DETAILED ACTION
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 .
Priority
Per MPEP 2152.01 the effective filing date of claims in a continuation-in-part application is determined on a claim by claim basis. Priority therefore belongs to a claim and not to an individual limitation within a claim. (see e.g., Santarus, Inc. v. Par Pharmaceutical, Inc., 694 F.3d 1344 (Fed. Cir. 2012)). “If the application is a continuation-in-part of an earlier U.S. application or international application, any claims in the new application not supported by the specification and claims of the parent application have an effective filing date equal to the actual filing date of the new application. Any claims which are fully supported under 35 U.S.C. 112 by the earlier parent application have the effective filing date of that earlier parent application”. MPEP 2152.01. In this case, all of the claims in the present application are not supported by the specification or claims of any parent application. As a result, all claims in the present application have an effective filing date equal to the actual filing date of this application (28 February 2025).
CLAIM INTERPRETATION
Claims in this application are not interpreted under 35 U.S.C. §112(f).
Claim Rejections - 35 USC § 103
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.
Claims 1-3, 5-6, 8-10, 12-13, 15-17 and 19-20 are rejected under 35 U.S.C. 103 as being unpatentable over US Patent Application Publication No. US 2023/0067208 A1 (Dedrick) in view of the paper titled “Health Status Assessment and Failure Prediction for Hard Drives with Recurrent Neural Networks” by Chang Xu et. al., from the November 2016 issue of IEEE Transactions on Computers, Vol. 65, No. 11., pgs. 3502-3508 (Xu).
Regarding claim 1 and analogous claims 8 and 15:
Dedrick discloses:
a method comprising: obtaining first time series data for a plurality of metrics associated with operation of a plurality of flash storage devices, wherein the first time series data comprise internally tracked operational metrics associated with operation of the plurality of flash storage devices (by disclosing tracking the telemetry data of multiple NAND SSDs, including historical and timestamped usage records [0018] [0023] [0025-0029] [Figs. 1A-3]. The data includes host usage data (310), media usage data (312) (first time series data) and wear data (314) [Fig. 3]. The media usage data (312) may be tracked by the monitor (230) (internally tracked) of each of the respective SSDs [Fig. 2] [0027]).
the internally tracked operational metrics including metrics tracked by one or more storage device controllers during operation of flash memory of the plurality of flash storage devices (by disclosing that controller (212) implements the monitor (230), which may record the various host usage data (310), media usage data (312) (first time series data) and wear data (314) [0025-0027] [Fig. 2]).
obtaining second time series data for one or more health metrics associated with operation of the plurality of storage devices, the one or more health metrics being distinct from the internally tracked operational metrics (by disclosing tracking the telemetry data of multiple NAND SSDs, including historical and timestamped usage records [0018] [0023] [0025-0029] [Figs. 1A-3]. The data includes host usage data (310), media usage data (312) and wear data (314) (second time series data) [Fig. 3]).
providing the first time series data for the plurality of metrics associated with operation of the plurality of flash storage devices and the second time series data for the one or more health metrics associated with the plurality of flash storage devices to estimate time to failure of a flash storage device based on the first time series data and the second time series data (by disclosing that the telemetry data including the host usage data (310), media usage data (312) and wear data (314) may be provided to an analytics engine (120) [Fig. 1B]. The analytics engine may use telemetry data to predict a remaining service life for each SSD (412) (time to failure) [0020] [0023]).
Dedrick does not explicitly disclose, but Xu teaches:
providing time series telemetry data as training data to a machine learning model, and training the machine learning model to estimate a time to failure of a flash storage device based on the time series telemetry data (by teaching that a RNN is an effective tool to model temporal dependency with time series data. It is an especially effective tool for modeling long-range dependencies, which is a natural and appropriate fit to assess the health status and predict failures of a storage device via a sequence of time series data [§1 Introduction, ¶¶2-4, pg. 3402]. The RNN may be trained, for example, on time series attributes of storage devices via backpropagation through time, can model temporal behaviors, and has the temporal dependence embedded into the RNN structure. An RNN is easy to implement and train [§3.1 Model: ¶1-2, pg. 3503] [§1 Introduction, ¶¶2-4, pg. 3402] [§3.3 Training, ¶¶1-4, pgs. 3504-3505]. The RNN may be trained to indicate the remaining life until failure of the storage devices and may be used to raise alarms if needed so that the storage devices can be replaced [§1 Introduction, ¶¶2-4, pg. 3402] [§3.2 Health Degree, ¶2-3, pg. 3504]. The model can be deployed and is suitable for on-line real-time monitoring of large-scale data centers with input of the time series data from the storage devices for prediction of the time to storage device failure [§4.4 Experimental Setup, ¶1-4, pg. 3506]).
