# US 2020/0394104 A1 · Techniques for generating symbol-preserving error correction codes

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2020/0394104 A1
- Document ID: US20200394104A1
- Status: Published application
- Kind: A1
- Application number: 16/900,754
- Filed: 2020-06-12
- Published: 2020-12-17
- Earliest claimed: 2019-06-13
- Assignees: NVIDIA Corporation
- Applicants: Nvidia Corporation
- Inventors: John Brooks
- Web page: https://patentgrep.com/patent/US20200394104A1

## CPC

- G06F 11/1012: {using codes or arrangements adapted for a specific type of error (G06F11/1048 takes precedence)}
- G06F 11/1068: {in sector programmable memories, e.g. flash disk (G06F11/1072 takes precedence)}
- G06F 11/102: {Error in check bits}
- G06F 11/1044: {with specific ECC/EDC distribution}
- G06F 11/1096: {Parity calculation or recalculation after configuration or reconfiguration of the system}
- H03M 13/19: Single error correction without using particular properties of the cyclic codes, e.g. Hamming codes, extended or generalised Hamming codes
- H03M 13/3723: {using means or methods for the initialisation of the decoder}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- H03M: CODING; DECODING; CODE CONVERSION IN GENERAL (using fluidic means F15C4/00; optical analogue/digital converters G02F7/00; coding, decoding or code conversion, specially adapted for particular applications, see the relevant subclasses, e.g. G01D, G01R, G06F, G06T, G09G, G10L, G11B, G11C, H04B, H04L, H04M, H04N; ciphering or deciphering for cryptography or other purposes involving the need for secrecy G09C)

## Abstract

Various embodiments include an on-die error correction code (ECC) system that preserves rectangular symbols of arbitrary size and shape, where the dimensions of the symbol are powers of two. Further, the on-die ECC system preserves symbols that include multiple rectangles of arbitrary size and shape, where the dimensions of each rectangle are powers of two, and where the vertical and horizontal offset between consecutive rectangles are also powers of two. If the on-die ECC system miscorrects a memory bit, then the miscorrection is constrained or restricted to the same symbol that includes the other error bits. Therefore, all error bits, including the miscorrected bit, are in the same symbol. As a result, a user ECC system, such as a symbol-based ECC system, can correct and detect any number of errors within a single symbol, even when the on-die ECC system miscorrects a memory bit.

## Claims

### Claim 1

A computer-implemented method for generating error correction codes within an error correction code system, the method comprising:
generating values for a plurality of on-die error correction code (ECC) bits based on a plurality of data bits and a parity-check matrix;
determining, based on the parity-check matrix, that a first value stored in a first bit included in at least one of the plurality of on-die ECC bits or the plurality of data bits is incorrect, wherein the parity-check matrix restricts the first bit to a location within a symbol that includes a second bit, and wherein a second value stored in the second bit is incorrect; and
modifying the first value stored in the first bit.

### Claim 2 (depends on 1)

The computer-implemented method of claim 1, wherein the data bits comprise one or more of user data bits or user ECC bits.

### Claim 3 (depends on 1)

The computer-implemented method of claim 1, wherein:
the data bits comprise user data bits and user ECC bits,
a number of user data bits comprises a power of two,
a number of user ECC bits comprises a power of two, and
a sum of the number of user data bits and the number of user ECC bits is not a power of two.

### Claim 4 (depends on 1)

The computer-implemented method of claim 1, wherein:
the data bits comprise user data bits and user ECC bits, and
a number of user ECC bits is not a power of two.

### Claim 5 (depends on 1)

The computer-implemented method of claim 1, wherein:
the symbol comprises a group of memory bits included in the at least one of the plurality of on-die ECC bits or the plurality of data bits;
the symbol comprises a rectangular shape within a memory device that includes the at least one of the plurality of on-die ECC bits or the plurality of data bits; and
a first dimension of the symbol and a second dimension of the rectangular shape are powers of two.

### Claim 6 (depends on 1)

The computer-implemented method of claim 1, wherein:
the symbol comprises a group of memory bits included in the at least one of the plurality of on-die ECC bits or the plurality of data bits,
the symbol comprises a plurality of rectangular shapes within a memory device that includes the at least one of the plurality of on-die ECC bits or the plurality of data bits,
for each rectangular shape included in the plurality of rectangular shapes, a first dimension of the rectangular shape and a second dimension of the rectangular shape are powers of two, and
at least one of a vertical offset or a horizontal offset between a first rectangular shape included in the plurality of rectangular shapes and a second rectangular shape included in the plurality of rectangular shapes consecutive to the first rectangular shape comprises a power of two.

