Reconfigurable memory with selectable error correction storage

[0002] The present invention relates to memory structures for computers, and more particularly, to error correction in computer memories.

[0003] Semiconductor memory systems are subject to errors. That is, data retrieved from the memory does not always match data that was originally written to the memory. Such errors can be caused by stray alpha particles, damage to the memory devices or by a variety of operating conditions, such as power supply fluctuations, noise, etc. Regardless of the source, such errors are clearly undesirable. Consequently, most modern memory systems include error is detection and/or error correction capabilities.

[0004] Typical approaches to detecting and correcting errors in memory rely upon some form of error correction code to identify and correct such data errors. Such error correction codes typically include a mathematical algorithm that is applied to the data to be checked and corrected, and additional error correction code (“ECC”) bits. Usually, the ECC bits are stored in a separate memory dedicated to the ECC bits. The amount of memory dedicated to storing the ECC bits can be significant. For example, the memory overhead for the ECC bits can often exceed 10%.

[0005] The amount of ECC bits required can depend upon the type of error correction code being utilized. In some applications, very little or no error correction is desired. For example, in video games, occasional image data errors are unlikely to significantly affect the images perceived by a user. Rather than devote processor power to error correction calculations and memory to ECC bits, such applications largely ignore image data errors to increase the speed of play.

[0006] Such applications will be referred to herein as error tolerant applications. Error tolerant applications typically use no error correction calculations or limited error correction algorithms that require little or no ECC memory.

[0007] Other applications can tolerate little or no data errors. For example, data errors can be extremely undesirable in accounting programs. Such applications will be referred to herein as error intolerant applications. Error intolerant applications usually utilize robust error correction algorithms requiring a substantial amount of ECC memory.

[0008] Typically, memory devices for storing ECC bits are segregated from memory devices for conventional data. For example, 144 pin 4-MB×64 double in-line memory module (“DIMM”) not used to store ECC bits could be implemented using 16 4-MB×4 dynamic random access memories (“DRAMs”). However, the same data storage capacity plus the capacity to store ECC bits would require a 4-MB×72 DIMM implemented using 18 4-MB×4 DRAMs. Thus, implementing ECC requires two additional DRAMs.

[0009] One problem with such memory architectures is that they do not fully utilize the available memory capacity. For example, error tolerant applications do not need nor use the extra memory provided to store ECC bits. Thus, valuable memory capacity is left unused. In the above example, 11% of the DRAMs on the DIMM are wasted when the DIMM is not used to store ECC bits.

[0010] On the other hand, error intolerant applications require more memory and are often limited by the amount of available ECC memory. Consequently, the speed with which the application runs can be increased by increasing the amount of available ECC memory. Adding such memory can be costly. Moreover, adding such memory capacity increases the amount of unused memory in error tolerant applications.

[0011] A software or hardware controlled reconfigurable memory system includes an auxiliary section of one or more data banks that can be selectively utilized as conventional memory or ECC memory, depending upon the particular application. In one embodiment, the auxiliary section is part of a memory module that includes a primary section directly coupled to an output data bus for conventional memory uses. A primary multiplexer selectively couples the auxiliary section to either the output data bus or to an error checking circuit, depending upon the selected configuration of the system. If the system runs an error intolerant application employing a robust error correction algorithm, the auxiliary section is coupled to the error correction circuit to store ECC data for ECC calculations. In error tolerant applications not requiring error correction, the auxiliary section is coupled to the output data bus to supplement the conventional memory, thereby providing increased memory capacity and improving speed of the system.

[0012] One embodiment of the invention also includes a dedicated ECC memory, which could be located on the motherboard. A secondary multiplexer receives data from the dedicated ECC memory at one input and data from the primary multiplexer at a second input. The primary and secondary multiplexers are controlled by software or hardware to establish the amount of ECC memory being used. For error intolerant applications, the primary multiplexer is activated to couple data from the auxiliary section to one input of the secondary multiplexer. The secondary multiplexer is then activated to couple data from both the primary multiplexer and the dedicated ECC memory to the error correction circuit. Thus, the auxiliary section is used to supplement the dedicated ECC memory in error intolerant applications where additional ECC memory is desirable.

[0013] In one embodiment, the second input of the secondary multiplexer is coupled to a set of memory sockets on the motherboard. The secondary multiplexer selectively couples only those sockets containing memory chips to the error correction circuit. Also, the primary and secondary multiplexers are controlled to select an appropriate portion of the auxiliary section to supplement the dedicated ECC memory, according to the ECC data requirements of an application and the amount of available dedicated ECC memory.

[0014] In one embodiment, the auxiliary section is segmented into two sections. The first section is used to supplement the dedicated ECC memory from the motherboard. The second section is used as a supplement to the conventional memory. To accommodate the difference in word length caused by segmenting of the auxiliary section, the second section is “double-written” and “double-read” so that data is written to and read from the second section in two or more pieces. When reading the data, the two or more pieces are combined to form the complete written data.