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the analytics engine, which is input with the time series telemetry data including host usage data, media usage data, and wear data used to predict the remaining service life as taught by Dedrick to be a RNN model that is trained with the time series data (as taught by Dedrick) input sequentially with backpropagation through time (BPTT) to then be deployed and used to classify a health status indicating the remaining time to failure of a storage device based on input time-series data as taught by Xu.
One of ordinary skill in the art would have been motivated to make this modification because a recurrent neural network is a natural fit for the progressive deterioration of storage devices, a RNN preserves history and builds sequential dependence into the model while supporting long-term dependencies, an RNN is simple to implement and train, the model is suitable for on-line real-time monitoring of large-scale data centers, the model provides improved failure detection rates and false alarm rates compared to others, it can provide warning lead times that are sufficient for data to be migrated and the storage devices serviced by technicians, and the model outperforms sequence independent and short-term dependency models, as taught by Xu in [§1 Introduction, ¶¶2-4, pg. 3502] [§3.1 Model, ¶¶1-4, pgs. 3503-3504] [§3.2 Health Degree, ¶1-4, pg. 3504] [§4.4 Experimental Setup, ¶1, pg. 3506] [§4.5 Failure Based Prediction for Hard Drives, ¶¶1-4, pgs. 3507-3508] [Table 3] [Table 4].
Regarding claim 2 and analogous claims 9 and 16:
The method of claim 1 is made obvious by Dedrick in view of Xu (Dedrick-Xu).
Dedrick further discloses:
and wherein the plurality of storage devices are managed flash storage devices (by teaching that the SSDs (plurality of storage devices) include a controller (212) for managing the communication between the host and the NAND devices (218) managing the write, read, and erase operations to the NAND (218), as well as implementing the flash translation layer (FTL), and the monitoring function (230) [Fig. 2] [0025]).
Dedrick does not explicitly disclose, but Xu teaches:
wherein the machine learning model comprises a time series capable machine learning model (by teaching that the model is a Recurrent Neural Network [§3.1 Model: ¶1-2, pg. 3503] [§1 Introduction, ¶¶2-4, pg. 3402] [§3.3 Training, ¶¶1-4, pgs. 3504-3505] as analyzed with the modifications and motivations performed for claim 1 (also see Applicant’s specification [00167] and [00178]– a RNN is a time-series capable machine learning model)).
Regarding claim 3 and analogous claims 10 and 17:
The method of claim 1 is made obvious by Dedrick-Xu.
Dedrick does not explicitly disclose, but Xu teaches:
wherein the machine learning model comprises a recurrent neural network (as analyzed with the modifications and motivations performed for claim 2).
Regarding claim 5 and analogous claims 12 and 19:
The method of claim 1 is made obvious by Dedrick-Xu.
Dedrick further discloses:
wherein the one or more health metrics comprise one or more of device wear levels, block wear levels, storage component failures, storage device failures, or voltage tuning metrics for pages of flash storage (by teaching that the wear data (314) may include the RBER distribution for each block or page and a number of bad blocks [0027-0028]).
Regarding claim 6 and analogous claims 13 and 20:
The method of claim 1 is made obvious by Dedrick-Xu.
Dedrick does not explicitly disclose, but Xu teaches:
further comprising: deploying the trained machine learning model to monitor a flash storage device (the RNN may be trained to indicate the remaining life until failure of the storage devices, and may then be used to raise alarms if needed so that the storage devices can be replaced [§1 Introduction, ¶¶2-4, pg. 3402] [§3.2 Health Degree, ¶2-3, pg. 3504]. The model can be deployed and is suitable for on-line real-time monitoring of large-scale data centers with input of the time series data from the storage devices for prediction of the time to storage device failure [§4.4 Experimental Setup, ¶1-4, pg. 3506]).
collecting operating metrics and health metrics for the flash storage device; providing the time series data (i.e., operating metrics and the health metrics as taught by Dedrick) for the flash storage device to the trained machine learning model (the model can be deployed and is suitable for on-line real-time monitoring of large-scale data centers with input of a sequence of the most-recent time series data from the storage devices for prediction of the time to storage device failure [§4.4 Experimental Setup, ¶1-4, pg. 3506])
and receiving, from the machine learning model, an estimated time to failure for the flash storage device (by teaching that the output from the model is a classified health status indicative of the time to failure of the storage device [§3.2 Health Degree, ¶1-4, pg. 3504] [§4.4 Experimental Setup, ¶1-4, pg. 3506] [Fig. 3, pg. 3505] [§4.5 Failure Prediction for Hard Drives ¶1-4, pgs. 3507-3508]).
