Circuit and method for reading a non-volatile memory

[0002] The present invention relates to a circuit and a method for reading a non-volatile memory. More particularly, the invention relates to a reading circuit of a non-volatile memory and an associated reader protocol.

[0003] It is known that in a conventional non-volatile memory, a random reading method is normally used in which word lines and bit lines are selected randomly and independently. This procedure entails reading times which are the sum of a sequence of events which can be summarized substantially as follows: propagation of the word line and bit line selection paths; precharge activity; a data evaluation interval; and propagation for data transfer to the output buffer and transition time of the buffer.

[0004] The total reading time, given by the sum of the times of the individual operations listed above, depends not only on the dimensions and architecture of the memory but also on the type of device that is present in the memory, i.e., on the technology that is used. The advantages of the conventional architecture are that it is simple (a minimal number of memory structures), it allows easy timing (because all reading cycles are equal), it provides maximum efficiency for redundancy structures, and it has significantly low consumption because the number of current-consuming circuits is kept to the lowest possible value.

[0005] However, the limitations of the conventional architecture reside in the reading speed, which is very important in large memories. One approach for improving the reading speed is the use of “page mode” reading, in which a plurality of words in parallel, for example eight, are read in a first cycle. Then, the content of the reading operation is stored in a suitable register and the portion of the register that relates to the selected word within the packet of read words is displayed externally. In this manner it is possible to scan the words contained in the packet of read words in a significantly shorter time than with random reading. However, the first reading cycle continues to be dependent on the time of a random selection.

[0006] In any case, page-mode reading is affected by the following drawbacks: a large number of reader circuits (e.g., if eight words of eight bits each are read in parallel, sixty-four sense amplifiers are required); high current absorption during parallel reading; the provision of word registers; the provision of register decoding; a reduction in the effectiveness of redundancy structures (one line at a time can be replaced with a redundant line, but in the case of multiple reading, the possibilities of line fault increase, yet there is still only one redundant line available); and the reading protocol entails a double cycle time, random or page.

[0007] An object of the present invention is to provide a circuit and a method for reading a non-volatile memory which allow performance advantages to be achieved, with respect to the conventional reading approach, and yet are simplified with respect to the page-mode approach. Thus, an object of the present invention is to provide a circuit and a method for reading a non-volatile memory which keep the number of structures of the memory to a minimum.

[0008] Another object of the present invention is to provide a circuit and a method for reading a non-volatile memory which do not require the reduction of the effectiveness of redundancy or the implementation of additional reader and decoder registers. Another object of the present invention is to provide a circuit and a method for reading a non-volatile memory which avoid additional current consumption. Another object of the present invention is to provide a circuit and a method for reading a non-volatile memory which are highly reliable, relatively easy to provide and at competitive costs.

[0009] These objects and others which will become apparent hereinafter are achieved by a method for reading a non-volatile memory including the steps of: providing a first random memory reading cycle; performing, at the end of the random reading cycle, a collective page precharge; subsequently performing a reading cycle of the page or random type, depending on whether the subsequent reading must be performed within the same page or not; and if a page reading cycle is performed, executing, when the data item is captured, a page precharge step in preparation both for page reading and for random reading.

[0010] Further characteristics and advantages of the invention will become apparent from the description of preferred but not exclusive embodiments of the circuit and of the method according to the invention, illustrated only by way of non-limitative examples in the accompanying drawings, wherein:

[0011] FIG. 1 is a schematic view of a first embodiment of an organizational structure of a non-volatile memory according to the present invention;

[0012] FIG. 2 is a schematic view of a second embodiment of an organizational structure of a non-volatile memory according to the present invention;

[0013] FIG. 3 is a chart of the reading timings of a non-volatile memory according to a known reading cycle, which points out the regularity of the timing events (reading cycles), as in the prior art;

[0014] FIG. 4 is a chart of the read timings of a non-volatile memory according to a reading cycle provided with the circuit and the method according to the present invention; and

[0015] FIG. 5 is a chart of the reading timings of a non-volatile memory provided according to the invention and with continuous page changing after an initial random cycle.

