System and method for correcting error of high bandwidth memory

The error correction system addresses the challenge of TSV-induced errors in high-bandwidth memory by employing a die controller and Hamming processing to replace zero blocks with metadata and check bits, enhancing error correction capabilities.

KR103047903B1Active Publication Date: 2026-11-11IND ACADEMIC COOP FOUND YONSEI UNIV
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
KR1020240179546
Authority / Receiving Office
KR · KR
Patent Type
Patents
Current Assignee / Owner
Filing Date
2024-12-05
Publication Date
2026-11-11
Estimated Expiration
2044-12-05

AI Technical Summary

Technical Problem

Existing high-bandwidth memory technologies lack effective methods to correct errors caused by Through Silicon Vias (TSVs).

Method used

An error correction system utilizing a die controller, zero flag generator, and Hamming processing unit to identify and correct errors by replacing zero blocks with metadata and Hamming check bits, enhancing error correction capabilities.

Benefits of technology

The system effectively corrects errors caused by TSVs, improving the reliability of high-bandwidth memory systems by utilizing zero blocks for error correction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure R1020240179546_ABST
    Figure R1020240179546_ABST
Patent Text Reader

Abstract

An error correction system for high-bandwidth memory is disclosed. The error correction system for high-bandwidth memory according to the present invention includes: a die controller that searches for a zero block of a data word among a first code word comprising a data word and metadata for the data word; a zero flag generator that generates zero flag information, which is information regarding the existence and location of a zero block, when one or more zero blocks exist; and a Hamming processing unit that generates a Hamming check bit for at least a portion of the data word and the zero flag information.
Need to check novelty before this filing date? Find Prior Art

Description

Technology Field

[0001] The present invention relates to an error correction system and method for high-bandwidth memory. Background Technology

[0003] The demand for high-bandwidth memory is increasing due to the advancement of artificial intelligence (AI) technology, as deeper neural networks require more computation and larger feature maps. While high-bandwidth memory utilizes I / O interfaces that support high bandwidth and 3D stacking technology, each die is susceptible to errors due to its high density. To address this, On-die Error Correction Code (On-die ECC) is used to correct soft errors occurring on each die.

[0004] The standard specification for high-bandwidth memory HBM3 uses a 304-bit codeword with a 32-bit on-die ECC check bit added to 272-bit codeword data including a CRC-16 check bit.

[0005] Conventional on-die ECC technology used in the HBM3 specification is Reed Solomon code (RS) based on on-die ECC, using one 16-bit RS (19,17) engine or two 8-bit RS (19,17) engines.

[0006] When using a single 16-bit based engine, the original dataword can be recovered only if the error in the DRAM is contained in a single 16-bit symbol.

[0007] Existing technologies have focused on enhancing the reliability of on-die ECC and approaches for each die, but approaches to correcting errors caused by Through Silicon Via (TSV) are lacking.

[0008] delete Prior art literature

[65535] Published Patent Application KR10-2018-0059149 Published Patent Application KR10-2020-0130113 Published Patent Application KR10-2023-0008706 Paper, Configurable-ECC: Architecting a Flexible ECC Scheme to Support Different Sized Accesses in High Bandwidth Memory Systems / Hsing-Min Chen, et al. / IEEE Transactions on Computers, vol. 68, no. 5, pp. 646-659, 1 May 2019. The problem to be solved

[0009] The technical problem that the present invention aims to solve is to provide an error correction system and method for high-bandwidth memory capable of correcting even errors caused by TSVs by utilizing zero blocks. means of solving the problem

[0011] To solve the above technical problem, an error correction system for a high-bandwidth memory according to an embodiment of the present invention comprises: a die controller that searches for a zero block of a data word among a first code word including a data word and metadata for the data word; a zero flag generator that generates zero flag information, which is information regarding the existence and location of a zero block when one or more zero blocks exist; and a Hamming processing unit that generates a Hamming check bit for at least a part of the data word and the zero flag information; wherein the die controller can generate a second code word in which the zero block within the data word among the first code words is replaced by the metadata and the Hamming check bit.

