Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 3/9/2026 has been entered.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1-3 and 7-8 are rejected under 35 U.S.C. 103 as being unpatentable over Chang et al., US Patent Application Publication Number 20240021264 (herein “CHANG”) in view of You, US Patent Application Publication Number 20100161880 (herein “YOU”), further in view of Li et al., US Patent Application Publication Number 20230315330 (herein “LI”). “Program/erase Cycling Endurance and Data Retention of Macronix SLC NAND Flash Memories” (herein “Macronix”) is cited as an evidentiary reference for program erase cycles being a measure of data retention degradation.
Regarding independent claim 1, CHANG discloses a method (FIGs. 4-8, methods) comprising:
determining, using a machine learning model, reliability characteristic data associated with memory cells of a non-volatile memory device (FIG. 1, [0037], “a memory system 110 may be or include …, or a non-volatile DIMM (NVDIMM)”, e.g. non-volatile memory device. FIG. 4, [0053]-[0061], “the features (denoted xi) used by the machine learning model to predict read window values at time t1 (denoted y) may include read windows from a Vt distribution curve as described herein…. At 415, the machine learning model may split the training dataset by word line group (WLG).” E.g. using machine learning model to train /determine read windows /Vts / reliability characteristics of memory cells in wordlines groups);
grouping, based on the reliability characteristics regarding data retention degradation, a first portion of the memory cells of the non-volatile memory device in a first management group and a second portion of the memory cells of the non-volatile memory device in a second management group (FIG. 4, [0053]-[0061], “the features (denoted xi) used by the machine learning model to predict read window values at time t1 (denoted y) may include read windows from a Vt distribution curve as described herein…. At 415, the machine learning model may split the training dataset by word line group (WLG).” E.g. using machine learning model to train /determine read windows /Vts / reliability characteristics of memory cells in wordlines groups) Paragraph 0042 shows that the threshold voltage shift and read window compression experienced by a set of memory cells varies with the number of erase operations performed on the cells. Macronix beginning page 1 in the section titled Cycling Endurance and Data Retention explains that data retention time is a function of Program Erase cycles, as the dielectric layers of the floating gate in the cell may suffer degradation over time as the number of Program Erase cycles increases. Accordingly, an erase count is a characteristic regarding data retention degradation.; and
managing, based on the reliability characteristic data, scanning or logical to physical mapping of the first management group of memory cells and the second management group of memory cells (FIG. 5, [0064], “At 505, the memory system may determine that a trigger condition for performing read window management has been satisfied. In some examples, the trigger condition may be the set of memory cells reaching a threshold quantity of PECs.” E.g. perform background scan of read windows, based on program/erase cycle counts / reliability characteristics of groups).
CHANG does not explicitly disclose grouping, based on the reliability characteristic data, a first portion of the memory cells of the non-volatile memory device in a first management group, and a second portion of the memory cells of the non-volatile memory device in a second management group.
YOU discloses grouping, based on the reliability characteristic data, a first portion of the memory cells of the non-volatile memory device in a first management group, and a second portion of the memory cells of the non-volatile memory device in a second management group ([0034]-[0035], “In an embodiment of the invention, at least two WBGs (Worn Block Groups) are designated in an IWL wear leveling process. A first group is referred to as the "high group". The high group will have relatively high PBEC and high LBEC values, and may include any spare blocks because reclaim cycles happen frequently for space blocks of non-volatile memories. A second group is referred to as the "low group" and will have relatively low PBEC and low LBEC, with spare blocks excluded.” E.g. grouping blocks of memory cells into 2 “worn block groups” based on cycle counts /reliability characteristics of the blocks). Macronix beginning page 1 in the section titled Cycling Endurance and Data Retention explains that data retention time is a function of Program Erase cycles, as the dielectric layers of the floating gate in the cell may suffer degradation over time as the number of Program Erase cycles increases. Accordingly, an erase count is a characteristic regarding data retention degradation
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, to further include YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to extend lifespan of memory device while balancing against performance cost of wear leveling (see YOU [0043]).
CHANG and YOU do not explicitly disclose wherein the reliability characteristic data identify reliability characteristics with respect to storing data, and wherein the reliability characteristics include reliability characteristics regarding data retention degradation, reliability characteristics regarding read disturb, and reliability characteristics regarding cross temperature; and background scanning of memory cells.
LI discloses wherein the reliability characteristic data identify reliability characteristics with respect to storing data, and wherein the reliability characteristics include reliability characteristics regarding data retention degradation, reliability characteristics regarding read disturb, and reliability characteristics regarding cross temperature ([0150] “… in-situ data collection 714 is performed to gather one or more metrics describing current operation of the non-volatile storage system 100. Examples of metrics include … read disturb, data retention issues, …, and cross temperature Vt budget. Other metrics can also be gathered.”);
background scanning of memory cells ([0150], “in-situ data collection 714 is performed to gather one or more metrics describing current operation of the non-volatile storage system 100. Examples of metrics include data word line threshold voltage distribution, non-data word line threshold voltage distribution, failed bit count (“FBC”);….” [0174], “In one embodiment, the in-situ gathering of metrics is performed in the field, by the storage system, while the storage system is being operated by a user.” E.g. in-situ scan of reliability characteristics of memory cells, as background scan, while user is operating the storage system).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, with YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to further include LI’s different types of reliability characteristics data for storing data and background scanning and determining memory cells groups with different marginalities, to perform more detailed reliability data collection only when needed (see LI Abstract).
Regarding claim 2, CHANG discloses the method of claim 1, but does not explicitly disclose wherein the first management group comprises first memory cells of the non-volatile memory device that were identified by the reliability characteristic data as having a first marginality, wherein the second management comprises second memory cells of the non-volatile memory device of the second management group that were identified by the reliability characteristic data as having a first marginality, and wherein the first marginality and the second marginality indicate that the first memory cells are closer to an operational limit than the second memory cells.
YOU discloses wherein the first management group comprises first memory cells of the non-volatile memory device that were identified by the reliability characteristic data as having a first marginality, wherein the second management comprises second memory cells of the non-volatile memory device of the second management group that were identified by the reliability characteristic data as having a first marginality, and wherein the first marginality and the second marginality indicate that the first memory cells are closer to an operational limit than the second memory cells ([0034]-[0035], “In an embodiment of the invention, at least two WBGs (Worn Block Groups) are designated in an IWL wear leveling process. A first group is referred to as the "high group". The high group will have relatively high PBEC and high LBEC values, and may include any spare blocks because reclaim cycles happen frequently for space blocks of non-volatile memories. A second group is referred to as the "low group" and will have relatively low PBEC and low LBEC, with spare blocks excluded.” E.g. grouping blocks of memory cells into 2 “worn block groups” based on cycle counts /reliability characteristics of the blocks. Inherently, 1 group would have marginality exceeds the marginality of the other group).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, to further include YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to extend lifespan of memory device while balancing against performance cost of wear leveling (see YOU [0043]).
Regarding claim 3, CHANG and YOU disclose the method of claim 2, but do not explicitly disclose wherein managing the background scanning or the logical to physical mapping comprises: performing first background scanning of the first portion of the first memory cells of the non-volatile memory device more frequently than performing second background scanning of the second portion of the memory cells of the non-volatile memory device.
