1. Field of the Invention
The present invention relates to non-volatile memory arrays. More particularly, the present invention relates to a NAND-like memory array employing high- density NOR-like memory devices.
2. The Prior Art
Non-volatile memory devices are widely employed in the market of portable systems such as laptops, PDAs, mobile phones and others. Non-volatile memories allow storage of binary information regardless of whether or not power is applied to the device. This is a very useful feature for portable systems.
Non-volatile memories are realized using floating-gate devices that can change their physical state. Two or more physical states are obtained by changing the threshold of the floating gate device by means of injection and extraction of electrons in the floating gate. The injection of electrons will result in a higher threshold, corresponding for instance to a programmed state. The successive extraction of electrons will bring the threshold to a lower value, corresponding for instance to an erase state.
Recent applications require higher memory capacity at the lowest price, pushing the developing of new architectures and scaling down of the technologies.
In the present market there are two kinds of flash memories, intended to satisfy different requirements and applications. NAND flash memories have very high density, but poor random-access read performance. Such memories are suitable for data storage where sequential access is the major application. A very high number of read circuit blocks are used to sustain the sequential read throughput.
The architecture of traditional NAND flash memories is focused on the efficiency of the chip from the die size point of view only, accepting very poor performance for random-access read operations. NAND flash memories have a Tacc (random read access) in the range of 5-10 us, while the performance needed for code execution is on the order of 100 ns, which is the typical Tacc of NOR flash memories.
As shown in
This prior-art approach has several disadvantages: bit lines are drawn at the same pitch of the cells (˜2F) and they are as long as the chip height, therefore the capacitance of the bit line is huge compared to the cell current and the resistivity, combined with this capacitance, results in an RC time-constant in the order of micro-seconds; word lines 250, driven by the row decoder 200, have the same pitch of the cell (2F) and, as for the bit lines, they have an RC in the order of micro-seconds. For these reasons the read access of a memory cell requires a time in the order of microseconds.
On the other hand, NOR flash memories have lower density but very high performance for the read random access; they are suitable for code execution rather then data storage. The cost per bit of the NOR flash memory devices is many times higher than the NAND flash memory devices, while the random read access is up to 100 times faster.
Many factors contribute to these different features of NAND flash memory devices and NOR flash memory devices, but the main difference is the memory cell itself. Floating gate devices of NOR flash memories have the source line and the bit line plug shared by two cells. Thus, as shown in
As shown in
A flash memory integrated circuit includes a plurality of flash memory arrays. A global word line driver is associated with each array, each global word line driver coupled to a plurality of select lines. A plurality of sense amplifiers are individually coupled to a plurality of bit lines. A plurality of sub arrays each include a plurality of NAND flash memory cells coupled to local word lines and local bit lines. A local word line driver is associated with each sub-array and coupled to the plurality of select lines and configured to drive ones of the local word lines in its sub array associated with selected ones of the plurality of NAND flash memory cells in its sub-array. A local bit line driver is coupled between selected ones of the local bit lines in each sub array and selected ones of the plurality of bit lines.
Persons of ordinary skill in the art will realize that the following description of the present invention is illustrative only and not in any way limiting. Other embodiments of the invention will readily suggest themselves to such skilled persons.
The present invention is an architecture for flash memories that combines the advantages of the chip size of NAND flash memory architectures and the performance of NOR flash memory architectures. The flash memory architecture of the present invention ameliorates the problems of the prior-art arrays, achieving a read performance in the order of ˜100 ns with a small increase in chip area.
Referring now to
Referring now to
Referring now to
The gates of memory cell transistors 110, 112, 114, and 116 and select transistors 118 and 120 are, respectively, coupled to word lines 124, 126, 128, 130, 132, and 134 through local selector transistors 136, 138, 140, 142, 144, and 146. The gates of local selector transistors 136, 138, 140, 142, 144, and 146 are coupled to a select line 92 common to the entire string. Select line 92 is driven by a global word line driver 80. The selector transistors 136, 138, 140, 142, 144, and 146 charge up the word lines 124, 126, 128, 130, 132, and 134 by means of global supply lines s<i> as indicated at the top of
The selection time of the local word lines 96 and local bit lines 98 is extremely faster than in a conventional NAND array. If “n” is the number of local bit-lines or local word-lines, the RC time-constant of the local connection is n2 times smaller than the one of the global connections. For instance, the word lines and the bit lines are split into four local sub connections as shown in
Referring now to
It should be noted that unselected ones of local bit lines 100a through 100d cannot be left floating, since, as known in the art, the program operation of a NAND memory implies the control of the unselected strings to prevent an unwanted programming of the cells that share the word line of the cell that is supposed to be programmed. In order to inhibit the program operation, the bit lines of the unselected strings are biased at a proper voltage (e.g. VCC) to prevent the formation of the electric field necessary to the program operation. A hierarchical column decoder for a NAND memory according to the present invention also includes a set of selector transistors 162a through 162d that connect the unselected bit lines to a bias supply line 164 by the use of complementary selection signals designated by sel<i> with an overlying bar to indicate signal inversion. The bias supply line 164 biases the unselected bit-lines during any operation that cannot leave the bit lines left floating.
The additional area required due to the sub-array architecture is not significant compared to the great advantage of area given by the NAND memory cell versus the NOR memory. Global word lines and global bit lines are drawn with a pitch more relaxed then the local ones, then the RC time-constant related to their selection does not impact the overall performance. Moreover, by placing the sense amplifiers at the center of the array, the length of the bit lines connection is reduced.
The above illustrated architecture improves the performance of traditional NAND memory with a small drawback in terms of additional die area. Since the area of a NAND cell is about 60% of a NOR cell (6 F2 vs 10 F2), the area efficiency is really improved with respect to a traditional NOR memory, obtaining the same read access performance.
While embodiments and applications of this invention have been shown and described, it would be apparent to those skilled in the art that many more modifications than mentioned above are possible without departing from the inventive concepts herein. The invention, therefore, is not to be restricted except in the spirit of the appended claims.