[0012] In one embodiment of the present invention, the second codeword is the metadata replaced with the zero flag information, and the die controller can transmit the second codeword to the memory die.

[0013] In one embodiment of the present invention, the memory die includes an RS engine, which is a Reed-Solomon-based error correction module; and a zero masking unit; wherein the zero masking unit can mask the zero block where the symbol error occurred as zero when a symbol error derived by the RS engine occurs in any one of the zero blocks.

[0014] In one embodiment of the present invention, the die controller can correct an error in a codeword transmitted from a memory die using the Hamming check bit.

[0015] In one embodiment of the present invention, the die controller can restore the zero block by zero-masking after error correction using the Hamming check bit, and restore the first codeword by replacing the zero flag information with the metadata.

[0016] In one embodiment of the present invention, the first codeword includes metadata that is a 256-bit data word and a 16-bit Cyclic Redundancy Check (CRC) check bit, and the die controller searches for a 16-bit zero block in the data word area for each of the first codewords divided into two, and the zero flag information includes a zero mode 1 bit and a zero position 8 bits, and when there is only one zero block, the Hamming check bit can be generated with respect to the upper 95-bit data word of the divided first codeword and the 9-bit zero flag information.

[0017] In one embodiment of the present invention, the first codeword includes metadata that is a 256-bit data word and a 16-bit Cyclic Redundancy Check (CRC) check bit, and the die controller searches for a 16-bit zero block in the data word area for each of the first codewords divided into two, and the zero flag information includes a zero mode 1 bit and a zero position 8 bits, and if there are two or more zero blocks, the Hamming check bit may include a first Hamming check bit generated for the upper 68-bit data word of the divided first codeword, and a second Hamming check bit generated for the lower 60-bit data word and 9-bit zero flag information.

[0018] In addition, to solve the above technical problem, an error correction method for a high-bandwidth memory according to an embodiment of the present invention may be performed by an error correction system for a high-bandwidth memory comprising a die controller, a zero flag generator, and a Hamming processing unit, and may include the steps of: the die controller receiving a first codeword comprising a data word and metadata for the data word; the die controller searching for a zero block of the data word; if one or more zero blocks exist, the zero flag generator generating zero flag information which is information regarding the existence and location of the zero block; the Hamming processing unit generating a Hamming check bit for at least a part of the data word and the zero flag information; and the die controller generating a second codeword in which the zero block within the data word among the first codeword is replaced with the metadata and the Hamming check bit. Effects of the invention

[0020] The present invention has the effect of being able to correct even errors caused by TSV by utilizing zero blocks. Brief explanation of the drawing

[0022] FIG. 1 shows an error correction system for a high-bandwidth memory according to an embodiment of the present invention. FIG. 2 shows an error correction system for high-bandwidth memory and its operation according to an embodiment of the present invention. FIG. 3 illustrates an error correction method for a high-bandwidth memory according to an embodiment of the present invention. FIG. 4 illustrates an error correction method for a high-bandwidth memory according to an embodiment of the present invention. FIG. 5 illustrates an error correction method for a high-bandwidth memory according to an embodiment of the present invention. Specific details for implementing the invention

[0023] The present invention is capable of various modifications and may have various embodiments, and specific embodiments are illustrated in the drawings and described in detail. However, this is not intended to limit the present invention to specific embodiments, and it should be understood that it includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the invention.

[0024] In describing the present invention, if it is determined that a detailed description of related known technology may obscure the essence of the present invention, such detailed description is omitted.

[0025] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

[0027] FIG. 1 shows an error correction system for a high-bandwidth memory according to an embodiment of the present invention.

[0028] FIG. 2 shows an error correction system for high-bandwidth memory and its operation according to an embodiment of the present invention.

[0029] With reference to FIGS. 1 and FIGS. 2, the configuration and operation of an error correction system for high-bandwidth memory will be described.

[0030] An error correction system for a high-bandwidth memory according to one embodiment of the present invention includes a zero flag generator (110), a Hamming processing unit (120), a die controller (130), an RS engine (210), and a zero masking unit (220).