LI discloses wherein managing the background scanning or the logical to physical mapping comprises: performing first background scanning of the first portion of the first memory cells of the non-volatile memory device more frequently than performing second background scanning of the second portion of the memory cells of the non-volatile memory device (FIG. 15, [0162]-[0166], “the second level prediction uses more detailed indicators that require more time to collect but provide higher accuracy for the inference, such as running additional self-tests to get current and/or voltage leakage values, RC values, erase/program loop count (see FIG. 6), fail bit count, threshold voltage tail, etc…. In step 1506, in response to more than a threshold number of groupings of the non-volatile memory cells being categorized as potentially defective, inference circuit 800 uses the second model with a second set of one or more metrics describing current operation of the non-volatile storage apparatus in order to make a second level prediction as to whether defects exist in one or more groupings of the non-volatile memory cells that are categorized as potentially defective.” E.g. 1st group of block as good block pool, 2nd group of blocks as categorized as potentially defective are further analyzed by second level prediction with threshold voltage tail /marginality, whether marginality is worse compared to the normal blocks, if actual defects exists, place in 3rd group of actual bad block pool, see claim 16. Thus, 2nd group of potentially defective blocks are background scanned more frequently than the 1st group of good block pool, which is subject to less intensive 1st level prediction data collection. 2nd level prediction is provided with additional in-situ reliability data collection, thus more often collected than for the good blocks, see [0163], “While second level prediction is more reliable (e.g., more accurate), it takes more time to perform and requires a more intensive (time and compute resources) data collection, so it is only used when needed.” E.g. more intensive (time and compute resources) is interpreted as more frequent data collection in background scans, because intensity of data collection here means more frequent sampling of data, larger sample of data collected for compute and predictions).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, with YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to further include LI’s different types of reliability characteristics data for storing data and background scanning and determining memory cells groups with different marginalities, to perform more detailed reliability data collection only when needed (see LI Abstract).
Regarding claim 7, CHANG discloses the method of claim 2, but does not explicitly disclose wherein grouping the first portion of the memory cells in the first management group and the second portion of the memory cells in the second management group comprises: grouping the first portion of the memory cells in the first management group and the second portion of the memory cells in the second management group based on a data structure, wherein the data structure is generated based on the reliability characteristic data.
YOU discloses wherein grouping the first portion of the memory cells in the first management group and the second portion of the memory cells in the second management group comprises: grouping the first portion of the memory cells in the first management group and the second portion of the memory cells in the second management group based on a data structure, wherein the data structure is generated based on the reliability characteristic data ([0034]-[0035], “In an embodiment of the invention, at least two WBGs (Worn Block Groups) are designated in an IWL wear leveling process. A first group is referred to as the "high group". The high group will have relatively high PBEC and high LBEC values, and may include any spare blocks because reclaim cycles happen frequently for space blocks of non-volatile memories. A second group is referred to as the "low group" and will have relatively low PBEC and low LBEC, with spare blocks excluded.” See also FIG. 8, [0067], “As illustrated the memory device 810 includes a registry of physical block erase counts 830 and a registry of logical block erase counts 835.” See also FIG. 8, [0067], “As illustrated the memory device 810 includes a registry of physical block erase counts 830 and a registry of logical block erase counts 835.” E.g. grouping blocks of memory cells into 2 “worn block groups” based on data structure /registry of physical block erase counts and logical block erase counts, cycle counts /reliability characteristics of the blocks).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, to further include YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to extend lifespan of memory device while balancing against performance cost of wear leveling (see YOU [0043]).
Regarding claim 8, CHANG discloses the method of claim 2, wherein the first memory cells are included in one or more first wordlines of the non-volatile memory device, wherein the second memory cells are included in one or more second wordlines of the non-volatile memory device, and wherein the one or more first wordlines and the one or more second wordlines are contiguous (FIG. 4, [0053]-[0061], “the features (denoted xi) used by the machine learning model to predict read window values at time t1 (denoted y) may include read windows from a Vt distribution curve as described herein…. At 415, the machine learning model may split the training dataset by word line group (WLG).” E.g. memory cells are groups by wordlines, and the wordlines are each contiguous).
Claims 15-17 are rejected under 35 U.S.C. 103 as being unpatentable over CHANG in view of YOU, further in view of LI, and further in view of Zuolo et al., US Patent Application Publication Number 20220027083 (herein “ZUOLO”). “Program/erase Cycling Endurance and Data Retention of Macronix SLC NAND Flash Memories” (herein “Macronix”) is cited as an evidentiary reference for program erase cycles being a measure of data retention degradation.
Regarding independent claim 15, CHANG discloses a non-transitory computer-readable medium storing a set of instructions, the set of instructions comprising: one or more instructions that, when executed by one or more processors of one or more devices, cause the one or more devices to (FIG. 1, [0037] “the host system 105 (e.g., a host system controller 106), the memory system 110 (e.g., a memory system controller 115), or a memory device 130 (e.g., a local controller 135) may include or otherwise may access one or more non-transitory computer readable media storing instructions (e.g., firmware, logic, code) for performing the functions ascribed herein to the host system 105, the memory system 110, or a memory device 130. For example, such instructions, if executed by the host system 105 (e.g., by a host system controller 106), by the memory system 110 (e.g., by a memory system controller 115), or by a memory device 130 (e.g., by a local controller 135), may cause the host system 105, the memory system 110, or the memory device 130 to perform associated functions as described herein”):
determine, using a machine learning model, reliability characteristic data associated with wordlines of a non-volatile memory device (FIG. 1, [0015], “a memory system 110 may be or include …, or a non-volatile DIMM (NVDIMM)”, e.g. non-volatile memory device. FIG. 4, [0053]-[0061], “the features (denoted xi) used by the machine learning model to predict read window values at time t1 (denoted y) may include read windows from a Vt distribution curve as described herein…. At 415, the machine learning model may split the training dataset by word line group (WLG).” E.g. using machine learning model to train /determine read windows /Vts / reliability characteristics of memory cells in wordlines groups);
Wherein the reliability characteristic data includes reliability characteristics regarding data retention degradation Paragraph 0042 shows that the threshold voltage shift and read window compression experienced by a set of memory cells varies with the number of erase operations performed on the cells. Macronix beginning page 1 in the section titled Cycling Endurance and Data Retention explains that data retention time is a function of Program Erase cycles, as the dielectric layers of the floating gate in the cell may suffer degradation over time as the number of Program Erase cycles increases. Accordingly, an erase count is a characteristic regarding data retention degradation.
determine a first group of one or more first wordlines of the non-volatile memory device and a second group of one or more second wordlines of the non-volatile memory device (FIG. 4, [0053]-[0061], “the features (denoted xi) used by the machine learning model to predict read window values at time t1 (denoted y) may include read windows from a Vt distribution curve as described herein…. At 415, the machine learning model may split the training dataset by word line group (WLG).” E.g. using machine learning model to train /determine read windows /Vts / reliability characteristics of memory cells in wordlines groups); and
perform, based on the reliability characteristic data, at least one of: scanning of the first group of one or more first wordlines and the second group of one or more second wordlines, or logical to physical mapping of the first group of one or more first wordlines and the second group of one or more second wordlines (FIG. 5, [0064], “At 505, the memory system may determine that a trigger condition for performing read window management has been satisfied. In some examples, the trigger condition may be the set of memory cells reaching a threshold quantity of PECs.” E.g. perform background scan of read windows, based on program/erase cycle counts / reliability characteristics of groups).
CHANG does not explicitly disclose determine, based on the reliability characteristic data, a first group of one or more first wordlines of the non-volatile memory device and a second group of one or more second wordlines of the non-volatile memory device.
YOU discloses determine, based on the reliability characteristics regarding data retention degradation, a first group of one or more first wordlines of the non-volatile memory device and a second group of one or more second wordlines of the non-volatile memory device ([0034]-[0035], “In an embodiment of the invention, at least two WBGs (Worn Block Groups) are designated in an IWL wear leveling process. A first group is referred to as the "high group". The high group will have relatively high PBEC and high LBEC values, and may include any spare blocks because reclaim cycles happen frequently for space blocks of non-volatile memories. A second group is referred to as the "low group" and will have relatively low PBEC and low LBEC, with spare blocks excluded.” E.g. grouping blocks of memory cells into 2 “worn block groups” based on cycle counts /reliability characteristics of the blocks. Blocks have groups of word lines). Macronix beginning page 1 in the section titled Cycling Endurance and Data Retention explains that data retention time is a function of Program Erase cycles, as the dielectric layers of the floating gate in the cell may suffer degradation over time as the number of Program Erase cycles increases. Accordingly, an erase count is a characteristic regarding data retention degradation.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, to further include YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to extend lifespan of memory device while balancing against performance cost of wear leveling (see YOU [0043]).
