Method for storing and reading data in a multilevel nonvolatile memory, and architecture therefor

1. Field of the Invention

The present invention relates to a method for storing and reading data in a multilevel nonvolatile memory and to an architecture therefor.

2. Description of the Related Art

According to the most recent market surveys, the most promising applications of nonvolatile memories, in particular EPROM and FLASH memories, in the coming years will mainly regard data storage in the digital audio/video sector, which is currently undergoing a marked expansion.

It is known that the aforesaid applications require increasingly large memories, for example to enable storage of several music pieces on a same medium or to increase the photographic quality (for example, by increasing the number of pixels).

An important design technique therefor includes the possibility of programming each memory cell at a level chosen from among a plurality of levels. At present, the voltage levels usable for programming a cell are binary levels (equal to m, with m=2 n , where n is the number of information bits that can be stored in the cell). In practice, the law that governs multilevel reading and writing is of a binary type, in so far as memories handle binary data, i.e., at two voltage levels (either high or low, corresponding, from an electrical standpoint, to ground voltage and supply voltage).

Currently memories with two bits per cell, i.e., four-level memories, are in an advanced stage of development and enable doubling of the capacity of the memory. In addition, memories with an even higher number of bits per cell, namely with three or even four bits per cell, corresponding to eight and sixteen levels, are under study. For these memories, above all in the case of sixteen levels, it is very difficult to use the same circuits as for four-level memories; consequently, the time spent in developing products increases considerably, and the know-how acquired remains unexploited. In fact, from the standpoints of the design and engineering development, multilevel architecture is very burdensome.

In order to prevent the need for technical staff to continue to develop products that involve so much expenditure, it is therefore preferable to develop architectures that exploit prior know-how to the full, enabling the design of multilevel memories to advance by short steps and departing as little as possible from the prior art.

An embodiment of the present invention provides a management method and an architecture that allow an increase in the storage capacity of a nonvolatile memory, without requiring complete re-design of the memory with respect to the prior art. The method for stores data in a multilevel storage device that includes a plurality of memory locations, each of which can be programmed at a plurality of voltage levels. The method includes programming each of the memory locations at any of N voltage levels, where N is a non-power of two, depending on where a value for storage in the memory location falls among N- 1 thresholds.

For a better understanding of the present invention, a preferred embodiment thereof is now described, purely by way of non-limiting example, with reference to the attached drawings, wherein:

FIGS. 1 a and 1 b show the organization of the voltage levels in two adjacent memory cells in two different cases;

FIGS. 2 a and 2 b show the position of the reference voltages that can be used with the organization of the levels illustrated in FIGS. 1 a and 1 b;

FIG. 3 shows a read circuit for pairs of adjacent cells programmed according to the organization shown in FIGS. 1 a and 1 b and using the references of FIGS. 2 a and 2 b;

FIG. 4 shows the waveforms of the timing signals during reading with the present organization;

FIG. 5 shows a flow-chart for storing voltage levels in the two cases of FIGS. 1 a and 1 b ; and

FIG. 6 is a flow-chart for reading the voltage levels in the two cases illustrated of FIGS. 1 a and 1 b.

According to an embodiment of the invention, in each cell of a multilevel memory a non-binary number of levels is stored, in particular, a number of levels m=(2 n +2), for example six, is stored. Since in any case the datum to be supplied to the memory must be of a binary type, an encoding method is provided that enables association of the stored non-binary levels to binary codes.

To this aim, contiguous pairs of cells are associated together, so that they can be read and written simultaneously. In the case, provided by way of example, of storing six levels per cell, which is described hereinafter, without the invention being limited to the specific case, this corresponds to the possibility of storing twelve different levels, which are the information content of each pair of cells. Since 12=8+4, this information content corresponds to having 3+2 bits, namely, 5 bits every two memory cells. For example, in a 128-Mcell memory, the binary information content is (5/2)*128=320 Mbits, instead of 256 Mbits that may be obtained with a four-level memory (wherein each cell can be programmed at four different voltage levels).

With an organization of this sort, the memory cells are read in pairs using read circuits associated to each memory cell and able to discriminate at least 6 levels. Since the memory cells are read in pairs, a pair of memory cells or adjacent physical cells—hereinafter also referred to as adjacent memory locations—forms a virtual cell and is addressed by a single address (Ax, Ay).

