Apparatus and method for detecting and assessing a spatially discrete dot pattern

• US 5,950,181 A
• Filed: 01/06/1997
• Issued: 09/07/1999
• Est. Priority Date: 01/04/1996
• Status: Expired due to Term

• Current Assignee: Polaris Innovations Limited (Wi-LAN Inc.)
• Original Assignee: Siemens AG
• Inventors: Federl, Peter
• Primary Examiner(s): Hafiz, Tariq R.
• Assistant Examiner(s): Rhodes, Jason W.
• Application Number:
  US08/779,365
• Time in Patent Office: 974 Days
• Field of Search: 706/15, 706/16, 706/20, 706/25, 706/26, 706/27, 706/40, 706/916
• US Class Current: 706/15
• CPC Class Codes: G01R 13/206 Arrangements for obtaining a 3- dimensional representation stereoscopic T.V. H04N13/00
• CPC Class Codes: G01R 31/31912 Tester/user interface
• CPC Class Codes: H01L 22/20 Sequence of activities consisting of a plurality of measurements, corrections, marking or sorting steps
• CPC Class Codes: H01L 2924/00 Indexing scheme for arrangements or methods for connecting or disconnecting semiconductor or solid-state bodies as covered by H01L24/00
• CPC Class Codes: H01L 2924/0002 Not covered by any one of groups H01L24/00, H01L24/00 and H01L2224/00
• CPC Class Codes: Y10S 706/916 Electronic or computer, internal or network, circuit

Abstract

In an apparatus and a method for detecting and assessing a spatially discrete dot pattern disposed in a multidimensional coordinate system, each dot in the pattern assumes at least two differentiatable status values. A measuring device records the coordinate values and status values of each dot of the multidimensional spatial dot pattern. A memory stores data that correspond to the recorded coordinate values and status values of each dot of the multidimensional spatial dot pattern. A computer into which the stored data are entered and in which a coordinate counter for each coordinate value of a coordinate axis is determined from the stored data, is associated with the memory. The value of the coordinate counter is formed from the number of detected dots of the coordinates that have a predetermined status value. A neural network is associated with the computer. An n-dimensional input vector with components formed from the calculated coordinate counters of each dot of the spatially discrete dot pattern is entered in the neural network. The neural network calculates an output vector by comparing the calculated input vector of the measured dot pattern with stored set-point vectors obtained on the basis of exemplary dot patterns. The neural network assigns a classification value of the measured dot pattern from the output vector ascertained and outputs it.

Claims

1. An apparatus for detecting and assessing a spatially discrete dot pattern disposed in a multidimensional coordinate system, in which each dot in the pattern assumes at least two differentiatable status values, the apparatus comprising:

a measuring device for recording coordinate values and status values of each dot of a multidimensional spatial dot pattern;

a memory associated with said measuring device for storing data corresponding to the recorded coordinate values and status values of each dot of the multidimensional spatial dot pattern;

a computer associated with said memory for receiving the stored data, for determining a coordinate counter for each coordinate value of a coordinate axis from the stored data, and for forming a value of the coordinate counter from a number of detected dots of coordinates having a predetermined status value; and

a neural network associated with said computer for receiving an n-dimensional input vector with components formed from the calculated coordinate counters of each dot of the spatially discrete dot pattern, for calculating an output vector by comparing the calculated input vector of the measured dot pattern with stored set-point vectors obtained on the basis of exemplary dot patterns, and for assigning a classification value of the measured dot pattern from the ascertained output vector and outputting the classification value, said neural network having three hidden layers.

2. The apparatus according to claim 1, wherein said three hidden layers are an input layer assigned to the input vector, an output layer assigned to the output vector, and a hidden layer disposed between said input layer and said output layer.

3. The apparatus according to claim 1, wherein at least one of the following numerical values are assigned to a respective one of n components of the n-dimensional input vector:

a proportion of coordinate values having more than a predetermined number of dots of the dot pattern having a predetermined status value;

a proportion of echo dots of dots of the dot pattern having a predetermined status value;

a parameter of the fluctuations of dots of adjacent coordinate values of the dot pattern having a predetermined status value; and

a proportion of coordinate values having fewer than a predetermined number of the corresponding coordinates of dots of the dot pattern having a predetermined status value.

