Combined Program and Data Nonvolatile Memory With Concurrent Program-Read/Data Write Capability

The present invention relates to nonvolatile semiconductor memories (e.g., EPROMs, Flash Memory, E.sup.2 PROMs), and in particular relates to a plural memory bank architecture with addressing and read/write circuitry that enables simultaneous read and write operations.

In U.S. Pat. No. 5,307,314, Lee describes a memory device that is split into banks with separate write enable inputs (WEH and WEL). Both memory banks can be simultaneously accessed, but only at the same address. When both write enable signals are active, the device carries out a word write to both memory banks at the same address location. When both write enable signals are inactive and the single output enable signal (OE) is active, the device carries out a word read from the same address location in both banks. When only one of the write enable signals is active, the device carries out a byte write to only one bank. The logic circuitry of the device also allows a split read/write operation when one of the two write enable signals is active and the output enable signal is also active. Then, a byte read is carried out from the bank corresponding to the inactive write enable signal, while a byte write is carried out on the other bank corresponding to the active write enable signal, where again both banks are accessed at the same address. Each bank has its own byte wide set of data input/output lines.

In U.S. Pat. No. 5,513,139, Butler describes a memory that has two address decoders, one decoder for read operations and a separate decoder for write operations. The memory also has separate input and output data buses. Both address decoders receive sequentially ordered addresses from an address counter, but decode the address bits so as to scan the rows of memory cells in the opposite direction. Thus, the rows of memory may be written sequentially in ascending order and read from in descending order (or vice versa). Further, the address counter is a binary counter whose outputs supplied to the write decoder are shifted by one bit from the outputs supplied to the read decoder, so that read cycle occurs at twice the rate of the write cycle. The number of memory rows is a multiple of (n+1), where "n" is the ratio of the read frequency to the write frequency (e.g., 2:1), to ensure that no row of memory will be both written to and read from in the same interval.

In U.S. Pat. No. 5,502,683, Marchioro describes a dual-port data cache memory that uses two row decoders, one on each side of the word lines or rows. Each row decoder enables a given word line when its address input indicated that the corresponding row is to be accessed (read from or written to). The storage cells of the activated word line are then accessed by bit lines connected to sense amplifiers or write control circuitry, as appropriate. The rows are organized into words of a given bit length, with column boundaries dividing each row into four words. There are two data ports, each accessing through data multiplexers a different one of the four possible word columns. In order to prevent conflicts between the two row decoders attempting to drive the same word line, access switches are placed in each of the rows at the boundaries between the words. These access switches are normally closed, so that each word line is fully connected. However, when the same row is to be accessed by both row decoders, an access switch is opened by control logic (which includes an address comparator) so as to disconnect that row's word line into two separate sections. In this way, the memory circuit allows independent concurrent access to two different words in the memory array.

In U.S. Pat. No. 5,367,494, Shebanow et al. describe a memory with a plurality of memory banks, each with its own address latch and decoder, its own data input latch and driver, and its own data output latch and driver. Control signals include a read/write signal, address strobe, data in strobe and data out strobe, which are sent to all memory banks, and multi-bit address bank, data in and data out bank address signals, which are first decoded then sent to only a selected memory bank. These control signals enable a selected memory bank to latch in an address, latch in data and drive output data, as requested. This scheme thereby allows time-overlapping memory access of the different memory banks.

In U.S. Pat. No. 5,361,343, Kosonocky et al. describe a system with two nonvolatile memory arrays. Each array has its own address register, decoders, and gating; as well as its own erase and program voltage switches. The arrays share input and output data paths through a common data-in latch, data output multiplexer and input/output buffer. The various registers, switches, multiplexers and enables are controlled by logic that includes an array select circuit which can select one array for a write operation and the other array for a simultaneous read operation.