In the above situation, according to a possible solution, each set of five bits to be written or read is broken down into a first set of bits made up of three bits and into a subsequent set of bits made up of two bits. The first set of bits requires eight different levels to be stored. These eight different levels are stored as follows: the first location stores the first six levels, and the second location stores the remaining two levels, which are preferably chosen from between the two higher levels usable in the second location; the second set of bits requires four different levels, which must be stored in a location other than that used for the first set of bits.

Consequently, when the first set of bits encodes one of the first six levels of the first set of bits (which, as has been said, can be stored in the first location), the second set of bits can be stored in the second location. This is the situation shown in FIG. 1 a , which illustrates the six levels usable in the first location and the four levels usable in the second location. In this case, the two higher levels of the second location are “forbidden” levels; i.e., they cannot be programmed.

When, instead, the seventh level and the eighth level of the first set of bits (which, as has been said, are stored in the second location) are to be stored, the second set of bits cannot be stored in the second location, and thus the first location is used. This is the situation shown in FIG. 1 b , which illustrates level 7 and level 8 of the first set of bits stored in the second location and the four levels of the second set of bits stored in the first location. In this case, the four lower levels of the second location and the two higher levels of the first location are “forbidden” levels; i.e., they cannot be used.

In practice, with the encoding method shown in FIGS. 1 a and 1 b , reading of the two higher levels of the second location indicates which of the following two situations is present: if the last two levels of the second location have not been used, this means that reading the first location supplies the first set of three bits and reading the second location supplies the second set of two bits; if, instead, the two higher levels of the second location have been used, this means that reading the first location supplies the second set of two bits and reading the second location supplies the first set of three bits.

In either case, the last two levels of the two locations cannot be used simultaneously.

FIGS. 2 a and 2 b show the arrangement of the reference levels used for reading the memory locations in the two cases represented in FIGS. 1 a and 1 b . As for circuits according to the prior art, for discriminating six levels, five references REF 1 -REF 5 are used, which are arranged approximately halfway between pairs of successive levels.

FIG. 3 shows a block diagram of the circuitry for decoding the information contained in two adjacent memory locations, designated by 10 a and 10 b , organized in the way described above and belonging to a memory device 5 . The circuitry of FIG. 3 implements a parallel sensing which, as is known, is the simplest and provides the best performance from an electrical standpoint.

In the diagram of FIG. 3, the memory locations 10 a , 10 b are each connected to a respective current-to-voltage converter 11 a , 11 b via an addressing circuit (known and not illustrated). The output of the current-to-voltage converter 11 a is connected to a first input of five sense amplifiers 12 a 1 - 12 a 5 , which have a second input that receives one of five references REF 1 -REF 5 via respective current-to-voltage converters 13 a . Likewise, the output of the current-to-voltage converter 11 b is connected to a first input of five sense amplifiers 12 b 1 - 12 b 5 , the second input whereof receive one of five references REF 1 -REF 5 via respective current-to-voltage converters 13 b.

The outputs of the sense amplifiers 12 b 4 , 12 b 5 connected to the second location 10 b and receiving the references REF 4 and REF 5 are connected to respective inputs of a NOR gate 15 , which is cascade-connected to an inverter 16 supplying a signal en 2 , of a high logic level only when at least one of the two sense amplifiers 12 b 4 , 12 b 5 connected thereto has a high output. In practice, the signal en 2 is high when the second memory location stores the seventh or the eighth level of the first set of bits; otherwise, it is zero. Consequently, its value indicates whether the first set of bits is to be read in the first memory location 10 a and the second set of bits is to be read in the second memory location 10 b (en 2 =0), or vice versa (en 2 =1).

The outputs of the sense amplifiers 12 a 1 - 12 a 5 are connected to a 3-bit encoder 20 via respective CMOS switches 21 , controlled by the signal en 2 so as to be closed when en 2 =0. The outputs of the sense amplifiers 12 b 1 - 12 b 5 are connected to the 3-bit encoder 20 via respective CMOS switches 22 , controlled by the signal en 2 so as to be closed when en 2 =1. Thereby, when en 2 =0, the 3-bit encoder 20 receives the outputs of the sense amplifiers 12 a 1 - 12 a 5 , and when en 2 =1, the 3-bit encoder 20 receives the outputs of the sense amplifiers 12 b 1 - 12 b 5 . The 3-bit encoder 20 moreover receives the signal en 2 so as to discriminate the fifth and sixth levels from the seventh and eighth levels.