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have performed the modifications according to the analysis and motivations discussed with respect to claim 1.
Claims 4, 11 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Dedrick-Xu in further view of in view of US Patent Application Publication No. US 2017/0131948 A1 (Hoang).
Regarding claim 4 and analogous claims 11 and 18:
The method of claim 1 is made obvious by Dedrick-Xu.
Dedrick does not explicitly disclose, but Hoang teaches,
wherein the plurality of metrics associated with operation of the plurality of flash storage devices comprise one or more of voltage table changes, read errors that can be corrected using ECC, read errors that can be corrected after adjusting voltage levels, a history of specific voltage levels for writes and for reads, read patterns, power failure/restarts with included clock values or with measured durations of run-time and down time, programming modes, patterns of programming the flash, patterns of erases, latency jitters in programming or erase operations or in read requests, wear level imbalances, interrupted programs or erases, monitored temperatures, or voltage fluctuations on a flash storage device (by teaching that the device usage information used in the training data may include monitored temperatures of the SSDs (flash storage device) [Fig. 7]. The logged temperature permits analysis of temperature dependent SSD behavior [0120]. It accounts for different conditions when calculating remaining lifetime [0122-0125]. The temperature may be logged by SSD-resident (internal) monitoring firmware along with time-stamp information as time series data [0126-0129]. High temperature operation may shorten lifetime [0241-0243] [0247] [0250]).
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the time series media usage information logged by the controller for input to the RNN as taught by Dedrick-Xu to include logging time series data of the temperature for input to the model as taught by Hoang.
One of ordinary skill in the art would have been motivated to make this modification because it would allow the prediction system to account for the temperature conditions, which allows for consideration of the temperature conditions in the lifetime predictions, which would be useful as high-temperature operation can shorten device lifetimes as taught by Hoang in [0120] [0247].
Claims 7 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Dedrick-Xu in further view of US Patent Application Publication No. US 2026/0023483 A1 (Redaelli).
Regarding claim 7 and analogous claim 14:
The method of claim 6 is made obvious by Dedrick-Xu.
Dedrick does not explicitly disclose, but Redaelli teaches:
further comprising: providing a notification of a potential failure of the flash storage device based on the estimated time to failure being below a threshold time to failure (by teaching that a host may present an alert to a user, such as on a display device, when the predicted remaining life of an SSD falls below a user-specified threshold [0086]).
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the predicted time before failure as taught by Xu to include being used with a threshold to then present an alert to a user when the predicted remaining time falls below the threshold as taught by Redaelli.
One of ordinary skill in the art would have been motivated to make this modification because the alert being presented to the user would allow a user to replace the non-volatile memory device before it fails as taught by Redaelli in [0086].
Response to Arguments/Amendments
Applicant’s arguments with respect to the claims have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
US Patent Application Publication No. US 2026/0133697 A1 (Woolweaver) – teaches using drive attributes (236) including endurance data (270), specification data (272), and SMART attribute data (274) to predict a life of a solid-state drive using an aging model [Fig. 2F] [0106-0109].
US Patent Application Publication No. US 2023/0141749 A1 (Hao) – teaches training a RNN LSTM model using a plurality of SMART training sets from different SSDs to predict SSD failures [Fig. 2] [Fig. 7] [Fig. 11] [0070-0072].
US Patent Application Publication No. US 2024/0370177 A1 (Kamarthi) – teaches collecting historical SMART data, training a machine learning model, collecting SMART data of operational drives, and then predicting failures using the trained machine learning model [Figs. 4-5] [0103-0111].
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to CURTIS JAMES KORTMAN whose telephone number is (303)297-4404. The examiner can normally be reached Monday through Friday 7:30 AM through 4:00 PM MT.
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, Reginald Bragdon can be reached at (571) 272-4204. 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.
/CURTIS JAMES KORTMAN/Primary Examiner, Art Unit 2139
Prosecution Insights