[0016] With reference to the above figures, the circuit for reading a non-volatile memory according to the invention operates with a memory which is divided into a right half-matrix 1 and a left half-matrix 2 which are arranged on two opposite half-planes. Each half-matrix has a plurality of memory cells arranged in rows and columns in addition to a plurality of redundancy memory cells. The memory cells are designated by the references ROW-MAT-R and ROW-MAT-L for the right half-matrix 2 and for the left half-matrix 1 respectively.

[0017] Sense amplifiers 3 are arranged between the two half-matrices and comparison bit lines are arranged on different half-matrices. A reference word line is provided for each one of the two half-matrices and each cell selected on the word line of one half-matrix is matched by the selection of a similar cell on the reference word line arranged on the opposite half-matrix. Column multiplexers are arranged symmetrically in each one of the two half-matrices. The two multiplexers are activated simultaneously so that each multiplexer selects, in its respective half-plane, two bit lines to be compared. Control buses YN-YM for the two multiplexers are provided to select a packet of bit lines (the bus YM) and a particular bit line within the packet (the bus YN).

[0018] It is known that a reading cycle for a conventional non-volatile memory has the characteristic of re-stimulating, at each address transition, all the selection paths and then continuing, as timing progresses, until external transfer of the data read from the memory is reached. This necessitates that each reading cycle is provided over a long time period because the time period is a sum of all the events, as mentioned above.

[0019] The invention, in particular, provides a dual reading cycle, following the model of page reading: a random cycle; and a page cycle. The expression “random cycle” is used to define every first reading cycle and every reading operation that does not act within a packet of memory locations which can be selected more quickly than locations belonging to different packets. The expression “page-type read cycle” is instead used to define any reading cycle that acts within a same packet of locations which can be selected more quickly than locations outside the packet.

[0020] A random cycle is performed with the same method as in a conventional-read memory, but an immediate page precharge activity is superimposed thereon. Page precharge occurs after the data item has been captured. Each random cycle is therefore performed as follows. First, a specific word line and a specific bit line are activated at the beginning of a reading cycle. At this point, there is reading timing progress for reading and capturing the data item, with subsequent transfer and display of the data read at the beginning of the memory. As soon as the read data has been captured, page precharge activity is initialized.

[0021] A generic random access occurs only if a subsequent reading must be performed in memory locations other than those of the page, i.e., if the bit line to be selected is outside the packet of bit lines selected by page precharge. A page reading cycle is performed starting from a preceding random cycle (which, as mentioned, already performs page precharge), thus maintaining the word line selection. Then, a page location is selected and the data item of the memory cell is extracted and transferred and thus displayed externally. Finally, page precharge is initialized as soon as the read data item has been captured.

[0022] A set of lines is precharged simultaneously in the time interval between the capture of the current data item and the beginning of a new read cycle. The selection operation occurs very quickly and isolates, among the selected bit lines, the one on which the cell to be read is to be identified. The selected bit line (whose precharge has already been performed in the previous reading cycle) remains under evaluation only for the time required for residual and optionally finer precharge, construction of the cell signal, and evaluation/capture of the cell data item.

[0023] The above-described reading protocol allows path selection time and precharge time to be recovered, thus allowing a faster page reading cycle, to freeze the number of necessary memory structures and to contain peak consumption. In addition, this protocol allows the full utilization of the effectiveness of the redundancy structures.

[0024] During reading (programming), the control buses YN-YM of the two multiplexers have only one active line and can be in two conditions during page precharge. The bus YM is forced fully to logic level “1” by a forcing signal (PAGE) and the bus YN is set to univocal selection, indicating a page read mode; or, both buses are in the univocal selection configuration but can, in the case of FIG. 2, be bypassed by devices which are stimulated with the PAGE control signal, which is synchronous with the precharge signal.

[0025] If passage from one bit line selection to the other occurs only by changing the configuration of the bus YM but not of the bus YN, after an initial reading activity, this indicates that the page is the same and can be advantageously read at the chosen location. If instead there is even just one change among the lines YN, then this indicates that a page change has occurred and that a random reading activity must therefore be performed.

[0026] The architecture shown in FIG. 1 shows the sense amplifiers 3 connected to tristate type structures 4, which are connected to a data bus designated by the reference sign DATA-bus. The architecture shown in FIG. 2 differs from the architecture of FIG. 1 in that the signal PAGE is sent directly to the gate terminals of transistors 5 which are connected to the decoding buses YM and YN.

