Western Digital Technologies, Inc.

Sandisk Technologies Inc.

Stella Achtenberg

Eran Sharon

Omer Fainzilber

Alexander Bazarsky

Ran Zamir

Idan Alrod

Dudy David Avraham

Idan Goldenberg

Ofir Pele

Ariel Navon

David Rozman

Alon Eyal

Dana Lee

Rami Rom

Karin Inbar

Yehuda Hahn

Ramanathan Muthiah

Tianhong Yan

David Robison Hall

Yoav Weinberg

Amir Shaharabany

Hadas Oshinsky

Chun Sei Tsai

Tz-Yi Liu

Ishai Ilani

Einav Zilberstein

Michael Micha Ionin

Arthur Shulkin

Nika Yanuka

Lakshmi Sowjanya Sunkavelli

David Gur

Jonathan Lloyd

Amit Chattopadhyay

Aimamorn Suvichakorn

Gilat Flaishman

Matar Krizhak

Anand Gupta

Krzysztof Gladysz

Kameron Jung

Tommer Kuper Lotan

Methods are provided for tactically deploying machine learning operations within existing storage devices without the need for additional capital investment. Machine learning operations are specifically designed to locate and evaluate multiple types of data to complete an operation, including synthesizing missing data. These operations may be processed within a SoC of a storage device as embedded software. Storage devices designed to utilize machine learning methods within existing configurations can include a non-volatile memory for storing data, executable instructions, and a processor to conduct a variety of steps. The steps can include executing a plurality of applications stored in the non-volatile memory, and receiving a request for data, including measurements, from at least one of the applications. The steps can further determine if the requested data is suitable for substitution by an inference and subsequently select at least one machine learning model for generating a suitable inference.

Embedded multi-attribute machine learning for storage devices

The present disclosure generally relates to data storage devices, such as solid state drives. A data storage device includes a controller, one or more volatile memory locations, and one or more non-volatile memory locations. Computations, including reinforcement learning algorithms, may be completed by the controller using the one or more non-volatile memory locations. Data associated with reinforcement learning is stored in a table on one or more planes of the non-volatile memory, where the results from the computations update the table with the relevant values. The data in the table are aligned to one or more wordlines, such that sensing the wordline senses all the data stored in the table.

Memory device with reinforcement learning with Q-learning acceleration

A stream set classification process may be implemented to classify streams opened by a host device on a data storage device. The data storage device may internally classify the streams into different stream classifications using a set of performance metrics. Stream classifications that cause the data storage device to show the greatest gains when compared with a set of baseline performance metrics for the data storage device and/or when compared with other stream classifications, may be selected by the data storage device and/or the host device for subsequent write operations.

Adaptive multi-stream classification for data storage systems

Advanced ultra-low power error correcting codes are generated using soft quantization and lattice interpolation based on clock and Syndrome Weight. Reinforcement learning may be used to generate threshold values for flipping bits for low density parity check Ultra-Low Power error correction codes. The threshold values can be generated offline and downloaded to a storage device or generated while the storage device is in use.

Advanced ultra low power error correcting code encoders and decoders

A data storage device includes a hard disk drive coupled to a printed circuit board (PCB), a volatile memory device coupled to the PCB, a non-volatile memory device coupled to the PCB, and a controller coupled to the PCB, such that the controller is in communication with the hard disk drive, the volatile memory device, and the non-volatile memory device. The controller is configured to identify patterns and/or structures of metadata for the hard disk drive, perform one or more of the following to the metadata to tailor the metadata: data shaping, content aware decoding, adaptive data trimming, and/or adaptive metablock sizing, and write the tailored metadata to the non-volatile memory device. The metadata is at least one of repeatable run out metadata, positioning error signal metadata, adjacent track interference metadata, and/or emergency power off metadata.