[0031] An error correction system for high-bandwidth memory according to one embodiment of the present invention may be applied, in particular, during write and read operations for high-bandwidth memory of the HBM3 standard. However, the concept of the present invention is not necessarily limited to operations according to the HBM3 standard.

[0032] A high-bandwidth memory comprises a base die (100) and a plurality of memory dies (200), and data transfer between the base die (100) and each memory die (200) utilizes a Through Silicon Via (TSV).

[0033] A zero flag generator (110), a Hamming processing unit (120), and a die controller (130) can be installed on a base die (100), and an RS engine (210) and a zero masking unit (220) can be installed on each memory die (200).

[0034] Meanwhile, the operation of the present invention does not necessarily have to be performed by the same components. For example, the operation of the die controller (130) may be handled by the zero flag generator (110).

[0035] The die controller (130) receives a first codeword containing a data word and metadata from the memory controller (50).

[0036] A data word is 256 bits of DQ data that is actually stored in memory. Metadata can be 16 bits of Cyclic Redundancy Check (CRC) check bits for the 256-bit data word.

[0037] The die controller (130) determines whether the data word contains a zero block. At this time, the first codeword may be divided into two parts: 128 bits and 8 bits of metadata. The divided codeword may be referred to as the first-1 codeword and the first-2 codeword. Dividing the first codeword into the above units may be because, when using the RS engine (19, 17), which is a Reed Solomon-based Error Correction Code engine, ECC error checking for the first-1 codeword and the first-2 codeword is scheduled to be performed by two 8-bit based RS engines (19, 17), respectively. The two RS engines (19, 17) can independently correct 8-bit error symbols. 19 is the total length of the codeword, and 17 is the length of the message. Two are used for error correction.

[0038] In the following, unless otherwise specified, the descriptions will be understood as explanations of divided codewords.

[0039] The die controller (130) determines whether a zero block is included in the first-1 codeword and the first-2 codeword, respectively. The search for a zero block may be to determine whether there is a zero block that is consecutive in 16 bits within a data block separated into 8-bit units. In FIG. 2, an 8-bit block is represented as a single square. The zero block is hatched as a diagonal line, the Hamming check bit as a wave, and the CRC-16 check bit as a brick shape.

[0040] The die controller (130) generates a second codeword that modifies the first codeword. The second codeword includes modified data for the zero block and zero flag information.

[0041] The zero flag generator (110) generates zero flag information, which is information regarding whether a zero block exists and the location of the zero block. The upper 1 bit of the zero flag information is information regarding whether a zero block exists; if it is 1, a zero block exists, and if it is 0, a zero block does not exist.

[0042] Zero flag information is 9 bits of information, where the upper 1 bit indicates whether a zero block exists and the lower 8 bits indicate the location of the zero block. However, if there is no zero block, 8 bits of metadata are written to the lower 8 bits.

[0043] The Hamming processing unit (120) generates a Hamming check bit for at least part of the data word.

[0044] The Hamming processing unit (120) does not generate a Hamming check bit when there is no zero block.

[0045] The Hamming processing unit (120) generates a Hamming check bit for the upper 95 bits of data word (DQ data) and 9 bits of zero flag information (total 104 bits) when there is only one zero block.

[0046] When generating Hamming check bits for 104 bits using Hsiao code for hardware optimization, the following conditions must be met.

[0047] 1) There are no columns in the H-matrix that are all zeros.

[0048] 2) All columns of the H-matrix must be different

[0049] 3) Each column has an odd number of 1s

[0050] 4) The number of 1s in the rows of the H-matrix must be the same.

[0051] To satisfy the above conditions, a Hamming check bit of up to 104 bits can be generated. In this case, a 1-bit error can be corrected. The error to be corrected by the Hamming check bit may be a TSV error.

[0052] When there are two or more zero blocks, the Hamming processing unit (120) generates a Hamming check bit for the upper 68-bit data word, the lower 60-bit data word, and the 9-bit zero flag information, respectively. In this case, a 2-bit error can be corrected. The error to be corrected by the Hamming check bit may be a TSV error.

