled, comparing all bytes in the host data to stored bytes for the matching entry in the volatile memory and signaling a data match when all byte match;

when the data match is signaled, discarding the host data and not writing the host data to the volatile memory,

whereby writing duplicate host data to the NVM is avoided by comparison of the stored CRC.

6. The smart-storage switch solid-state-drive of claim 5 wherein the protocol converter is coupled to the host through a Serial AT-Attachment (SATA) bus, a Serial small-computer system interface (SCSI) (SAS) bus, a fiber-channel (FC) bus, an InfiniBand bus, an integrated device electronics (IDE) bus, a Peripheral Components Interconnect Express (PCIe) bus, a compact flash (CF) bus, a Universal-Serial-Bus (USB), a Secure Digital Bus (SD), a MultiMediaCard (MMC), or a LBA bus protocol which transfers read and write commands, a starting page address with a sector offset, and a sector count.

7. The smart-storage switch solid-state-drive of claim 5 wherein the NVM comprises a plurality of flash modules each comprising a plurality of flash memory devices.

8. A smart-storage switch solid-state-drive (SSD) comprising:

a Non-Volatile Memory (NVM) that retains data when power is lost;

a volatile memory that loses data when power is lost;

a smart storage switch which comprises;

a protocol converter, coupled to a host, for receiving host commands to access the NVM and for receiving host data and a host logical block address (LBA);

a LBA range first-in-first-out (FIFO) having entries for storing the host LBA and a length of the host data;

a volatile memory controller for interfacing to the volatile memory;

a flash controller for interfacing to the NVM;

a processor for controlling transfer of host data among the protocol converter, the LBA range FIFO, the volatile memory controller, and the flash controller;

a sector data buffer, in the volatile memory, for storing the host data before writing to the NVM; and

a lookup table, in the volatile memory, that maps pages in the host LBA to physical pages in a physical-block address (PBA) in the NVM;

wherein the lookup table is page-mapped;

wherein the processor operates to reduce writes to the NVM to reduce write acceleration;

a pattern detector for comparing an initial byte of the host data to subsequent bytes of the host data and signaling a pattern match when the initial byte is repeated for the subsequent bytes;

a meta-data code generator for generating a meta-data code from the initial byte when the pattern detector signals the pattern match; and

a meta-data mapping table storing meta-pointer entries;

wherein the processor loads an entry in the lookup table with a pointer to a meta-pointer entry in the meta-data mapping table when the pattern match is signaled;

wherein the meta-pointer entry identifies a meta entry in a meta-pattern block, the meta entry containing the meta-data code generated by the meta-data code generator.

9. The smart-storage switch solid-state-drive of claim 8 wherein the host data is not stored to the NVM when the pattern match is signaled, wherein the meta-data code is written to the NVM.

10. The smart-storage switch solid-state-drive of claim 9 wherein the meta-data codes are packed into a meta page that stores a plurality of meta-data codes, wherein writes to the NVM are reduced for pattern-matched host data.

11. The smart-storage switch solid-state-drive of claim 10 wherein the host data is a sector of 512 bytes;

wherein the meta-data code is 32 bytes or less in size.

12. The smart-storage switch solid-state-drive of claim 10 further comprising:

a meta-pattern cache that caches pages of the meta-pattern block;

wherein a plurality of the meta entries are loaded into a meta page in the meta-pattern cache before the meta page is written to the NVM,

wherein writes to the NVM are reduced by loading the meta page with a plurality of the meta entries each representing pattern-matched host data before the meta page is written to the NVM.

13. The smart-storage switch solid-state-drive of claim 12 wherein the meta-pattern cache is stored in the volatile memory.

14. The smart-storage switch solid-state-drive of claim 12 wherein the meta entry in the meta-pattern block further comprises the pointer to a meta-pointer entry in the meta-data mapping table and the host LBA.

15. The smart-storage switch solid-state-drive of claim 12 wherein the meta-pattern cache is periodically flushed to the meta-pattern block in the NVM.