When designing memory devices that are capable of simultaneous read and write operations considerable duplication of circuitry is provided if addressing flexibility is required. The simpler devices with a minimum of control logic and addressing circuitry are normally limited in their concurrent read/write capability, e.g. to same address access of two memory banks or sequential scanning of addresses. Those memory devices capable of truly random independent access of two or more memory addresses simultaneously typically have at least duplicated addressing circuitry, such as two or more row decoders for the separate read and write operations or for the different memory banks. Often there are separate data input and data output pathways or duplicated data paths for the different requested memory accesses. One reason for the complexity is due to the fact that memory reads take less time to complete than memory write operations. In nonvolatile memory devices, a read operation may take only 150 to 200 ns per address, while a write operation may have a byte load cycle time of about 150 .mu.s and a complete page write may take as much as 10 ms. Thus, it can be seen that one can read from many locations in the time it takes to write just one byte or page of data. The added logic, addressing (and data) circuitry allows one to free up the address lines (and data lines) for read operations, while a write operation is going on at another address with a different set of data bits.

There are some applications which require both a high density program memory which stores program instruction codes that are relatively permanent and a smaller data memory which stores parameters that may need to be frequently updated. A memory device combining both types of memory in a single chip would be desirable. In order to be practicable, such a device would need to be able to read from the program memory while a write operation is being carried out in the data memory.

It is an object of the present invention to provide a combined program and data nonvolatile memory device with concurrent program read and data write capability, where as much of the circuitry as possible is shared without sacrificing independent addressing and data access to the program and data memory arrays.

The object is met by a nonvolatile memory device comprising two memory arrays therein, for example, one intended to be used as a relatively permanent program memory and the other intended to be used as a more frequently updated data memory. The device further includes a single set of address lines and a single set of data lines which are used for both memory arrays. The memory device also has address decoding means that includes a shared row decoder common to both memory arrays, with row address latch means associated with at least the data memory for holding a decoded row address during a write operation so as to free the shared row decoder for use in read operations with the program memory. While both arrays have their own separate column decoders, column select circuits and data latches, they share not only a common row decoder, but also common sense amplifiers, data I/O buffers and control logic. The control logic is responsive to input control signals and controls the various elements of the device to carry out a selected read or write operation in a selected memory array.

FIG. 1 is a schematic block level plan view of a nonvolatile memory device in accord with the present invention.

FIGS. 2A and 2B are a timing diagram illustrating read and write operations for each memory array of the device of FIG. 1, including concurrent reading of program memory during a write cycle of the data memory of the device.

With reference to FIG. 1, a nonvolatile memory device of the present invention combines a program memory 11 with a data memory 13. The program memory 11 may be a flash memory array, which is meant to be initially programmed with program instruction codes and only infrequently updated (or not at all). The data memory 13 may be an E.sup.2 PROM array which is programmed and more frequently updated with data parameters. The two memory arrays 11 and 13 need not be the same size and normally the program memory 11 is much larger than the data memory 13. For example, the program memory 11 might be a 512K.times.8 flash array (i.e., 4 Mbits) divided for write purposes into 2K 256-byte sectors, while the data memory 13 might be a 32K.times.8 full-featured E.sup.2 PROM capable of both single byte and 16-byte page writes. Other memory sizes and data widths are possible.

The two memories 11 and 13 in the device share common address input lines A.sub.i, a considerable amount of address decoding circuitry, in particular, a common row decoder 15, common data input/output lines D.sub.j, and considerable data circuitry, including shared sense amplifiers 17 and I/O buffer circuitry 19. Control signals for read or output enable OE and write enable WE are also shared, but there are separate chip enable signals CEF and CEE corresponding to the flash and E.sup.2 PROM memory arrays 11 and 13. Control logic 21 for the device is essentially shared by both memory arrays. This degree of resource sharing enables a smaller device to be built with fewer pins required for the address and data, since unnecessary duplication of circuitry and signal paths is avoided. Yet the device is still capable of concurrent access to the two memory arrays 11 and 13. Concurrent access is achieved despite shared addressing and data resources by providing the data memory 13 with a number of latch circuits 23, 24 and 27, especially row address latches 24 that free up the shared row decoder 15 for use by the program memory 11 during write operations to the data memory 13.