CHANG and YOU do not explicitly disclose perform background scanning of memory cells.
LI discloses background scanning of memory cells ([0150], “in-situ data collection 714 is performed to gather one or more metrics describing current operation of the non-volatile storage system 100. Examples of metrics include data word line threshold voltage distribution, non-data word line threshold voltage distribution, failed bit count (“FBC”);….” [0174], “In one embodiment, the in-situ gathering of metrics is performed in the field, by the storage system, while the storage system is being operated by a user.” E.g. in-situ scan of reliability characteristics of memory cells, as background scan, while user is operating the storage system).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, with YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to further include LI’s background scanning and determining memory cells groups with different marginalities, to perform more detailed reliability data collection only when needed (see LI Abstract).
CHANG, YOU, and LI do not explicitly disclose wherein the machine learning model is trained using characterization data generated for different types of non-volatile memory devices manufactured by different manufacturers.
ZUOLO discloses wherein the machine learning model is trained using characterization data generated for different types of non-volatile memory devices manufactured by different manufacturers ([0047] “FIG. 4A shows a diagram of an exemplary RNN inference model 40a that includes a plurality of input neurons 41, a plurality of output neurons 45a and layers 42-44 of hidden neurons. The input for the RNN inference model includes a reference value (REFERENCE VALUE) that identifies one or more characteristic of the flash memory device that is to be read, such as for example, a manufacturer, a part number, and/or a manufacturing lot number, without limitation. By including a reference value identifying the type of flash memory device, an RNN inference model can be used that can identify a TVS-RE curve for different types of flash memory devices.” E.g. reference values for manufacturers are used to train a RNN machine learning model using different corresponding characteristics data such as TVS-RE curves).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, with YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, and LI’s background scanning and determining memory cells groups with different marginalities, to further include ZUOLO’s training of machine learning model using characterization data generated for different types of non-volatile memory devices manufactured by different manufacturers, to reduce number of read errors (see ZUOLO [0011]).
Regarding claim 16, CHANG and YOU disclose the non-transitory computer-readable medium of claim 15, but do not explicitly disclose wherein the one or more instructions to perform at least one of the background scanning or the logical to physical mapping comprise: one or more instructions to perform, based on the reliability characteristic data, first background scanning of the first group of one or more first wordlines at a first frequency that is different than a second frequency of performing second background scanning of the second group of one or more second wordlines.
LI discloses wherein the one or more instructions to perform at least one of the background scanning or the logical to physical mapping comprise: one or more instructions to perform, based on the reliability characteristic data, first background scanning of the first group of one or more first wordlines at a first frequency that is different than a second frequency of performing second background scanning of the second group of one or more second wordlines (FIG. 15, [0162]-[0166], “the second level prediction uses more detailed indicators that require more time to collect but provide higher accuracy for the inference, such as running additional self-tests to get current and/or voltage leakage values, RC values, erase/program loop count (see FIG. 6), fail bit count, threshold voltage tail, etc…. In step 1506, in response to more than a threshold number of groupings of the non-volatile memory cells being categorized as potentially defective, inference circuit 800 uses the second model with a second set of one or more metrics describing current operation of the non-volatile storage apparatus in order to make a second level prediction as to whether defects exist in one or more groupings of the non-volatile memory cells that are categorized as potentially defective.” E.g. 1st group of block as good block pool, 2nd group of blocks as categorized as potentially defective are further analyzed by second level prediction with threshold voltage tail /marginality, whether marginality is worse compared to the normal blocks, if actual defects exists, place in 3rd group of actual bad block pool, see claim 16. Thus, 2nd group of potentially defective blocks are background scanned more frequently than the 1st group of good block pool, which is subject to less intensive 1st level prediction data collection. 2nd level prediction is provided with additional in-situ reliability data collection, thus more often collected than for the good blocks, see [0163], “While second level prediction is more reliable (e.g., more accurate), it takes more time to perform and requires a more intensive (time and compute resources) data collection, so it is only used when needed.” E.g. more intensive (time and compute resources) is interpreted as more frequent data collection in background scans, because intensity of data collection here means more frequent sampling of data, larger sample of data collected for compute and predictions).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, with YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to further include LI’s background scanning and determining memory cells groups with different marginalities, to perform more detailed reliability data collection only when needed (see LI Abstract).
Regarding claim 17, CHANG discloses the non-transitory computer-readable medium of claim 16, but does not explicitly disclose wherein a first marginality of the first group of one or more first wordlines and a second marginality of the second group of one or more second wordlines indicate that the first group of one or more first wordlines are closer to an operational limit than the second group of one or more second wordlines, and wherein the first frequency exceeds the second frequency based on the first marginality and the second marginality indicating that the first group of one or more first wordlines are closer to the operational limit than the second group of one or more second wordlines.
YOU discloses wherein a first marginality of the first group of one or more first wordlines and a second marginality of the second group of one or more second wordlines indicate that the first group of one or more first wordlines are closer to an operational limit than the second group of one or more second wordlines ([0034]-[0035], “In an embodiment of the invention, at least two WBGs (Worn Block Groups) are designated in an IWL wear leveling process. A first group is referred to as the "high group". The high group will have relatively high PBEC and high LBEC values, and may include any spare blocks because reclaim cycles happen frequently for space blocks of non-volatile memories. A second group is referred to as the "low group" and will have relatively low PBEC and low LBEC, with spare blocks excluded.” E.g. grouping blocks of memory cells into 2 “worn block groups” based on cycle counts /reliability characteristics of the blocks. Inherently, 1 group would be closer to an operational limit of erase cycles of the other group).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, to further include YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to extend lifespan of memory device while balancing against performance cost of wear leveling (see YOU [0043]).
LI discloses wherein the first frequency exceeds the second frequency based on the first marginality and the second marginality indicating that the first group of one or more first wordlines are closer to the operational limit than the second group of one or more second wordlines (FIG. 15, [0162]-[0166], “the second level prediction uses more detailed indicators that require more time to collect but provide higher accuracy for the inference, such as running additional self-tests to get current and/or voltage leakage values, RC values, erase/program loop count (see FIG. 6), fail bit count, threshold voltage tail, etc…. In step 1506, in response to more than a threshold number of groupings of the non-volatile memory cells being categorized as potentially defective, inference circuit 800 uses the second model with a second set of one or more metrics describing current operation of the non-volatile storage apparatus in order to make a second level prediction as to whether defects exist in one or more groupings of the non-volatile memory cells that are categorized as potentially defective.” E.g. 1st group of block as good block pool, 2nd group of blocks as categorized as potentially defective are further analyzed by second level prediction with threshold voltage tail /marginality, whether marginality is worse compared to the normal blocks, if actual defects exists, place in 3rd group of actual bad block pool, see claim 16. Thus, 2nd group of potentially defective blocks are background scanned more frequently than the 1st group of good block pool, which is subject to less intensive 1st level prediction data collection. 2nd level prediction is provided with additional in-situ reliability data collection, thus more often collected than for the good blocks, see [0163], “While second level prediction is more reliable (e.g., more accurate), it takes more time to perform and requires a more intensive (time and compute resources) data collection, so it is only used when needed.” E.g. more intensive (time and compute resources) is interpreted as more frequent data collection in background scans, because intensity of data collection here means more frequent sampling of data, larger sample of data collected for compute and predictions).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, with YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to further include LI’s background scanning and determining memory cells groups with different marginalities, to perform more detailed reliability data collection only when needed (see LI Abstract).
Claims 4-6 are rejected over CHANG in view of YOU, further in view of LI, and further in view of Helmick et al., US Patent Application Publication Number 20240012580 (herein “HELMICK”). “Program/erase Cycling Endurance and Data Retention of Macronix SLC NAND Flash Memories” (herein “Macronix”) is cited as an evidentiary reference for program erase cycles being a measure of data retention degradation.
