Patent Application
20080019198
The present invention relates to an improvement in memory modules. The following description is presented to enable one of ordinary skill in the art to make and use the invention as provided in the context of a particular application and its requirements. Various modifications to the preferred embodiment will be apparent to those with skill in the art, and the general principles defined herein may be applied to other embodiments. Therefore, the present invention is not intended to be limited to the particular embodiments shown and described, but is to be accorded the widest scope consistent with the principles and novel features herein disclosed.
The inventors have realized that a spare memory may be integrated with the Advanced Memory Buffer (AMB) of a fully-buffered memory module. This spare memory on the AMB may be accessed rather than a defective memory location on a memory chip. The AMB can compare incoming addresses to a list or map of defective addresses, and re-route the memory access to the spare memory when a hit occurs.
The spare memory on the AMB allows the memory chips mounted onto the fully-buffered memory module to contain a few defects. Manufacturing costs may be reduced since a limited pre-screen test of incoming memory chips may be performed, rather than a more exhaustive pre-screen test. Memory modules with a single-bit defect in one of its memory chips do not have to be reworked, eliminating the time-consuming de-soldering and soldering of the defective memory chip.
Serial packets are received and retransmitted over southbound lanes for packets generated by the host. Serial packets are received and retransmitted over northbound lanes for packets generated by the memory modules. Serializer-deserializer 504 is an interface to the northbound and southbound lanes, and can examine serial packets and extract address, data, and commands from the packets. Serializer-deserializer 504 can also create serial packets for transmission back to the host, such as packets containing data read from local memory chips.
When serializer-deserializer 504 determines that an incoming packet is destined for the local memory module, the information in the packet is extracted and any address or data is converted from serial to a parallel format. The extracted address is sent to address router 502, which examines the address to determine if the address is for a defective memory location. When the address is not to a defective location, address router 502 passes the address and the request to DRAM controller 50.
DRAM controller 50 generates local control signals such as RAS, CAS, WE and sends these signals to DRAM chips on the memory module. State machines may be used by DRAM controller 50 to generate these signals with appropriate pulse widths and timings to properly access the DRAM chips. Chip-select and byte-enable signals to the DRAM chips may be generated from the address as well as from these control signals. Data is applied to the DRAM chips for a write, or read from the DRAM chips being accessed for a read. The read data is then passed back to serializer-deserializer 504, converted to serial packets and transmitted back to the host over the northbound lanes.
Sometimes the address being accessed is for a defective location. The memory chip is unreliable for that defective address location. Rather than access the defective location in the memory chip, a spare memory is accessed. For addresses matching defective addresses, address router 502 sends the request to memory buffer 506 rather than to DRAM controller 50. DRAM controller 50 is effectively disabled for these defective memory locations.
Memory buffer 506 contains spare memory that may be accessed rather than defective memory in the defective memory chip. Data that is to be written to the defective memory chip is instead written to a memory location within memory buffer 506. When the defective memory location is read, the read data is read from a location in memory buffer 506 rather than from the defective memory chip.
Memory buffer 506 may contain just one memory location, allowing only one defect to be repaired in the entire memory module. Alternately, memory buffer 506 may contain several locations, allowing several different defects to be repaired. Two or more memory chips could be defective, or there could be two defects in one chip. Each spare memory location in memory buffer 506 could be a byte or larger word that replaces an entire byte or word in the defective memory chip, even when only one bit is defective. A larger block could also be replaced with the spare memory.
Repair address buffer 68 contains a list of one or more defective memory locations. The entire address of the defective location could be stored in repair address buffer 68, or just a portion of the defective address. The address could be divided into tag and index portions when repair address buffer 68 is arranged as a cache, or repair address buffer 68 could be a fully-associative buffer.
Address router 502 compares addresses of requests from serializer-deserializer 504 to repair addresses in repair address buffer 68 to determine when the request is to a defective memory location. The address lookup could delay processing of the request, or pipelining may be able to reduce or eliminate this lookup delay. When repair address buffer 68 contains few entries, the lookup may be very short. DRAM controller 50 could be activated for all requests, and then de-activated when a defective address hit is detected by address router 502 in repair address buffer 68.
