# US 2022/0027083 A1 · Regression Neural Network for Identifying Threshold Voltages to be Used in Reads of Flash Memory Devices

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

- Number: US 2022/0027083 A1
- Document ID: US20220027083A1
- Status: Published application
- Kind: A1
- Application number: 17/089,891
- Filed: 2020-11-05
- Published: 2022-01-27
- Earliest claimed: 2020-07-27
- Assignees: Microchip Technology Incorporated
- Applicants: Microchip Technology Inc.
- Inventors: Lorenzo Zuolo; Rino Micheloni
- Web page: https://patentgrep.com/patent/US20220027083A1

## CPC

- G06N 3/08: Learning methods
- G06F 3/0679: {Non-volatile semiconductor memory device, e.g. flash memory, one time programmable memory [OTP]}
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/09: Supervised learning
- G06N 5/04: Inference or reasoning models
- G11C 11/54: using elements simulating biological cells, e.g. neuron
- G11C 16/26: Sensing or reading circuits; Data output circuits
- G11C 29/021: {in voltage or current generators}
- G11C 29/028: {with adaption or trimming of parameters}
- G06N 20/00: Machine learning
- G11C 11/5642: {Sensing or reading circuits; Data output circuits}
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G11C: STATIC STORES (semiconductor memory devices H10B)

## Abstract

A method and apparatus for reading a flash memory device are disclosed. A Regression Neural Network (RNN) inference model is stored on a flash controller. The RNN inference model is configured for identifying at least one Threshold-Voltage-Shift Read-Error (TVS-RE) curve that identifies a number of errors as a function of Threshold Voltage Shift Offset (TVSO) values. The operation of a flash memory device is monitored to identify usage characteristic values. A neural network operation of the RNN inference model is performed to generate a TVS-RE curve corresponding to the usage characteristic values. The input for the neural network operation includes the usage characteristic values. A TVSO value is identified corresponding to a minimum value of the TVS-RE curve. A read of the flash memory device is performed using a threshold-voltage-shift read at the TVSO value.

## Claims

### Claim 1

A method for reading a flash memory device comprising:
storing a Regression Neural Network (RNN) inference model on a flash controller that is coupled to the flash memory device or on a memory device that is coupled to the flash controller, the RNN inference model configured for identifying at least one threshold-voltage-shift read-error curve, each threshold-voltage-shift read-error curve identifying a number of errors as a function of threshold voltage shift offset values;
monitoring the operation of a flash memory device to identify one or more usage characteristic values of the flash memory device;
performing a neural network operation of the stored RNN inference model to generate a threshold-voltage-shift read-error curve corresponding to the identified usage characteristic values, the input for the neural network operation including the identified one or more usage characteristic values;
identifying a threshold voltage shift offset value corresponding to a minimum value of the generated threshold-voltage-shift read-error curve; and
performing a read of the flash memory device using a threshold-voltage-shift read at the identified threshold voltage shift offset value.

### Claim 2 (depends on 1)

The method of claim 1 wherein the input for the neural network operation includes one or more value identifying a page that is to be read.

### Claim 3 (depends on 1)

The method of claim 1 wherein the stored RNN inference model is a single RNN inference model that predicts a threshold-voltage-shift read-error curve for different points in the lifetime of the flash memory device.

### Claim 4 (depends on 1)

The method of claim 1 wherein the stored RNN inference model is a single RNN inference model that predicts a threshold-voltage-shift read-error curve for different points in the lifetime of the flash memory device, and
wherein the monitoring the operation of a flash memory device further includes monitoring the operation of the flash memory device to determine a current point in the lifetime of the flash memory device; and
wherein the input for the neural network operation includes a value identifying the determined current point in the lifetime of the flash memory device.

### Claim 5 (depends on 4)

The method of claim 4 wherein the value identifying the current point in the lifetime of the flash memory device is a number of program and erase cycles.

### Claim 6 (depends on 1)

The method of claim 1 wherein the storing the RNN inference model on the flash controller comprises storing a plurality of RNN inference models, each of the plurality of RNN inference models identifying threshold-voltage-shift read-error curves for a different period in the lifetime of the flash memory device, and wherein the monitoring the operation of the flash memory device comprises monitoring the operation of the flash memory device to determine the current period in the lifetime of the flash memory device, the method further comprising:
selecting the one of the plurality of RNN inference models corresponding to the determined current period in the lifetime of the flash memory device, and
wherein the performing the neural network operation uses the selected one of the plurality of RNN inference models.

### Claim 7 (depends on 6)

The method of claim 6 wherein the monitoring the operation of the flash memory device further comprises counting a number of program and erase cycles of each block of the flash memory device, the determined current period in the lifetime of the flash memory device comprising the counted number of program and erase cycles of the block containing the page that is to be read.

### Claim 8 (depends on 1)

The method of claim 1 wherein the monitoring the operation of the flash memory device further comprises counting a number of closed-block reads, the input for the neural network operation including a value indicating the counted number of closed block reads of the block containing the page that is to be read.

### Claim 9 (depends on 1)

The method of claim 1 wherein the monitoring the operation of the flash memory device further comprises determining a closed-block retention time, the input for the neural network operation including a value indicating the determined closed-block retention time of the block containing the page that is to be read.

### Claim 10 (depends on 1)

The method of claim 1 further comprising:
generating the RNN inference model, the RNN inference model configured for performing a regression to identify coefficients of the at least one threshold-voltage-shift read-error curve.

