Patent Application
20150206569
The present disclosure relates generally to computing architectures, and in particular, to a system and a method of managing memory.
Spin transfer torque magnetoresistive random-access memory (STT-MRAM) is a non-volatile memory technology. The probability of writing a correct value in an STT-MRAM generally increases with the duration of the write pulse, or write pulse width. Put another way, the more time that is taken to write a value into the memory cell, the more likely that the written value will be correct. There is an upper limit of the write pulse width, beyond which there is either no additional benefit in terms of write reliability. Increased reliability via write pulse width increase is generally obtained at the cost of increased write latency and energy consumption.
In a particular embodiment, a method of managing memory includes determining a temperature associated with the memory and determining a level of utilization associated with the write queue associated with the memory. A write operation may be performed based on the level of said utilization and the temperature. Where advantageous, the overall write apparatus and method may include a false write operation.
The level of memory write queue utilization may be determined based on a high water mark indicator. For instance, the high water mark indicator may be compared to a utilization threshold value. The high water mark indicator may be incremented in response to a received write request and decremented in response to a write request being written into the actual memory data array.
The level of write queue utilization may be determined to be above or below a programmable utilization threshold value, and the temperature may be compared to a latency determining threshold value. The temperature may be continuously monitored. A multiplexer may be configured to receive a first input from a high water mark indicator and a second input from a false write queue. The programming pulse width may be adjusted as part of the write operation. The write operation may further include a bulk write operation that services multiple write requests in a single burst of write cycles.
According to another particular embodiment, an apparatus includes a memory storing data and program code. A temperature sensor may be configured to determine a temperature associated with the memory. A controller may be in communication with the memory and the temperature sensor. The controller may be configured to receive the temperature and to execute the program code to determine a level of queue utilization associated with the memory, as well as to perform a write operation based on the level of said utilization and the temperature.
A false write queue may be configured to generate a false write request during the write operation. A false write operation in may be halted in response to at least one of: a detection of a desired temperature, an expiration of a time period, and a queue full indication. The controller may further be configured to determine a level of write queue utilization based on a high water mark indicator. The controller may adjust a programming pulse width as part of the write operation. A multiplexer may be configured to receive a first input from a high water mark indicator and a second input from a false write queue.
According to another particular embodiment, a computer readable storage medium includes instructions, that when executed by a processor, cause the processor to determine a temperature associated with the memory, to determine a level of queue utilization associated with the memory, and to determine a write operation based on the level of said utilization and the temperature.
An embodiment may manage write operations in an STT-MRAM in a manner that reduces power (or energy) consumption and processing latency. For example, variably adjusting a length of a programming pulse width may reduce power consumption by using shorter lengths when higher temperatures are present. Processes may be transparent to a main memory controller, e.g., only the microchip may be aware of a changing latency. False writes may be used to increase the temperature in the STT-MRAM. The increased temperature may result in decreased write latency.
Features and other benefits that characterize embodiments are set forth in the claims annexed hereto and forming a further part hereof. However, for a better understanding of the embodiments, and of the advantages and objectives attained through their use, reference should be made to the Drawings and to the accompanying descriptive matter.
Programming pulse widths may be continuously adjusted based on sensed temperatures in an STT-MRAM. One or more temperature sensors on the STT-MRAM microchip may be continuously monitored to adjust the programming pulse width. Write latency may be improved by using shorter programming pulse widths in response to higher temperatures. Longer programming pulse widths may be used at lower temperatures according to the adaptive scheme. An embodiment capitalizes on the fact that the write latency required to achieve a targeted level of reliability is inversely related to the operating temperature of the memory data array and associated logic.
An STT-MRAM microchip with temperature sensors may enable the automatic adjustment of microchip-wide programming pulse width writes. A cache system of the STT-MRAM may cluster write operations to a physical section of the cache. The clustering may increase a probability that shorter writes are performed (e.g., Most Recently Written, Most Frequently Written, or Most Recently Used sets of a cache may all be physically adjacent locations).
False writes, or dummy writes, may be introduced within or close to the STT-MRAM with the purpose of intentionally increasing temperature. Three-dimensional integration of the writes may be performed on a separate layer (e.g., STT-MRAM could be placed directly above processor the cores).