The defective address locations stored in repair address buffer 68 may be programmed or written through SM-bus interface 134. SM-bus interface 134 receives serial commands from SM bus 192, which is a system management bus. Repair address buffer 68 may be a volatile memory such as a register, static RAM, or dynamic RAM, and have to be loaded at each power-up. Alternately, repair address buffer 68 could be a non-volatile memory such as electrically-erasable programmable read-only memory (EEPROM) or fuses that could be blown.
SPD-EEPROM 130 also stores locations of defective memory on the memory module. During manufacture, defective locations are identified and their addresses are written as repair addresses 132 in SPD-EEPROM 130. Each time that the memory module is powered up or re-initialized, repair addresses 132 are transferred to repair address buffer 68. SM-bus interface 134 reads repair addresses 132 from SPD-EEPROM 130 over SM bus 192.
Repair controller 69 compares addresses of requests to the repair addresses stored in repair address buffer 68 to determine when a repair hit occurs. For repair-address hits, spare memory is accessed by repair controller 69 rather than the defective memory in the DRAM chips through DRAM controller 50. The spare memory may be within repair controller 69 or may be attached to repair address buffer 68.
Repairing AMB 100 contains DRAM controller 50. For non-repair addresses, DRAM controller 50 generates DRAM control signals to read and write data to and from DRAM chips 22 on memory module 10 (
Frames that are destined for the current memory module are copied into FIFO 58 and processed by repairing AMB 100. For example, for a write frame, the data from FIFO 58 is written to DRAM chips 22 on the memory module by repairing AMB 100. For a read, the data read from DRAM chips 22 is stored in FIFO 58. Repairing AMB 100 forms a frame and sends the frame to northbound re-timing and re-synchronizing buffers 64 and out over the northbound lanes from differential output buffer 62. Input buffers 66 and output buffers 64 contain differential receivers and transmitters for the northbound lanes that are buffered by re-timing and re-synchronizing buffers 64.
Memory chips passing the reduced pre-screen test are assembled onto substrates of memory modules, step 302. A repairing AMB chip is also soldered onto the memory module substrate, and other components such as capacitors may be added. The memory modules are tested, step 304, to detect any defects. When a memory module passes the test at step 304 with no defects, it can be sorted and sold as a good memory module, step 308.
If a defect is found at steps 304, 306, then the module has failed. The locations of the bad memory locations are determined, step 310. This may require further testing, perhaps on a different test machine than the tester for step 304. The number of defects is compared to a limit, step 312. When there are more defects than the limit, the memory module is sorted as a bad module and reworked or discarded, step 320.
When the number of defects is below the limit, step 312, then the locations of the defects are written to SPD-EEPROM 130 on the module as the repair addresses, step 314. Since SPD-EEPROM 130 is non-volatile, these repair addresses are not lost when power is removed from the memory module.
A repair flag may be set in SPD-EEPROM 130, step 316. The repair flag may be read during power-up initialization of the memory module to determine when the repair controller needs to be activated. The repair addresses are then copied to repair address buffer 68 in repairing AMB 100 when the repair flag is in the set state.
The repaired memory modules are sorted and can be sold as repaired memory modules, step 318. Further testing of the repaired memory module may also be performed. Since the repaired memory modules and the good memory modules from step 308 have the same chips, and operate the same from the motherboard's viewpoint, both kinds of memory modules could be sold.
The limit N of allowable defects (step 312) may be set to a low value such as 1, 2, or 8 when few defects are likely. Memory modules with more defects are likely to have serious problems and may be better to be reworked or discarded. Rather than have a numerical limit to the number of defects, the locations of the defects may be considered. For example, a direct-mapped repair address buffer 68 may only allow one defect with a particular index address. Any other defects with the same index address cannot be repaired. Thus while several defects may be allowed, no two defects can map to the same index. Other limitations on defect locations could correspond to shortcuts in decoding made by the repair controller.
