# US 2021/0173560 A1 · Non-volatile memory with on-chip principal component analysis for generating low dimensional outputs for machine learning

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

- Number: US 2021/0173560 A1
- Document ID: US20210173560A1
- Status: Published application
- Kind: A1
- Application number: 16/706,618
- Filed: 2019-12-06
- Published: 2021-06-10
- Earliest claimed: 2019-12-06
- Assignees: Western Digital Technologies, Inc.
- Applicants: Western Digital Technologies, Inc.
- Inventors: Won Ho Choi; Yongjune Kim; Martin Lueker-Boden
- Web page: https://patentgrep.com/patent/US20210173560A1

## CPC

- G06F 3/0604: {Improving or facilitating administration, e.g. storage management}
- G06F 11/1048: {using arrangements adapted for a specific error detection or correction feature}
- G06F 3/0659: {Command handling arrangements, e.g. command buffers, queues, command scheduling}
- G06F 3/0679: {Non-volatile semiconductor memory device, e.g. flash memory, one time programmable memory [OTP]}
- G06F 11/1068: {in sector programmable memories, e.g. flash disk (G06F11/1072 takes precedence)}
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/063: using electronic means
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G06N 3/10: Interfaces, programming languages or software development kits, e.g. for simulating neural networks
- G06F 3/0664: {at device level, e.g. emulation of a storage device or system}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Methods and apparatus are disclosed for implementing principal component analysis (PCA) within a non-volatile memory (NVM) die of solid state drive (SSD) to reduce the dimensionality of machine learning data before the data is transferred to other components of the SSD, such as to a data storage controller equipped with a machine learning engine. The machine learning data may include, for example, training images for training an image recognition system in which the SSD is installed. In some examples, the on-chip PCA components of the NVM die are configured as under-the-array or next-to-the-array components. In other examples, one or more arrays of the NVM die are configured as multiplication cores for performing PCA matrix multiplication. In still other aspects, multiple NVM dies are arranged in parallel, each with on-chip PCA components to permit parallel concurrent on-chip processing of machine learning data.

## Claims

### Claim 1

A memory die comprising:
a non-volatile memory (NVM) array;
a principal component analysis (PCA) component configured in the memory die to process input data obtained from the NVM array to generate PCA output data; and
an output device configured to output the PCA output data from the memory die.

### Claim 2 (depends on 1)

The memory die of claim 1, wherein the PCA component is configured to generate the PCA output data to have a lower dimensionality than a dimensionality of the input data.

### Claim 3 (depends on 1)

The memory die of claim 1, wherein the PCA component is configured to generate the PCA output data by generating eigenvalues and eigenvectors for a first set of input data and generating eigenvalues without eigenvectors for a plurality of subsequent sets of input data.

### Claim 4 (depends on 1)

The memory die of claim 1, wherein the PCA component is configured to generate the PCA output data by generating eigenvalues without corresponding eigenvectors.

### Claim 5 (depends on 1)

The memory die of claim 1, further comprising bypass circuitry configured to bypass the PCA component and control the output device to output the input data stored in the NVM array rather than output the PCA output data.

### Claim 6 (depends on 1)

The memory die of claim 1, wherein the PCA component comprises a matrix multiplication core formed within a portion of the NVM array.

### Claim 7 (depends on 1)

The memory die of claim 1, further comprising an error correction code (ECC) component configured to apply ECC to the input data before the input data is processed by the PCA component.

### Claim 8 (depends on 1)

The memory die of claim 1, further comprising an error correction code (ECC) component configured to apply ECC to the PCA output data before the PCA output data is output from the memory die by the output device.

### Claim 9 (depends on 1)

The memory die of claim 1, wherein the output device of the memory die is configured to output the PCA output data to a data storage device (DSD) controller.

### Claim 10 (depends on 1)

The memory die of claim 1, wherein NVM array comprises one or more of NAND storage elements, phase-change memory (PCM) elements, magneto-resistive random access memory (MRAM) elements and resistive random access memory (ReRAM) elements.

### Claim 11

A method for use with a memory die comprising a non-volatile memory (NVM) array, the method comprising:
storing data within the NVM array;
determining one or more principal components of the data using a principal component analysis (PCA) component formed within the memory die; and
outputting the one or more principle components of the data from the memory die.

### Claim 12 (depends on 11)

The method of claim 11, wherein determining the one or more principal components comprises determining eigenvalues and eigenvectors for a first set of data and determining eigenvalues without eigenvectors for a plurality of subsequent sets of data.

### Claim 13 (depends on 11)

The method of claim 11, wherein determining the one or more principal components comprises generating eigenvalues without generating corresponding eigenvectors.

### Claim 14 (depends on 11)

The method of claim 11, further comprising receiving a bypass signal and, in response to the bypass signal, outputting the data from the memory die rather than outputting the one or more principal components.

### Claim 15 (depends on 11)

The method of claim 11, wherein determining the one or more principal components comprises:
reading the data from NVM array;
storing at least some of the data in a page buffer of the memory die; and
determining the one or more principal components of the data by performing PCA operations on the data in the page buffer.

### Claim 16 (depends on 11)

The method of claim 11, further comprising applying an error correction code (ECC) to the data before determining the one or more principal components.

### Claim 17 (depends on 11)

The method of claim 11, further comprising applying an error correction code (ECC) to the one or more principle components before outputting the one or more principal components from the memory die.

### Claim 18

An apparatus comprising:
means, formed on a memory die having a non-volatile memory (NVM) array, for performing principal component analysis (PCA) on data obtained from the NVM array to determine principal components of the data; and
means, formed on the memory die, for outputting principal components of the data from the memory die.

### Claim 19 (depends on 18)

The apparatus of claim 18, wherein the means for performing PCA comprises means for determining eigenvalues and eigenvectors for a first set of data and determining eigenvalues without eigenvectors for a plurality of subsequent sets of data.

### Claim 20 (depends on 18)

The apparatus of claim 18, wherein determining the one or more principal components comprises means for generating eigenvalues without generating corresponding eigenvectors.

### Claim 21

A data storage device comprising:
a storage device controller;
a memory array comprising a plurality of memory dies;
wherein at least one of the memory dies comprises
a non-volatile memory (NVM) array;
a principal component analysis (PCA) component in the at least one of the memory dies and configured to process input data obtained from the NVM array to generate PCA output data; and
an output device configured to output the PCA output data to the storage device controller.

### Claim 22 (depends on 21)

The data storage device of claim 21, wherein the PCA component is configured to generate the PCA output data by generating eigenvalues and eigenvectors for a first set of input data and generating eigenvalues without eigenvectors for a plurality of subsequent sets of input data.

### Claim 23 (depends on 21)

The data storage device of claim 21, wherein the PCA component is configured to generate the PCA output data by generating eigenvalues without corresponding eigenvectors.

### Claim 24 (depends on 21)

The data storage device of claim 21, wherein the PCA component comprises a matrix multiplication core formed within a portion of the NVM array.

### Claim 25 (depends on 21)

The data storage device of claim 21, further comprising an error correction code (ECC) component configured to apply ECC to the input data before the input data is processed by the PCA component.

### Claim 26 (depends on 21)

The data storage device of claim 21, further comprising an error correction code (ECC) component configured to apply ECC to the PCA output data before the PCA output data is output to the storage device controller.
