Method and system for hiding refreshes in a dynamic random access memory

The present invention relates to dynamic random access memories, and more particularly to a method and system for hiding the refresh of such memories thereby allowing such memories to be accessed at any time.

Dynamic random access memories (“DRAMs”) are commonly used as the system memory in computer systems. The DRAMs are typically coupled to a central processing unit (“CPU”) through a DRAM controller or bus bridge which is, in turn, coupled to the CPU through a processor bus. The DRAMs include one or more memory arrays each having a large number of memory cells arranged in rows and columns. Each memory cell of the DRAM requires only a single transistor and capacitor, thus making DRAMs very compact. The compact nature of DRAMs makes it possible for them to provide high memory capacities at relatively little cost.

Although DRAMs have the advantage of providing relatively high memory capacities at relatively little cost, they have the disadvantage of requiring periodic refresh of their memory cells. More particularly, since memory cells store data by charging their respective memory cell capacitors to predetermined voltages, their capacitors must be recharged or “refreshed” before excessive charge has leaked from the capacitors. During the time that the memory cells are being refreshed, data generally cannot be written to either the row of memory cells being refreshed or any other memory cells in the array. Further, although data may be read from a row of memory cells being refreshed, data generally cannot be read from other memory cells in the array. For many applications of DRAMs, such as for use as system memory, the need to periodically refresh the memory cells does not significantly limit their usefulness. However, the need to periodically refresh DRAMs generally makes them unsuitable for other applications, as described below.

Computer systems using DRAMs for system memory also frequently use a cache memory, which is normally coupled directly to the processor bus. The cache memory allows the CPU to more quickly access frequently or recently used instructions and data. Cache memories are generally implemented using static random access memories (“SRAMs”) that also have at least one array of memory cells. Each memory cell in the array is generally implemented by a pair of cross-coupled transistors, and each cell is accessed through a pair of access transistors. Each SRAM memory cell thus generally contains at least 4 transistors. As a result, SRAMs are far less compact than DRAMs and therefore significantly more expensive. However, SRAMs have a significant advantage over DRAMs in that their memory cells do not need to be periodically refreshed. As a result, SRAMs can be accessed by the CPU or other device at any time. In contrast, it would sometimes be necessary for the CPU or other device accessing a DRAM cache memory to wait until a refresh has been completed before the cache memory could be accessed. Therefore, despite their significantly higher expense, SRAM's are generally considered more suitable than DRAMs for use as cache memory.

It would be desirable to use DRAMs for cache memory so that high capacity cache memory could be provided at relatively little cost. Attempts have been made to use DRAMs for cache memory, but these attempts have generally proven unsuccessful because of the inability to access them during periodic refreshes. For example, attempts have been made to implement cache memory using a high capacity DRAM and a significantly smaller SRAM to temporarily store write data during a refresh of the DRAM when the data cannot be written to the DRAM. While these attempts have been able to “hide” refreshes of cache DRAMs under some circumstances, they have been unable to hide refreshes of DRAMs under at least some circumstances for a variety of reasons. Thus, these prior art cache DRAMs constrained computer systems using them in various ways, and their use often required that the computer system be specially modified for use with such cache DRAMs.

It would therefor be desirable to provide cache DRAMs that could truly hide the periodic refresh of their memory cells and which did not require special modifications of computer system, thus allowing such DRAMs to be pin compatible with cache SRAMS.

A dynamic random access memory (“DRAM”), includes first and second arrays of memory cells each containing substantially the same data. A refresh circuit is structured to refresh the memory cells in either the first array or the second array, but not both. The refresh circuit is also structured to control access to one of the arrays as a function of which if any of the arrays is being refreshed. For a write access, the refresh circuit determines if either of the arrays is being refreshed. If one of the arrays is being refreshed, the refresh circuit allows data to be written to the array that is not being refreshed. The write data are also buffered until the refresh is completed, and are then written to the array that was being refreshed. Alternatively, the data can be later read from the array to which it was written and then written to the array that was being refreshed. If neither of the arrays is being refreshed, the refresh circuit allows data to be written to both arrays simultaneously. In either case, the refresh circuit also retains an indication of to which array(s) the data have been written, such as by setting a flag in a register. For a read access, the refresh circuit determines if either or both of the arrays contain valid data at locations corresponding to an address from which the data are to be read. The refresh circuit then allows data to be read from at least one of the arrays containing valid data. Since the refreshes are substantially hidden, the DRAM can be advantageously used as a cache memory in a computer system.