Repair address buffer 68 can be arranged as a cache. The incoming address of a memory request extracted from a serial packet received on the southbound lanes can be split into an index portion and a tag portion. For example, the least-significant-bits (LSBs) can be the index and the most-significant-bits (MSBs) be the tag. The index portion of the address selects one of the entries in repair address buffer 68 as selected entry 70. Address tag 72 stored in selected entry 70 is read out and compared to the tag portion of the incoming address by comparator 63 to determine if a match occurred. A valid bit (not shown) may also be stored with selected entry 70 and used to validate the match.
When comparator 63 finds a valid match, a repair hit is signaled. Mux 65 selects repair data 74 stored with corresponding address tag 72 in selected entry 70, rather than data read from the defective memory chip by DRAM controller 50. When no repair hit occurs, data is transferred from the memory chip by DRAM controller 50.
Repair address buffer 68 could be arranged as a direct-mapped cache with only one entry per index address, or as a set-associative cache with 2, 4, or more entries per index address. If the number of defects is expected to be small, then a direct-mapped cache is acceptable. If two or more defect addresses map to the same index, then the memory module could be re-worked or discarded as an un-repairable module.
With memory module 10 inserted into memory module socket 26, memory controller 38 on PC motherboard 28 receives data from a CPU or bus master and generates control signals that pass through memory module socket 26 to memory module 10. Clock generator 35 generates a clock that is also passed through as a clock to repairing AMB 100 on memory module 10.
Repair addresses are stored in SPD-EEPROM 130 during manufacture. These repair addresses are copied to repairing AMB 100 so that repair controller 69 can compare addresses from the southbound lanes to the local copy of the repair addresses. Thus one or more of DRAM chips 22 may have a defective memory location and still be repairable using repair controller 69 in repairing AMB 100.
Memory controller 38 on motherboard 28 does not have to support repair re-mapping. Indeed, memory controller 38 is not aware of repair remapping by repair controller 69, since repair remapping is transparent to motherboard 28.
BIST controller 30 on motherboard 28 activates test modes of repairing AMB 100 using the SM bus. Errors detected by the internal BIST circuitry in repairing AMB 100 can be passed through to BIST controller 30 to signal an error to an operating system or boot routine running on motherboard 28.
SPD EEPROM 130 stores configuration and repair-address information about memory module 10 that is sent over serial data line SPD_D synchronized to SPD clock SPD_CLK. Address inputs to SPD EEPROM 130 are carried from motherboard 28 on address lines SPD_A[2:0], which may be hard wired on motherboard 28. The wiring configuration of SPD_A[2:0] on motherboard 28 determines the device address (memory-module slot number) of memory module 10. Data sent over serial data line SPD_D is a series of frames consisting of device address, device type (repairing AMB 100 or SPD EEPROM 130), register location, and register data. Test mode is activated on repairing AMB 100 by writing to the AMB test-mode control registers. Repairing AMB 100 and SPD EEPROM 130 can share clock, address, and serial data lines, but respond to different device types at the same device address.
Repair controller 69 is within repairing AMB 100. Repair addresses stored in SPD-EEPROM 130 can be copied to a repair address buffer in repairing AMB 100 as serial data sent over serial data line SPD_D synchronized to serial clock SPD_CLK. SM-bus interface 134 in repairing AMB 100 can drive the device address of SPD-EEPROM 130 onto serial address lines SPD_A[2:0] to read SPD-EEPROM 130 over serial data lines SPD_D. SM-bus interface 134 on repairing AMB 100 could generate the serial clock, or a free-running serial clock generated on motherboard 28 could be used.
Northbound lane inputs NB_IN[13:0], #NB_IN[13:0] to repairing AMB 100 are connected to northbound lane motherboard outputs 86, NB_OUT[13:0], #NB_OUT[13:0] on motherboard 28. These 14 northbound lanes carry frames generated by downstream memory modules that are being sent to the processor, perhaps through upstream memory modules (not shown).
Northbound lane outputs NB_OUT[13:0], #NB_OUT[13:0] from repairing AMB 100 are connected to northbound lane motherboard inputs 84 (NB_IN[13:0], #NB_IN[13:0]) on motherboard 28. These 14 northbound lanes carry frames generated by memory module 10 or generated by downstream memory modules that are being sent to the processor. Northbound lane inputs NB_IN[13:0], #NB_IN[13:0] on motherboard 28 could connect to the memory controller and to the processor directly, or could connect to an upstream memory module (not shown).
