The invention relates to a semiconductor device having a byte-erasable EEPROM memory comprising a matrix of rows and columns of memory cells.
There are two main types of electrically erasable non-volatile memories. Byte-erasable EEPROM memories are used for persistent data storage. To this purpose the memory can be programmed and erased per byte. Program and erase times are fast and reliability requirements are high. This is achieved by designing a robust cell that is insensitive to disturbs. As a consequence the cell size is relatively large.
The most commonly used non-volatile memory nowadays is known as Flash. Flash EEPROMs are used to store persistent code and are typically programmed per byte but erased per block or per sector. Because of the sector or block erase, the disturb conditions are relatively relaxed. Also the reliability requirements (in terms of the number of program/erase actions) are relaxed. This allows making much smaller cells than byte-erasable EEPROM cells.
Most commonly used byte-erasable EEPROM cells are based on the FLOTOX principle. FLOTOX cells consist of a double gate structure with a floating gate and a control gate separated by a dielectric layer. The floating gate is located on a relatively thick gate oxide (20-40 nm) in which a tunnel window is made with a thinner oxide layer (typically 7-10 nm). Below the tunnel oxide is a highly n-type doped injector region. Applying a large positive voltage (10-15 V) to the injector region with the control gate grounded performs programming. Electrons are ejected from the floating gate to the injector region by Fowler-Nordheim (FN) tunneling. Applying a large positive voltage to the control gate and grounding the injector region erases cells. Electrons are injected into the floating gate by FN tunneling.
The use of FN tunneling for write and erase is very efficient and consumes very little power. This makes a FLOTOX type of byte-erasable EEPROMs very suitable for portable applications and (contact-less) smart cards. The disadvantage of FN tunneling is that relatively high voltages are required.
The reliability of FLOTOX type of EEPROMs is achieved by separating the tunnel window both on the source side as well as the drain side of the cell by a high-voltage transistor. The high-voltage transistor on the source side is also used as a read transistor. A high-voltage transistor on the drain side is used to isolate the injector region from high bit line potential for non-selected cells.
In IC manufacturing there is also a strong desire to reduce the size of non-volatile cells. Because of the reliability restrictions for these memories the tunnel oxide thickness cannot be decreased below 7 to 8 nm. As a consequence, voltages needed to program and erase more or less remain constant from generation to generation. It becomes more and more difficult to support these voltages when all other dimensions on the chip become smaller. Particularly the FLOTOX type of byte-erasable EEPROM is very hard to scale as the device consists of two high-voltage transistors in series with an injector region. Moreover, in the operation of this device high voltages are used on the bit lines (drains) of the cells. This also puts a restriction on the cell dimension because high-voltage parasites have to be suppressed in the memory array.
Another trend in modem IC processes is to combine the two types of non-volatile devices, i.e. flash-EEPROM and byte-erasable EEPROM, on one chip. Generally, however, flash and EEPROM are made in different processes. Combination of these processes is not trivial and will lead to an increase of process complexity and cost.
One way to diminish the cell size of byte-erasable EEPROMs and to combine them with flash EEPROMs is to try to make a byte-erasable EEPROM starting from a flash-EEPROM cell. The usual flash EEPROM is of the ETOX type. It consists of a single floating-gate cell. Programming is achieved by channel hot electron injection at the drain side of the cell by applying a moderate voltage (4-6 V) to the drain and a high voltage to the control gate (8-12 V). Erasing is achieved by either applying a large positive voltage to the source (10-15 V) and grounding the control gate or by applying a moderate voltage to the source (4-6V) with a negative voltage (−6 to −10 V) on the control gate. Erasing is done by FN tunneling. The programming is very fast (10-100 microseconds) but very power consuming (i.e. with a very high drain current (0.1-0.5 mA) per cell). Erasing is very slow (seconds) but at very low currents (nano-Amperes per cell).
However, ETOX cells are not very attractive for byte-erasable EEPROM applications for the following reasons:
WO 00/75994 A1 describes an alternative 2-transistor flash cell that is optimized for low-voltage and low-power embedded flash applications. One cell consists of a stacked gate transistor with a floating gate and a control gate on top and an isolation transistor at the source side. The cell can be programmed and erased by FN tunneling from the floating gate to the channel of the stacked gate transistor. The drain and source junctions can be very soft to suppress disturbs and to optimize for short channel effects. This allows making a relative short stacked gate and isolation transistor. Because of the presence of the isolation transistor the cell can be erased to a negative threshold voltage. Reading can thus be done at low voltages, similar to a FLOTOX EEPROM. A 2-transistor flash cell is thus ideally suited to use for byte-erasable EEPROMs.
A byte-erasable EEPROM having a matrix of memory cells each comprising a select transistor and a floating gate transistor is known from GB 2 321 738 A. The memory cells are coupled to bit lines and common control lines. Programming transistors are coupled between an erase line and a common control line of one row of memory cells.