FIG. 1 is a schematic top plan view of a semiconductor wafer showing the layout of a plurality of arrays of a memory device.

FIG. 2 is a block diagram and schematic showing a portion of the circuitry that may be used to access the memory arrays shown in FIG. 1 .

FIG. 3 is a block diagram of the memory device shown in FIG. 1 .

FIGS. 4A-4B is a flow chart showing the operation of a refresh controller used in the memory device of FIG. 3 .

FIG. 5 is a block diagram of a computer system containing the memory device of FIGS. 3 and 4 .

The following is a description of a DRAM and method that may be accessed during refreshes of the memory cells in the DRAM. In this description, certain details are set forth in order to provide a thorough understanding of various embodiments of the present invention. It will be clear to one skilled in the art, however, that the present invention may be practiced without these details. In other instances, well-known circuits, circuit components, and control signals and associated timing protocols have not been shown or described in detail in order to avoid unnecessarily obscuring the description of the various embodiments of the invention.

The preferred embodiment of the invention is a DRAM formed on a semiconductor die 10 , as illustrated in FIG. 1 . The semiconductor die 10 contains a plurality of memory banks, BANK 0 , BANK 1 , BANK 2 . . . BANK N. Each memory bank contains two memory arrays 12 a,b having the same size and storing the same data bits. Thus, two memory cells exist for each memory address. As a result, if one memory array 12 a is being refreshed, a memory access to the DRAM can occur in the other memory array 12 b . Although the layout of memory arrays 12 a,b shown in FIG. 1 is preferred, it will be understood that the memory arrays 12 a,b may be laid out in other arrangements.

The manner in which access to corresponding memory cells in the arrays 12 a,b will now be explained with reference to FIG. 2 . As is well known in the art, each array 12 a,b contains a large number of memory cells (not shown) arranged in rows and columns. A row line (not shown) is provided for each row of memory cells which, when active, allows access to all of the memory cells in the row. A particular memory cell in the accessed row is accessed by coupling a data bit to or from the memory cell through a pair of complimentary digit lines, one pair of which 14 a,b and 16 a,b is shown in FIG. 2 . The digit lines 14 a,b couple data to or from one column of the respective arrays 12 a,b , and the digit lines 16 a,b couple the compliment of the data to or from the same column of the arrays 12 a,b . The digit lines for the other columns of the arrays 12 a,b have been omitted in the interest of clarity and brevity.

As shown in FIG. 2, the digit lines 14 a,b and 16 a,b are coupled to common I/O circuitry 20 through a respective set of isolation transistors 22 a,b , negative sense amps (“NSA”) 24 a,b , positive sense amps (“PSA”) 26 a,b and a respective set of transfer gate transistors 28 a,b .The operation of the isolation transistors 22 a,b and the sense amps 24 , 26 is conventional, and a complete explanation of them will therefore be omitted in the interest of brevity. During a write memory access, data is coupled through a data path (not shown) to the I/O circuitry 20 . Either or both of the sets of transfer gate transistors 28 a,b are turned ON to couple the data to memory cells in an addressed column of an addressed row in either or both of the arrays 12 a,b . If the set of transfer gate transistors 28 a are turned ON, data is written to the memory array 12 a . If the set of transfer gate transistors 28 b are turned ON, data is written to the memory array 12 b . If both sets of transfer gate transistors 28 a,b are turned ON, data is written to both memory arrays 12 a,b . The I/O 20 circuit preferably contains a write buffer 18 that is capable of storing a row of write data. If one of the arrays 12 a,b is being refreshed during a write access to the array, the data are temporarily stored in the write buffer 18 of the I/O circuit 20 . After the refresh has been completed, the write data are transferred from the write buffer 18 to the array 12 a,b that was being refreshed. Alternatively, data may be transferred from the array 12 a,b to which it was written to the other array 12 b,a at a later time after the refresh has been completed. For a read access, one of the arrays, e.g., array 12 a , can be considered a primary array from which the data will be read if the array 12 a is not being refreshed. If the array 12 a is being refreshed, then data can be read from the array 12 b , which is then considered to be a secondary array. Alternatively, each of the arrays 12 a,b can alternatively be considered a primary array, with the other array being considered a secondary array. For example, the array 12 a may be considered the primary array and the array 12 b may be considered the secondary array for a period of 64 ms, which is the duration of a typical refresh period. For the next 64 ms, the array 12 b may be considered the primary array and the array 12 a may be considered the secondary array, and so forth. In either case, data will be read from the primary array if the primary array is not being refreshed. If the primary array is being refreshed, the data are read from the secondary array. The transfer gate transistors 28 a,b are preferably operated by a refresh controller (not shown), which will be explained with reference to FIGS. 3 and 4.