Southbound lane inputs SB_IN[9:0], #SB_IN[9:0] to repairing AMB 100 are connected to southbound lane motherboard outputs 87, SB_OUT[9:0], #SB_OUT[9:0] on motherboard 28. These 10 southbound lanes carry frames generated by the processor that are being sent to memory module 10 or to downstream memory modules in the daisy chain.
Southbound lane outputs SB_OUT[9:0], #SB_OUT[9:0] from repairing AMB 100 are connected to southbound lane motherboard inputs 85 (SB_IN[9:0], #SB_IN[9:0]) on motherboard 28. These 10 southbound lanes carry frames generated by the processor that are being sent to downstream memory modules. Southbound lane outputs SB_OUT[9:0], #SB_OUT[9:0] on motherboard 28 could be driven by the memory controller directly, or could connect to an upstream memory module (not shown).
Several other embodiments are contemplated by the inventors. For example the various functions may be partitioned into a variety of kinds and numbers of blocks. Functions may be implements in hardware, software, firmware, or various combinations. For example, basic functions such as address comparisons may be implemented in hardware logic gates, while more complex functions such as error handling may be assisted by execution of program instructions.
SPD-EEPROM 130 could be integrated into repairing AMB 100. Memory buffer 506 could be a memory external to repairing AMB 100 rather than be integrated with repairing AMB 100. Repair address buffer 68 could be non-volatile memory on repairing AMB 100 and could be directly programmed once, eliminating the need to transfer repair addresses from SPD-EEPROM 130 at each initialization. Memory buffer 506 could be an extension of repair address buffer 68, or could be part of a larger on-chip memory that includes FIFO 58.
Memory for the repair address in repair address buffer 68 may be flip-flops, registers, latches, SRAM, DRAM, non-volatile memory, or other kinds of memory. Likewise, spare repair memory that stores the replacement data may be flip-flops, registers, latches, SRAM, DRAM, non-volatile memory, or other kinds of memory. The repair addresses and/or the repair data may be internal to repairing AMB 100 or may be external to repairing AMB 100.
BIST controller 30 could be BIOS codes that are tightly linked to the operating system. It could also be an application program which is run during system maintenance. Other arrangements of repair address buffer 68 may be used, such as a linked list, a tree lookup structure, or a simple list of repair addresses. A few bits of the address could be compared, and if a match occurred, then more bits are compared, and delay time added. Many optimizations are possible.
The number of northbound and southbound lanes may vary. Different control signals may be used. Traces may be formed from metal traces on surfaces of the memory module, or on interior traces on interior layers of a multi-layer PCB. Vias, wire jumpers, or other connections may form part of the electrical path. Resistors, capacitors, or more complex filters and other components could be added. For example, power-to-ground bypass capacitors could be added to the memory module.
Signals may be half swing with source termination (output buffer) and load termination (input buffer). A series resistor or a shunt resistor in the path attenuates the signal. Shunt resistance may be around 500 ohms with a line impedance of 50 ohms.
Muxes and switches could be added to allow for loop-back testing as well as standard operation. Future memory module standards and extensions of the fully-buffered DIMM standard could benefit from the invention.
Any advantages and benefits described may not apply to all embodiments of the invention. When the word “means” is recited in a claim element, Applicant intends for the claim element to fall under 35 USC Sect. 112, paragraph 6. Often a label of one or more words precedes the word “means”. The word or words preceding the word “means” is a label intended to ease referencing of claim elements and is not intended to convey a structural limitation. Such means-plus-function claims are intended to cover not only the structures described herein for performing the function and their structural equivalents, but also equivalent structures. For example, although a nail and a screw have different structures, they are equivalent structures since they both perform the function of fastening. Claims that do not use the word “means” are not intended to fall under 35 USC Sect. 112, paragraph 6. Signals are typically electronic signals, but may be optical signals such as can be carried over a fiber optic line.
The foregoing description of the embodiments of the invention has been presented for the purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise form disclosed. Many modifications and variations are possible in light of the above teaching. It is intended that the scope of the invention be limited not by this detailed description, but rather by the claims appended hereto.