It is an object of the present invention to provide a semiconductor device having a byte-erasable EEPROM which has a reduced chip size and increased density and which is suitable for low-power applications.
This object is achieved according to the invention by a semiconductor device as claimed in claim 1, wherein the memory cells each comprise a selection transistor having a selection gate and, arranged in series therewith, a memory transistor having a floating gate and a control gate, the selection transistor being further connected to a source line of the byte-erasable EEPROM memory, which source line is common for a plurality of memory cells, and the memory transistor being further connected to a bit line of the byte-erasable EEPROM memory, wherein the columns of memory cells are located in separate p-type wells separated by n-type wells.
The invention is based on the idea to use a 2-transistor flash cell also for byte-erasable EEPROM applications. Because of this, combinations of flash and byte-erasable EEPROM can be made on one chip without process adaptations. Both flash and EEPROM are processed on a triple well structure; i.e. the non-volatile cells are processed in an isolated p-type Well (i-pWell) that is isolated from the environment by a buried n-type Well. This allows the use of negative voltages during program and erase. To suppress disturbs the memory array is divided in byte columns located each in an i-pWell and separated by nWells. The semiconductor device according to the invention thus has a very high density and a very high performance.
In contrast to standard FLOTOX memory cells where the bit line is contacted at the side of the access transistor, the bit line is connected to the memory transistor according to the present invention. This leads to a different write operation as will be explained below. Further, since the second transistor does not have to be switch through a high voltage, it can be processed on the tunnel oxide rather than on a special high voltage oxide, which would make the memory cell very much larger. Still further, the byte-erasable EEPROM according to the invention is suitable for combination with a flash EEPROM on the same chip.
Preferred embodiments of the invention are included in the dependent claims. To make the memory cells erasable per byte (or word) without getting major gate disturbs in non-selected words the global control gate lines are divided in local control gate lines that run over one byte. Switching elements achieve control gate separation. As the switching elements have to switch through a high voltage they must be high voltage (HV) devices. HV transistors can be very large, depending on the voltages they have to handle: the larger the voltage the larger the size. Therefore, it is preferred to keep these voltages as low as possible.
In the known byte-erasable EEPROM memories of the FLOTOX type a HV nMOS switching element is used. According to the present invention the switching elements can be HV-nMOS or HV-pMOS or even HV-CMOS switching elements. Either of these elements can be used having advantages and disadvantages. An evaluation of all possibilities shows that a HV-pMOS switch is preferably used because:
Addressing of said switching elements, i.e. addressing of single bytes, is preferably done using byte select gate lines to which said switching elements are connected.
According to the preferred embodiment of claim 5 sectors are defined by dividing global bit lines into to local bit lines for each sector and by using sector select gate lines for addressing of said sectors. Thus, program disturbs in non-selected sectors can be avoided and bit line capacitance can be reduced during read making the memory much faster.
The semiconductor device according to the invention is suitable for combining the byte-erasable EEPROM with a flash EEPROM on one chip as defined in claim 6.
Preferably, said p-type wells are isolated by buried n-type wells. Preferred ways of applying a write, read and erase operation to the semiconductor device according to the invention are defined in claims 8 to 10.
The invention will now be explained in more detail with reference to the drawings, in which
The whole memory 1 is located in one large isolated p-type well 3 (i-pWell) with a buried n-type well 2 (BNW) beneath. The memory 1 is organized in sectors Sm, e.g. S1-S2, by dividing the global bit lines BLj, e.g. BL1-BL4, into local bit lines BLjm, e.g. BL11-BL42, that are addressed by a sector select gate line SSGm. As the SSGm only has to switch through a relatively low voltage (typically 5V) no high voltage (HV) transistor is needed. Therefore, it can be a nMOS device processed on the same oxide as the memory cells Mij. The function of the sector segmentation is to suppress program disturbs in non-selected sectors and to reduce the capacitive load of the bit lines during read. The latter makes the memory much faster.
Compared to the matrix of the flash EEPROM shown in
Byte selection gate lines BSGn, e.g. BSG1-BSG2, that run parallel to the bit lines BLj addresses the HV-pMOS devices T3. Although not strictly necessary, sectors Sm, e.g. S1-S2, are defined by using sector select gate lines SSGm, e.g. SSG1-SSG2, as for the flash memory to reduce bit line capacitance during read.
To write, read and erase only data in the memory cells Mij of the EEPROMs shown in
Table 1 gives an example for an operation table for the flash memory as shown in
Table 2 gives an example for the byte-erasable EEPROM memory as shown in
A comparison of the two tables shows that the operation schemes for the two memories are very similar. This particular combination of flash and byte-erasable EEPROM is therefore not only attractive because they can be combined in one technology but also from an operation and application point of view.