During a read operation, the sense amplifiers 24 a,b and 26 a,b develop a complimentary voltage across the digit lines 14 , 16 corresponding to the data being read from an addressed column in an addressed row of the memory arrays 12 a,b . One of the sets of transfer gate transistors 28 a,b are turned ON to couple the complimentary voltage to the I/O circuitry 20 , and it is then coupled from the I/O circuitry through the data path. If the set of transfer gate transistors 28 a are turned ON, data is read from the memory array 12 a . If the set of transfer gate transistors 28 b are turned ON, data is read from the memory array 12 b . Although data are not simultaneously read from both memory arrays 12 a,b , both transfer gate transistors 28 a,b may be turned ON to perform a write to one array 12 a,b following a read from the other array 12 b ,a. More specifically, for example, the transfer gate transistors 28 a may be turned ON to read data from the memory array 12 a . Then, after the voltage corresponding to the read data has been developed across the digit lines 14 a and 16 a , the transfer gate transistors 28 b may be turned ON to write the data being read from the memory array 12 a to the memory array 12 b.

Although the embodiment of FIG. 2 includes transfer gate transistors 28 a,b specifically provided for coupling data to or from the memory arrays 12 a,b , it will be understood that the transistors 28 a,b may also be part of column addressing circuitry. If so, the transistors 28 a,b are turned ON based on a column address as well as which array 12 a,b is to be accessed. Other circuitry for controlling access to either of the arrays 12 a,b may, of course, also be used.

One embodiment of a DRAM 40 having a hidden refresh is shown in FIG. 3 . As in the DRAM of FIG. 1, the DRAM 40 includes N memory banks each having two arrays 12 a,b . Row and column addresses for accessing data in the arrays 12 a,b are applied through an address bus 42 to an address multiplexer 44 , and command signals are applied through a control bus 46 to a control circuit 48 that includes an internal command decoder (not shown). The control circuit 48 generates a variety of control signals for controlling the operation of the DRAM 40 , some of which are discussed below. The address multiplexer 44 couples row addresses to a row decoder 50 and column addresses to respective column decoders 52 a,b for the arrays 12 a,b . The DRAM 40 also includes respective sense amplifiers 56 a,b like the negative and positive sense amplifiers 24 , 26 shown in FIG. 2, and respective D.C. sense amplifiers (“DCSA”) and write drivers (“WRITE”) 58 a,b (“DCSA/WRITE”). The DC sense amplifiers 58 a,b couple read data from the respective sense amplifiers 56 a,b to a data path circuit 60 . The data path circuit 60 , in turn, couples the read data to a data bus 64 . Write data is coupled from the data bus 64 to the write drivers 58 a,b through the data path circuit 60 . The write drivers 58 a,b ,in turn, couple that data to complimentary pairs of digit lines (not shown in FIG. 3) as determined by the column address. The structure and operation of D.C. sense amplifiers, write drivers, and data path circuits of the type that may be used in the embodiment of FIG. 3 are conventional.