4. The apparatus according to claim 1, wherein at least one of the following numerical values are assigned to a respective one of n components of the n-dimensional input vector:

an equidistance of dots of the dot pattern having a predetermined status value in a periodic partial region of the dot pattern; and

a number of at least partially cohesive partial regions of dots of the dot pattern having a predetermined status value in predetermined coordinate directions.

5. The apparatus according to claim 1, wherein numerical values assigned to n components of the n-dimensional input vector are normalized in a numerical range of numbers between -1 and +1.

6. An apparatus for detecting and assessing a spatially discrete dot pattern disposed in a multidimensional coordinate system, in which each dot in the pattern assumes at least two differentiatable status values, the apparatus comprising:

a measuring device for recording coordinate values and status values of each dot of a multidimensional spatial dot pattern representing a failure pattern of a physically cohesive block of memory cells of a semiconductor memory of a plurality of semiconductor memories constructed on a main surface of a semiconductor wafer;

a memory associated with said measuring device for storing data corresponding to the recorded coordinate values and status values of each dot of the multidimensional spatial dot pattern;

a computer associated with said memory for receiving the stored data, for determining a coordinate counter for each coordinate value of a coordinate axis from the stored data, and for forming a value of the coordinate counter from a number of detected dots of coordinates having a predetermined status value; and

a neural network associated with said computer for receiving an n-dimensional input vector with components formed from the calculated coordinate counters of each dot of the spatially discrete dot pattern, for calculating an output vector by comparing the calculated input vector of the measured dot pattern with stored set-point vectors obtained on the basis of exemplary dot patterns, and for assigning a classification value of the measured dot pattern from the ascertained output vector and outputting the classification value.

7. The apparatus according to claim 6, wherein each dot of the failure pattern has one of three status values, of which one status value is assigned to a "0" failure of the memory cell, one status value is assigned to a "1" failure of the memory cell, and one status value is assigned to a "0" -and-"1" failure of the memory cell.

8. The apparatus according to claim 6, wherein the semiconductor memory represents a 4-MBit memory with physically cohesive memory blocks, each made up of 256 rows and 1024 columns.

9. The apparatus according to claim 6, wherein the memory cells of the semiconductor memory constructed on the main surface of the semiconductor wafer are disposed in a matrix in bit lines or columns and in rows or word lines, and the coordinate counters have a column counter and a row counter.

10. The apparatus according to claim 6, wherein components of the n-dimensional input vector are assigned at least one of the following numerical values:

a deviation from a center of distribution of failure cells of a physical block of the semiconductor memory having a predetermined fail status in at least one of a bit line and word line direction;

a second moment of the mass distribution of failure cells of a physical block of the semiconductor memory with a predetermined fail status in at least one of the bit line and word line direction;

a parameter for a global monotonous distribution of failure cells of a physical block of the semiconductor memory having a predetermined fail state; and

a parameter for a periodicity of the fail pattern of failure cells of a physical block of the semiconductor memory having a predetermined fail state.

11. The apparatus according to claim 6, wherein components of the n-dimensional input vector are assigned at least one of the following numerical values:

a proportion of at least one of bit lines and word lines of failure cells of a physical block of the semiconductor memory having a predetermined fail state with more than one specified first limit value on an affected row;

a proportion of at least one of bit lines and word lines of failure cells of a physical block of the semiconductor memory having a predetermined fail state with more than one specified second limit value on the affected row, wherein at least one of the affected word lines and bit lines have an echo in adjacent physical blocks;

a parameter for fluctuations of failure cells of a physical block of the semiconductor memory having a predetermined fail state between at least one of adjacent bit lines and adjacent word lines;

a proportion of at least one of bit lines and word lines having failure cells of a physical block of the semiconductor memory having a predetermined fail state with fewer than a third limit value of failure bits on at least one of the affected word line and bit line;

a parameter for equidistance of at least one of failure-tainted bit lines and word lines having a periodic pattern;

a parameter for a number of larger groups of failure cells in at least one of a bit line and word line direction;

a proportion of at least one of bit lines and word lines having intermediate-layer short circuits in at least one of the bit line and word line direction;

a total number of failure cells;

a parameter for failure cells occurring at an increased rate on a peripheral region of the wafer; and

a ratio between at least one of longer failure-tainted word lines and bit lines in at least one of the bit line and word line direction.

12. The apparatus according to claim 6, wherein all of the semiconductor memories of a wafer are measured and examined by said measuring device, and said neural network outputs a classification of memory cell failure patterns applied to the entire wafer, from calculated output vectors of data of all of the physical blocks or all of the memory chips.