[0053] The die controller (130) may replace some of the zero blocks of the first codeword with metadata and replace other parts with Hamming check bits. Additionally, the 8-bit metadata of the first codeword may be replaced with 9-bit zero flag information. Specifically, the first-1 codeword is converted into a second-1 codeword containing 128-bit data words and 9-bit zero flag information, and the first-2 codeword is converted into a second-2 codeword containing 128-bit data words and 9-bit zero flag information. The data word of the second codeword may include a block in which the zero blocks are replaced with Hamming check bits and metadata.

[0054] In the example of FIG. 2, the data word of the 1-1 codeword includes two zero blocks, and the data word of the 1-2 codeword includes one zero block.

[0055] The base die (100) generates Hamming check bits for the upper data word of 68 bits and the lower data word of 60 bits and the 9-bit zero flag information of the first-1 codeword, respectively. The base die (100) replaces the upper 8 bits of the first zero block with metadata, replaces the lower 8 bits with one of the two generated Hamming check bits, and replaces the lower 8 bits of the second zero block with the other of the two generated Hamming check bits. The upper 8 bits of the second zero block are left as they are.

[0056] The base die (100) generates 9-bit zero flag information for the first-1 codeword. The zero flag information can be 11010000. This means that zero blocks exist in the first and third 16-bit blocks.

[0057] The base die (100) generates a second-1 codeword containing 128 bits of data and 9 bits of zero flag information, in which some of the zero blocks of the first-1 codeword are replaced with Hamming check bits and metadata, and transmits it to the memory die (200).

[0058] The base die (100) generates a Hamming check bit for the upper data word of the 96-bit first-second codeword and the 8-bit metadata. The base die (100) replaces the upper 8 bits of the zero block with the metadata and replaces the lower 8 bits with the generated Hamming check bit.

[0059] The base die (100) generates 9-bit zero flag information for the first-second codeword. The zero flag information can be 10000001. This means that there is a zero block in the 8th 16-bit block.

[0060] The base die (100) generates a second-second codeword containing 128 bits of data and 9 bits of zero flag information in which the zero block of the first-second codeword is replaced with a Hamming check bit and metadata, and transmits it to the memory die (200).

[0061] Generating a second codeword and transmitting it to the memory die (200) may mean a write operation to the memory.

[0062] Each memory die (200) may include two RS engines (210). The memory die (200) may also include a zero masking portion (220).

[0063] The RS engine (210) generates a third codeword by adding a 16-bit ECC check bit to each of the second codewords during a write operation. The 2-1 codeword and the 2-2 codeword are stored as the 3-1 codeword and the 3-2 codeword, each containing 128-bit DQ data, 9-bit zero flag information, and a 16-bit ECC check bit.

[0064] The RS engine (210) corrects errors in the 128 bits and 136 bits for the 8-bit zero flag information, excluding the 1-bit information indicating whether a zero block exists in the 3-1 codeword and the 3-2 codeword, respectively, during a read operation.

[0065] After error correction, the memory die (200) transmits the 2-1 codeword and 2-2 codeword, which are 137 bits (128 bits + 9 bits) excluding the ECC check bit, to the base die (100).

[0066] Meanwhile, if there are two or more zero blocks, the zero masking unit (220) can correct the symbol error by masking the symbol to zero when a symbol error occurs at a zero block location that has not been replaced by a Hamming check bit or metadata. Depending on the error location, when the symbol error is masked to zero, two symbol errors can be reduced to one symbol error, thereby enhancing the performance of the RS engine.

[0067] When reading, the 2-1 codeword and the 2-2 codeword are transmitted from the memory die (200) to the base die (100). The 2-1 codeword and the 2-2 codeword may each have errors corrected by the RS engine (210) or zero block errors corrected by the zero masking unit (220).

[0068] The die controller (130) corrects up to 2 bit errors using the Hamming check bit included in the second codeword. The error may be a TSV transmission error.

[0069] In the example of FIG. 2, the die controller (130) corrects a 2-bit error in the 128-bit data word and 9-bit zero flag information using two Hamming check bits included in the 2-1 codeword.

