Granted Patent B2

US 10,879,940 · App. 16/407,671 · Granted Dec 29, 2020

Abstract

Claims (57)

1. A method for decoding data, comprising:

receiving, from a non-volatile memory, an encoded, mapped data file comprising an original data file mapped via a one-to-one mapping based on a characteristic of a data type of the original data file, and encoded;

decoding the encoded, mapped data file with a soft decoder and thereby obtaining a mapped data file;

de-mapping the mapped data file based on the characteristic of the data type of the original data file and thereby obtaining a recreation of the original data file; and

outputting the recreation of the original data file,

wherein the characteristic of the data type is one of a structure of the data type and a pattern of the data type.

2. The method of claim 1 , wherein:

the characteristic of the data type is an inherent redundancy of the data type; and

the de-mapping comprises reversing a transformation of the original data file that was mapped based on the inherent redundancy of the data type.

3. The method of claim 2 , wherein the de-mapping further comprises reversing the transformation of the original data file according to a probability of a data value to occur in the data type.

4. The method of claim 1 , wherein the decoding comprises transmitting learned information L in from the soft decoder to a content aware module, and transmitting information L out from the content aware module to the soft decoder.

5. The method of claim 4 , wherein the decoding further comprises the content aware module processing L in and additional information about the original data file to produce L out , wherein the additional information about the original data file is at least one of: information received from an offline source and information learned from a current estimation of the original data file.

6. The method of claim 1 , wherein the a size of the mapped data file and a size of the original data file are equal.

7. The method of claim 6 , the characteristic of the data type of the original data file is common to all data files of the data type.

8. A method, comprising:

mapping an original data file using a one-to-one mapping based on a characteristic of a data type of the original data file, thereby obtaining a mapped data file;

encoding the mapped data file, thereby obtaining an encoded, mapped data file;

storing the encoded, mapped data file in a non-volatile memory;

reading the encoded, mapped data file;

decoding the encoded, mapped data file using error correction, thereby obtaining a decoded, mapped data file; and

de-mapping the decoded, mapped data file, thereby obtaining a recreation of the original data file;

wherein the characteristic of the data type is one of a structure of the data type and a pattern of the data type.

9. The method of claim 8 , further comprising:

determining the one-to-one mapping based on an improvement in error correction of the decoding the encoded, mapped data file.

10. The method of claim 8 , further comprising:

determining the one-to-one mapping using an iterative process in which each iterative step improves improves mutual information of a content aware module's output (L out ) and the mapped data file.

11. The method of claim 8 , wherein:

the original data set is a non-uniformly distributed data set;

the mapping comprises: performing in a processor:

creating estimated data according to a distribution of a temporary underlying statistic based on a probable noise in the non-volatile memory;

calculating information L out based on the estimated data and feedback from a content aware module;

calculating mutual information between L out and the original data file before noise;

choosing a highest mutual information between L out and the original data file before noise; and

setting, as a map to be used to map the original data file, a map with a highest mutual information.

12. The method of claim 8 ,

wherein the original data file is a text file; and

wherein the mapping comprises: performing in a processor:

creating estimated data according to a distribution of a temporary underlying statistic based on a probable noise in the non-volatile memory;

calculating information L out based on the estimated data and feedback from a content aware module;

calculating mutual information between L out and the original data file before noise;

choosing a highest mutual information between L out and the original data file before noise; and

setting, as a map to be used to map the original data file, a map with a highest mutual information.

13. A memory controller for non-volatile memory, comprising:

a controller memory storing a plurality of data maps; and

a processor operatively coupled with the controller memory and configured to:

determine a data type of an original data file to be stored in nonvolatile memory based on a characteristic of the original data file;

select, from the plurality of data maps, a data map corresponding to the data type of the original data file;

map the original data file using the data map corresponding to the data type, thereby obtaining a mapped data file;

encode the mapped data file, thereby obtaining an encoded, mapped data file; and

store the encoded, mapped data file in the non-volatile memory;

wherein the characteristic of the original data file is one of a structure of the original data file and a pattern of the original data file.

14. The memory controller of claim 13 , wherein the processor is further configured to map a bijection of the original data file.

15. The memory controller of claim 13 , wherein the processor is further configured to: read the encoded, mapped data file from the non-volatile memory, decode, with a soft decoder, the encoded, mapped data file using an iterative content aware module, thereby obtaining the mapped data file, and de-map the decoded, mapped data file set using the data map corresponding to the data type, thereby obtaining a recreation of the original data file.

16. The memory controller of claim 15 , wherein the processor is further configured to feedback data characteristics learned from the soft decoder to the content aware module and, using the content aware module, process historical data characteristics for the data type with the feedback data characteristics.

17. The memory controller of claim 16 , wherein a size of the mapped data file and a size of the original data file are equal.

18. The memory controller of claim 13 , wherein each of the plurality of data maps is a data table that transforms each input byte of the original data file into an output byte of the mapped data file.

19. The memory controller of claim 18 , wherein the processor is further configured to store, in memory, the data type of the original data file, and select the data map corresponding to the data type using the stored data type.

Assignments (10)