The DRAM 40 also includes a set of transfer circuits 68 , which may be implemented using the transfer gate transistors 28 shown in the embodiment of FIG. 2 . The transfer circuit 68 allow data to be transferred from a column of one array 12 a,b to the corresponding column of the other array 12 b,a , as explained above with reference to FIG. 2 .

As mentioned earlier, the memory cells in the memory arrays 12 a,b must be periodically refreshed to avoid a loss of data. A refresh controller 70 is provided for this purpose. As in a conventional refresh controller, the refresh controller 70 ensures that each row of memory cells in each array 12 a,b is refreshed at a rate sufficient to prevent a loss of data. However, the refresh controller 70 is capable of individually refreshing either memory array 12 a,b , although it preferably does not refresh both memory arrays 12 a,b at the same time. As explained in detail below, the operation of the refresh controller 70 is a function of whether a memory access is either a write or a read, and thus receives control signals from the control circuit 48 , including a read/write* (“R/W*”) signal . The refresh controller 70 also interfaces with a data valid register 74 that, as explained in detail below, provides an indication of whether the data in each row of the arrays 12 a,b are valid. The register 74 receives row addresses from the address multiplexer 44 via the refresh controller 70 so that, as each row of the memory arrays 12 a,b is addressed, the register 74 can provide an indication of whether the data in the addressed row of each array 12 a,b is valid.

The operation of the refresh controller 70 in conjunction with the operation of the DRAM 40 will now be explained with reference to a flow chart 80 , which is shown in FIG. 4 . Based on the flow chart 80 and a knowledge of conventional refresh controller circuitry, the refresh controller 70 used in the embodiment of FIG. 3 can be easily implement by one skilled in the art using either a properly programmed microprocessor (not shown) or hard-wired logic circuitry (not shown). It should be noted, however, that a large number of operating steps of the DRAM 40 have been omitted from the flow chart 80 and the following explanation because they are conventional and somewhat peripheral to the inventive concepts described herein.

With reference to FIGS. 4A-4B the process is entered at 100 with the refresh controller 70 detecting a memory access request from a CPU (not shown in FIG. 3) or other memory access device. The control circuit 48 (FIG. 3) applies suitable control signals to the refresh controller 70 to allow the refresh controller 70 to make this determination. The refresh controller 70 then checks the status of the R/W* signal (FIG. 3) at 102 to determine if the memory access request is for a “Write” memory access. If a “Write” request is detected, the refresh controller 70 determines at 104 if the primary array is currently being refreshed. As explained above, either of the arrays 12 a,b can be used as the primary array, or the array that is used as the primary array can either alternate between the arrays 12 a,b . The refresh controller 70 is able to determine if either of the arrays 12 a,b is being refreshed without any external inputs because the refresh controller 70 itself causes the arrays 12 a,b to be periodically refreshed. If it is determined at 104 that the primary array is being refreshed, then, at 106 , the write data is written to Row R of the secondary array, where R is the row address applied to the address multiplexer 44 . As also explained above, either of the arrays 12 a,b can be used as the primary array, or the array that is used as the primary array can either alternate between the arrays 12 a,b . (It is assumed herein that memory accesses are to an entire row of data in the arrays 12 a ,b). The writing of data to the secondary array in this manner is carried out by circuitry other than the refresh controller 70 , as explained above. However, as also explained above, the refresh controller 70 selects either the primary array or the secondary array for the memory access by suitable means, such as by controlling the transfer gate transistors 28 a,b shown in FIG. 2 . At 108 , the write data is stored in the write buffer 18 , which, as explained above, is internal to the I/O circuitry 20 (FIG. 2 ). The refresh controller 70 then waits in a loop at 110 for the refresh of the primary array to be completed before proceeding to 112 where the data stored in the write buffer 18 are transferred to Row R of the primary array. As a result, data can be written to the DRAM 40 without the need for a CPU or other memory access device to wait for the refresh to be completed. The process then returns to the entry point 100 via step 116 .