### Claim 7 (depends on 1)

The computer-implemented method of claim 1, wherein:
the parity-check matrix includes an on-die ECC section and a user section, and
no column of the on-die ECC section replicates any column of the user section.

### Claim 8 (depends on 1)

The computer-implemented method of claim 1, wherein the plurality of on-die ECC bits and the plurality of data bits are included in a high-bandwidth memory (HBM) device or a low power double data rate (LPDDR) memory device.

### Claim 9 (depends on 1)

The computer-implemented method of claim 1, wherein:
the first bit is included in the plurality of on-die ECC bits, and
the parity-matrix restricts the second bit to a location that is included in the plurality of on-die ECC bits.

### Claim 10 (depends on 1)

The computer-implemented method of claim 1, wherein:
the first bit is included in the plurality of data bits, and
the parity-matrix restricts the second bit to a location that is included in the plurality of data ECC bits.

### Claim 11

One or more non-transitory computer-readable media storing program instructions that, when executed by one or more processors, cause the one or more processors to perform steps of:
generating values for a plurality of on-die error correction code (ECC) bits based on a plurality of data bits and a parity-check matrix;
determining, based on the parity-check matrix, that a first value stored in a first bit included in at least one of the plurality of on-die ECC bits or the plurality of data bits is incorrect, wherein the parity-check matrix restricts the first bit to a location within a symbol that includes a second bit, and wherein a second value stored in the second bit is incorrect; and
modifying the first value stored in the first bit.

### Claim 12 (depends on 11)

The one or more non-transitory computer-readable media of claim 11, wherein the data bits comprise one or more of user data bits or user ECC bits.

### Claim 13 (depends on 11)

The one or more non-transitory computer-readable media of claim 11, wherein:
the data bits comprise user data bits and user ECC bits,
the user data bits, user ECC bits, and on-die ECC bits are arranged into a plurality of codewords, and
each codeword includes a user data array of eight rows by thirty-two columns, a user ECC array of eight rows by four columns, and an on-die ECC array of eight rows by two columns.

### Claim 14 (depends on 11)

The one or more non-transitory computer-readable media of claim 11, wherein:
the data bits and on-die ECC bits are arranged into a plurality of codewords, and
each codeword includes a data array of sixteen rows by sixteen columns and an on-die ECC array of one row by sixteen columns.

### Claim 15 (depends on 11)

The one or more non-transitory computer-readable media of claim 11, wherein:
the symbol comprises a group of memory bits included in the at least one of the plurality of on-die ECC bits or the plurality of data bits;
the symbol comprises a rectangular shape within a memory device that includes the at least one of the plurality of on-die ECC bits or the plurality of data bits; and
a first dimension of the symbol and a second dimension of the rectangular shape are powers of two.

### Claim 16 (depends on 11)

The one or more non-transitory computer-readable media of claim 11, wherein:
the symbol comprises a group of memory bits included in the at least one of the plurality of on-die ECC bits or the plurality of data bits,
the symbol comprises a plurality of rectangular shapes within a memory device that includes the at least one of the plurality of on-die ECC bits or the plurality of data bits,
for each rectangular shape included in the plurality of rectangular shapes, a first dimension of the rectangular shape and a second dimension of the rectangular shape are powers of two, and
at least one of a vertical offset or a horizontal offset between a first rectangular shape included in the plurality of rectangular shapes and a second rectangular shape included in the plurality of rectangular shapes consecutive to the first rectangular shape comprises a power of two.

### Claim 17 (depends on 11)

The one or more non-transitory computer-readable media of claim 11, wherein:
the parity-check matrix includes an on-die ECC section and a user section, and
no column of the on-die ECC section replicates any column of the user section.

### Claim 18

A system, comprising:
on-die error correction code (ECC) memory that includes a plurality of on-die ECC bits;
data memory that includes a plurality of data bits; and
an on-die ECC subsystem that:
generates values for the plurality of on-die ECC bits based on the plurality of data bits and a parity-check matrix,
determines, based on the parity-check matrix, that a first value stored in a first bit included in at least one of the plurality of on-die ECC bits or the plurality of data bits is incorrect, wherein the parity-check matrix restricts the first bit to a location within a symbol that includes a second bit, and wherein a second value stored in the second bit is incorrect, and
modifies the first value stored in the first bit.

### Claim 19 (depends on 18)

The system of claim 18, wherein the on-die ECC memory and the user memory are included in a high-bandwidth memory (HBM) device or a low-power double data rate (LPDDR) memory device.

### Claim 20 (depends on 18)

The system of claim 18, wherein the data memory comprises a user data memory that includes a first portion of the plurality of data bits and a user ECC memory that comprises a second portion of the plurality of data bits.