13. The apparatus according to claim 12, including wafer distributions in which the failure patterns of physical blocks of the memory cells are constructed at an increased rate at selected regions on the surface of the wafer, for training the neural network to make a classification of memory cell failure patterns applying to the entire wafer.

14. A method for detecting and assessing a spatially discrete dot pattern disposed in a multidimensional coordinate system, in which each dot in the pattern assumes at least two differentiatable status values, the method which comprises:

recording coordinate values and status values of each dot of a multidimensional spatial dot pattern with a measuring device;

storing in a memory data corresponding to the recorded coordinate values and status values of each dot of the multidimensional spatial dot pattern;

entering the stored data in a computer associated with the memory;

determining a coordinate counter for each coordinate value of a coordinate axis from the stored data with the computer, and forming a value of the coordinate counter from a number of detected dots of the coordinates having a predetermined status value;

forming n components of an n-dimensional input vector from the calculated coordinate counters of each dot of the spatially discrete dot pattern;

assigning at least one of the following numerical values to a respective one of n components of the n-dimensional input vector;

a proportion of coordinate values having more than a predetermined number of dots of the dot pattern having a predetermined status value;

a proportion of echo dots of dots of the dot pattern having a predetermined status value;

a parameter of fluctuations of dots of adjacent coordinate values of the dot pattern having a predetermined status value; and

a proportion of coordinate values having fewer than a predetermined number of corresponding coordinates of dots of the dot pattern having a predetermined status value;

entering the n-dimensional input vector into a neural network;

calculating and outputting an output vector with the neural network by comparing the calculated input vector of the measured dot pattern with stored set-point vectors obtained on the basis of exemplary dot patterns; and

assigning and outputting a classification value of the measured dot pattern from the output vector ascertained with the neural network.

15. A method for detecting and assessing a spatially discrete dot pattern disposed in a multidimensional coordinate system, in which each dot in the pattern assumes at least two differentiatable status values, the method which comprises:

recording coordinate values and status values of each dot of a multidimensional spatial dot pattern with a measuring device;

storing in a memory data corresponding to the recorded coordinate values and status values of each dot of the multidimensional spatial dot pattern;

entering the stored data in a computer associated with the memory;

determining a coordinate counter for each coordinate value of a coordinate axis from the stored data with the computer, and forming a value of the coordinate counter from a number of detected dots of the coordinates having a predetermined status value;

forming n components of an n-dimensional input vector from the calculated coordinate counters of each dot of the spatially discrete dot pattern;

assigning at least one of the following numerical values to a respective one of n components of the n-dimensional input vector;

an equidistance of dots of the dot pattern having a predetermined status value in a periodic partial region of the dot pattern; and

a number of at least partially cohesive partial regions of dots of the dot pattern having a predetermined status value in predetermined coordinate directions;

entering the n-dimensional input vector into a neural network;

calculating and outputting an output vector with the neural network by comparing the calculated input vector of the measured dot pattern with stored set-point vectors obtained on the basis of exemplary dot patterns; and

assigning and outputting a classification value of the measured dot pattern from the output vector ascertained with the neural network.

16. A method for detecting and assessing a spatially discrete dot pattern disposed in a multidimensional coordinate system, in which each dot in the pattern assumes at least two differentiatable status values, the method which comprises:

recording coordinate values and status values of each dot of a multidimensional spatial dot pattern with a measuring device;

storing in a memory data corresponding to the recorded coordinate values and status values of each dot of the multidimensional spatial dot pattern;

entering the stored data in a computer associated with the memory;

determining a coordinate counter for each coordinate value of a coordinate axis from the stored data with the computer, and forming a value of the coordinate counter from a number of detected dots of the coordinates having a predetermined status value;

forming n components of an n-dimensional input vector from the calculated coordinate counters of each dot of the spatially discrete dot pattern;

normalizing numerical values assigned to n components of the n-dimensional input vector in a numerical range of numbers between -1 and +1;

entering the n-dimensional input vector into a neural network;

calculating and outputting an output vector with the neural network by comparing the calculated input vector of the measured dot pattern with stored set-point vectors obtained on the basis of exemplary dot patterns; and

assigning and outputting a classification value of the measured dot pattern from the output vector ascertained with the neural network.