As the temperature in an STT-MRAM increases, writes to the STT-MRAM may be performed in less time. Additionally, writes in an STT-MRAM have a heating effect. Reads in an STT-MRAM, incidentally, are accomplished at a relatively low voltage and do not have a heating effect. Because increasing the temperature may negatively affect system retention and reliability, the number of false write requests issued may be regulated in terms of a detected temperature, an expired time limit, or a detected memory capacity.
Local temperature sensors of an STT-MRAM may enable an automatic adjustment of write pulse length, or programming pulse length. As such, a length of a programming pulse width may be determined based on the sensed temperature. For example, shorter programming pulse widths may be used in connection with higher temperatures, and longer programming pulse widths may be used in the presence of lower temperatures.
Embodiments of methods have application in page-based designs. For instance, where slow writes have been performed to multiple Column Address Strobe (CAS) latency locations in a page, subsequent writes to that page may be performed with fast writes, since prior activity may have warmed the page.
Write latency may be improved with dynamic temperature adjustment in an STT-MRAM. Embodiments of methods exploit the natural physical phenomena that as temperature increases, writes to STT-MRAM become faster. Benefits from higher temperatures (e.g., associated with higher voltages) may be realized at a cost of reliability. Retention time in the STT-MRAM may be reduced with higher temperatures. The reduced retention time may translate into diminished reliability.
Write performance in an STT-MRAM may depend on a length of a programming pulse width, or the length of the pulse used to write the data. For example, longer programming pulse widths may result in higher power consumption and greater accuracy. The amount of time elapsing during the write operation to the STT-MRAM is referred to as write pull switch, write pull length, or write latency.
Embodiments may include adaptive programming pulse widths to write to memory according to stored relationships. The relationships may associate one or more of a bit error rate, a temperature, a voltage, historical predictive data, and coding. Coding may include additional error handling and redundancy processes (e.g., parity coding, error correction code (ECC), and memory mirroring) used to increase reliability given a particular programming pulse width and varying parameters.
Methods may vary write latency and programming pulse width capability. For example, an internal write latency of a microchip may defer, while an external write latency remains constant. A microchip may include an internal thermal sensor. Where the microchip determines a high temperature condition is present, a short programming pulse width may be adequate and consequently used. Where the microchip is alternatively at a low temperature, a long programming pulse width may be used. Determining whether to use a long or short programming pulse width based on a detected microchip temperature may preserve power because shorter write latencies (which use relatively less power) are used when higher temperatures are present. Processes may be transparent to a main memory controller, e.g., only the microchip may be aware of a changing latency.
Embodiments of methods exploit the natural physical phenomena that as temperature increases, writes to STT-MRAM become faster. Benefits from higher temperatures (e.g., associated with higher voltages) may be realized at a cost of reliability. Retention time in the STT-MRAM may be reduced with higher temperatures. The reduced retention time may translate into diminished reliability. Known relationships may be leveraged to provide a desired level of reliability.
Turning more particularly to the drawings,
The computer 110 generally includes one or more physical processors 111, 112, 113 coupled to a memory subsystem including a main storage 116. The main storage 116 may include one or more STT-MRAMs. Another or the same embodiment may contain a main storage having a static random access memory (SRAM), a dual in-line memory module (DIMM) a flash memory, a hard disk drive, and/or another digital storage medium. The processors 111, 112, 113 may be multithreaded and/or may have multiple cores. A cache subsystem 114 is illustrated as interposed between the processors 111, 112, 113 and the main storage 116. The cache subsystem 114 typically includes one or more levels of data, instruction and/or combination caches, with certain caches either serving individual processors or multiple processors.
The main storage 116 may be coupled to a number of external input/output (I/O) devices via a system bus 118 and a plurality of interface devices, e.g., an I/O bus attachment interface 120, a workstation controller 122, and/or a storage controller 124 that respectively provide external access to one or more external networks 126, one or more workstations 128, and/or one or more storage devices 130, such as a direct access storage device (DASD). In a particular embodiment, the storage device 130 may include an STT-MRAM. The system bus 118 may also be coupled to a user input (not shown) operable by a user of the computer 110 to enter data (i.e., the user input sources may include a mouse, a keyboard, etc.) and a display (not shown) operable to display data from the computer 110 (i.e., the display may be a CRT monitor, an LCD display panel, etc.). The computer 110 may also be configured as a member of a distributed computing environment and communicate with other members of that distributed computing environment through a network 126.