Regarding claim 4, CHANG, YOU, and LI disclose the method of claim 2, but do not explicitly disclose wherein managing the background scanning or the logical to physical mapping comprises: storing data in the second memory cells of the non-volatile memory device when the data is a first type of data; and storing the data in the first memory cells of the non-volatile memory device when the data is a second type of data.
HELMICK discloses wherein managing the background scanning or the logical to physical mapping comprises: storing data in the second memory cells of the non-volatile memory device when the data is a first type of data; and storing the data in the first memory cells of the non-volatile memory device when the data is a second type of data ([0040], “For example, if the information indicates that the data is likely to be written and/or re-written frequently (which may be referred to as hot data), the storage device may select and/or compose a reclaim unit with storage media that may have a relatively large number of P/E cycles remaining in its expected lifetime. As another example, if the information indicates that the data has a relatively high tolerance for errors, for example, because the data may be stored in the storage device as part of a redundant storage scheme (e.g. a redundant array of independent drives RAID) scheme), the storage device may select and/or compose a reclaim unit with storage media that may have a relatively small number of P/E cycles remaining in its expected lifetime. As a further example, a storage device may use information about data to store at the device to select and/or compose a reclaim unit based on one or more of the following attributes of storage media at the drive: wafer production information, erase block location on a storage medium die, voltage and/or speed characteristics of read, write (program), and/or erase operations, bit error accumulation rate, current and/or historical temperature exposure. access activity, and/or the like of the storage media and/or neighboring media, and/or the like.” [0118], “Access information may include access information, for example, historical access information such as how recently and/or frequently the data has been written, read, re-written, and/or the like. For example, if data has been written and then not read or re-written in a relatively long time, the data may be considered relatively cold data. Historical access information may be sent from the host 902 to the storage device 904 which may use the historical access information to determine whether the data 921 to store in the one or more reclaim units 914 may be considered hot, cold, and/or the like.” E.g. selecting different portions of memory to store data, based on data types (hot or cold, etc.). Composing reclaim unit is mapping of blocks, see [0061], “a reclaim unit handle 236 may identify (e.g., map to) one or more reclaim units 216 in one or more (e.g., each) reclaim group 218”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, with YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, and LI’s background scanning and determining memory cells groups with different marginalities, to further include HELMICK’s data type based target memory storage portion selection/mapping, to improve or optimize the usage of different reclaim units, erase blocks, and/or the like, in a storage device (see HELMICK [0037]).
Regarding claim 5, CHANG, YOU, and LI disclose the method of claim 4, but do not explicitly disclose wherein the first type of data is more frequently accessed than the second type of data, and wherein storing the data in the second memory cells of the non-volatile memory device comprises storing the data in the second memory cells when the data is the first type of data based on: the first type of data being more frequently accessed than the second type of data, and the first marginality and the second marginality indicating that the first memory cells are closer to the operational limit than the second memory cells.
HELMICK discloses wherein the first type of data is more frequently accessed than the second type of data, and wherein storing the data in the second memory cells of the non-volatile memory device comprises storing the data in the second memory cells when the data is the first type of data based on: the first type of data being more frequently accessed than the second type of data, and the first marginality and the second marginality indicating that the first memory cells are closer to the operational limit than the second memory cells ([0040], “For example, if the information indicates that the data is likely to be written and/or re-written frequently (which may be referred to as hot data), the storage device may select and/or compose a reclaim unit with storage media that may have a relatively large number of P/E cycles remaining in its expected lifetime. As another example, if the information indicates that the data has a relatively high tolerance for errors, for example, because the data may be stored in the storage device as part of a redundant storage scheme (e.g. a redundant array of independent drives RAID) scheme), the storage device may select and/or compose a reclaim unit with storage media that may have a relatively small number of P/E cycles remaining in its expected lifetime. As a further example, a storage device may use information about data to store at the device to select and/or compose a reclaim unit based on one or more of the following attributes of storage media at the drive: wafer production information, erase block location on a storage medium die, voltage and/or speed characteristics of read, write (program), and/or erase operations, bit error accumulation rate, current and/or historical temperature exposure. access activity, and/or the like of the storage media and/or neighboring media, and/or the like.” [0118], “Access information may include access information, for example, historical access information such as how recently and/or frequently the data has been written, read, re-written, and/or the like. For example, if data has been written and then not read or re-written in a relatively long time, the data may be considered relatively cold data. Historical access information may be sent from the host 902 to the storage device 904 which may use the historical access information to determine whether the data 921 to store in the one or more reclaim units 914 may be considered hot, cold, and/or the like.” E.g. storing different data types based on access frequencies (hot or cold, etc.) and also based on memory block marginality/reliability characteristics/PE cycles left).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, with YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, and LI’s background scanning and determining memory cells groups with different marginalities, to further include HELMICK’s data type based target memory storage portion selection/mapping, to improve or optimize the usage of different reclaim units, erase blocks, and/or the like, in a storage device (see HELMICK [0037]).
Regarding claim 6, CHANG, YOU, and LI disclose the method of claim 4, but do not explicitly disclose wherein the first type of data is received from a host device, wherein the second type of data is obtained as part of a garbage collection operation or a wear leveling operation, and wherein storing the data in the second memory cells of the non-volatile memory device comprises storing the data in the second memory cells of the non-volatile memory device when the data is the first type of data based on: the first type of data being received from the host device, and the marginality of the first memory cells exceeding the marginality of the second memory cells of the non-volatile memory device.
HELMICK discloses wherein the first type of data is received from a host device, wherein the second type of data is obtained as part of a garbage collection operation or a wear leveling operation, and wherein storing the data in the second memory cells of the non-volatile memory device comprises storing the data in the second memory cells of the non-volatile memory device when the data is the first type of data based on: the first type of data being received from the host device, and the marginality of the first memory cells exceeding the marginality of the second memory cells of the non-volatile memory device ([0119]-[0121], “Additionally, or alternatively, access information sent from the host 902. to the storage device 904 may include a determination made by the host-side information logic 963-1. For example, the host 902 may store a set of data in the storage device 902 using a first set of LBAs. The host 902 may read a portion of the set of data from the storage device 904, for example, as part of a file system compaction operation (e.g., a garbage collection operation to remove one or more gaps between remaining valid portions of the set of data), and write the portion of the set of data to the storage device 904 using a second set of LBAs. The host-side information logic 963-1 may send information 961 to the device-side information logic 965-1 indicating that the data 921 may be relatively cold data (e.g., the data 921 was not re-written recently and, therefore, was compacted as part of a garbage collection operation). Thus, the host-side information logic 963-1 may make a prediction of future access of the data 921 based on historical access information…. Additionally, or alternatively, the information logic 963-1 and/or 965-1 may enable the host 902 to provide, and/or the storage device 904 to use, information about one or more acceptable characteristics of storage media that may be used to store the data 921 that the information 961 is provided for. Acceptable characteristics may include error tolerance information such as an acceptable bit error rate (BER), which in turn may be based on a bit error accumulation. For example, a BER and/or a bit error accumulation may be specified based on a length of time the data is stored, a number of read operations performed on the data, a number of re-write operations (e.g., refresh operations) performed on the data, one or more data movement operations, (e.g., garbage collection (GC) operations) performed on the data, and/or the like. As another example, a BER and/or a bit error accumulation may be correlated to a number of remaining PE/E cycles in the expected and/or estimated lifetime of a reclaim unit and/or erase blocks used in the reclaim unit.” E.g. storing host data type vs. garbage collection data type, also based on memory block marginality/reliability characteristics/PE cycles left).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, with YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, and LI’s background scanning and determining memory cells groups with different marginalities, to further include HELMICK’s data type based target memory storage portion selection/mapping, to improve or optimize the usage of different reclaim units, erase blocks, and/or the like, in a storage device (see HELMICK [0037]).
Claims 9-14 and 18-20 are rejected over CHANG in view of YOU, further in view of LI, further in view of ZUOLO, and further in view of HELMICK. “Program/erase Cycling Endurance and Data Retention of Macronix SLC NAND Flash Memories” (herein “Macronix”) is cited as an evidentiary reference for program erase cycles being a measure of data retention degradation.