17. A method for detecting and assessing a spatially discrete dot pattern disposed in a multidimensional coordinate system, in which each dot in the pattern assumes at least two differentiatable status values, the method which comprises:

recording coordinate values and status values of each dot of a multidimensional spatial dot pattern with a measuring device;

storing in a memory data corresponding to the recorded coordinate values and status values of each dot of the multidimensional spatial dot pattern;

entering the stored data in a computer associated with the memory;

determining a coordinate counter for each coordinate value of a coordinate axis from the stored data with the computer, and forming a value of the coordinate counter from a number of detected dots of the coordinates having a predetermined status value;

forming n components of an n-dimensional input vector from the calculated coordinate counters of each dot of the spatially discrete dot pattern;

entering the n-dimensional input vector into a neural network;

calculating and outputting an output vector with the neural network by comparing the calculated input vector of the measured dot pattern with stored set-point vectors obtained on the basis of exemplary dot patterns; and

assigning and outputting a classification value of the measured dot pattern from the output vector ascertained with the neural network; and

representing a failure pattern of a physically cohesive block of memory cells of a semiconductor memory of a plurality of semiconductor memories constructed on the main surface of a semiconductor wafer, with the dot pattern.

18. The method according to claim 17, which comprises assigning one of three status values to each dot of the failure pattern, of which one status value is assigned to a "0" failure of the memory cell, one status value is assigned to a "1" failure of the memory cell, and one status value is assigned to a "0" -and-"1" failure of the memory cell.

19. The method according to claim 17, which comprises representing a 4-MBit memory with physically cohesive memory blocks each made up of 256 rows and 1024 columns, with the semiconductor memory.

20. The method according to claim 17, which comprises placing the memory cells of the semiconductor memory constructed on the main surface of the semiconductor wafer in a matrix in bit lines or columns and in rows or word lines, and assigning a column counter and a row counter to the coordinate counters.

21. The method according to claim 17, which comprises assigning components of the n-dimensional input vector at least one of the following numerical values:

a deviation from a center of distribution of failure cells of a physical block of the semiconductor memory having a predetermined fail status in at least one of a bit line and word line direction;

a second moment of a mass distribution of failure cells of a physical block of the semiconductor memory with a predetermined fail status in at least one of the bit line and word line direction;

a parameter for a global monotonous distribution of failure cells of a physical block of the semiconductor memory having a predetermined fail state; and

a parameter for a periodicity of the fail pattern of failure cells of a physical block of the semiconductor memory having a predetermined fail state.

22. The method according to claim 17, which comprises assigning the components of the n-dimensional input vector at least one of the following numerical values:

a proportion of at least one of bit lines and word lines of failure cells of a physical block of the semiconductor memory having a predetermined fail state with more than one specified first limit value on the affected row;

a proportion of at least one of bit lines and word lines of failure cells of a physical block of the semiconductor memory having a predetermined fail state with more than one specified second limit value on the affected row, wherein at least one of the affected word lines and bit lines have an echo in adjacent physical blocks;

a parameter for the fluctuations of failure cells of a physical block of the semiconductor memory having a predetermined fail state between at least one of adjacent bit lines and adjacent word lines;

a proportion of at least one of bit lines and word lines having failure cells of a physical block of the semiconductor memory having a predetermined fail state with fewer than a third limit value of failure bits on at least one of the affected word line and bit line;

a parameter for the equidistance of at least one of failure-tainted bit lines and word lines having a periodic pattern;

a parameter for a number of larger groups of failure cells in at least one of the bit line and word line direction;

a proportion of at least one of bit lines and word lines having intermediate-layer short circuits in at least one of the bit line and word line direction;

a total number of failure cells;

a parameter for failure cells occurring at an increased rate on a peripheral region of the wafer; and

a ratio between at least one of longer failure-tainted word lines and bit lines in at least one of the bit line and word line direction.

23. The method according to claim 17, which comprises measuring and examining all of the semiconductor memories of a wafer with the measuring device, and outputting a classification of memory cell failure patterns applying to the entire wafer with the neural network from calculated output vectors of data of all of the physical blocks or all of the memory chips.

24. The method according to claim 23, which comprises training the neural network to make a classification of memory cell failure patterns applying to the entire wafer by providing wafer distributions in which failure patterns of physical blocks of the memory cells are constructed at an increased rate at selected regions on the surface of the wafer.