NON-VOLATILE MEMORY AND METHOD WITH ADAPTIVE LOGICAL GROUPS

• US 20150186270A1
• Filed: 12/30/2013
• Published: 07/02/2015
• Est. Priority Date: 12/30/2013
• Status: Active Grant

• Current Assignee: SanDisk Technologies LLC (Western Digital Corporation)
• Original Assignee: Sandisk Technologies Incorporated (Western Digital Corporation)
• Inventors: Peng, Yong, Nagabhirava, Rajeev
• Granted Patent: US 9,466,383 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 12/0246 in block erasable memory, e.g. flash memory
• CPC Class Codes: G06F 12/10 Address translation
• CPC Class Codes: G06F 2212/2022 Flash memory
• CPC Class Codes: G06F 2212/7201 Logical to physical mapping or translation of blocks or pages
• CPC Class Codes: G06F 2212/7202 Allocation control and policies
• CPC Class Codes: G06F 2212/7205 Cleaning, compaction, garbage collection, erase control
• CPC Class Codes: G11C 16/16 for erasing blocks, e.g. arrays, words, groups

Abstract

A nonvolatile memory is organized into blocks as erase units and physical pages as read/write units. A host addresses data by logical pages, which are storable in corresponding physical pages. Groups of logical pages are further aggregated into ogical groups as addressing units. The memory writes host data in either first or second write streams, writing to respective blocks either logical-group by logical-group or logical-page by logical-page in order to reduce the size of logical-to-physical-address maps that are cached in a controller RAM. Only one block at a time needs be open in the second stream to accept logical pages from multiple logical groups that are active. Garbage collection is performed on the blocks from each write stream independently without data copying between the two streams.

Claims

1. A nonvolatile memory, comprising:

an array of memory cells organized into blocks as erase units and physical pages as read/write units, each physical page for storing a logical page of data, each logical page having a logical address assigned by a host;

a plurality of logical groups provided by partitioning a logical address space of the host into non-overlapping groups of sequential logical addresses, each logical group consisting a plurality of corresponding logical pages;

a controller, for writing data from a host write in either a first or second write stream;

a pool of erase blocks for allocating multiple blocks in the first write stream, each of the multiple blocks being open in parallel in the first write stream for writing logical group by logical group, and for allocating blocks sequentially in the second write stream, each of the blocks being open one at a time for writing logical page by logical page;

said controller, responsive to the data having at least a predetermined size or its logical addresses following sequentially a previous write in a partially filled block in the first stream, for writing the data to the first write stream, otherwise, for writing the data to the second write stream;

a group-level map for tracking logical groups stored among the plurality of blocks in the first write stream, said group-level map being maintained in said non-volatile memory with a copy cached in a RAM of said controller;

a page-level map for tracking obsolete pages among the logical groups written in the first write stream and updated pages of the obsolete pages written in the first and second write streams, said page-level map being maintained in said non-volatile memory with at least a portion of a copy cached in the RAM of said controller; and

said controller performing garbage collection independently on each of the first and second write streams without data being copied between each of the first and second write streams.

2. A nonvolatile memory as in claim 1, wherein:

a block containing obsolete and valid logical pages to be reclaimed in a given write stream, has the valid logical pages salvaged to another block of the same given write stream in a garbage collection before being erased and added to the pool of erase blocks.

3. The nonvolatile memory as in claim 1, wherein:

the predetermined size is such to be conducive to writing the blocks logical group by logical group in the first write stream without the need to allocate a new block beyond a predetermined frequency.

4. The nonvolatile memory as in claim 1, wherein the predetermined size is 64 KB.

5. The nonvolatile memory as in claim 1, wherein:

the data has a starting logical address that follow sequentially on a previous write in a partially filled block in the first stream; and

said controller has the data appended after the previous write in the partially filled block.

6. The nonvolatile memory as in claim 1, wherein:

the data does not have a starting logical address that follow sequentially on a previous write in a partially filled block in the first stream but is at least of the predetermined size; and

said controller has the data written to a new block allocated to the first stream.

7. The nonvolatile memory as in claim 1, wherein the memory cells each stores one bit of data.

8. The nonvolatile memory as in claim 1, wherein the memory cells each stores more than one bit of data.

9. The nonvolatile memory as in claim 1, wherein:

said group-level map includes a flag for each logical group to indicate whether or not the logical group contains an obsolete logical page.

10. The nonvolatile memory as in claim 8, wherein:

when the flag indicates that the logical group in a block contains an obsolete logical page, any obsolete logical pages in the logical group is located in the block by said page-level map.

11. A method of operating a non-volatile memory, comprising:

organizing the non-volatile memory into blocks of memory cells that are erasable together, each block containing a plurality of physical pages, each physical page containing a page of memory cells that are read or written in parallel and for storing a logical page of data, each logical page having a logical address assigned by a host;

providing a plurality of logical groups provided by partitioning a logical address space of the host into non-overlapping groups of sequential logical addresses, said plurality of logical groups further partitioned into a plurality of logical pages;

writing the data from a host write either to a first write stream or, otherwise, to a second write stream;

providing a pool of erase blocks for allocating multiple blocks in the first write stream, each of the multiple blocks being open in parallel in the first write stream for writing logical group by logical group, and for allocating blocks sequentially in the second write stream, each of the blocks being open one at a time for writing logical page by logical page;

responsive to the data having at least a predetermined size or its logical addresses following sequentially a previous write in a partially filled block in the first stream, for writing the data to the first write stream, otherwise, for writing the data to the second write stream;

tracking with a group-level map logical groups stored among the plurality of blocks in the first write stream, said group-level map being maintained in said non-volatile memory with a copy cached in a RAM of said controller;

tracking with a page-level map obsolete pages among the logical groups written in the first write stream and updated pages of the obsolete pages written in the first and second write streams, said page-level map being maintained in said non-volatile memory with at least a portion of a copy cached in the RAM of said controller; and

performing garbage collection independently on each of the first and second write streams without data being copied between each of the first and second write streams.

12. The method as in claim 11, further comprising:

reclaiming a block containing obsolete and valid logical pages in a given write stream in a garbage collection by salvaging the valid logical pages to another block of the same given stream;

erasing the block thereafter; and

adding the block to the pool of erase blocks.

13. The method as in claim 12, wherein:

the predetermined size is such to be conducive to writing the blocks logical group by logical group in the first write stream without the need to allocate a new block beyond a predetermined frequency.

14. The method as in claim 11, wherein the predetermined size is 64 KB.

15. The method as in claim 11, wherein:

the data has a starting logical address that follow sequentially on a previous write in a partially filled block in the first stream; and

said writing data of the host write appends the data after the previous write in the partially filled block.

16. The method as in claim 11, wherein:

the data does not have a starting logical address that follow sequentially on a previous write in a partially filled block in the first stream but is at least of the predetermined size; and

said writing data of the host write writes the data to a new block allocated to the first stream.

17. The method as in claim 11, wherein the memory cells each stores one bit of data.

18. The method as in claim 11, wherein the memory cells each stores more than one bit of data.

19. The method as in claim 11, wherein:

said first map includes a flag for each logical group to indicate whether or not the logical group contains an obsolete logical page.

20. The method as in claim 19, wherein:

when the flag indicates that the logical group in a block contains an obsolete logical page, any obsolete logical pages in the logical group is located in the block by said second map.