Memory module and method having improved signal routing topology

[0001] This application is a continuation of U.S. patent application Ser. No. 11/973,684, filed Oct. 9, 2007, which is a continuation of U.S. patent application Ser. No. 11/238,900, filed Sep. 28, 2005, U.S. Pat. No. 7,282,947, which is a divisional of U.S. patent application Ser. No. 10/932,477, filed Sep. 1, 2004, U.S. Pat. No. 7,242,213, which is a continuation of U.S. patent application Ser. No. 10/460,588, filed Jun. 11, 2003, U.S. Pat. No. 7,245,145. These applications are incorporated by reference herein.

[0002] The present invention relates to memory systems, and, more particularly, to memory modules having transmission lines coupled to a large number of memory devices.

[0003] Memory devices are in widespread use in a variety of processor-based systems, such as computer systems. In some cases, memory devices can be mounted on the same circuit board as a processor and a memory controller, which is generally used to couple the memory devices to the processor. However, in most cases, the memory devices are part of a memory module in which several memory devices are mounted on a common substrate, such as a printed circuit board. The memory modules are generally plugged into sockets mounted on a motherboard to establish communication with a memory controller and processor. Such memory modules are commonly used in computer systems as system memory in which the memory modules are dynamic random access memory (“DRAMs”) devices.

[0004] Although the memory devices in a memory module may be coupled directly to a memory controller, in one type of memory module, known as a “registered” memory module, the memory devices are coupled to a register that is, in turn, coupled to the memory controller. More specifically, in a registered DRAM module, the command and address lines from the memory controller are coupled to a register. The register stores memory commands and addresses coupled through the command and address lines, respectively, and then couples the commands and addresses to the memory devices. Data signals are typically coupled directly to and from the memory devices without being registered. By registering the command and address signals, they can be coupled to the memory module for a relatively short period of time since it is not necessary to wait for the memory devices to latch the command and address signals. Also, registering the command and address signals avoids excessive loading of the command and address lines because the command and address lines are coupled to only a single device, i.e., the register, rather than to multiple devices, i.e., all of the memory devices.

[0005] The manner in which each of the command and address lines are routed from the register to the memory devices can significantly affect the performance of the memory module. One coupling topology, known as a “daisy chain” topology, is shown in FIG. 1A. In a daisy chain topology, a first transmission line 10, which may be a command signal line or an address signal line, extends from a register 16 to one end of a second transmission line 20 (which is a single conductor, but functions as separate segments or transmission lines 20a-g). Respective transmission lines 30a-h are coupled to spaced apart locations of the transmission lines 20(a-g). The transmission lines 30a-h each extend to an input terminal of respective memory devices 36a-h, which, in this case are DRAM devices.

[0006] A “hybrid tree” topology shown in FIG. 1B differs from the daisy chain topology of FIG. 1A in that the first transmission line 10 is coupled to the center of the second transmission line 20a rather than to one of its ends. In alternative embodiments of a hybrid tree topology, the first transmission line 10 may be coupled to locations of the second transmission lines 20(a-d) other than either one end or the center of the second transmission line 20. Like the daisy chain topology shown in FIG. 1A, respective transmission lines 30a-h are coupled to spaced apart locations of the transmission lines 20a-d and extend to input terminals of the respective memory devices 36a-h.

[0007] Still another hybrid tree topology shown in FIG. 1C uses two separate transmission lines 20a,b coupled to the transmission lines 30a-d, 30e-h. Transmission lines 20a and 20b are made up of transmission lines 20aa, 20ab, 20ac, 20ad, 20 ba, 20 bb, 20 bc and 20 bd respectively. The transmission lines 20a,b are, in turn, coupled to the first transmission line 10 through a third transmission line 40, which joins the first transmission line 10 at the center of the transmission line 40. Transmission line 40 is made up of transmission lines 44a and 44b. Transmission lines 44a and 44b are coupled to the transmission lines 20a,b at the ends of the transmission line 40 by respective transmission lines 44a,b.