[0070] The die controller (130) uses one Hamming check bit included in the second-2 codeword to correct a 1-bit error in the upper 95-bit data word and 9-bit zero flag information.

[0071] When correcting errors, the position of the Hamming check bit can refer to the zero flag information.

[0072] After correcting the TSV error, the die controller (130) restores the original data word by masking the zero block of the second codeword. Simultaneously, sequentially, or out of order, the die controller (130) restores the 9-bit zero flag information as metadata.

[0073] Specifically, the die controller (130) masks both zero blocks of the 2-1 codeword with zeros and restores the 1-1 codeword by restoring the metadata written to the upper 8 bits of the first zero block. The die controller (130) masks one zero block of the 2-2 codeword with zeros and restores the 1-2 codeword by restoring the metadata written to the upper 8 bits of the zero block.

[0074] The die controller (130) combines the first-1 codeword and the first-2 codeword to transmit the first codeword, which includes a 256-bit data word and 16-bit CRC check data, to the memory controller (50).

[0076] FIGS. 3 to 5 illustrate an error correction method for a high-bandwidth memory according to an embodiment of the present invention. The error correction method for a high-bandwidth memory according to an embodiment of the present invention is described as being implemented by an error correction system for a high-bandwidth memory according to an embodiment of the present invention.

[0077] Figure 3 shows the operation of the base die (100) when writing to memory.

[0078] Referring to FIG. 3, in step S101, the base die (100) receives a first codeword containing a 256-bit data word and 16-bit metadata. However, the description here is limited to the first-1 codeword containing a divided 128-bit data word and 8-bit metadata.

[0079] In step S102, the base die (100) determines whether there is a zero block in the data word of the first-1 codeword. If there is, proceed to step S104, and if there is no block, proceed to step S103.

[0080] In step S103, the base die (100) generates a second-1 codeword containing zero flag information recorded as 1-bit zero mode = 0. At this time, the second-1 codeword includes a 128-bit data word and 9-bit zero flag information, and the zero flag information includes 1-bit zero mode data and 8-bit metadata.

[0081] In step S104, the base die (100) determines whether the number of zero blocks is one or two or more. If it is one, proceed to step S105, and if it is not one, proceed to step S109.

[0082] In step S105, the base die (100) stores 8 bits of metadata, specifically CRC-16 check bits, in the upper 8 bits of the zero block.

[0083] In step S106, the base die (100) replaces the metadata with zero flag information containing position data indicating a 1-bit zero mode = 1 and an 8-bit zero block position.

[0084] In step S107, the base die (100) generates 95 bits of DQ data and a Hamming check bit for 9 bits of zero flag information.

[0085] In step S108, the base die (100) stores the Hamming check bit generated in the lower 8 bits of the zero block. Then, proceed to step S116.

[0086] When there is one zero block, the 2-1 codeword contains 128-bit DQ data in which the zero block is converted into metadata and a Hamming check bit, and 9-bit zero flag information.

[0087] In step S109, if the number of zero blocks is 2 or more, proceed to step S109.

[0088] Steps S110 through S115 describe the case where the number of zero blocks is 2, and when the number of zero blocks is 3 or more, the second zero block is processed in the same way, but the zero blocks from the third zero block onwards are maintained as they are.

[0089] In step S110, the base die (100) stores the CRC bits in the upper 8 bits of the first zero block.

[0090] In step S111, the base die (100) replaces the metadata with zero flag information containing position data indicating a 1-bit zero mode = 1 and an 8-bit zero block location. The 2-1 codeword contains 128-bit DQ data in which the zero block is converted into metadata and a Hamming check bit, and 9-bit zero flag information.

[0091] In step S112, the base die (100) generates a first Hamming check bit for the upper 68 bits of the DQ data.

[0092] In step S113, the base die (100) stores the first Hamming check bit generated in the lower 8 bits of the first zero block.

[0093] In step S114, the base die (100) generates a second Hamming check bit for the lower 60 bits of DQ data and 9 bits of zero flag information.

[0094] In step S115, the base die (100) stores the second Hamming check bit in the second zero block lower 8 bits.