Regarding independent claim 9, CHANG discloses a solid-state drive (SSD), comprising: a non-volatile memory device; and a controller to (FIG. 1, [0015], “A memory system 110 may be or include any device or collection of devices, where the device or collection of devices includes at least one memory array. For example, a memory system 110 may be or include a Universal Flash Storage (UFS) device, an embedded Multi-Media Controller (eMMC) device, a flash device, a universal serial bus (USB) flash device, a secure digital (SD) card, a solid-state drive (SSD), a hard disk drive (HDD), a dual in-line memory module (DIMM), a small outline DIMM (SO-DIMM), or a non-volatile DIMM (NVDIMM), among other possibilities.” [0037] “the host system 105 (e.g., a host system controller 106), the memory system 110 (e.g., a memory system controller 115), or a memory device 130 (e.g., a local controller 135)…”):
determine, using a machine learning model, reliability characteristic data associated with wordlines of the non-volatile memory device (FIG. 1, [0037], “a memory system 110 may be or include …, or a non-volatile DIMM (NVDIMM)”, e.g. non-volatile memory device. FIG. 4, [0053]-[0061], “the features (denoted xi) used by the machine learning model to predict read window values at time t1 (denoted y) may include read windows from a Vt distribution curve as described herein…. At 415, the machine learning model may split the training dataset by word line group (WLG).” E.g. using machine learning model to train /determine read windows /Vts / reliability characteristics of memory cells in wordlines groups);
determine a first group of one or more first wordlines of the non-volatile memory device and a second group of one or more second wordlines of the non-volatile memory device (FIG. 4, [0053]-[0061], “the features (denoted xi) used by the machine learning model to predict read window values at time t1 (denoted y) may include read windows from a Vt distribution curve as described herein…. At 415, the machine learning model may split the training dataset by word line group (WLG).” E.g. using machine learning model to train /determine read windows /Vts / reliability characteristics of memory cells in wordlines groups); and
perform at least one of: first background scanning of the first group of one or more first wordlines and the second group of one or more second wordlines, or logical to physical mapping of the first group of one or more first wordlines and the second group of one or more second wordlines (FIG. 5, [0064], “At 505, the memory system may determine that a trigger condition for performing read window management has been satisfied. In some examples, the trigger condition may be the set of memory cells reaching a threshold quantity of PECs.” E.g. perform background scan of read windows, based on program/erase cycle counts / reliability characteristics of groups).
CHANG does not explicitly disclose wherein the reliability characteristic data includes reliability characteristics regarding data retention degradation and to determine, based on the reliability characteristic data, a first group of one or more first wordlines of the non-volatile memory device and a second group of one or more second wordlines of the non-volatile memory device.
YOU discloses wherein the reliability characteristic data includes reliability characteristics regarding data retention degradation. Paragraph 0025 shows that the erase count is tracked. Macronix beginning page 1 in the section titled Cycling Endurance and Data Retention explains that data retention time is a function of Program Erase cycles, as the dielectric layers of the floating gate in the cell may suffer degradation over time as the number of Program Erase cycles increases. Accordingly, an erase count is a characteristic regarding data retention degradation.
determine, based on the reliability characteristics regarding data retention degradation, a first group of one or more first wordlines of the non-volatile memory device and a second group of one or more second wordlines of the non-volatile memory device ([0034]-[0035], “In an embodiment of the invention, at least two WBGs (Worn Block Groups) are designated in an IWL wear leveling process. A first group is referred to as the "high group". The high group will have relatively high PBEC and high LBEC values, and may include any spare blocks because reclaim cycles happen frequently for space blocks of non-volatile memories. A second group is referred to as the "low group" and will have relatively low PBEC and low LBEC, with spare blocks excluded.” See also FIG. 8, [0067], “As illustrated the memory device 810 includes a registry of physical block erase counts 830 and a registry of logical block erase counts 835.” E.g. grouping blocks of memory cells into 2 “worn block groups” based on cycle counts /reliability characteristics of the blocks. Blocks have groups of word lines).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, to further include YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to extend lifespan of memory device while balancing against performance cost of wear leveling (see YOU [0043]).
CHANG and YOU do not explicitly disclose perform background scanning of memory cells.
LI discloses background scanning of memory cells ([0150], “in-situ data collection 714 is performed to gather one or more metrics describing current operation of the non-volatile storage system 100. Examples of metrics include data word line threshold voltage distribution, non-data word line threshold voltage distribution, failed bit count (“FBC”);….” [0174], “In one embodiment, the in-situ gathering of metrics is performed in the field, by the storage system, while the storage system is being operated by a user.” E.g. in-situ scan of reliability characteristics of memory cells, as background scan, while user is operating the storage system).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, with YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to further include LI’s background scanning and determining memory cells groups with different marginalities, to perform more detailed reliability data collection only when needed (see LI Abstract).
CHANG, YOU, and LI do not explicitly disclose wherein the machine learning model is trained using characterization data generated for different types of non-volatile memory devices manufactured by different manufacturers.
ZUOLO discloses wherein the machine learning model is trained using characterization data generated for different types of non-volatile memory devices manufactured by different manufacturers ([0047] “FIG. 4A shows a diagram of an exemplary RNN inference model 40a that includes a plurality of input neurons 41, a plurality of output neurons 45a and layers 42-44 of hidden neurons. The input for the RNN inference model includes a reference value (REFERENCE VALUE) that identifies one or more characteristic of the flash memory device that is to be read, such as for example, a manufacturer, a part number, and/or a manufacturing lot number, without limitation. By including a reference value identifying the type of flash memory device, an RNN inference model can be used that can identify a TVS-RE curve for different types of flash memory devices.” E.g. reference values for manufacturers are used to train a RNN machine learning model using different corresponding characteristics data such as TVS-RE curves).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, with YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, and LI’s background scanning and determining memory cells groups with different marginalities, to further include ZUOLO’s training of machine learning model using characterization data generated for different types of non-volatile memory devices manufactured by different manufacturers, to reduce number of read errors (see ZUOLO [0011]).
CHANG, YOU, LI, and ZUOLO do not explicitly disclose perform at least one of: first background scanning of the first group of one or more first wordlines at a first frequency that is different than a second frequency of performing second background scanning of the second group of one or more second wordlines, or logical to physical mapping of data to the first group of one or more first wordlines or to the second group of one or more second wordlines based on the data being a first type of data or a second type of data.