A processor microchip 202 of the system 200 may continuously send memory requests to the STT-MRAM 208. As shown in
The STT-MRAM 208 may include multiple memory banks 224, 226, 228, 230, 232, 236. Each memory bank 224, 226, 228, 230, 232, 236 may be associated with a respective one of a number of queues 262. The memory requests from the processor microchip 202 may be addressed for a specific memory bank 224, 226, 228, 230, 232, 236 and an associated one of the queues 262. A steering module 210 of the system 200 may direct the memory requests to the appropriate queue 262 according to an address of the destination memory bank 224, 226, 228, 230, 232, 236.
A high water mark module 264 may facilitate management of incoming memory requests by accounting and adjusting for the incoming and outgoing traffic. For example, the high water mark module 264 may indicate the fullness (e.g., of waiting write requests) of a particular queue 262. The high water mark module 264 may be incremented in response to an incoming memory request and may be decremented whenever a memory request at the head of the queue 262 is issued to an appropriate memory bank 224, 226, 228, 230, 232, 236. When a queue 262 associated with a memory bank 224 becomes full, it may no longer receive new memory requests.
The memory banks 224, 226, 228, 230, 232, 236 may include one or more temperature sensors 244, 246, 248, 250, 252, 256. The temperature sensors 244, 246, 248, 250, 252, 256 may individually or collectively detect an average temperature of a memory bank or array of memory banks 224, 226, 228, 230, 232, 236. One skilled in the art will appreciate that the array may include any number of memory banks, including a number that is a power of two.
A pulse width determiner 258 may adjust the programming pulse width for writing to the specific memory bank or the multiple memory banks 224, 226, 228, 230, 232, 236 based on the detected temperature 216, as well as on other inputs. Illustrative such inputs may include one or more of a desired coding strength 218, an application requirement 220, and a history based predictive performance parameter 220 associated with past performance of the system.
Relationship information 212 may be stored in a manner that is accessible to the pulse width determiner 258. In a particular embodiment, the relationship information 212 includes bit error rates for the STT-MRAM 208 stored versus one or more of: programming pulse width, temperature, coding, and voltage. Voltage in some cases may vary proportionally with temperature. Incidentally, comparable relationships do not exist for dynamic random-access memory (DRAM). While such relationship information 212 is shown in
The programming pulse width determined by the pulse width determiner 258 may be provided as an input to a write enable module 260. Another input to the write enable module 260 may be provided by the high water mark module 264. Based on these inputs, the write enable module 260 may write the memory request sitting at the head of the queue 262 to the designated bank 232 with the determined programming pulse width. The write process, including the programming pulse width, may be continuously adjusted in this manner based on the temperature and other inputs from the stored relationship information 212.
Similar to the system 200 of
The multiplexer 368 may select one or more inputs to write to a memory bank 324, 326, 328, 330, 332, 336. A first input at the multiplexer 368 may be received from bank queue 362. The memory bank queue 362 may temporarily store received memory requests. Another input received at the multiplexer 368 may originate from the false write queue 366. The input (i.e., a false request) may be a non-functional write request used to increase the temperature in the STT-MRAM 308. The increased temperature may result in decreased write latency.
In one example, false requests may be sent when the arrival rate of legitimate, real write requests is relatively low. For instance, the high water mark module 364 may indicate that there is not a backlog of write requests in the memory bank queue 362. For example, a threshold number 370 of write requests may be present in the memory bank queue 362. The false requests may thus decrease latency while not significantly disrupting write operations. The false request may continue to be sent (and the latency may be reduced) until the high water mark module 364 indicates an increased number of write requests in the bank queue 362, until the expiration of a preset period, or until a desired temperature (e.g., stored as a threshold value 370) is detected.