Turning first to the particular address circuits of the device, the address inputs A.sub.i are received by address buffers 25, 27 and 29. In the case of the column address bits for the data memory 13, the address buffer 27 may take the form of read-transparent latches to hold these address bits for the duration of a byte write cycle of the data memory. Alternatively, the latching function for the data memory column address could be made part of the column select circuitry 37 after decoding. The other address buffers 25 and 29 could also take the form of latching circuits or may be simple tristate buffers that hold the address signals only for as long as they remain asserted on the address input lines A.sub.i. In either case, all address buffers are enabled in response to control signals C.sub.1 -C.sub.3 from the control logic 21. The control logic 21 in turn derives its control signals C.sub.k from the signals CEE, CEF, OE and WE received from the input pins of the device. The row address buffer 25 is enabled by signal C.sub.1 whenever either chip enable CEE or CEF (but not both) is active (low) and either output enable or write enable signal OE or WE (but not both) is also active (low). The column address buffer 27 for the data memory 13 is enabled by signal C.sub.2 whenever the E.sup.2 PROM chip enable signal CEE is active and either output enable or write enable signal OE or WE is also active. The column address buffer 29 for the program memory 11 is enabled by signal C.sub.3 whenever the Flash memory chip enable signal CEF is active and either output enable or write enable signal OE or WE is also active. The address information is input into the buffers 23, 27, 29 on the falling edge of appropriate chip enable signal CEE or CEF and appropriate output enable or write enable signal OE or WE, whichever occurs last. The condition where both chip enable signals are low or both output and write enable signals are low is invalid and generates no control signals from the logic 21. The row address buffer 25 is assigned the bulk of the address bits, e.g. bits A4 to A14 corresponding to a sector of the 512K.times.8 flash memory array 11 and a page of the 32K.times.8 E.sup.2 PROM array 13. The column address buffers 27 and 29 are assigned the remaining address bits, e.g. bits A0 to A3 of the E.sup.2 PROM array, and bits A0 to A3 and A15 to A18 of the flash array. Other arrangements of row and column address bits are possible, depending on the size and configuration of the respective memory arrays.

The address information is supplied by the buffers 25, 27 and 29 to address decoding circuitry 15, 31 and 33. Typically, address decoding is carried out in two or more stages, including a first pre-decoding stage followed by a final decoding stage. For simplicity, all stages have been shown lumped together in FIG. 1 into corresponding single decoder circuits 15, 31, and 33. The shared row decoder 15 connects to both memory arrays 11 and 13. In the case of the program memory 11, the row decoder 15 communicates directly with row drivers 32 that activate a selected word line or row corresponding to the decoded row address bits. In the case of the data memory 13, the row decoder 15 connects to a read-transparent latching circuit 24. The latch circuit 24 in turn connects to row drivers 34 that activate a selected word line corresponding to the decoded row address received from the row decoder. During a read operation, the latch circuit 24 is effectively transparent so that the row decoder 15 communicates directly with the selected row driver. But during a write operation, the decoded row address is latched into the row address latches 24 and isolated from the row decoder 15 by pass gates controlled by control signal C.sub.7 from the control logic 21 (this occurs when WE=low and CEE=low). This frees the row decoder 15 for decoding addresses to be read from the program memory 11. The latches 24 hold the decoded row address for writing to the data memory 13, so that the selected word line continues to be asserted with the programming voltage V.sub.pp.