HELMICK discloses perform at least one of: first background scanning of the first group of one or more first wordlines at a first frequency that is different than a second frequency of performing second background scanning of the second group of one or more second wordlines ([0040], “For example, if the information indicates that the data is likely to be written and/or re-written frequently (which may be referred to as hot data), the storage device may select and/or compose a reclaim unit with storage media that may have a relatively large number of P/E cycles remaining in its expected lifetime. As another example, if the information indicates that the data has a relatively high tolerance for errors, for example, because the data may be stored in the storage device as part of a redundant storage scheme (e.g. a redundant array of independent drives RAID) scheme), the storage device may select and/or compose a reclaim unit with storage media that may have a relatively small number of P/E cycles remaining in its expected lifetime. As a further example, a storage device may use information about data to store at the device to select and/or compose a reclaim unit based on one or more of the following attributes of storage media at the drive: wafer production information, erase block location on a storage medium die, voltage and/or speed characteristics of read, write (program), and/or erase operations, bit error accumulation rate, current and/or historical temperature exposure. access activity, and/or the like of the storage media and/or neighboring media, and/or the like.” [0118], “Access information may include access information, for example, historical access information such as how recently and/or frequently the data has been written, read, re-written, and/or the like. For example, if data has been written and then not read or re-written in a relatively long time, the data may be considered relatively cold data. Historical access information may be sent from the host 902 to the storage device 904 which may use the historical access information to determine whether the data 921 to store in the one or more reclaim units 914 may be considered hot, cold, and/or the like.” E.g. storing different data types based on access frequencies (hot or cold, etc.) and also based on memory block marginality/reliability characteristics/PE cycles left), or
logical to physical mapping of data to the first group of one or more first wordlines or to the second group of one or more second wordlines based on the data being a first type of data or a second type of data ([0040], “For example, if the information indicates that the data is likely to be written and/or re-written frequently (which may be referred to as hot data), the storage device may select and/or compose a reclaim unit with storage media that may have a relatively large number of P/E cycles remaining in its expected lifetime. As another example, if the information indicates that the data has a relatively high tolerance for errors, for example, because the data may be stored in the storage device as part of a redundant storage scheme (e.g. a redundant array of independent drives RAID) scheme), the storage device may select and/or compose a reclaim unit with storage media that may have a relatively small number of P/E cycles remaining in its expected lifetime. As a further example, a storage device may use information about data to store at the device to select and/or compose a reclaim unit based on one or more of the following attributes of storage media at the drive: wafer production information, erase block location on a storage medium die, voltage and/or speed characteristics of read, write (program), and/or erase operations, bit error accumulation rate, current and/or historical temperature exposure. access activity, and/or the like of the storage media and/or neighboring media, and/or the like.” [0118], “Access information may include access information, for example, historical access information such as how recently and/or frequently the data has been written, read, re-written, and/or the like. For example, if data has been written and then not read or re-written in a relatively long time, the data may be considered relatively cold data. Historical access information may be sent from the host 902 to the storage device 904 which may use the historical access information to determine whether the data 921 to store in the one or more reclaim units 914 may be considered hot, cold, and/or the like.” E.g. selecting different portions of memory to store data, based on data types (hot or cold, etc.). Composing reclaim unit is mapping of blocks, see [0061], “a reclaim unit handle 236 may identify (e.g., map to) one or more reclaim units 216 in one or more (e.g., each) reclaim group 218”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, with YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, and LI’s background scanning and determining memory cells groups with different marginalities, and ZUOLO’s training of machine learning model using characterization data generated for different types of non-volatile memory devices manufactured by different manufacturers, to further include HELMICK’s data type based target memory storage portion selection/mapping, to improve or optimize the usage of different reclaim units, erase blocks, and/or the like, in a storage device (see HELMICK [0037]).
Regarding claim 10, CHANG disclose the SSD of claim 9, but does not explicitly disclose wherein a first marginality of the first group of one or more first wordlines and a second marginality of the second group of one or more second wordlines indicate that the first group of one or more first wordlines are closer to an operational limit than the second group of one or more second wordlines, and wherein, to perform the first background scanning, the controller is to: perform the first background scanning at the first frequency that exceeds the second frequency based on the first marginality and the second marginality indicating that the first group of one or more first wordlines are closer to an operational limit than the second group of one or more second wordlines.
YOU discloses wherein a first marginality of the first group of one or more first wordlines and a second marginality of the second group of one or more second wordlines indicate that the first group of one or more first wordlines are closer to an operational limit than the second group of one or more second wordlines ([0034]-[0035], “In an embodiment of the invention, at least two WBGs (Worn Block Groups) are designated in an IWL wear leveling process. A first group is referred to as the "high group". The high group will have relatively high PBEC and high LBEC values, and may include any spare blocks because reclaim cycles happen frequently for space blocks of non-volatile memories. A second group is referred to as the "low group" and will have relatively low PBEC and low LBEC, with spare blocks excluded.” E.g. grouping blocks of memory cells into 2 “worn block groups” based on cycle counts /reliability characteristics of the blocks. Inherently, 1 group would be closer to the operational limit of erase cycles than the other group).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, to further include YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to extend lifespan of memory device while balancing against performance cost of wear leveling (see YOU [0043]).
LI discloses wherein, to perform the first background scanning, the controller is to: perform the first background scanning at the first frequency that exceeds the second frequency based on the first marginality exceeding the second marginality (FIG. 15, [0162]-[0166], “the second level prediction uses more detailed indicators that require more time to collect but provide higher accuracy for the inference, such as running additional self-tests to get current and/or voltage leakage values, RC values, erase/program loop count (see FIG. 6), fail bit count, threshold voltage tail, etc…. In step 1506, in response to more than a threshold number of groupings of the non-volatile memory cells being categorized as potentially defective, inference circuit 800 uses the second model with a second set of one or more metrics describing current operation of the non-volatile storage apparatus in order to make a second level prediction as to whether defects exist in one or more groupings of the non-volatile memory cells that are categorized as potentially defective.” E.g. 1st group of block as good block pool, 2nd group of blocks as categorized as potentially defective are further analyzed by second level prediction with threshold voltage tail /marginality, whether marginality is worse compared to the normal blocks, if actual defects exists, place in 3rd group of actual bad block pool, see claim 16. Thus, 2nd group of potentially defective blocks are background scanned more frequently than the 1st group of good block pool, which is subject to less intensive 1st level prediction data collection. 2nd level prediction is provided with additional in-situ reliability data collection, thus more often collected than for the good blocks, see [0163], “While second level prediction is more reliable (e.g., more accurate), it takes more time to perform and requires a more intensive (time and compute resources) data collection, so it is only used when needed.” E.g. more intensive (time and compute resources) is interpreted as more frequent data collection in background scans, because intensity of data collection here means more frequent sampling of data, larger sample of data collected for compute and predictions).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, with YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to further include LI’s background scanning and determining memory cells groups with different marginalities, to perform more detailed reliability data collection only when needed (see LI Abstract).
Regarding claim 11, CHANG discloses the SSD of claim 9, but does not explicitly disclose wherein the first type of data is received from a host device, wherein a first marginality of the first group of one or more first wordlines and a second marginality of the second group of one or more second wordlines indicate that the first group of one or more first wordlines are closer to an operational limit than the second group of one or more second wordlines, and wherein when performing the logical to physical mapping of data to the first group of one or more first wordlines or to the second group of one or more second wordlines, the controller is to map the data to the second group of one or more second wordlines cause the data to be stored in the second group of one or more second wordlines based on: the first type of data being received from the host device and the first marginality and the second marginality indicating that the first group of the one or more fist wordlines are closer to an operational limit than the second group of one or more second wordlines.
CHANG paragraph 0030 shows that cells in a page are grouped by a common word line, and paragraph 0031 shows that blocks are made of pages.YOU discloses wherein a first marginality of the first group of one or more first wordlines and a second marginality of the second group of one or more second wordlines indicate that the first group of one or more first wordlines are closer to an operational limit than the second group of one or more second wordlines ([0034]-[0035], “In an embodiment of the invention, at least two WBGs (Worn Block Groups) are designated in an IWL wear leveling process. A first group is referred to as the "high group". The high group will have relatively high PBEC and high LBEC values, and may include any spare blocks because reclaim cycles happen frequently for space blocks of non-volatile memories. A second group is referred to as the "low group" and will have relatively low PBEC and low LBEC, with spare blocks excluded.” E.g. grouping blocks of memory cells into 2 “worn block groups” based on cycle counts /reliability characteristics of the blocks. Inherently, 1 group be closer to the operational limit of erase cycles than the other group).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, to further include YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to extend lifespan of memory device while balancing against performance cost of wear leveling (see YOU [0043]).