[0027] On the contrary, in the architecture of FIG. 1, the page control signal PAGE is sent in input to a NOR gate 6 together with a column selection signal, i.e., a bit line selection signal (the word line WL always remains the same in the case of page reading, while the bit line BL changes every time) obtained by inputting column addresses to an AND gate 7. If the column address corresponds to the bit line group, i.e., to the packet of bit lines grouped under the PAGE control, then the control bus YM becomes entirely high and the simultaneous selection of bit lines is activated. Thus, the signal PAGE forces all the lines YM to 1, regardless of the output of the logic gate 7. If the signal PAGE is low, the AND gate is selected univocally and selects a single line.

[0028] If even just one YN changes, then this indicates that the selected bit line does not belong to the packet of bit lines that are precharged together and it is therefore necessary to perform a random reading activity again and then attempt a subsequent page reading. With reference now to the timing charts of FIGS. 3, 4 and 5, FIG. 3 plots the reading timing for successive conventional random cycles, in which the random cycle is designated by the reference numeral 10 and the reference signs arranged laterally to the chart are to be interpreted as follows. The reference ADD designates the address transition, i.e., the address change. The reference ATD designates address transition detection. The reference WL designates the word line. The reference YM designates the control bus described above, which is suitable to select a bit line packet. The reference BL designates the bit line; the reference PC designates the bit line precharge step; and the reference EQ designates the equalization step. The reference LOAD designates the step for loading the data into the output buffers; the reference PAGE designates the signal for activating the page reading mode; and the reference OUT designates the output of the read data from the memory.

[0029] In FIG. 3 the reference numeral 11 designates the bit line selection time at each random reading cycle. FIG. 4 instead plots the reading cycle that uses a random reading mode followed by a page reading mode. In this case, the random cycle is again designated by the reference numeral 10, while the page-mode reading cycle is designated by the reference numeral 12. These page-mode reading cycles 12 have a shorter duration than the random reading cycles 10. The random cycle is never affected by the stimulus of the signal PAGE in the initial step; it is affected only at the end. At the references YM indicating the control buses, the portions shown in dashed lines correspond to the behavior of the bus YM if a random reading cycle were used each time instead of a page cycle after an initial random cycle.

[0030] Regarding the behavior of the bit line BL, FIG. 4 shows that after the initial random cycle 10, which follows a preceding page cycle 12 (not shown in the figure for the sake of convenience), all the bit lines BL that are not selected are brought to zero, while only the bit line BL that must be selected remains high. Therefore, different from the conventional random cycle in which a single bit line is selected, when page mode reading, all the bit lines that belong to a same packet are precharged in advance and then only the bit line that must actually be selected is kept high, while the remaining bit lines are discharged. This behavior is pointed out by the chart showing the development of the bit lines BL in FIG. 4. In this manner it is possible to have an additional selection time 13 which is saved with respect to the bit line selection that occurs in the conventional random cycle.

[0031] Finally, FIG. 5 illustrates the case of the present invention, in which a random cycle is followed by a page cycle but in which a page precharge is performed at the end of the random cycle in order to prepare the memory for the subsequent page-mode reading. FIG. 5 shows that the page cycle is performed starting from a preceding random cycle which, as mentioned, already performs a page precharge at the end. In this case, the word lines WL are maintained, since the activity occurs within a same page, and a page location is selected. In this manner, page precharging occurs collectively and simultaneously on the same number of matrix and reference lines arranged in the two opposite half-planes, while the evaluation step is also conducted symmetrically and is of course limited to its line of the data item to be read.

[0032] The above-described architecture, together with the reading protocol, limits the page-mode reading time to the time required for in-page selection and evaluation/capture and transfer of the data item. In-page selection time is reduced by performing selection exclusively via the bus lines YM (which are less capacitive than the corresponding lines YN), eliminating any resistive extra load during execution and assigning appropriate dimensions to the drivers of the lines. The reading protocol according to the invention therefore entails a succession of random and page cycles, depending on whether the selected line belongs to a same packet of bit lines.

[0033] In practice, it has been observed that the circuit and the method for reading a non-volatile memory according to the invention fully achieve the intended objects, because they increase the reading speed of a non-volatile memory by utilizing a hybrid random and page reading. The circuit and the method thus described can include numerous modifications and variations, all of which are within the intended scope of the invention. Also, all the details may be replaced with other technically equivalent elements.