The outputs of the sense amplifiers 12 a 1 - 12 a 3 are connected to a 2-bit encoder 23 via respective CMOS switches 24 , controlled by the signal en 2 so as to be closed when en 2 =1. The outputs of the sense amplifiers 12 b 1 - 12 b 3 are connected to the 2-bit encoder 23 via respective CMOS switches 25 , controlled by the signal en 2 so as to be closed when en 2 =0. Thereby, when en 2 =0, the 2-bit encoder 23 receives the outputs of the sense amplifiers 12 b 1 - 12 b 3 , and when en 2 =1, the 2-bit encoder 23 receives the outputs of the sense amplifiers 12 a 1 - 12 a 3 .

The 3-bit encoder 20 and the 2-bit encoder 23 are connected to respective output lines 26 , along which switches 27 are arranged and controlled by the output signals of an address encoder, which in turn receives three address signals Az( 0 ), Az( 1 ) and Az( 2 ).

Reading the information contained in the two memory locations 10 a , 10 b takes place whenever there is a variation in the addresses {Ax, Ay} (FIG. 4) which causes, in a known way, generation of a pulse of an address-transition detection signal ATD (not shown), which in turn generates all the read sync signals known to the persons skilled in the art. Consequently, the contents of the two memory locations 10 a , 10 b addressed are read and encoded by the circuitry of FIG. 3, taking the values of the first set of bits and of the second set of bits in the way previously described. The bits thus encoded are outputted sequentially using an appropriate combination of addresses {Az n }, as shown in the timing of FIG. 4 .

The essential operations for writing and reading a virtual cell of address (Ax, Ay) are shown in FIGS. 5 and 6.

According to FIG. 5, which regards writing, initially a set of bits that are to be stored is divided into two or more sets, step 30 . Then, the virtual cell of address (Ax, Ay) is addressed, step 31 . It is verified whether the first set of bits (in the example considered above, three bits) encodes a number higher than the number of levels that can be stored in a single location (in the example, whether level 7 or level 8 is to be stored), step 32 . If not, the first set of bits is stored in a first location (in the example, in the first memory location 10 a ); i.e., the first location is programmed at a level corresponding to the value encoded by the first set of bits, step 33 , and the second set of bits is stored in a second location (in the example shown, in the second memory location 10 b ); i.e., the second location is programmed at a level given by the value encoded by the second set of bits, step 34 . In the affirmative, the first set of bits is stored in the second location, step 35 , and the second set of bits is stored in the first location (programming of the first location at the value encoded by the second set of bits), step 36 .

According to FIG. 6 and to what explained for the circuit of FIG. 3, during reading, first the virtual cell of address (Ax, Ay) is addressed, step 40 . Then the cell is read, step 41 . It is verified whether one of the levels of the second location reserved to the first set of bits has been written (in the example, whether level 5 and/or level 6 of the second memory location 10 b has been written), step 42 . If not, the read circuits of the first location are connected to the encoder of the first set of bits (in the example shown, to the 3-bit encoder 20 ), step 43 , and the read circuits of the second location are connected to the encoder of the second set of bits (in the example to the 2-bit encoder 23 ), step 44 . In the affirmative, the read circuits of the first location are connected to the encoder of the second set of bits, step 45 , and the read circuits of the second location are connected to the encoder of the first set of bits, step 46 .

The solution described herein affords the important advantage of implementing a non-binary architecture for reading the information stored in nonvolatile memory cells, with minimal overall dimensions and a minimal logic complexity, exploiting known multilevel techniques.

Finally, it is clear that numerous modifications and variations may be made to the method and memory described and illustrated herein, all falling within the scope of the invention, as defined in the attached claims. In particular, the same architecture may be applied for storing any non-binary number of levels in each memory location. The choice of the particular location to be used for storing each set of bits is arbitrary. The levels associated to each set of bits can be chosen in a way different from the one illustrated. The levels encoded by the first set of bits and stored in the second memory location may be other than the ones illustrated (for instance, the second memory location 10 b could store levels 1 and 2, instead of 7 and 8, of the first set of bits; in this case, the first memory location should store levels 3-8 of the first set of bits). Finally, reading may be performed via a different type of multilevel sensing, such as merely serial, dichotomous, mixed parallel-serial sensing, etc. Furthermore, from a theoretical standpoint it is possible to store more than two sets of bits in more than two adjacent memory locations.

All of the above U.S. patents, U.S. patent application publications, U.S. patent applications, foreign patents, foreign patent applications and non-patent publications referred to in this specification and/or listed in the Application Data Sheet, are incorporated herein by reference, in their entirety.