16. The smart-storage switch solid-state-drive of claim 12 wherein the processor sets a meta flag in the lookup table when the pattern detector signals the pattern match, and loads the pointer to the meta-pointer entry in the meta-data mapping table;

wherein the processor clears the meta flag in the lookup table when the pattern match is not signaled, a loads an identifier to physical block and an identifier to a physical page into an entry in the lookup table, wherein the host data is stored at the physical page in the physical block,

whereby entries in the lookup table store physical mappings and pointers to the meta-data mapping table.

17. A computer-implemented method for reducing writes to flash memory in a flash-memory system comprising:

receiving host data for a host write command and storing the host data in a sector data cache at a location identified by a buffer pointer;

receiving the host write command from a host and loading a starting logical block address (LBA), a length of the host transfer in sectors, and the buffer pointer into a current entry in a LBA range FIFO;

detecting a range overlap of the current entry with a prior entry in the LBA range FIFO;

sub-dividing the prior entry into sub-divided entries in the LBA range FIFO when the range overlap is detected by dividing the prior entry at the starting LBA and at an end that is the starting LBA plus the length of the current entry;

invalidating a sub-divided entry that is completely overlapped by the current entry in the LBA range FIFO;

delaying writing the host data from the sector data cache to a flash memory until after a host read command is received from the host;

wherein the host tread command triggers writing to the flash memory.

18. The computer-implemented method for reducing writes to flash memory of claim 17 further comprising page aligning by:

detecting when the starting LBA in the current entry is a multiple of a page size;

pre-fetching data sectors from the flash memory that are before the starting LBA until a boundary that is a multiple of the page size is reached;

storing the data sectors that are prefetched in the sector data cache for the current entry;

calculating an ending LBA by adding the starting LBA to the length for the current entry;

detecting when the ending LBA in the current entry is a multiple of the page size;

post-fetching data sectors from the flash memory that are after the ending LBA until a boundary that is a multiple of the page size is reached;

storing the data sectors that are post-fetched in the sector data cache for the current entry.

19. A solid-state-drive (SSD) comprising:

Non-Volatile Memory (NVM) means for storing data that is retained when power is lost;

volatile memory means for storing data that is lost when power is lost;

a smart storage switch which comprises;

protocol converter means, coupled to a host, for receiving host commands to access the NVM means and for receiving host data and a host logical block address (LBA);

a LBA range first-in-first-out (FIFO) having entries for storing the host LBA and a length of the host data;

volatile memory controller means for interfacing to the volatile memory means;

flash controller means for interfacing to the NVM means;

processor means for controlling transfer of host data among the protocol converter means, the LBA range FIFO, the volatile memory controller means, and the flash controller means;

sector data buffer means, in the volatile memory means, for storing the host data before writing to the NVM means; and

lookup table means, in the volatile memory means, for mapping pages in the host LBA to physical pages in a physical-block address (PBA) in the NVM means;

wherein the lookup table means is page-mapped;

wherein the LBA range FIFO comprises sub-divided entries that are generated by the processor means when a new host command has new host data that overlaps host data for a prior entry in the LBA range FIFO, the processor means further for dividing the prior entry into a plurality of sub-divided entries in the LBA range FIFO;

wherein sub-divided entries that are completely overlapped by new host data are invalidated and not written to the NVM means;

wherein the processor means is further for delaying writing the host data for entries in the LBA range FIFO into the NVM means until a host read is received by the protocol converter means,

wherein the processor means operates to reduce writes to the NVM means to reduce write acceleration.

20. The solid-state-drive of claim 19 wherein the processor means further comprises:

pre-fetch means for pre-fetching data from the NVM means and for loading the pre-fetched data before a start of the host data when the start of the host data is not aligned to a boundary;

post-fetching means for post-fetching data from the NVM means and for loading the post-fetched data after an end of the host data when the end of the host data is not aligned to a boundary,

whereby host data is aligned to boundaries by pre-fetching and post-fetching.

21. The solid-state-drive of claim 19 wherein the protocol converter means further comprises:

vendor identifier means for reading a vendor identifier and a volume capacity and for sending the vendor identifier and the volume capacity to the host;

wherein a device driver in the host uses a Non-Volatile Memory Host Controller Interface (NVMHCI) interface to identify a register set based on the vendor identifier and the volume capacity;

wherein the device driver hierarchically configures registers based on the vendor identifier and the volume capacity read from the solid-state-drive.