In one example, while waiting for regular requests, a sitting module of the system may rewrite values that have already been sent and stored to a memory bank 324, 326, 328, 330, 332, 336. No extra storage space may be used and baseline performance may not be significantly affected because there are not that many real write requests waiting in the memory bank queue 362. Where the high water mark module 364 indicates an increased number of write requests in the memory bank queue 362, where a preset period has expired, or where the desired temperature is detected, the system 300 may use the multiplexer 368 to select the bank queue 362 as the source for populating a memory bank 324, 326, 328, 330, 332, 336. Additionally, the extra waiting time resultant from a sequence of non-functional write requests may facilitate writing numerous real write requests in a burst, due to accumulation of said write requests. As in the system of
At 408, the system may determine if the queue has reached a high water mark. As discussed herein, the high water mark may include a programmable value or threshold that is indicative of the how much of the memory write queue resource is being utilized (e.g., available memory queue capacity). Where the high water mark indicator has not been reached at 408, the system may receive a next memory request at 402.
Alternatively, where the high water mark has not been reached at 408, the system may at 410 monitor the bank memory temperature. The system may determine at 412 whether the detected temperature is lower than a latency determining threshold. Where the temperature has fallen below the latency determining threshold, the system may invoke at 414 a false write process. The false write(s) may increase the temperature, and in so doing, lower the latency in the STT-MRAM. Put another way, a false write request may be initiated if the temperature is too low. The false write(s) may be issued before an actual write request goes through because the temperature is too low before the false write request to gauge the regular write process (e.g., at the desired temperature and latency). The system may return to monitoring the memory bank temperature at 410.
Where the temperature is determined at 412 to be higher than the latency determining threshold, the programming pulse width associated with the temperature may be determined at 416. A few actual write requests that accumulate as the temperature is raised may be delivered under desired temperature and latency conditions in bulk at 418. Bulk writes have the added advantage of further increasing bank temperature without recourse to false writes.
Particular embodiments described herein may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment containing both hardware and software elements. In a particular embodiment, the disclosed methods are implemented in software that is embedded in processor readable storage medium and executed by a processor, which includes but is not limited to firmware, resident software, microcode, etc.
Further, embodiments of the present disclosure, such as the one or more embodiments may take the form of a computer program product accessible from a computer-usable or computer-readable storage medium providing program code for use by or in connection with a computer or any instruction execution system. For the purposes of this description, a non-transitory computer-usable or computer-readable storage medium may be any apparatus that may tangibly embody a computer program and that may contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
In various embodiments, the medium may include an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system (or apparatus or device) or a propagation medium. Examples of a computer-readable storage medium include a semiconductor or solid state memory, magnetic tape, a removable computer diskette, a random access memory (RAM), a read-only memory (ROM), a rigid magnetic disk and an optical disk. Current examples of optical disks include compact disk—read only memory (CD-ROM), compact disk—read/write (CD-R/W) and digital versatile disk (DVD).
A data processing system suitable for storing and/or executing program code may include at least one processor coupled directly or indirectly to memory elements through a system bus. The memory elements may include local memory employed during actual execution of the program code, bulk storage, and cache memories which provide temporary storage of at least some program code in order to reduce the number of times code must be retrieved from bulk storage during execution.
Input/output or I/O devices (including but not limited to keyboards, displays, pointing devices, etc.) may be coupled to the data processing system either directly or through intervening I/O controllers. Network adapters may also be coupled to the data processing system to enable the data processing system to become coupled to other data processing systems or remote printers or storage devices through intervening private or public networks. Modems, cable modems, and Ethernet cards are just a few of the currently available types of network adapters.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the disclosed embodiments. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the scope of the disclosure. Thus, the present disclosure is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope possible consistent with the principles and features as defined by the following claims.
This application is a continuation application and claims priority from U.S. patent application Ser. No. 14/159,798, entitled “Dynamic Temperature Adjustments In Spin Transfer Torque Magnetoresistive Random-Access Memory (STT-MRAM),” by the same inventors, filed on Jan. 21, 2014, and which is incorporated herein in its entirety.
| Number | Date | Country | |
|---|---|---|---|
| Parent | 14159178 | Jan 2014 | US |
| Child | 14320219 | US |