The column decoder 31 for the data memory 13 is connected to a column select circuit 35. Likewise, the column decoder 33 for the program memory 11 is connected to a column select circuit 37. The column select circuits 35 and 37 are bidirectional multiplexing and gating circuits that control access of the data paths to a selected column of eight bit lines of the respective memory arrays 11 and 13. The operation of the select circuits 35 and 37 are controlled by signals C.sub.4 and C.sub.5 from the control logic 21. During a read operation (OE=low, WE=high), the selected column of bit lines corresponding to the column address in the enabled memory array 11 or 13 (either CEE or CEF is low) is connected to sense amplifiers 17. A data input/output buffer 19 responsive to another control signal C.sub.6 outputs the sensed byte of data into the data lines D.sub.j. During a write operation (WE=low, OE=high), the enable column select circuit 35 or 37 (either CEE or CEF is low) connects the input/output buffer 19 to data latches 23 or 39 for the selected column of bit lines in the enabled memory array 11 or 13. Data received from the data lines D.sub.j is loaded into the selected data latches 23 or 39 from which they can be loaded into the row and column of memory corresponding to the received and decoded address bits A.sub.i.

The control logic 21, in addition to generating appropriate control signals C.sub.k in response to the input signals CEE, CEF, OE and WE, also controls generation of the high voltage V.sub.pp for programming data into the memory cells. In particular, the device may include JEDEC standard software data protection (W.P.). In this scheme, each programming sequence for a sector of flash memory or a byte or page of the E.sup.2 PROM, must be preceded by a three-byte program command sequence in order for actual programming to take place. This sequence might consist of a specific combination of data bits D.sub.j and address bits A.sub.i, typically alternating zeros and ones.

Referring now to FIGS. 2A and 2B, a timing diagram illustrates the main operational features of the present invention. A write operation to the program memory (i.e., the flash memory array 11 in FIG. 1) begins with a three-byte write enable code to override the software write protection. The chip enable signal CEF for the flash memory and the write enable signal WE pulse low, while a sequence of three addresses and three corresponding sets of data are input into the device. Typically, address bits A18-A15 are ignored, since the write protect logic is also shared by the smaller E.sup.2 PROM array 13. No actual writing to the flash memory occurs yet, but the sequence causes the control logic to begin allowing generation of a program voltage V.sub.pp and starts internal write timers. The flash memory is programmed in 256-byte sectors. An entire sector is erased upon receipt of the sector address on bits A.sub.14 to A.sub.4 prior to programming. No special erase command is needed. Any bytes in the sector that are subsequently left unprogrammed will be indeterminate. The sector address A.sub.14 to A.sub.4 remains unchanged for the duration of the sector write, while the byte address A.sub.18 to A.sub.15 and A.sub.3 to A.sub.0 within the sector changes. While the byte address is typically varied sequentially, this is not essential and byte programming of a flash sector can proceed in any order. In FIGS. 2A and 2B, the byte address for a sector is carried out sequentially from a start address ADDR to a finish address ADDR+255. The corresponding data DATA-IN loaded into the memory is indicated by BYTE 0 to BYTE 255 on the data lines.

A byte load is performed by applying a low pulse on CEF and WE with a CEE and OE high. The address is latched on the falling edge of CEF or WE whichever occurs last, while data is latched on the first rising edge of CEF or WE. Once a byte has been loaded into the data latches of the flash memory array, they are programmed into the memory cells during an internal programming period. The byte write cycle time typically is about 150 .mu.s, although actual loads may take less time. After the first data byte has been programmed, successive bytes are entered in the same manner. Each new byte to be programmed must have its WE (or CEF) high to low transition within 150 .mu.s of the preceding bytes low to high transition of WE (or CEF) or the load period will end. The total write cycle time for a sector of flash memory is typically about 10 ms. No read operations on the E.sup.2 PROM array are allowed during flash memory's write period and attempts to read the flash memory will effectively be polling operations of the current byte being loaded. Alternatively, modifications to the device of FIG. 1 which include decoded row address latching circuits for the flash memory 11, as well as the E.sup.2 PROM 13, would allow E.sup.2 PROM reads during flash memory write operations, if desired.