### Claim 11 (depends on 10)

The method of claim 10 wherein the generating the RNN inference model further comprises:
receiving as input one or more configuration file defining an initial RNN inference model;
receiving as input one or more sets of threshold-voltage-shift read test results;
creating a threshold-voltage-shift-read training data set, a threshold-voltage-shift-read validation data set and a threshold-voltage-shift-read test data set from the received input one or more sets of threshold-voltage-shift read test results;
performing a smoothing function on the threshold-voltage-shift read training data set to obtain a smoothed threshold-voltage-shift read training data set;
training the initial RNN inference model using the smoothed threshold-voltage-shift read training data set to obtain a trained RNN inference model;
validating the trained RNN inference model using the threshold-voltage-shift read validation data set, and
testing the trained and validated RNN inference model using the threshold-voltage-shift read testing data set.

### Claim 12 (depends on 1)

The method of claim 1 wherein the storing the RNN inference model on the flash controller comprises storing a plurality of RNN inference models, each of the plurality of RNN inference models identifying threshold-voltage-shift read-error curves corresponding to one or more grouping of factors that affect threshold voltage, wherein the one or more grouping of factors that affect threshold voltage are selected from the group consisting of a group of wordlines of the flash memory device, a group of blocks of the flash memory device, periods of time, a group of read disturbs and a group of retention times.

### Claim 13

A flash controller including a write circuit, a read circuit configured to perform a read of a flash memory device using a threshold-voltage-shift read operation and a decoder configured to decode the results of the read, the flash controller comprising:
a status circuit for monitoring the operation of the flash memory device to identify one or more usage characteristic values;
a data storage configured for storing a Regression Neural Network (RNN) inference model, the regression neural network inference model configured to identify at least one threshold-voltage-shift read-error curve, each threshold-voltage-shift read-error curve identifying a number of errors as a function of threshold voltage shift offset values;
a neural processing module coupled to the data storage, the neural processing module configured to perform a neural network operation of the stored RNN inference model to generate a threshold-voltage-shift read-error curve corresponding to the identified usage characteristic values, the input for the neural network operation including the identified one or more usage characteristic values; and
a minimum function module coupled to the neural processing module, the minimum function module configured to determine a minimum value for the generated threshold-voltage-shift read-error curve and identifying a threshold voltage shift offset value corresponding to the determined minimum value; and
wherein the threshold-voltage-shift read operation uses the identified threshold voltage shift offset value to perform the read.

### Claim 14 (depends on 13)

The flash controller of claim 13 wherein the input for the neural network operation includes one or more value identifying a page that is to be read.

### Claim 15 (depends on 13)

The flash controller of claim 13 wherein the data storage is configured to store a plurality of RNN inference models, each of the plurality of RNN inference models corresponding to one or more grouping of factors that affect threshold voltage, wherein the one or more grouping of factors that affect threshold voltage are selected from the group consisting of a group of wordlines of the flash memory device, a group of blocks of the flash memory device, periods of time, a group of read disturbs and a group of retention times.

### Claim 16 (depends on 13)

The flash controller of claim 13 wherein the data storage is configured to store a plurality of RNN inference models, each of the plurality of RNN inference models identifying threshold-voltage-shift read-error curves for a group of wordlines of the flash memory device, wherein the neural processing module is configured to select the one of the plurality of RNN inference models corresponding to the wordline of the read that is to be performed and perform the neural network operation on the selected one of the plurality of RNN inference models.

### Claim 17 (depends on 13)

The flash controller of claim 13 wherein the data storage is configured to store a plurality of RNN inference models, each of the plurality of RNN inference models identifying threshold-voltage-shift read-error curves for a group of blocks of the flash memory device, wherein the neural processing module is configured to select the one of the plurality of RNN inference models corresponding to the block of the read that is to be performed and perform the neural network operation on the selected one of the plurality of RNN inference models.

### Claim 18 (depends on 13)

The flash controller of claim 13 wherein the data storage is configured to store a plurality of RNN inference models, each of the plurality of RNN inference models identifying threshold-voltage-shift read-error curves for a group of read disturbs of the flash memory device,
wherein the neural processing module is configured to select the one of the plurality of RNN inference models corresponding to a number of closed-block reads for the block of the page that is to be read and perform the neural network operation on the selected one of the plurality of RNN inference models.

### Claim 19 (depends on 13)

The flash controller of claim 13 wherein the data storage is configured to store a plurality of RNN inference models, each of the plurality of RNN inference models identifying threshold-voltage-shift read-error curves for one or more of a group of wordlines of the flash memory device a group of blocks of the flash memory device, periods of time, a group of read disturbs, a group of retention times.

### Claim 20

A Solid State Drives (SSD) comprising:
a plurality of flash memory devices;
a flash controller that is coupled to the flash memory devices, the flash controller comprising:
a write circuit configured to program codewords into each of the flash memory devices;
a read circuit configured to perform reads of the flash memory devices using a threshold-voltage-shift read operation;
a decoder coupled to the read circuit and configured to decode the results of the read;
a status circuit for monitoring the operation of the flash memory devices to identify, for each flash memory device one or more usage characteristic value;
a data storage configured for storing a Regression Neural Network (RNN) inference model, the RNN inference model configured to identify at least one threshold-voltage-shift read-error curve, each threshold-voltage-shift read-error curve identifying a number of errors as a function of threshold voltage shift offset values;
a neural processing module coupled to the data storage, the neural processing module configured to perform a neural network operation of the stored RNN inference model to generate a threshold-voltage-shift read-error curve corresponding to the identified usage characteristic values, the input for the neural network operation including the identified one or more usage characteristic values; and
a minimum function module coupled to the neural processing module, the minimum function module configured to determine a minimum value for the generated threshold-voltage-shift read-error curve and identifying a threshold voltage shift offset value corresponding to the determined minimum value, and
wherein the threshold-voltage-shift read operation uses the identified threshold voltage shift offset value to perform the read.