Methods for writing HDD metadata in NAND flash

A method and apparatus for identifying data that is to be accessible in a low power state of a data storage device, and store this data in a physical (or logical) block that will be accessible in a low power state of the data storage device. Low power accessible data may be identified by host metadata of the data, indicating access is needed in a low power state. In other embodiments, the data storage device may learn the power state in which data should be accessible. In these embodiments, a controller stores information regarding the power state of a namespace in which the data is stored as an indicator to make the data accessible in a low power state. Alternatively, the controller stores a previous power state in which the data was accessed as an indicator to make the data accessible in a low power state.

Data management for efficient low power mode handling in a storage device

A method of temperature compensation to read a flash memory device includes determining a state of the flash memory device. An action is selected with a maximum Q-value from a Q-table for the current state during exploitation. A read operation of a code word from the flash memory device is conducted using one or more parameters according to the selected action. The code word is decoded with an error correction code (ECC) process.

Machine learning for temperature compensation

A method and apparatus for obtaining data from a memory, estimating a probability of data values of the obtained data based on at least one of a source log-likelihood ratio and a channel log-likelihood ratio, wherein each bit in the obtained data has an associated log-likelihood ratio, determining at least one data pattern parameter for the data and performing a decoding process using the at least one data pattern parameters to determine a decoded data set.

Content aware decoding method and system

A data storage device includes a non-volatile memory and a controller coupled to the non-volatile memory. The controller is operable to measure a first threshold voltage (VT) of a memory cell under a first parameter at a read temperature and measure a second VT of the memory cell under a second parameter at the read temperature in which the first parameter is different from the second parameter. A VT correction term for the memory cell is determined based upon the first VT measurement and the second VT measurement. A read VT of the memory cell is adjusted by using the VT correction term.

Temperature variation compensation

Systems and methods are described for reducing error rates on data storage devices by applying data shaping to data written to such devices in order to avoid error-prone states on cells within the devices. Different states of individual cells (such as those representing different bit patterns) may have different propensities for error, and these propensities may vary during operation of a device. Thus, a device as disclosed herein may determine error-prone states for a cell or group of cells, and apply data shaping to data written to such cells to reduce the likelihood that writing the data places the cell or cells into an error-prone state. Data shaping may be used, for example, to increase the occurrence of '0' bits within input data, thus avoiding error-prone low voltage states that may be used to represent a series of '1' bits.

Data shaping to reduce error rates in solid state memory devices

Disclosed herein are memory devices, systems, and methods of content-aware decoding of encoded data. In one aspect, an encoded data chunk is received and one or more characteristics, such as source statistics, are determined. A similar data chunk (that may, e.g., contain data of a similar type) with comparable statistics may be sought. The similar data chunk may, for example, have source statistics that are positively correlated to the source statistics of the encoded data chunk to be decoded. Decoder parameters for the encoded data may be set to correspond with decoder parameters suited to the similar data chunk. The encoded data chunk is decoded using the new decoder parameters. Decoding encoded data based on content can enhance performance, reducing decoding latency and/or power consumption.

Memory device with enhanced error correction

Disclosed herein are memory devices, systems, and methods of encoding and decoding data. In one aspect, an encoded data chunk is received and segmented into data segments with similar features. Each segment can be decoded based on its features. Data can also be rearranged and partitioned so as to minimize an entropy score that is based on the size and entropy of the data partitions. The approach is capable of enhancing performance, reducing decoding latency, and reducing power consumption.

Memory device with enhanced error correction via data rearrangement, data partitioning, and content aware decoding

A storage device may include a decoder configured to connect bits to a content node based on content-aware decoding process. The content-aware decoding process may be dynamic and determine connection structures of bits and content nodes based on patterns in data. In some cases, the decoder may connect non-adjacent bits to a content node based on a content-aware decoding process. In other cases, the decoder may connect a first number of bits to a first content node and a second number of bits to a second content node. In such cases, the first number of bits and the second number of bits are a different number.

Data driven ICAD graph generation

Growing community of inventors

Netanya, Israel

Stella Achtenberg