[0008] The daisy chain and hybrid tree topologies shown in FIGS. 1A-1C can provide adequate performance at relatively slow speeds, but they provide less than optimum performance at higher operating speeds. In particular, signals coupled through the transmission line 20 (or transmission lines 20a,b in the case of the topology shown in FIG. 1C) reflect from the junctions with the transmission lines 30a-h as well as from the junction between each of the transmission lines 30a-h and its respective memory devices 36a-h, respectively. These reflections produce destructive and constructive interference at each of the junctions between the transmission line 20 and the transmission lines 30a-h that can seriously degrade signals coupled to the memory devices 36. The hybrid tree topologies shown in FIGS. 1B and 1C place the transmission lines 30a-h closer to the ends of the transmission line 20 compared to the daisy chain topology shown in FIG. 1A. Therefore, the hybrid tree topologies tend to provide better performance than the daisy chain topology. However, the hybrid tree topologies still provide less than optimum performance. Further, the hybrid tree topology shown in FIG. 1C also produces reflections from the junctions between the transmission lines 20a,b and the transmission lines 44a,b.

[0009] Although the signal reflection problems have been described in the context of registered memory modules, the same problem can exist in other types of memory modules. For example, in a memory hub module, signals are coupled from a memory hub in the module to each of several memory devices in the module. Also, in a buffered memory module, signals are coupled from respective buffers in the module to each of several memory devices in the module. Reflections produced in these types of memory modules can degrade performance in essentially the same manner as described above.

[0010] There is therefore a need for a connection topology for routing signals to memory devices in memory modules that can avoid signal degradation caused by reflections generated at the junctions between transmission lines and memory device terminals and between different transmission lines.

[0011] A signal routing topology and method couples signals between an active memory component, such as a register, and a plurality of memory devices using transmission lines. The transmission lines are each connected at only its ends to either an input terminal of one of the memory devices or to an end of another of the transmission lines. The transmission lines form a symmetrical tree having several levels of hierarchy, with the transmission lines in the same level of hierarchy having the same length. The transmission lines are preferably impedance matched to pairs of transmission lines to which they are coupled. In the event the memory module contains an odd-numbered memory device, such as an error checking memory device, the odd-numbered memory device is coupled to the active memory component through a dedicated transmission line.

[0016] A topology for routing signals from a register to memory devices in a registered memory module according to one embodiment of the invention is shown in FIG. 2. The memory module 50 includes a register 58 and four dynamic random access memory (“DRAM”) devices 60a-d. A large number of address and command lines are coupled from the register 58 to the DRAM devices 60a-d, although the signal lines for routing only one of these signals are shown in FIG. 2. The register 58 includes an output buffer 62 for each signal that is coupled to a first transmission line 64. The end of the first transmission line 64 is coupled to a first branch 70 that consists of two transmission lines 72, 74. The branch 70 is symmetrical in that both of the transmission lines 72, 74 have the same electrical and physical characteristics, particularly the same length. The ends of the transmission lines 72, 74 are coupled to a second pair of branches 80, 82, respectively. The branches 80, 82 are symmetrical internally and with each other as they are each formed by two transmission lines 86, 88 and 90, 92, respectively, having the same length. Although it is necessary that the transmission lines 86, 88 and 90, 92 of the respective branches 80, 82 all be of the same length, it is not necessary that the lengths of these transmission lines 86, 88, 90, 92 be equal to the lengths of the transmission lines 72, 74 in the branch 70. Finally, the end of each of the transmission lines 86, 88, 90, 92 is coupled to a respective one of the DRAMs 60a-d.

