GB2279786A - Associative memory systems - Google Patents

Abstract

An associative memory system comprising an M by N array of memory cells forming M words of N trit length are disclosed having single word alterability. Each memory cell further comprises first and second nonvolatile storage devices 25, 26 without the need for additional switching devices. Generally, complementary data is stored in the two storage devices and the cell is interrogated by applying complementary data to the storage devices. A match or mismatch is achieved by looking for current through either of the storage devices. Preferably, the storage devices are one of two types of flash EEPROM transistors, or a SONOS transistor. Each transistor has a gate, drain and source and has write and rewrite capabilities that allows single word alterability in the memory array to be accomplished by applying a high voltage to the transistor drain, or by applying a combination of voltages to the drain and gate. <IMAGE>

Description

OPERATION OF MEMORY SYSTEMS This invention relates to a method of storing, retrieving and rewriting information in an associative memory system.

Associative memories or associative content addressable memories (ACAM or CAM) are useful in many applications where a number of records must be searched to see if there is a match to another word. The utility of the ACAM is that the records are searched in parallel and consequently the search can be accomplished in a much shorter time than in a serial search through a conventional organized memory. Despite the apparent advantage of ACAMs for search applications, they have not been widely utilized because of their high cost relative to conventional memories.

In many applications it is desirable to alter one word in the memory without disturbing the other words. Previous implementations of nonvolatile ACAMs using simple two transistor cells have depended on various write mechanisms.

One type of write mechanism alters all cells on an integrated circuit by removing electrons from the storage gate by either exposure to ultraviolet light or application of voltage to the substrate. Another type of write mechanisms changes the stored information on an entire column by causing a charge (electron) to tunnel to the storage regions of the transistors in that column by applying a high voltage on the gate line.

Selected methods utilizing associative memory cells are illustrated in the following references. In United States Patent No. 3,693,174, issued to MacKnight, one associative memory array system is comprised of selected memory cells.

These memory cells are further comprised of at least two storage transistors and at least five semiconductor switches, preferably transistors. Both storage transistors are nonvolatile type transistors which allow the information stored in them to be read without deteriorating the signal stored therein.

In United States Patent No. 3,750,115, issued to Mundy, a basic memory cell for use in an associative array which consists of three transistors and two voltage variable capacitors is described. To use this cell in a parallel associative search, the flag line of all the cells to be searched in parallel must be common. This means that the flag line must run parallel to the data lines.

In United States Patent No. 3,800, 297 issued to Mundy et al., a basic memory cell is used in an associative memory array that is composed of four transistors: two storage transistors and two switching transistors. By applying a large negative pulse to a word line, writing is possible in all the cells connected to the word line.

In United States Patent No. 4,064,494, issued to Dickson et al. a memory cell that allows writing is disclosed but the method to write requires the erasure of all contents in the content addressable memory. Also, unselected, previously stored words are protected in their selected locations while new words are written into their respective cells.

To understand the operation of ACAMs and their application, reference may be made to Fig. 1 of the accompanying drawings, which shows a basic associative memory system, or ACAM 10. In this memory the contents of an "n" position register 12 are simultaneously compared with "m" words contained in array 14. If a match or matches are found, a flag is present on the appropriate y, output or outputs. For maximum utility, it is desirable to have each location in the input register and the memory capable of three state, "1", ot, and "don't care". Locations capable of storing these three states have been referred to as "trits".

In accordance with the invention, the method of storing, retrieving and rewriting information in an associative memory system comprises the steps of: providing an M by N array of memory cells in which each row of N memory cells represents one word of N trit length and each column of M memory cells represents a corresponding trit in each of the M words; storing information in the array in a predetermined pattern to selectively enable portions of the array; applying an input information to the array; sequentially performing read and then search operations to operate on the input information in accordance with the pattern of stored information; and applying a changing signal to a single N-trit word in the array to alter the contents of that word but not the contents of the remaining words.

The invention is further described below, by way of example, with reference to the remaining Figures of the accompanying drawings, in which: Fig. 2 is a schematic diagram of a memory cell operable in accordance with the present invention; Fig. 3 illustrates a schematic diagram of the memory cell of Fig. 2 incorporated into a word row; Fig. 4 shows a cross-section view of a storage transistor type used in the present invention; and Fig. 5 shows a cross-sectional view of an alternative storage transistor type which could be used in the present invention.

The memory cell 20 shown schematically in Fig. 2 comprises two transistors 21 and 22, both capable of non-volatile information storage. The dashed lines in gates 25 and 26 of transistors 21 and 22, ~~~~~~~~~~~~~~~~~~~~~~~~ respectively, indicate that charge may be stored in gate regions 28 and 29 to set the state of each transistor 23 and 24. This stored charge affects the conduction threshold (voltage) of each transistor 21 and 22 so that the charge state may be detected by sensing the conduction of the transistor. Preferably, only two states per transistor (e.g. on and off) are allowed; this choice greatly simplifies conduction sensing and contributes to a design highly tolerant to variation in circuit parameters. Furthermore, only a minimum of two transistors are needed to store a trit.

