JPS6045501B2 - associative memory device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an associative memory device, specifically, a pattern memory device.
plTEL1-se wo Lnwei n↓↓
This book concerns memory devices.

Conventional associative memory devices select and output only stored data that completely matches externally input search data, and do not take into account the possibility of selecting stored data based on ambiguity. .

Therefore, when trying to apply such an unambiguous associative memory to pattern recognition processing, the pattern to be recognized must completely match the standard pattern stored in the associative memory. There is a restriction. However, in general, the pattern to be recognized and the standard pattern rarely match perfectly due to noise, positional deviation, etc., and it is difficult to perform pattern recognition processing using conventional unambiguous associative memory. , we have to introduce ambiguity into the associative operation by using a special error-correcting code and its decoding process. In other words, an error-correcting decoding operation is performed on the principle pattern data that has been encoded to become an error-corrected perfect code, creating a degenerate data set that has ambiguity with respect to the original data, and degenerating data in the degenerate data space. It must be constructed by performing associative operations without conventional ambiguity. In this configuration using conventional associative memory,
The disadvantages are that it introduces ambiguity, increases the operating time for decoding, and increases the amount of hardware. This was done in view of the circumstances described above of the present invention, and by giving an associative memory an associative function with ambiguity using Hamming distance, it is possible to combine search data and Hamming data other than stored data that completely matches the search data. It is an object of the present invention to provide an associative memory device that can also select stored data related by distance.

Now, two data A = [Ala2a3a4], B = [b
1■B3b4] by modulo 2 to obtain.

), dl=A,+Bi (1=1, 2, 3, 4). At this time, Di is either 0 or 1, and the sum of 1 is called the Hamming distance.In other words, the Hamming distance d is expressed as.Of course, d=
0, two data A. ．． B are completely consistent, and as the value of d increases, they become less similar. FIG. 1 and FIG. 2 show an embodiment of the present invention, which is an association having an associative function with the simplest ambiguity using three types of relationships in which the Hamming distance d is 0, less than 2, and 2 or more. It's memory.

Figure 1 shows a unit cell with the necessary logic in a distributed manner.
FIG. 2 shows a cell array having a 4 word x 4 bit configuration, which is used for the unit cells shown in FIG. The memory targeted by the present invention may be a semiconductor memory or other memory. In FIG. 1, 1 is a storage element that stores information Mj, 2 is an input terminal for search data Si, 3 is an input terminal for propagation information Ei-1 from a lower unit cell, and 4 is a lower . This is the input terminal for the propagation information F,-1 from the unit cell.

5 is an exclusive OR gate, 6 and 7 are AND gates including an inverter as part of their input terminals, and 8 is an OR gate.

9 is the output end of the propagation information E, to the upper unit cell, and 10 is the output end of the propagation information E, to the upper unit cell! This is the output terminal for transport information Fi.

Reference numerals 11 to 26 in FIG. 2 are the unit cells shown in FIG.
, 10 correspond to the same symbols in FIG. The same applies to other unit cells. 27 to 30 are search data supply sources; 31 is a potential supply terminal corresponding to logic ゜゜1゛; 3
2 is a potential supply terminal corresponding to logic "0".

33 to 40 are output terminals for association results;
36 is the highest unit cell 14, 18, 22,
Similarly, 37 to 40 are output signals F from the most significant unit cells 14, 18, 22, and 26 of each word.

Now, the unit cell in Fig. 1 has the propagation information Ei-
1, F, -1 and the storage information Mil search data Si, propagation information EI and Fl expressed by the following logical formula are output.

