Read threshold prediction in memory devices using deep neural networks
Devices, systems and methods for improving performance of a memory device are described. An example method includes obtaining a plurality of samples corresponding to a probability distribution for each of a plurality of cell voltage distributions of the memory device, each of the plurality of cell voltage distributions corresponding to a read voltage, determining, based on the samples obtained for the plurality of cell voltage distributions, a number of first deep neural networks (DNNs), estimating, for each of the first DNNs, one or more parameters of the corresponding probability distribution based on the plurality of samples, training, based on the samples and the corresponding one or more parameters, each of the first DNNs, and training, based on the samples and the one or more parameters from each of the first DNNs, a second DNN to enable generation of an updated read voltage value for retrieving information from the memory device.
1 . A method for improving performance of a memory device, comprising:
obtaining a plurality of samples corresponding to a probability distribution for each of a plurality of cell voltage distributions of the memory device, wherein each of the plurality of cell voltage distributions corresponds to a read voltage;
determining, based on the samples obtained for the plurality of cell voltage distributions, a number of a plurality of first deep neural networks (DNNs);
estimating, for each of the plurality of first DNNs, one or more parameters of the corresponding probability distribution based on the plurality of samples;
training, based on the samples and the corresponding one or more parameters, each of the plurality of first DNNs; and
training, based on the samples and the one or more parameters from each of the plurality of first DNNs, a second DNN to enable generation of an updated read voltage value for retrieving information from the memory device.
2 . The method of claim 1 , wherein determining the number of the plurality of first DNNs comprises:
performing, for each pair of probability distributions, a comparison between the corresponding plurality of samples to generate a metric indicative of a distance between the pair of probability distributions; and
determining, based on the metric, the number of the plurality of first DNNs.
3 . The method of claim 2 , wherein the metric is based on a Fisher information metric.
4 . The method of claim 2 , wherein the metric is based on a Kullback-Leibler divergence or a Jensen-Shannon divergence.
5 . The method of claim 1 , wherein estimating the one or more parameters of the probability distribution comprises:
measuring an error between actual values of the one or more parameters and predicted values of the one or more parameters.
6 . The method of claim 5 , wherein the error is a mean square error.
7 . The method of claim 5 , further comprising:
using a stochastic gradient descent algorithm, based on the corresponding error, to determine one or more weights of the corresponding first DNN.
8 . A method for improving performance of a memory device, comprising:
obtaining samples corresponding to a plurality of probability distributions, wherein each of the plurality of probability distributions is associated with each of a plurality of cell voltage distributions of the memory device, and wherein each of the plurality of cell voltage distributions corresponds to a read voltage;
estimating, for each of a plurality of first deep neural networks (DNNs) associated with a corresponding probability distribution, one or more parameters of the corresponding probability distribution based on the samples, each of the first DNNs having been trained using the samples obtained from the corresponding probability distribution;
determining an updated read voltage based on an output of a second DNN, an input to the second DNN comprising the one or more parameters from each of the plurality of first DNNs, and the second DNN having been trained using the samples and the one or more parameters from each of the plurality of first DNNs; and
applying the updated read voltage to the memory device to retrieve information from the memory device.
9 . The method of claim 8 , wherein the samples correspond to cumulative distribution functions associated with the plurality of cell voltage distributions.
10 . The method of claim 8 , wherein the probability distribution is a skew-normal distribution.
11 . The method of claim 10 , wherein a probability density function ƒ(x) of the skew-normal distribution is given by:
f
(
x
;
Θ
)
=
2
ω
·
ϕ
(
x
-
ξ
ω
)
·
Φ
(
α
·
x
-
ξ
ω
)
.
wherein Θ=(ξ, ω, α) corresponds to the one or more parameters, and
wherein
ϕ
(
y
)
=
1
2
π
e
-
y
2
2
and Φ(y)=∫ −∞ y ϕ(t) dt.
12 . The method of claim 8 , wherein the probability distribution is a Gaussian distribution.
13 . The method of claim 8 , wherein the memory device is a triple level cell (TLC) NAND flash memory device.
14 . An apparatus for improving performance of a memory device, comprising:
a memory controller;
a plurality of first deep neural networks (DNNs), each first DNN being communicatively coupled to the memory controller; and
a second DNN comprising a plurality of inputs, each input being coupled to an output of a corresponding first DNN, and an output coupled to the memory controller,
wherein the memory controller is configured to:
obtain a plurality of samples corresponding to a probability distribution for each of a plurality of cell voltage distributions of the memory device, wherein each of the plurality of cell voltage distributions corresponds to a read voltage;
determine, based on the samples obtained for the plurality of cell voltage distributions, a number of a plurality of first deep neural networks (DNNs);
estimate, for each of the plurality of first DNNs, one or more parameters of the corresponding probability distribution based on the plurality of samples;
train, based on the samples and the corresponding one or more parameters, each of the plurality of first DNNs; and
train, based on the samples and the one or more parameters from each of the plurality of first DNNs, a second DNN to enable generation of an updated read voltage value for retrieving information from the memory device.
15 . The apparatus of claim 14 , wherein the memory controller is further configured, as part of determining the number of the plurality of first DNNs, to:
perform, for each pair of probability distributions, a comparison between the corresponding plurality of samples to generate a metric indicative of a distance between the pair of probability distributions; and
determine, based on the metric, the number of the plurality of first DNNs.
16 . The apparatus of claim 15 , wherein the metric is based on a Fisher information metric, a Kullback-Leibler divergence, or a Jensen-Shannon divergence.
17 . An apparatus for improving performance of a memory device, comprising:
a memory controller;
a plurality of first deep neural networks (DNNs), each first DNN being communicatively coupled to the memory controller; and
a second DNN comprising a plurality of inputs, each input being coupled to an output of a corresponding first DNN, and an output coupled to the memory controller,
wherein the memory controller is configured to:
obtain samples corresponding to a plurality of probability distributions, wherein each of the plurality of probability distributions is associated with each of a plurality of cell voltage distributions of the memory device, and wherein each of the plurality of cell voltage distributions corresponds to a read voltage;
estimate, for each of a plurality of first deep neural networks (DNNs) associated with a corresponding probability distribution, one or more parameters of the corresponding probability distribution based on the samples, each of the first DNNs having been trained using the samples obtained from the corresponding probability distribution;
determine an updated read voltage based on an output of a second DNN, an input to the second DNN comprising the one or more parameters from each of the plurality of first DNNs, and the second DNN having been trained using the samples and the one or more parameters from each of the plurality of first DNNs; and
apply the updated read voltage to the memory device to retrieve information from the memory device.
18 . The apparatus of claim 17 , wherein the samples correspond to cumulative distribution functions associated with the plurality of cell voltage distributions.
19 . The apparatus of claim 17 , wherein the probability distribution is a skew-normal distribution or a Gaussian distribution.
20 . The apparatus of claim 17 , wherein the memory device is a triple level cell (TLC) NAND flash memory device.