A flash memory read operation 52 occurs when the flash memory's chip enable signal CEF and the output enable signal OE pulse low, while CEE and WE remain high. The flash memory array 11 is read like a static RAM. Reading is carried out on individual bytes, rather than whole sectors. Moreover, for read operations the sectors appear seamless and sector boundaries need not be taken into account. That is, bytes from different sectors may be read successively. When CEF and OE are low, the data DATA-OUT to be output that is stored at the flash memory location determined by the address inputs to A.sub.0 is asserted on the data lines. The maximum read time is typically only 150 to 200 ns per byte.

The data memory 13, typically an E.sup.2 PROM array, is written to more frequently than the program memory 11. The E.sup.2 PROM write operation 54 in FIG. 2 illustrates the capability provided by the memory architecture of the present invention of concurrently reading flash memory 11 during the write cycle period of the E.sup.2 PROM array 13. Write protection is disabled by a 3-byte write enable code sequence, which is typically the same as that for the flash memory, except that now CEE pulses low, while CEF remains high. Read operations from the flash memory may interrupt loading of the write enable code sequence, as shown, provided the byte load cycle time of 150 .mu.s is not violated. Since a read operation generally takes less than 200 ns to complete, there can be many bytes read from the flash memory between each byte of the code sequence. Once the valid command sequence has been loaded, the write cycle is initiated by both CEE and WE going low. Once again, the address is latched by the falling edge of CEE or WE whichever occurs last, while data is latched into latches 23 of FIG. 1 on the rising edge of CEE or WE whichever occurs first.

All write operations to the E.sup.2 PROM array must conform to the limits of a page write. That is, while anywhere from a single byte of data to up to sixteen bytes of data may be written, all such bytes during a write cycle must reside on the same page, as defined by the address bits A14-A4. For each high to low transition of the signal WE, the bits A14-A4 must be the same (PAGE in FIG. 2). The A3 to A0 address bits are used to specify which bytes within the page are to be written. Address bits A18 to A15 do not apply to the smaller E.sup.2 PROM array and will be ignored. For an entire page write operation, the bytes are normally written sequentially as indicated in FIG. 2 by the input data BYTE 0 to BYTE 15 being loaded into sequential bytes beginning with start address ADDR and ending with address ADDR+15. However, the bytes can be loaded in any order and can be altered within the same load period, if desired. Only bytes which are specified for writing will be erased and written with new data held in the data latches.

Reads 56 from the flash memory array are allowed throughout the E.sup.2 PROM write cycle time (up to 10 ms) so long as the byte load cycle time of 150 .mu.s for the E.sup.2 PROM write is not violated. As before, a flash memory read occurs when CEF and OE pulse low. The data (BYTE) stored at the flash memory location (F.ADDR) determined by the address inputs A18 to A0 will be sensed and output on the data lines. Many reads can be performed in the time available. Attempts to read the E.sup.2 PROM array (CEE and OE low) during the E.sup.2 PROM write cycle will produce a polling operation of the data held in the latches 23.

The E.sup.2 PROM read operation 58 is performed just like the flash read operation 52 except that now the E.sup.2 PROM array is selected (CEE=low). The data (BYTE) stored at the memory location (E.sup.2 ADDR.) determined by the address inputs A14-A0 will be sensed and output onto the data lines. An E.sup.2 PROM read cannot be performed during any of the write cycles.

The device of the present invention is intended to be used to store infrequently updated program information in one memory array and more frequently updated data parameters in another memory array. The architecture allows concurrent reading of the program memory during writes to the data memory, while eliminating much duplication in addressing and data hardware. The data memory's address latches, free up the row decoder for performing read operation decoding for the program memory during the time period when data held in the data memory's data latches are actually programmed into memory cells. Thus, only one row decoder is needed. Alternative devices could have both memory arrays with independent address latches and drivers, so that read operations can be performed in either memory array while the other is completing a write operation. Only a single set of data and address inputs are needed for read and write of both arrays.