4 represents addition of modulo 2, that is, exclusive OR. According to the above equation (3), the propagation information E, takes logic 1 when Mi(5S,) matches in all lower unit cells.In other words, in each word, Mi
(5S, If there is even one unit cell that does not match, the output Ei of the highest unit cell becomes logic "゜0゛."On the other hand,
Among the unit cells in which the propagation information F, constitutes a word, Mi
When the number of unit cells in which and S, do not match is less than or equal to 1, the logic value is "0", and when the number of unit cells that are mismatched is greater than or equal to 2, the logic value is "1". That is, the output E1 of the most significant unit cell of each word is E. ．． Three types of relationships can be identified by the combination of Fi=F. That is, E=1, F=
0, the number of unmatched unit cells is zero, that is, the Hamming distance between the stored data and the search data is O, and E
= 0, F = 0, the number of unmatched unit cells is 1,
In other words, the Hamming distance between the stored data and the retrieved data is 1
When E=0 and F=1, the number of mismatched unit cells is 2 or more, that is, the Hamming distance between the stored data and the search data is 2 or more. In this way, the cell array shown in FIG. 2 is constructed by the unit cells shown in FIG.
When inputting search data to ~30 and performing an associative operation, E=1, F=0 based on the propagation information E, F output from the highest unit cell 14, 18, 22, 26 of each word.
When memory data of a word is selected, memory data that matches the search data is output, and E=
If the stored data of the word 0, F=0 is selected, the stored data whose Hamming distance with the search data is 1 will be output.

Furthermore, if we focus only on the propagation information F output from the highest unit cell of each word and select the stored data of the word where F=0, the stored data whose Hamming distance with the search data is less than 2 will be output. Therefore, when the stored data of the word F=1 is selected, the stored data having a Hamming distance of 2 or more with respect to the search data will be output.
Configurations that satisfy the logic function shown in the above logical formula (1) have been considered other than in Figure 1, but in any case, one unit cell uses a memory element and about 4 logic gates. As a result, an associative memory having an associative function with ambiguity using Hamming distance can be constructed very easily, making it possible to realize high-speed pattern recognition processing. In addition, in the embodiment, the configuration of an associative memory having an associative function with an ambiguity using a relationship with a Hamming distance of less than 012 and 2 or more was shown, but by further adding logic,
When n is any natural number greater than or equal to 3, the Hamming distance is 0. ．． An associative memory with ambiguity using a relationship of an arbitrary distance less than n, an arbitrary distance less than n, or a certain value or more can also be constructed based on a similar idea.

As an example, a logical formula for propagation information required when n=3 or 4 is shown below. As explained above, according to the present invention, an associative function with ambiguity, which was conventionally achieved using extremely complicated methods, can be realized using extremely small-scale unit cells based on a completely new perspective of using Hamming distance. Since it can be realized in hardware with a repeating structure, the following advantages arise. (1) Unlike the conventional complicated method, the memory device itself has an ambiguous associative function, and no logic device other than the memory device is required.

(2) Therefore, if this associative memory device is applied to a pattern recognition processing device, the processing speed can be increased, the processing device can be made smaller, and the cost can be reduced.

(3) The unit cell of the present associative memory device can be composed of a storage element and several logic gates, and is easy to design.

(4) Since the present content addressable memory device can be designed by arranging unit cells in an array, the present content addressable memory device can be designed with the same amount of effort as that required for designing conventional memory devices. Can be done.

(5) Since the logical operation functions necessary for calculating the Hamming distance are distributed in each unit cell, there is no need to provide a circuit for calculating the Hamming distance other than the unit cell.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing a unit cell of an embodiment of an associative memory device according to the present invention, and FIG. 2 is a diagram showing a memory cell array using the unit cell of FIG. 1. 1...Storage element, 2...Search data input terminal, 3, 4...Propagation information input terminal, 5...
...Exclusive OR gate, 6......AND gate, 7.......OR gate, 9, 10...
Propagation information output end, 11-26... unit cell, 27
~30...Search data supply line.

Claims (1)

[Claims] 1. Calculate the Hamming distance between each stored data and the search data, and calculate the memory that is related to the search data by the Hamming distance, other than the stored data that completely matches the search data. An associative memory device characterized by having a function of also selecting and outputting data. 2. In calculating the Hamming distance, a unit cell is constructed by distributing the necessary logic operation functions in a memory cell corresponding to each bit of stored data, and the unit cell is arranged in an array. 2. The associative memory device according to claim 1, wherein the Hamming distance is calculated by arranging the hamming distance. 3. When selecting and outputting stored data related to the search data by Hamming distance, the Hamming distance is zero, any distance less than a certain value, any distance less than a certain value, or more than a certain value. 2. The associative memory device according to claim 1, wherein storage data having a desired relationship with the search data is outputted from among these relationships.