[0095] If there are two or more zero blocks, the 2-1 codeword includes 128-bit DQ data in which the first zero block is converted into metadata and a first Hamming check bit, and the lower 8 bits of the second zero block are converted into a second Hamming check bit, and 9-bit zero flag information.

[0097] In step S116, the base die (100) transmits the 2-1 codeword to the memory die (200).

[0099] FIG. 4 illustrates the operation of the memory die (200) when reading from memory. In the write operation, it is assumed that the memory die (200) additionally stores a 16-bit ECC check bit in the 2-1 codeword.

[0100] Referring to FIG. 4, in step S201, the memory die (200) inputs the 128-bit DQ data, 8-bit zero flag information excluding zero mode information, and 16-bit ECC check bits from the 2-1 codeword to the RS engine (210).

[0101] Zero mode information is used to determine branches and may not be used for error checking.

[0102] In step S202, the memory die (200) determines whether 1-bit zero mode = 1. If it is 1, proceed to step S203, and if it is 0, proceed to step S206.

[0103] In step S203, the memory die (200) determines whether the number of zero blocks is 1. If it is 1, proceed to step S206, and if it is not 1, proceed to step S204.

[0104] In step S204, determine whether the number of zero blocks is 2 or more. If it is 2 or more, proceed to step S205.

[0105] In step S205, the memory die (200) masks zero blocks to zero based on zero flag information, excluding areas replaced by metadata or Hamming check bits. For example, the first zero block of the second zero block may be masked to zero. For example, if there is a fourth zero block, all zero blocks may be masked to zero.

[0106] Step S205 has the effect of correcting errors when an error occurs in a zero block replaced by metadata or a Hamming check bit. Therefore, even if an error of 2 symbols or more occurs implicitly, if it is resolved by Step S205 and reduced to a 1 symbol error, error correction can be performed by the RS engine (210).

[0107] In step S206, the RS engine (210) corrects errors in units of 8-bit symbols for 136-bit data containing 128-bit DQ data and 8-bit zero flag information. RS (19, 17) can correct errors in the case of a 1-symbol error. As described, in step S205, if one or more symbols of errors are resolved by zero masking, there may be cases where two or more symbols of errors can also be corrected.

[0108] In step S207, the memory die (200) transmits the error-corrected 2-1 codeword to the base die (100). At this time, a TSV error may occur.

[0110] Figure 5 shows the operation of the base die (100) when reading from memory.

[0111] Referring to FIG. 5, in step S301, the memory die (200) outputs a second-1 codeword containing 128-bit DQ data and 9-bit zero flag information to the base die (100).

[0112] In step S302, the base die (100) determines whether the 1-bit zero mode = 1. If it is 1, proceed to step S303, and if it is not 1, proceed to step S312.

[0113] In step S303, the base die (100) determines whether the number of zero blocks is 1. If it is 1, proceed to step S304, and if it is not 1, proceed to step S305.

[0114] In step S304, Hamming decoding is performed using the Hamming check bits for the upper 94 bits and 9 bits of zero flag information of the DQ data. Then, proceed to step S308.

[0115] In step S305, determine whether the number of zero blocks is 2 or more. If it is 2 or more, proceed to step S306.

[0116] In step S306, the base die (100) performs Hamming decoding using the first Hamming check bit for the upper 68 bits of the DQ data.

[0117] In step S307, the base die (100) performs Hamming decoding using a second Hamming check bit for the lower 60 bits and 9 bits of zero flag information of the DQ data.

[0118] In step S308, it is determined whether an error exists in the result of Hamming decoding. If an error exists, proceed to step S309; ​​if no error exists, proceed to step S310.

[0119] In step S309, the base die (100) corrects errors using a Hamming check bit.

[0120] In step S310, the base die (100) restores the zero block by masking it to 0 using the 8-bit zero flag information written at the metadata location in the first-1 codeword.

[0121] In step S311, the base die (100) restores the 1-1 codeword by writing the original metadata to the metadata location of the 1-1 codeword and restoring it.

[0122] The base die (100) combines the restored first-1 codeword and first-2 codeword to restore a first codeword consisting of 256-bit DQ data and 16-bit metadata (CRC check bit).