[0017] In operation, a signal is coupled through the first transmission line 64 to the transmission lines 72, 74 of the first branch 70. If these transmission lines are of the same impedance there is a mismatch and some of the energy reflected and some of the energy is passed. When the signal reaches the ends of the transmission lines 72, 74, it is again reflected from the junctions with the branches 80, 82, respectively. As a result, the signal is coupled to the branches 80, 82 with less magnitude. The reflections of the signal are then coupled back through the transmission lines 72, 74 to the junction of the transmission line 64. The reflected signals reach the junction with the transmission line 64 at the same time so that they appear as a single transmission line coupling to a higher impedance transmission line. This acts to reflect in phase increasing the voltage that eventually reaches the DRAMs. In a similar manner, the signal coupled to the branches 80, 82 have a relatively large magnitude at the ends of the transmission lines 86, 88 and 90, 92 because of their reflection from an input terminal of the DRAMs 60a-d. An open circuit reflects the signal resulting in an apparent doubling of the voltage. The signals reflected from the DRAMs 60a-d reach the junctions to the transmission lines 72, 74 at the same time because the lengths of the transmission lines 86, 88, 90, 92 are all identical. As a result, the impedance of the transmission lines 86, 88 and 90, 92 at their junction to the transmission lines 72, 74, respectively, appear to have a lower impedance coupling to a higher impedance resulting in an in phase reflection back to the DRAMs resulting in more signal at the DRAMs. The signals reflected through the transmission lines 86, 88 and 90, 92 are also coupled through the transmission lines 72, 74, respectively, of the first branch 70 where they reach the junction to the first transmission line 64 at the same time. Again another reflection back towards the DRAMs. The result of all these reflections is a stairstep of increasing voltage at the DRAMs.

[0018] Using a tree of transmission lines in which each branch is entirely symmetrical maximizes the magnitude of signals coupled to the DRAMs 60a-d and minimizes the magnitude of reflections reflected from the end of each transmission line. By coupling the DRAMs 60a-d only to the ands of the transmission lines, the destructive interference away from the ends of the transmission lines, which reduces signal amplitude, is avoided.

[0019] The symmetrical tree used in the memory module 50 of FIG. 2 includes branches with only two levels of hierarchy, namely the first branch 70 with a first level of hierarchy and the second branches 80, 82 with a second level of hierarchy. However, since each branch consists of two transmission lines each coupled to either a transmission line or a memory device at only its end, a memory module having 2^N of the memory devices will have N hierarchies of branches.

[0020] The reflections from the junctions between the transmission lines 86, 88 and 90, 92 and the transmission lines 72, 74, respectively, and from the junctions between the transmission lines 72, 74 and the transmission line 64 can be reduced even further by impedance matching the transmission lines. More specifically, as is well known in the art, each transmission line has a characteristic impedance. As is also well-known in the art, signals coupled through a transmission line are reflected from impedance discontinuities in the line. Therefore, such reflections can be avoided by avoiding impedance discontinuities. If the first transmission line 64 has an impedance of R (e.g., 17.5 ohms), the transmission lines 72, 74 in the first branch should each have an impedance of 2R (i.e., 35 ohms) since two resistors in parallel have an impedance of half the impedance of each resistor. Similarly, the transmission lines 86, 88, 90, 92 should each have an impedance of 4R (i.e., 70 ohms).

[0021] Although impedance matching of the transmission lines used in a symmetrical tree in accordance with the invention is desirable, it is not necessary. Furthermore, it may not be practical to provide impedance matching where the tree includes a large number of branch hierarchies because the practical range of impedance values that can be obtained is very limited. With commonly used conductive materials mounted on commonly used substrates, the transmission lines formed by the conductive materials become excessively wide at impedance values much less than 15 ohms, and they become excessively narrow at impedance values much larger than 80 ohms.

[0022] Data are commonly stored in a memory module using an even number of memory devices, and, more commonly, a number of memory devices equal to powers of two, i.e. 2, 4, 8, etc. However, some memory modules include error checking and/or correcting (“ECC”) capabilities, which generally requires an additional memory device. The use of an odd number of memory devices, such as an ECC memory device, precludes coupling the memory devices through a symmetrical tree. However, the benefits of a symmetrical tree can be obtained using the topography shown in FIG. 3. FIG. 3 shows the same components of a memory module that are shown in FIG. 3 with the addition of an ECC DRAM 90. Therefore, in the interest of brevity, all of the component shown in FIG. 3 have been provided with the same reference numerals, and an explanation of their structure in operation will not be repeated. Although ECC capabilities can be obtained using the odd-numbered ECC DRAM 90 as shown in FIG. 3, ECC capabilities can also be obtained by simply using one of the DRAMs 60a-d to store ECC data.