[0015] FIG. 1 is a block diagram of a memory system according to an embodiment of the invention on which a memory device is selectively coupled to either an output data bus or to an error correction circuit by a primary multiplexer.

[0016] FIG. 2 is a more detailed block diagram of the memory system of FIG. 1 coupled to a memory controller.

[0017] FIG. 3 is a block diagram of another embodiment of the memory system of FIG. 1.

[0018] FIG. 4 is a block diagram of still another embodiment of the memory system of FIG. 1.

[0019] FIG. 5 is a block diagram of a computer system according to one embodiment of the invention including input and output devices, a processor, and the memory system of FIG. 1.

[0020] As shown in FIG. 1, a memory system 40 according to an embodiment of the invention includes as its central storage element a memory array 42 having at least one low-order bank 44 and at least one high-order bank 46. The memory array 42 is formed from one or more integrated memory devices. The devices may be any suitable type of memory device, such as dynamic random access memories (DRAMs), static RAMs (SRAMs), or non-volatile memory devices. The memory devices may also be synchronous or asynchronous, or some other variety of memory device.

[0021] Data that are to be written to and read from the memory array 42 are coupled to and from the array 42 along a 64-bit primary data path 48 and a 64-bit auxiliary data path 50, respectively. The primary data path 48 extends directly from the low order banks 44 to a data bus 52. Although the embodiment of FIG. 1 uses 64-bit buses 48, 50, it will be better understood that higher or lower capacity buses may be used.

[0022] The auxiliary data path 50 is coupled to a primary multiplexer 54, which is a 64-bit, 1-to-2 multiplexer. In response to a mode select input MODE, the primary multiplexer 54 selectively couples data from the high order banks 46 to either the data bus 52 or to an error correction circuit 56. The high order banks 46 can therefore provide storage for ECC bits or can provide storage to supplement the primary section 44.

[0023] FIG. 2 shows the memory system 40 of FIG. 1 in greater detail in conjunction with a separate error correction memory (ECC memory) 58. In this embodiment, the memory array 42 includes eight memory banks 60₀-60₇where the first seven banks 60₀-60₆correspond to the low order banks of FIG. 1 and the eighth memory bank 60₇corresponds to the high order bank 46 of FIG. 1. The output of the eighth bank 60₇is input to the primary multiplexer 54 through the auxiliary data path 50 while the outputs of the first seven banks 60₀-60₆are connected directly to the data bus 52.

[0024] Switching of the primary multiplexer 54 is controlled by a memory controller 63 through a mode signal MODE, in response to either software or hardware commands. Depending upon the state of the mode signal MODE, the primary multiplexer 54 couples data from the eighth data bank 60₇to either the data bus 52 or to a secondary multiplexer 64. If the mode signal MODE is high, the primary multiplexer 54 directs data to the data bus 52. If the mode signal MODE is low, the primary multiplexer 54 directs data to the secondary multiplexer 64.

[0025] Like the primary multiplexer 54, the secondary multiplexer 64 includes eight, 8-bit 2-to-1 multiplexers, rather than eight, 8-bit 1-to-2 multiplexers. Thus, the 64 outputs of the primary multiplexer 54 are coupled to a first set of 64 inputs of the secondary multiplexer 64. The second set of 64 inputs of the secondary multiplexer 64 is coupled to the ECC memory 58. When enabled by the mode signal MODE and by an error correction enable signal ECCENABLE, the secondary multiplexer 64 couples one of eight 8-bit bytes of ECC data to the error correction circuit 56 as determined by a SELECT input from the memory controller. The first two 8-bit bytes of ECC data are supplied by respective ECC chips 68 in the ECC memory 58. The remaining six 8-bit bytes are supplied by the high order banks 46 through the primary multiplexer 54.

[0026] The error correction circuit 56 operates on the 8 bits of ECC data ECC₀-ECC₇to identify and correct errors according to conventional error correction techniques, such as Hamming code or similar correction algorithms. The error correction circuit 56 may be implemented as dedicated hardware or as a software program in a processor 210 (FIG. 5).

[0027] In operation, the memory system 40 can operate in either an error tolerant mode or an error intolerant mode. In the error tolerant mode, the primary multiplexer 54 couples the high order bank 60₇to the data bus 52 so that all 8 banks 60₀-60₇are used for storing data. In the error intolerant mode, the primary multiplexer 54 couples the high order bank 60₇to the secondary multiplexer 64, the secondary multiplexer 64 then couples the high order bank 60₇to the error correction circuit 56. Alternatively, the secondary multiplexer 64 can couple the ECC memory 68 to the error correction circuit 56. In either case, when data are written to the low order banks 60₀-60₆, 8 ECC check bits are applied to the error correction circuit 56. The error correction circuit 56 then couples 64 bits to the high order banks 60₇through the secondary multiplexer 64 and the primary multiplexer 54.