In one specific embodiment, a well known "floating gate transistor is used. In such a device, a polycrystalline silicon layer is sandwiched between a substrate and a control gate. This sandwiched layer is completely surrounded by a dielectric which insulates the gate such that when it is charged, the charge remains under normal operation and storage conditions. For example, in the case of an n-channel transistor, putting a negative charge on the floating gate raises the threshold conduction voltage, while removing the charge lowers the threshold conduction voltage. In normal operation, one state is set by putting enough electrons onto the floating gate such that it will not conduct when a read mode voltage (e.g. 5 volts (V)) is put on the control gate. The other state is established by removing enough electrons such that the transistor conducts when the read mode voltage is applied to the control gate.

When a deselect voltage, usually ground, is applied to the control gate, the transistor does not conduct in either case.

In cell 20 (Fig. 2), preferably the "1" state is defined as transistor 21 "on" and transistor 22 "off".

The "0 " state is then defined as transistor 22 "on" and transistor 21 "off". The "don't care" state is when both transistors are "off". A search is performed by applying a first and second voltage to gates 25 and 26, respectively. The two voltages applied to the two gates are normally complements, that is the read voltage (5 V) is applied to one gate and the deselect voltage (normally o V) is applied to the other. Thus, if cell 20 is in the "1" state and 5 V is applied to gate 25 of transistor 21 while gate 26 of transistor 22 is grounded, current flows through transistor 21. This is detected as a "mismatch".

If the gate voltages are reversed, no current flows through either transistor 21 or 22 because transistor 21 is deselected and transistor 22 is in the "off" state.

This is detected as a "match" on this trit.

A word of memory is shown schematically in Fig. 3. In this word, the drains of "n" cells are connected in parallel to a first sense line 31 and a second sense line 32 and all the sources are tied to a common line 33. The gates of each cell are connected to the corresponding cell of the input register. Each cell of the input register normally generates two complementary outputs Di and Di. If all cells match, no current flows between either of sense lines 31 or 32 and the common source line 33. In this case the word matches the input. Any column of trits can be masked by holding both of the output lines from the corresponding cell of the input register low. Similarly, if one of the trits in the jth word is in the "don't care" state, the determination of the match between the input register and that word is made based on the other trits in the word, since that cell never conducts.

Examination of Fig. 3 shows that the restriction of changing only one word constrains the memory to use a write mechanism which causes a change in the state of the memory transistors depending solely on the voltage applied on lines 31 or 32 to the transistor drains, or depending on a combination of the voltages applied on Di or Di on the gate and on lines 31 or 32 on 1 1 the drain.

In a preferred application, the write mechanism includes an erase mode and a program mode. The erase mode occurs by applying a high voltage to both drains 23 and 24 with gates 25 and 26 held at ground. The program mode occurs by applying a voltage to either gate 25 or 26, or both and a high voltage to their respective drain 23 or 24. In the erase mode, the common source line 33 is allowed to float. In the program mode, the common source line is held at ground.

A recently introduced memory technology called flash electrically erasable programmable read only memory (EEPROM) provides the desired write mechanism functionality. This technology uses injection of channel hot electrons onto the floating gate to charge the storage gate negative and removal of electrons from the floating gate by application of a high voltage on the drain to charge the storage gate positive. Two specific cases are shown by way of illustration in Figs. 4 and 5.

In Fig. 4, a storage transistor 40 is formed from a merged gate transistor with a source 43 and a drain 44. This transistor 40 is viewed as two transistors in series, one consisting of a stacked gate transistor formed by an overlap of a control gate 42 on a floating storage gate 41, and the other formed by an overlap of control gate 42 on a substrate, and connected by a virtual source-drain. Preferably, transistor 40 is designed such that when control gate 42 is taken to 5 V, a conducting path is formed between source 43 and drain 44 when floating gate 41 is charged positive. Otherwise, no conducting path is formed if floating gate 41 is charged negatively. Floating gate 41 may be charged positively by taking drain 44 to a high voltage, typically between 12 V and 20 V, while control gate 42 is held at ground. Conversely, floating gate 41 is charged negatively by taking control gate 42 to a high voltage, typically about 12 V, while drain 44 is taken to an intermediate high voltage, typically about 8 V. Current flows between source 43 and drain 44 and generates hot electrons in the channel region of the transistor near drain 44. Some of these electrons are attracted to the floating gate 41 which has been capacitively coupled to a positive voltage by the voltage applied to control gate 42. Because the hot electrons are generated by the channel current, programming occurs only at the intersection of large positive drain and control gate voltages.