[0023] The ECC DRAM 90 is coupled to the output of the buffer 62 through a dedicated transmission line 94 extending from a location at or near the buffer 62 to a terminal of the DRAM 90. By placing the junction between the transmission line 94 and the first transmission line 64 near the buffer 62, the magnitude of any reflections from the junction that are coupled through the tree is relatively small. The transmission line 94 preferably has a length that is equal to the combined length of the transmission lines between the buffer 62 and the DRAMs 60a-d so that a signal from the buffer 62 reaches the ECC DRAM 90 at the same time the signal reaches the DRAMs 60a-d. Also, any reflections from the DRAMs 60a-d reach the buffer 62 at the same time that any reflection from the ECC DRAM 90 reaches the buffer 62 so that the transmission lines to ECC DRAM 90 and DRAMs 60a-d can be considered to be in parallel. If the output impedance of buffer 62 is equal to this parallel impedance then signals reflected from the DRAMs 60a-d and 90 are not reflected from the output of the buffer 62.

[0024] A computer system 100 according to one embodiment of the invention is shown in FIG. 4. The computer system 100 includes a processor 104 for performing various computing functions, such as executing specific software to perform specific calculations or tasks. The processor 104 includes a processor bus 106 that normally includes an address bus, a control bus, and a data bus. The processor bus 106 is typically coupled to cache memory 108, which, is typically static random access memory (“SRAM”). Finally, the processor bus 106 is coupled to a system controller 110, which is also sometimes referred to as a bus bridge.

[0025] The system controller 110 serves as a communications path to the processor 104 for a variety of other components. More specifically, the system controller 110 includes a graphics port that is typically coupled to a graphics controller 112, which is, in turn, coupled to a video terminal 114. The system controller 110 is also coupled to one or more input devices 118, such as a keyboard or a mouse, to allow an operator to interface with the computer system 100. Typically, the computer system 100 also includes one or more output devices 120, such as a printer, coupled to the processor 104 through the system controller 110. One or more data storage devices 124 are also typically coupled to the processor 104 through the system controller 110 to allow the processor 104 to store data or retrieve data from internal or external storage media (not shown). Examples of typical storage devices 124 include hard and floppy disks, tape cassettes, and compact disk read-only memories (CD-ROMs).

[0026] The system controller 110 includes a memory controller 128 that is coupled to several registered memory modules 130a,b . . . n, which serve as system memory for the computer system 100. The memory modules 130 are coupled to the memory controller 128 through a bus system 134. The memory modules 130 are shown coupled to the memory controller 128 in a multi-drop arrangement in which the single bus system 134 is coupled to all of the memory modules 130. However, it will be understood that other topologies may also be used.

[0027] Each of the memory modules 130 includes a register 140 for storing command and address signals as well as eight memory devices 148, which, in the example illustrated in FIG. 4, are synchronous dynamic random access memory (“SDRAM”) devices. However, a fewer or greater number of memory devices 148 may be used, and memory devices other than SDRAM devices may also be used. The register 140 is coupled to each of the system memory devices 148 through symmetrical tree 150 in accordance with the present invention. An error checking and/or correcting memory device is included in the memory module 130n, in which the topography shown in FIG. 3 is preferably used.

[0028] From the foregoing it will be appreciated that, although specific embodiments of the invention have been described herein for purposes of illustration, various modifications may be made without deviating from the spirit and scope of the invention. For example, transmission line topologies according to the present invention can be used to couple signals to memory devices other than DRAMs and to memory devices from components other than registers. As previously mentioned, transmission line topologies according to the present invention can be used to route signals to memory devices from buffers or memory hubs, for example. Accordingly, the invention is not limited except as by the appended claims.