If the refresh controller 70 determines at 104 that the primary array is not being refreshed, then it checks at 120 to determine if the secondary array is being refreshed. If so, the write data is written to Row R of the primary array at step 122 , thereby allowing the refresh of the secondary array to continue. The write data is then stored in the write buffer 18 of the I/O circuitry 20 at 124 . The refresh controller 70 then waits in a loop at 126 for the refresh of the secondary array to be completed before proceeding to 128 where the data stored in the write buffer 18 are transferred to Row R of the secondary array. The process then again returns to 100 via 116 .

If the refresh controller 70 determines at 104 that the primary array is not being refreshed and it determines at 120 that the secondary array is not being refreshed, then the write data can be written to both arrays 12 a,b . Accordingly, the write data is written to Row R of both the primary array and the secondary array 12 b at 130 before returning to 100 via 116 .

If, after detecting a memory access request at 100 , the refresh controller 70 does not detect a “Write” memory access request at 102 , then it concludes the request must be for a “Read” memory access request and proceeds to step 140 . At 140 , the refresh controller 70 determines if the primary array is being refreshed. If not, the refresh controller 70 causes data to be read from Row R of the primary array at step 142 . The process then returns to step 100 via 144 If the refresh controller 70 determines at 140 that the primary array is not being refreshed, it causes data to be read from Row R of the secondary array at step 146 before returning via 144 . The refresh controller 70 is able to select either the primary array or the secondary array for the memory access by suitable means, such as by controlling the transfer gate transistors 28 a,b (FIG. 2) so that the data are read from the primary array rather than the secondary array or vice-versa.

The DRAM 40 described with reference to FIGS. 3 and 4 can be advantageously used in a computer system, such as a computer system 200 shown in FIG. 5 . The computer system 200 includes a processor 202 for performing various computing functions, such as executing specific software to perform specific calculations or tasks. The processor 202 includes a processor bus 204 that normally includes an address bus 206 , a control bus 208 , and a data bus 210 . In addition, the computer system 200 includes one or more input devices 214 , such as a keyboard or a mouse, to allow an operator to interface with the computer system 200 . Typically, the computer system 200 also includes one or more output devices 216 coupled to the processor 202 , such output devices typically being a printer or a video terminal. One or more data storage devices 218 are also typically coupled to the processor 202 to store data or retrieve data from external storage media (not shown). Examples of typical storage devices 218 include hard and floppy disks, tape cassettes, and compact disk read-only memories (“CD-ROMs”). The processor 202 is also typically coupled to system memory 228 through a memory controller 230 . The system memory 228 is typically composed of dynamic random access memories (“DRAMs”). Finally, the computer system 200 includes a cache memory 236 that is coupled to the processor 202 through the processor bus 204 . As explained above, the cache memory 236 is usually static random access memory (“SRAM”), and it is provided for the purpose of allowing the processor 202 to more quickly read frequently or recently accessed data or instructions. However, unlike conventional computer systems, the cache memory 236 used in the computer system 200 of FIG. 5 is a DRAM having hidden of the type shown in FIGS. 3 and 4. Even though the cache memory 236 requires twice as many memory cells as a conventional DRAM of the same capacity, it nevertheless uses about half the number of transistors as a conventional SRAM of the same capacity. Thus, the cache memory 236 is able to provide a relatively high capacity at relatively little cost compared to an SRAM used as the cache memory 236 .

It will be appreciated that, although specific embodiments of the present invention have been described above for purposes of illustration, various modifications may be made without deviating from the spirit and scope of the invention. For example, although the DRAM 40 is shown in FIG. 4 in somewhat basic form, it will be understood that the DRAM 40 may assume a variety of DRAM configurations, including synchronous DRAMs (“SDRAMs”), Rambus DRAMs (“RDRAMs”) and Synchronous Link DRAMs, (“SLDRAMs”). Further, although FIG. 5 shows the DRAM 40 being accessed by a processor 202 , it will be understood that other access devices, such as graphics processors and other direct memory access (“DMA”) devices, may also be used. Accordingly, the invention is not limited by the disclosed embodiments, but instead the scope of the invention is determined by the following claims.