HELMICK discloses wherein the first type of data is received from a host device, and wherein the controller is to cause the data to be stored in the second group of one or more second wordlines based on: the first type of data being received from the host device ([0119]-[0121], “Additionally, or alternatively, access information sent from the host 902. to the storage device 904 may include a determination made by the host-side information logic 963-1. For example, the host 902 may store a set of data in the storage device 902 using a first set of LBAs. The host 902 may read a portion of the set of data from the storage device 904, for example, as part of a file system compaction operation (e.g., a garbage collection operation to remove one or more gaps between remaining valid portions of the set of data), and write the portion of the set of data to the storage device 904 using a second set of LBAs. The host-side information logic 963-1 may send information 961 to the device-side information logic 965-1 indicating that the data 921 may be relatively cold data (e.g., the data 921 was not re-written recently and, therefore, was compacted as part of a garbage collection operation). Thus, the host-side information logic 963-1 may make a prediction of future access of the data 921 based on historical access information…. Additionally, or alternatively, the information logic 963-1 and/or 965-1 may enable the host 902 to provide, and/or the storage device 904 to use, information about one or more acceptable characteristics of storage media that may be used to store the data 921 that the information 961 is provided for. Acceptable characteristics may include error tolerance information such as an acceptable bit error rate (BER), which in turn may be based on a bit error accumulation. For example, a BER and/or a bit error accumulation may be specified based on a length of time the data is stored, a number of read operations performed on the data, a number of re-write operations (e.g., refresh operations) performed on the data, one or more data movement operations, (e.g., garbage collection (GC) operations) performed on the data, and/or the like. As another example, a BER and/or a bit error accumulation may be correlated to a number of remaining PE/E cycles in the expected and/or estimated lifetime of a reclaim unit and/or erase blocks used in the reclaim unit.” E.g. storing host data type vs. garbage collection data type, also based on memory block marginality/reliability characteristics/PE cycles left).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, with YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, and LI’s background scanning and determining memory cells groups with different marginalities, and ZUOLO’s training of machine learning model using characterization data generated for different types of non-volatile memory devices manufactured by different manufacturers, to further include HELMICK’s data type based target memory storage portion selection/mapping, to improve or optimize the usage of different reclaim units, erase blocks, and/or the like, in a storage device (see HELMICK [0037]).
Regarding claim 12, CHANG discloses the SSD of claim 9, but does not explicitly disclose wherein, to determine the first group of one or more first wordlines and the second group of one or more second wordlines, the controller is to: determine the first group of one or more first wordlines and the second group of one or more second wordlines based on a data structure, wherein the data structure is generated based on the reliability characteristic data.
YOU discloses wherein, to determine the first group of one or more first wordlines and the second group of one or more second wordlines, the controller is to: determine the first group of one or more first wordlines and the second group of one or more second wordlines based on a data structure, wherein the data structure is generated based on the reliability characteristic data ([0034]-[0035], “In an embodiment of the invention, at least two WBGs (Worn Block Groups) are designated in an IWL wear leveling process. A first group is referred to as the "high group". The high group will have relatively high PBEC and high LBEC values, and may include any spare blocks because reclaim cycles happen frequently for space blocks of non-volatile memories. A second group is referred to as the "low group" and will have relatively low PBEC and low LBEC, with spare blocks excluded.” See also FIG. 8, [0067], “As illustrated the memory device 810 includes a registry of physical block erase counts 830 and a registry of logical block erase counts 835.” E.g. grouping blocks of memory cells into 2 “worn block groups” based on data structure /registry of physical block erase counts and logical block erase counts, cycle counts /reliability characteristics of the blocks).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, to further include YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to extend lifespan of memory device while balancing against performance cost of wear leveling (see YOU [0043]).
Regarding claim 13, CHANG discloses the SSD of claim 12, but does not explicitly disclose wherein the data structure identifies different program/erase cycles associated with different wordlines with different marginalities.
YOU discloses wherein the data structure identifies different program/erase cycles associated with different wordlines with different marginalities ([0034]-[0035], “In an embodiment of the invention, at least two WBGs (Worn Block Groups) are designated in an IWL wear leveling process. A first group is referred to as the "high group". The high group will have relatively high PBEC and high LBEC values, and may include any spare blocks because reclaim cycles happen frequently for space blocks of non-volatile memories. A second group is referred to as the "low group" and will have relatively low PBEC and low LBEC, with spare blocks excluded.” See also FIG. 8, [0067], “As illustrated the memory device 810 includes a registry of physical block erase counts 830 and a registry of logical block erase counts 835.” E.g. grouping blocks of memory cells into 2 “worn block groups” based on data structure /registry of physical block erase counts and logical block erase counts, cycle counts /reliability characteristics of the blocks).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, to further include YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to extend lifespan of memory device while balancing against performance cost of wear leveling (see YOU [0043]).
Regarding claim 14, CHANG discloses the SSD of claim 9, wherein the first group of one or more first wordlines and the second group of one or more second wordlines are contiguous (FIG. 4, [0053]-[0061], “the features (denoted xi) used by the machine learning model to predict read window values at time t1 (denoted y) may include read windows from a Vt distribution curve as described herein…. At 415, the machine learning model may split the training dataset by word line group (WLG).” E.g. memory cells are groups by wordlines, and the wordlines are each contiguous).
Regarding claim 18, CHANG, YOU, and LI disclose the non-transitory computer-readable medium of claim 15, but do not explicitly disclose wherein the one or more instructions to perform the background scanning and the logical to physical mapping comprise: one or more instructions to perform the logical to physical mapping of data to the first group of one or more first wordlines or to the second group of one or more second wordlines based on the data being a first type of data or a second type of data.
HELMICK discloses wherein the one or more instructions to perform the background scanning and the logical to physical mapping comprise: one or more instructions to perform the logical to physical mapping of data to the first group of one or more first wordlines or to the second group of one or more second wordlines based on the data being a first type of data or a second type of data ([0040], “For example, if the information indicates that the data is likely to be written and/or re-written frequently (which may be referred to as hot data), the storage device may select and/or compose a reclaim unit with storage media that may have a relatively large number of P/E cycles remaining in its expected lifetime. As another example, if the information indicates that the data has a relatively high tolerance for errors, for example, because the data may be stored in the storage device as part of a redundant storage scheme (e.g. a redundant array of independent drives RAID) scheme), the storage device may select and/or compose a reclaim unit with storage media that may have a relatively small number of P/E cycles remaining in its expected lifetime. As a further example, a storage device may use information about data to store at the device to select and/or compose a reclaim unit based on one or more of the following attributes of storage media at the drive: wafer production information, erase block location on a storage medium die, voltage and/or speed characteristics of read, write (program), and/or erase operations, bit error accumulation rate, current and/or historical temperature exposure. access activity, and/or the like of the storage media and/or neighboring media, and/or the like.” [0118], “Access information may include access information, for example, historical access information such as how recently and/or frequently the data has been written, read, re-written, and/or the like. For example, if data has been written and then not read or re-written in a relatively long time, the data may be considered relatively cold data. Historical access information may be sent from the host 902 to the storage device 904 which may use the historical access information to determine whether the data 921 to store in the one or more reclaim units 914 may be considered hot, cold, and/or the like.” E.g. selecting different portions of memory to store data, based on data types (hot or cold, etc.). Composing reclaim unit is mapping of blocks, see [0061], “a reclaim unit handle 236 may identify (e.g., map to) one or more reclaim units 216 in one or more (e.g., each) reclaim group 218”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, with YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, and LI’s background scanning and determining memory cells groups with different marginalities, and ZUOLO’s training of machine learning model using characterization data generated for different types of non-volatile memory devices manufactured by different manufacturers, to further include HELMICK’s data type based target memory storage portion selection/mapping, to improve or optimize the usage of different reclaim units, erase blocks, and/or the like, in a storage device (see HELMICK [0037]).
Regarding claim 19, CHANG discloses the non-transitory computer-readable medium of claim 18, but does not explicitly disclose wherein a first marginality of the first group of one or more first wordlines and a second marginality of the second group of one or more second wordlines indicate that the first group of one or more first wordlines are closer to an operational limit than the second group of one or more second wordlines.
YOU discloses wherein a first marginality of the first group of one or more first wordlines and a second marginality of the second group of one or more second wordlines indicate that the first group of one or more first wordlines are closer to an operational limit than the second group of one or more second wordlines ([0034]-[0035], “In an embodiment of the invention, at least two WBGs (Worn Block Groups) are designated in an IWL wear leveling process. A first group is referred to as the "high group". The high group will have relatively high PBEC and high LBEC values, and may include any spare blocks because reclaim cycles happen frequently for space blocks of non-volatile memories. A second group is referred to as the "low group" and will have relatively low PBEC and low LBEC, with spare blocks excluded.” E.g. grouping blocks of memory cells into 2 “worn block groups” based on cycle counts /reliability characteristics of the blocks. Inherently, 1 group would have marginality exceeds the marginality of the other group).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, to further include YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to extend lifespan of memory device while balancing against performance cost of wear leveling (see YOU [0043]).