The operation scheme of the byte-erasable EEPROM according to the invention is now further described with reference to
The write operation is schematically shown in
The CG voltage of 10V is switched through the BSG by grounding the gate of this transistor. Program disturbs on cells of non-selected bytes (i.e. on the same row as the selected byte) are avoided by grounding the i-pWell of those bytes and by closing the BSG. This is achieved by putting the same voltage to the gate of this transistor as to the BNW; i.e. 10V in this example. By doing so, BSGs in non-selected columns are off and the local CGs in these bytes are floating (indicated by “fl”). Their potential will be capacitively coupled to their environment. The main contribution will come from the i-pWell potential. Therefore, the local CG potential will be close to the i-pWell potential and there will be only a very small disturbance. The local CG potential can further increase by junction leakage from the HNW and by sub-threshold leakage through the channel of the BSG. To avoid this, the BSG should be sufficiently long and the junction should have a high breakdown value. For the same reason, the positive programming voltage (10V) should be kept as low as possible. However, making this value too low will cause disturbs in selected columns, as will become clear.
It is also important that between multiple program actions the local CGs are reset to zero to avoid cumulative program disturbs.
In a selected column the local CGs of non-selected rows are closed by applying zero volt to the global CGs. Depending on the local CG potential, these BSGs can be open (positive potential) or off (negative potential). If for some reason, i.e. capacitive coupling or junction leakage from the HNW, the local CG potential becomes positive the potential will never increase to a value above one threshold voltage (with backbias) of the BSG. Therefore, the transistor will be always off and the local CG will be capacitively coupled to a negative value by the i-pWell potential and the drain potentials of bit lines that are either at 0V or −5V. Depending on the state of the memory cells (0 or 1) and the data on the bit lines, the local CGs will charge to a voltage between typically −1V and −4V. For cells with −5V on the drain, this will give a very small gate disturbance. For cells with 0V on the drain this can cause a drain disturb. Particularly for cells that are in the programmed state (i.e. with negative charge on the floating gate) the drain is positively biased compared to the floating gate. This may cause electron loss from the floating gate to the drain or hole injection caused by electron-hole formation due to substrate currents generated by the gated reverse-biased drain junction (GIDL: gate induced drain leakage). Several measures are taken to suppress this drain disturb:
The erase operation is shown schematically in
An alternative is to increase the CG to floating gate capacitive coupling in the cell so that the CG potential is more effectively coupled to the floating gate. This can be achieved by increasing the width of the floating gate thus increasing the area between floating gate and CG at the cost of a somewhat larger cell size. However, the cell layout will still be almost identical to that of the 2-transistor flash cell.
The negative CG voltage needed for erasing is switched through the BSG of a selected column by applying a sufficiently high negative voltage (i.e. −8V) to the gate of this HV-pMOS transistor.
Other potentials during an erase operation are carefully chosen to find an optimum between disturbs seen on non-selected pages in a selected column and bias conditions as experienced by the BSGs of non-selected columns. To avoid that the latter transistors have to handle the full erase voltage (i.e. 15V) that would make them very large, some disturbance in non-selected pages in a selected column is accepted.
An optimum is found by allowing a gate disturb of 4V on non-selected pages in a selected column and a voltage of −11V over the BSG transistors of a selected page in non-selected columns. The sum of these two voltages equals the total voltage swing (15V) needed for erasing. Of course, other voltage combinations can be chosen (i.e. 3V disturbance and −12V HV-pMOS stress) depending on how strong the HV-pMOS can be made.
Also the division of the total erasing voltage in a positive and a negative voltage can be optimized for optimal overall circuit performance. In the example of Table 2 and
Control gates on non-selected pages in a selected row are biased to +6V so that the cells in those bytes experience a gate disturbance of −4V because the i-pWell of the selected column is biased to +10V. This can give rise to soft erasure of programmed cells. It can be shown that intrinsically cells will not significantly be erased even when this disturbance is applied during many erase cycles. To avoid that some (extrinsic) cells might fail, it can be considered to add error correction to the memory.
Disturbance in non-selected columns is suppressed by biasing the i-pWell to +6V and also the gates of the BSGs to +6V. The HNW, which is at +10V and the i-pWell, which is at +6V, will capacitively charge the local CGs in non-selected columns up to a value slightly above +6V. Global CGs are at either −5V (selected page) or +6V (non-selected page) so that the BSGs are off in all cases and the local CGs are floating. Because the largest capacitive coupling to the cells comes from the i-pWell, cells will charge to a value slightly above this well potential. Hence, the disturbance experienced by the cells is very small.
Separated from the transistors T1 and T2 above the i-pWell area by a field oxide region FOX a n-type well, preferably a high voltage nWell HNW, is formed in the substrate 60 covered by high voltage oxide 50. A switching transistor T3 is formed by a source 51 connected to the local control gate line CGin and a drain 52 connected to the global control gate line CGi. Electrically insulated therefrom a byte selection gate BSG connected to a byte selection gate line is formed on top of the oxide 50.
Number | Date | Country | Kind |
---|---|---|---|
01204570.4 | Nov 2001 | EP | regional |
Filing Document | Filing Date | Country | Kind |
---|---|---|---|
PCT/IB02/04473 | 10/24/2002 | WO |