[0028] During a read operation, data are coupled from the low order banks 60₀-60₆to the data bus 52, and corresponding ECC data are coupled from the high order bank 60₇through the multiplexer 54, 64 to the error correction circuit 56. The error correction circuit 56 then checks the 8 ECC bits in a conventional manner to detect and correct errors in the data coupled to the data bus 52.

[0029] FIG. 3 shows the system of FIG. 2 where the error correction implementation is less robust, i.e., ECC data is supplied only by the ECC memory 58 on the mother board. Consequently, the two ECC chips 68 in the ECC memory 58 provide the ECC data for banks 60₀and 60₁only. The secondary multiplexer 64 thus outputs only the ECC data for the banks 60₀and 60₁to the error correction circuit 56. More memory chips would need to be added to ECC memory 58 in order to support more databanks 60₂-60₆.

[0030] FIG. 4 shows another embodiment of the invention in which the ECC memory 58 includes four ECC chips 68 and which uses an error correction algorithm involving 8-bits of ECC data ECC₀-ECC₇. The four ECC chips 68 supply the ECC data for banks 60₀-60₃. Therefore, only half of the high order bank 60₇is used for error correction of banks 60₄-60₇, leaving half of the high order bank 60₇free for conventional memory use. Instead of leaving half of the high order bank 60₇unused, the high order bank 60₇is broken into two subsections 46A, 46B where the first subsection 46A contains ECC data and the second subsection 46B is used as conventional memory to supplement the low order banks 60₀-60₆. The first subsection 46A provides the second four 8-bit bytes of error correction data used for memory bank 60₀-60₃to the secondary multiplexer 64 through the primary multiplexer 54 in a similar fashion to that described above for FIG. 2. The secondary multiplexer 64 selects either the on-board ECC data for banks 60₄-60₇or the auxiliary ECC data for banks 60₀-60₃.

[0031] The second subsection 46B of the high order bank 60₇is not wasted. Instead, the second subsection 46B provides data to the data bus 52 through the primary multiplexer 54. One skilled in the art will recognize that the second subsection 46B will only be comprised of a memory bank that is half the depth of the banks 60₀-60₆. Since the high order bank 60₇cannot both supply data and ECC bits simultaneously, the on board ECC memory 58 must supply the ECC data for the second subsection of the high order bank 60₇.

[0032] FIG. 5 is a block diagram of a computer system 200 that uses one of the embodiments of FIGS. 2-4. The computer system 200 includes a processor 210 for performing computer functions, such as executing software to perform desired calculations and tasks. The processor 210 accesses the memory module 40 and ECC memory 58 through the data bus 52 by activating the memory controller 63 which, in turn, controls the multiplexers 54, 64. The memory module 40 and the ECC memory 58 are preferably mounted at separate locations within the computer system 200 with the ECC memory 58 being mounted to a common board with the processor 210. The memory module 40 and ECC memory 58 are coupled to the processor 210 through the data bus 52. The processor 210 is also coupled to the error correction circuit 56 to receive error detection and correction information that the processor 210 uses to control the memory controller 63. One or more input devices 214, such as a keypad or a mouse, are coupled to the processor 210 through an I/O controller 216 and allow an operator (not shown) to manually input data thereto. One or more output devices 218 are coupled to the processor 210 through the I/O controller to provide to the operator data generated by the processor 210 or retrieved from the memory module 40. Examples of output devices 218 include a printer and a video display unit. One or more mass data storage devices 220 are preferably coupled to the processor 210 through the I/O controller 216 to store data in or retrieve data from the storage device 220. Examples of the storage devices 220 include disk drives and compact disk read-only memories (CD-ROMs).

[0033] While the present invention has been described herein by way of exemplary embodiments, various modifications may be made without departing from the scope of the invention. For example, the number of ECC chips 68 in the ECC memory 58 may be larger or smaller. Also, the secondary multiplexer 64 can be configured to vary the multiplexing of the ECC data in response to the enable signal ECC ENABLE, so that the combination of bits from the ECC memory 58 and auxiliary section 46 can be controlled remotely by the memory controller 63. Additionally, the high order bank 46 may include more than one bank of the memory array 42. Further, the memory array 42 may include fewer or more than eight banks. And, the ECC memory 58 can be located off the motherboard in some applications. Moreover, the error correction circuit 56 and/or the memory controller 63 can be implemented in whole or in part by the processor 210 in response to software. Also, the number of bits in each bank or on the data busses may be fewer than or more than the 64-bit bus structure described herein. Accordingly, the invention is not limited except as by the appended claims.