Training method and test apparatus using the same
Provided is a training method capable of reducing or minimizing a training time. The training method includes, for each of devices to be tested, calculating a first eye width at which a first signal and a second signal synchronize with each other at a first operation speed, and calculating a second eye width at which the first signal and the second signal synchronize with each other at a second operation speed different from the first operation speed; performing machine learning on the first eye width and the second eye width to derive a model showing a relation between operation speeds and eye widths; and calculating a third eye width corresponding to a third operation speed different from the first operation speed and the second operation speed, using the model.
1 . A training method comprising:
for each of first devices to be tested, calculating a first eye width at which a first signal and a second signal synchronize with each other at a first operation speed, and calculating a second eye width at which the first signal and the second signal synchronize with each other at a second operation speed different from the first operation speed;
deriving a model showing a relation between operation speeds and eye widths by performing machine learning on the first eye width and the second eye width; and
using the model to calculate a third eye width corresponding to a third operation speed different from the first operation speed and the second operation speed.
2 . The training method of claim 1 , wherein performing the machine learning comprises learning of a linear regression model.
3 . The training method of claim 2 , wherein a gradient descent method is used to calculate parameters of the linear regression model.
4 . The training method of claim 1 , wherein the calculating of the first eye width comprises:
while adjusting a delay of the first signal at the first operation speed, finding a first fail-to-pass point and a first pass-to-fail point, and
determining the first eye width based on an interval between the first fail-to-pass point and the first pass-to-fail point.
5 . The training method of claim 4 , wherein finding of the first fail-to-pass point comprises:
determining the first signal having a first delay, as a fail,
determining the first signal having a second delay greater than the first delay by a first value, as a pass,
determining the first signal having a third delay less than the second delay by a second value, as a fail, the second value being less than the first value, and
determining the first signal having a fourth delay greater than the third delay by a third value, as a pass, the third value being smaller than the second value.
6 . The training method of claim 4 , wherein finding of the first pass-to-fail point comprises:
determining the first signal having a first delay, as a fail,
determining the first signal having a second delay less than the first delay by a first value, as a pass,
determining the first signal having a third delay greater than the second delay by a second value, as a fail, the second value being less than the first value, and
determining the first signal having a fourth delay smaller than the third delay by a third value, as a pass, the third value being less than the second value.
7 . The training method of claim 1 , further comprising:
determining a position of a valid window margin at the third operation speed,
wherein determining of the position of the valid window margin at the third operation speed includes:
finding a third fail-to-pass point, while adjusting a delay of the first signal at the third operation speed, and
determining from the third fail-to-pass point to a point obtained by adding the third eye width to the third fail-to-pass point, as the position of the valid window margin.
8 . The training method of claim 1 , further comprising:
determining a position of a valid window margin at the third operation speed,
wherein determining of the position of the valid window margin at the third operation speed includes:
finding a third pass-to-fail point while adjusting a delay of the first signal at the third operation speed, and
determining from a point, which is obtained by subtracting the third eye width from the third pass-to-fail point, to the third pass-to-fail point, as the position of the valid window margin.
9 . The training method of claim 1 , wherein the first signal corresponds to a DQ signal, and the second signal corresponds to a DQS signal.
10 . The training method of claim 1 , wherein when second devices to be tested belong to a same lot as the first devices to be tested,
no model showing the relation between operation speeds and eye widths is separately derived for the second devices to be tested, and
the model derived for the first devices to be tested is used, when calculating eye widths associated with the second devices to be tested.
11 . The training method of claim 1 , wherein when third devices to be tested belong to a different lot from the first devices to be tested,
the model derived for the first devices to be tested is updated, and
the updated model is used, when calculating eye widths associated with the third devices to be tested.
12 . The training method of claim 1 , wherein
the calculating of the first eye width and the second eye width is performed by a first processor, and
the deriving of the model showing the relation between operation speeds and eye widths is performed by a second processor.
13 . The training method of claim 12 , wherein the first processor performs a DC test on each of the first devices to be tested, after calculating the first eye width and the second eye width.
14 . A training method comprising:
for each of first devices to be tested, calculating a first eye width at which a first signal and a second signal synchronize with each other at a first operation speed, and calculating a second eye width at which the first signal and the second signal synchronize with each other at a second operation speed different from the first operation speed;
performing machine learning on the first eye width and the second eye width to derive a linear regression model showing a relation between operation speeds and eye widths by using a gradient descent method to calculate parameters of the linear regression model;
calculating a third eye width corresponding to a third operation speed, using the linear regression model;
finding a third fail-to-pass point, while adjusting a delay of the first signal at the third operation speed; and
determining from the third fail-to-pass point to a point obtained by adding the third eye width to the third fail-to-pass point, as a position of a valid window margin.
15 . The training method of claim 14 , wherein when second devices to be tested belong to a same lot as the first devices to be tested,
no model showing the relation between operation speeds and eye widths is separately derived for the second devices to be tested,
the model derived for the first devices to be tested is used, when calculating eye widths associated with the second devices to be tested, and
wherein when third devices to be tested belong to a different lot from the first devices to be tested,
the model derived for the first devices to be tested is updated, and
the updated model is used, when calculating eye widths associated with the third devices to be tested.
16 . A test apparatus comprising:
output nodes configured to electrically connect to first devices to be tested;
a timing generator configured to provide a first signal and a second signal to the first devices to be tested through the output nodes; and
a controller configured to control the timing generator,
wherein the controller is configured to calculate a first eye width at which the first signal and the second signal synchronize with each other at a first operation speed, and calculate a second eye width at which the first signal and the second signal synchronize with each other at a second operation speed different from the first operation speed, for each of first devices to be tested,
wherein the controller is configured to perform machine learning on the first eye width and the second eye width to derive a model showing a relation between operation speeds and eye widths, and
wherein the controller is configured to calculate a third eye width corresponding to a third operation speed different from the first operation speed and the second operation speed by using the model.
17 . The test apparatus of claim 16 , wherein
the controller is configured to perform the machine learning by learning of a linear regression model, and
the controller is configured to calculate parameters of the linear regression model using a gradient descent method.
18 . The test apparatus of claim 16 , wherein the controller is configured to calculate the first eye width by,
finding a first fail-to-pass point and a first pass-to-fail point, while adjusting a delay of the first signal at the first operation speed, and
determining an interval between the first fail-to-pass point and the first pass-to-fail point as the first eye width.
19 . The test apparatus of claim 18 , wherein the controller is configured to find of the first fail-to-pass point by,
determining the first signal having a first delay, as a fail,
determining the first signal having a second delay greater than the first delay by a first value, as a pass,
determining the first signal having a third delay less than the second delay by a second value, as a fail, the second value being smaller than the first value, and
determining the first signal having a fourth delay greater than the third delay by a third value, as a pass, the third value being less than the second value.
20 . The test apparatus of claim 16 , wherein the controller is further configured to:
determine a position of a valid window margin at the third operation speed,
wherein the determining of the position of the valid window margin at the third operation speed includes:
finding a third fail-to-pass point, while adjusting a delay of the first signal at the third operation speed, and
determining from the third fail-to-pass point to a point obtained by adding the third eye width to the third fail-to-pass point, as the position of the valid window margin.