In another specific embodiment, storage transistor 50 (Fig. 5) is also composed of two transistors merged by a virtual source-drain. In this case a source 54 and a drain 55 are separated by a stacked gate transistor formed by an overlapping of a control gate 51 on a floating storage gate 52, and a "sidewall1' gate 53 formed adjacent to the stacked gate transistor with no intervening diffusion. Sidewall gate 53 may be a floating gate; in which case, its potential is set capacitively by the potential of adjacent potentials (preferably that of control gate 51), or a contact may be made to it. In either case of operation, the sidewall gate 53 potential is generally the same.

When control gate 51 goes to 5 V to read, the sidewall gate 53 transistor is in a conducting state and the integrated transistor has a conducting path between drain 55 and source 54 if storage gate 52 is charged positive.

If storage gate 52 is charged negatively, there is no conducting path between source 54 and drain 55. If control gate 51 is at ground, sidewall gate 53 is also at a low potential and no conduction can occur between source 54 and drain 55. The storage gate 52 is charged positively by applying a high voltage, typically between 12 V and 20 V, to drain 55 while control gate 51 is at ground. This causes electrons to tunnel from storage gate 52 to the drain 55 leaving gate 52 positively charged. Storage gate 52 is charged negatively by applying a positive voltage, typically 5 V, to drain 55 while control gate 51 is taken to a high voltage, typically in the range of 10 V to 15 V. Sidewall gate 53 goes to an intermediate voltage. There is a high field region occurring under the dielectric spacer which separates sidewall gate 53 from the stacked gates 51 and 52; this field generates a large density of hot electrons under the source end of floating gate 52. The positive potential arising on storage gate 52 because of capacitive coupling from the control gate attracts some of these hot electrons to storage gate 52.

These two described embodiments of the storage transistor are given by way of examples and are not intended to be an exhaustive listing. For example, an silicon oxide nitride oxide semiconductor (SONOS) transistor which is programmed by channel hot electrons and erased by a high drain voltage would also serve.

What is essential to this invention is two nonvolatile storage transistors connected in an OR configuration, as illustrated in Fig. 2, to form a cell as connected in Fig. 3. Generally, complimentary data is stored in the two storage transistors and the cell is interrogated by applying complementary data to the two transistor control gates and looking for current through either transistor.

A single word can be altered in the memory array by applying a high voltage to the transistor drains or by applying a combination of voltages to the drain and gate.

The invention has now been explained with reference to specific embodiments. Other embodiments will be apparent to those of ordinary skill in the art.

It is therefore not intended that the invention be limited except as indicated by the appended claims.

Claims (6)

1. A method of storing, retrieving and rewriting information in an associative memory system comprising the steps of: providing an M by N array of memory cells in which each row of N memory cells represents one word of N trit length and each column of M memory cells represents a corresponding trit in each of the M words; storing information in the array in a predetermined pattern to selectively enable portions of the array; applying input information to the array; sequentially performing read and then search operations to operate on the input information in accordance with the pattern of stored information; and applying a changing signal to a single N-trit word in the array to alter the contents of that word but not the contents of the remaining words.

2. A method as claimed in claim 1 wherein the memory cells comprise first and second semiconductor memory devices each having electrically controllable first and second threshold voltage levels and each having an input electrode, an output electrode and a common electrode, wherein the input electrodes of the first semiconductor memory devices in the same column of the array are coupled to a first input line in each of the columns of the array, and the input electrodes of the second semiconductor memory devices are coupled to a second input line in each of the columns of the array, and the output electrodes of the first semiconductor devices in the same row are coupled to a first output line in each row of the array, and the output electrodes of the second semiconductor devices of the same row are coupled to a second output line in each row of the array, and the common electrodes of the electrode devices in the same row are coupled to a common line in the row of the array.

3. A method as claimed in claim 2 wherein the step of applying a changing signal comprises the steps of: applying a first signal to the input line of the first semiconductor devices in the same row containing the word; and applying a second signal to the output line of the first semiconductor devices in the same row containing the word.

4. A method as claimed in claim 2 or 3 wherein the step of applying a changing signal comprises the steps of: applying a first signal to the input lines of the semiconductor devices in the same row containing the word; and applying a second signal to the output line of the second semiconductor devices in the same row containing said word.

5. A method as claimed in claim 2, 3 or 4 wherein the step of applying a changing signal comprises the steps of applying a signal to each of the output lines in a select row containing said word.

6. A method of storing, retrieving and rewriting information in an associative memory system substantially as herein described with reference to Figs. 2, 3 & 4 or Figs. 2, 3 & 5 of the accompanying drawings.