[0123] In step S312, the base die (100) transmits a first codeword containing restored 256-bit DQ data and 16-bit metadata to the host.

[0125] The terms used in this application are used merely to describe specific embodiments and are not intended to limit the invention. In this application, terms such as “comprising” or “having” are intended to indicate the presence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof. Explanation of the symbols

[0127] 50: Memory controller 100: Base Die 110: Zero Flag Generator 120: Hamming processing unit 130: Die Controller 200: Memory die 210: RS engine 220: Zero Masking Section

Claims

Claim 1 An error correction system for a high-bandwidth memory comprising: a die controller that searches for a zero block of a data word among a first codeword including a data word and metadata for the data word; a zero flag generator that generates zero flag information regarding the existence and location of a zero block, wherein if one or more zero blocks exist, the zero flag information regarding the location of the zero block is generated; and a Hamming processing unit that generates a Hamming check bit for at least a portion of the data word and the zero flag information; wherein the die controller generates a second codeword in which the zero block within the data word among the first codeword is replaced by the metadata and the Hamming check bit. Claim 2 An error correction system for a high-bandwidth memory according to claim 1, wherein the second codeword is the metadata replaced with the zero flag information, and the die controller transmits the second codeword to the memory die. Claim 3 An error correction system for a high-bandwidth memory according to claim 2, wherein the memory die comprises an RS engine, which is a Reed-Solomon-based error correction module; and a zero masking unit; wherein the zero masking unit masks the zero block where the symbol error occurred to zero when a symbol error derived by the RS engine occurs in any one of the zero blocks. Claim 4 An error correction system for a high-bandwidth memory according to claim 2, wherein the die controller corrects an error in the second codeword transmitted from the memory die using the Hamming check bit. Claim 5 An error correction system for a high-bandwidth memory according to claim 4, wherein the die controller restores the data word by zero-masking the zero block after error correction using the Hamming check bit, and restores the first code word by replacing the zero flag information with the metadata. Claim 6 An error correction system for a high-bandwidth memory according to claim 1, wherein the first codeword includes metadata consisting of a 256-bit data word and a 16-bit Cyclic Redundancy Check (CRC) check bit, the die controller searches for a 16-bit zero block in the data word area for each of the first codewords divided into two, the zero flag information includes a zero mode bit and a zero position bit of 8 bits, and when there is only one zero block, the Hamming check bit is generated with respect to the upper 95-bit data word of the first codeword divided into two and the 9-bit zero flag information. Claim 7 An error correction system for a high-bandwidth memory according to claim 1, wherein the first codeword includes metadata consisting of a 256-bit data word and a 16-bit Cyclic Redundancy Check (CRC) check bit, the die controller searches for a 16-bit zero block in the data word area for each of the first codewords divided into two, the zero flag information includes a zero mode bit and a zero position bit of 8 bits, and if there are two or more zero blocks, the Hamming check bit includes a first Hamming check bit generated for the upper 68-bit data word of the first codeword divided into two, and a second Hamming check bit generated for the lower 60-bit data word and 9-bit zero flag information. Claim 8 A method for correcting errors in a high-bandwidth memory, comprising a die controller, a zero flag generator, and a Hamming processor, wherein the die controller receives a first codeword including a data word and metadata for the data word; the die controller searches for a zero block of the data word; the zero flag generator generates zero flag information which is information regarding the existence and location of a zero block, wherein if one or more zero blocks exist, the zero flag generator generates zero flag information which is information regarding the location of the zero block; the Hamming processor generates a Hamming check bit for at least a portion of the data word and the zero flag information; and the die controller generates a second codeword in which the zero block within the data word among the first codeword is replaced by the metadata and the Hamming check bit.

Citation Information

Patent Citations

  • Non-volatile memory error correction system and method

    KR1020090028507A

  • Data Processing System, Memory Controller and Operating Method Thereof

    KR1020210124688A

  • Apparatus and method for error correction code

    KR1020240022905A

  • Use of error correcting code to carry additional data bits

    US20160365876A1

  • Multi-page parity protection with power loss handling

    US20200210280A1