CHANG, YOU, and LI do not explicitly disclose wherein the first type of data is received from a host device, and wherein the one or more instructions to perform the background scanning and the logical to physical mapping comprise: one or more instructions to store the data in the second group of one or more second wordlines based on: the data being the first type of data, and the first marginality and the second marginality indicating that the first group of one or more first wordlines are closer to the operational limit than the second group of one or more second wordlines.
HELMICK discloses wherein the first type of data is received from a host device, and wherein the one or more instructions to perform the background scanning and the logical to physical mapping comprise: one or more instructions to store the data in the second group of one or more second wordlines based on: the data being the first type of data, and the first marginality and the second marginality indicating that the first group of one or more first wordlines are closer to the operational limit than the second group of one or more second wordlines ([0119]-[0121], “Additionally, or alternatively, access information sent from the host 902. to the storage device 904 may include a determination made by the host-side information logic 963-1. For example, the host 902 may store a set of data in the storage device 902 using a first set of LBAs. The host 902 may read a portion of the set of data from the storage device 904, for example, as part of a file system compaction operation (e.g., a garbage collection operation to remove one or more gaps between remaining valid portions of the set of data), and write the portion of the set of data to the storage device 904 using a second set of LBAs. The host-side information logic 963-1 may send information 961 to the device-side information logic 965-1 indicating that the data 921 may be relatively cold data (e.g., the data 921 was not re-written recently and, therefore, was compacted as part of a garbage collection operation). Thus, the host-side information logic 963-1 may make a prediction of future access of the data 921 based on historical access information…. Additionally, or alternatively, the information logic 963-1 and/or 965-1 may enable the host 902 to provide, and/or the storage device 904 to use, information about one or more acceptable characteristics of storage media that may be used to store the data 921 that the information 961 is provided for. Acceptable characteristics may include error tolerance information such as an acceptable bit error rate (BER), which in turn may be based on a bit error accumulation. For example, a BER and/or a bit error accumulation may be specified based on a length of time the data is stored, a number of read operations performed on the data, a number of re-write operations (e.g., refresh operations) performed on the data, one or more data movement operations, (e.g., garbage collection (GC) operations) performed on the data, and/or the like. As another example, a BER and/or a bit error accumulation may be correlated to a number of remaining PE/E cycles in the expected and/or estimated lifetime of a reclaim unit and/or erase blocks used in the reclaim unit.” E.g. storing host data type vs. garbage collection data type, also based on memory block marginality/reliability characteristics/PE cycles left).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, with YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, and LI’s background scanning and determining memory cells groups with different marginalities, and ZUOLO’s training of machine learning model using characterization data generated for different types of non-volatile memory devices manufactured by different manufacturers, to further include HELMICK’s data type based target memory storage portion selection/mapping, to improve or optimize the usage of different reclaim units, erase blocks, and/or the like, in a storage device (see HELMICK [0037]).
Regarding claim 20, CHANG discloses the non-transitory computer-readable medium of claim 15, but does not explicitly disclose wherein the one or more instructions to determine the first group of one or more first wordlines and the second group of one or more second wordlines comprise: one or more instructions to determine the first group of one or more first wordlines and the second group of one or more second wordlines based on a data structure, wherein the data structure is generated based on the reliability characteristic data.
YOU discloses wherein the one or more instructions to determine the first group of one or more first wordlines and the second group of one or more second wordlines comprise: one or more instructions to determine the first group of one or more first wordlines and the second group of one or more second wordlines based on a data structure, wherein the data structure is generated based on the reliability characteristic data ([0034]-[0035], “In an embodiment of the invention, at least two WBGs (Worn Block Groups) are designated in an IWL wear leveling process. A first group is referred to as the "high group". The high group will have relatively high PBEC and high LBEC values, and may include any spare blocks because reclaim cycles happen frequently for space blocks of non-volatile memories. A second group is referred to as the "low group" and will have relatively low PBEC and low LBEC, with spare blocks excluded.” See also FIG. 8, [0067], “As illustrated the memory device 810 includes a registry of physical block erase counts 830 and a registry of logical block erase counts 835.” E.g. grouping blocks of memory cells into 2 “worn block groups” based on data structure /registry of physical block erase counts and logical block erase counts, cycle counts /reliability characteristics of the blocks).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify CHANG’s machine learning based reliability characteristics and management of memory cells, to further include YOU’s grouping of memory cells into at least 2 groups based on reliability characteristics, to extend lifespan of memory device while balancing against performance cost of wear leveling (see YOU [0043]).
Response to Arguments
Applicant's arguments filed 3/9/2026 have been fully considered but they are not fully persuasive.
Applicant’s arguments, see the third paragraph beginning on page 13 of the remarks filed 3/9/2026 with respect to the objection to the specification have been fully considered and are persuasive. The objection to the specification has been withdrawn.
In the fourth paragraph beginning on page 13 of the remarks submitted 3/9/2026, Applicant argues:
“CHANG, YOU, and LI do not disclose or suggest one or more of the features recited in claim 1, amended as proposed. For example, CHANG, YOU, and LI do not disclose or suggest "grouping, based on the reliability characteristics regarding data retention degradation, a first portion of the memory cells of the non-volatile memory device in a first management group and a second portion of the memory cells of the non-volatile memory device in a second management group," as recited in claim 1, amended as proposed.”
Applicant continues in the first and second paragraphs beginning on page 14:
“At paragraph 0034 and 0035, not reproduced here for brevity, YOU discloses that "at least two WBGs (Worn Block Groups) are designated in an IWL wear leveling process," that a "first group is referred to as the "high group" [that] will have relatively high PBEC and high LBEC values, and may include any spare blocks because reclaim cycles happen frequently for space blocks of non-volatile memories," and that a "second group is referred to as the "low group" [that] have relatively low PBEC and low LBEC, with spare blocks excluded."
While YOU discloses two WBGs (Worn Block Groups), namely a high group and a low group, Applicant initially notes that this section of YOU specifically refers to blocks. Additionally, while YOU discloses a high group and a low group, YOU does not disclose or suggest that the high group and the low group are grouped based on the reliability characteristics regarding data retention degradation. Instead, YOU specifically discloses that the groups are based on PBEC (physical block erase count) and LBEC (logical block erase count).”
You paragraph 0020 shows that a block contains multiple cells. Accordingly, grouping blocks is grouping the cells of the block.
Paragraph 0017 of the specification of the instant application states “As used herein, “data retention degradation” may be used to refer to a degraded (or decreased) data retention of the non-volatile memory device due to loss of electrons occurring during a power-off condition of the memory device.” Macronix beginning page 1 in the section titled Cycling Endurance and Data Retention explains that data retention time is a function of Program Erase cycles, as the dielectric layers of the floating gate in the cell may suffer degradation over time as the number of Program Erase cycles increases. Accordingly, an erase count is a characteristic regarding data retention degradation.
Therefore, the combination of Chang, You, and Li teaches “grouping, based on the reliability characteristics regarding data retention degradation, a first portion of the memory cells of the non-volatile memory device in a first management group and a second portion of the memory cells of the non-volatile memory device in a second management group," as recited in claim 1.
Accordingly, Applicant’s arguments are not found persuasive and the rejection of claim 1 is maintained.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Jared Ian Rutz whose telephone number is (571)272-5535. The examiner can normally be reached Monday-Friday, 8:00 AM to 4:00 PM.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, John Cottingham can be reached at (571)272-1400. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/JARED I RUTZ/ Supervisory Patent Examiner, Art Unit 2135
Prosecution Insights