How to Compensate for Voltage Temperature (VT) Drift in Memory Interfaces
The data processing system addresses VT drift in DRAM by dynamically adjusting timing parameters, reducing the need for retraining and improving performance and user experience.
Patent Information
- Application Number
- JP2024526595
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2022-06-30
- Filing Date
- 2022-10-27
- Publication Date
- 2024-11-19
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Modern dynamic random-access memory (DRAM) systems experience voltage-temperature (VT) drift, necessitating periodic retraining that halts system operations and impairs performance.
A data processing system with a compensation circuit that adjusts timing parameters based on measured temperature and voltage changes, reducing the need for periodic retraining by incorporating a calibration controller and compensation circuit to dynamically update delay values.
Reduces the frequency and eliminates the need for high-speed link phase retraining, enhancing system performance and user experience by maintaining accurate data transmission without halting operations.
Smart Images

Figure 2024543033000001_ABST
Abstract
Description
[Background technology]
[0001] Modern dynamic random-access memory (DRAM) provides high memory bandwidth by increasing the speed of data transmission on a bus connecting the DRAM and one or more data processors, such as a graphics processing unit (GPU) or a central processing unit (CPU). In one example, graphics double data rate (GDDR) memory has pushed the boundaries of data transmission rates to accommodate the high bandwidth required for graphics applications. To ensure accurate reception of data, modern GDDR memories have required extensive training prior to operation to ensure that the receiving circuitry can accurately capture the data. However, over time, GDDR data transmission systems experience voltage temperature (VT) drift that changes the optimal point of delay, necessitating periodic retraining, which requires the system to be stopped while the retraining is performed. [Brief description of the drawings]
[0002] [Figure 1] 1 is a block diagram of a data processing system that compensates for VT drift in accordance with some embodiments. [Diagram 2] 2 is a block diagram of a GDDR PHY-DRAM link of the data processing system of FIG. 1 in accordance with some embodiments. [Diagram 3] 3 is a block diagram of an annotated GDDR PHY-DRAM link corresponding to the GDDR PHY-DRAM link of FIG. 2. [Figure 4] 3 is a timing diagram useful in understanding the operation of the data processing system of FIG. 2; [Diagram 5] 3 is another timing diagram useful in understanding the operation of the data processing system of FIG. 2; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0003] In the following description, the use of the same reference numbers in different figures indicates similar or identical items. Unless otherwise noted, the word "coupled" and its related verb forms include both direct and indirect electrical connections by means well known in the art, and unless otherwise noted, any description of a direct connection also refers to alternative embodiments using a suitable form of indirect electrical connection.
[0004] The data processing system includes a data processor coupled to a memory, the data processor including a reference clock generation circuit for providing a reference clock signal, a first delay circuit for delaying the reference clock signal by a first amount to provide command and address signals, a second delay circuit for delaying the reference clock signal by a second amount to provide a read data signal, a calibration circuit for determining current values of the first and second quantities, and a compensation circuit for calculating drift in the first and second quantities based on a measured temperature change, at least one voltage sensitivity coefficient and at least one temperature sensitivity coefficient, and updating the first and second quantities according to the drift.
[0005] A data processor adapted to be coupled to a memory includes a reference clock generation circuit, a first delay circuit, a second delay circuit, a calibration circuit, and a compensation circuit. The reference clock generation circuit provides a reference clock signal. The first delay circuit delays the reference clock signal by a first amount to provide the command and address signals. The second delay circuit delays the reference clock signal by a second amount to provide the read data signal. The calibration circuit determines current values of the first and second quantities. The compensation circuit calculates drift in the first and second quantities based on the measured temperature change, at least one voltage sensitivity coefficient, and at least one temperature sensitivity coefficient, and updates the first and second quantities according to the drift.
[0006] A method for updating timing values for a data processor to access a memory to compensate for voltage temperature (VT) drift during operation without retraining the link includes generating a reference clock signal. The reference clock signal is delayed a first amount using a first delay circuit to provide command and address signals. The reference clock signal is delayed a second amount using a second delay circuit to provide a read data signal. Current values of the first amount and the second amount are determined using a calibration circuit. Drift in the first amount and the second amount is calculated based on a measured temperature change, at least one voltage sensitivity coefficient and at least one temperature sensitivity coefficient using a compensation circuit.
[0007] 1 is a block diagram of a data processing system 100 that compensates for VT drift, according to some embodiments. Data processing system 100 generally includes data processors in the form of a graphics processing unit (GPU) 110, a host central processing unit (CPU) 120, a double data rate (DDR) memory 130, and a graphics DDR (GDDR) memory 140.
[0008] GPU 110 is a discrete graphics processor that has very high performance for optimized graphics processing, rendering and display, but requires high memory bandwidth to perform these tasks. GPU 110 generally includes a set of command processors 111, a graphics single instruction, multiple data (SIMD) core 112, a set of caches 113, a memory controller 114, a DDR physical interface circuit (DDR PHY) 117, and a GDDR PHY 118.
[0009] Command processor 111 is used to interpret high-level graphics instructions such as those specified in the OpenGL programming language. Command processor 111 has a bidirectional connection to memory controller 114 for receiving high-level graphics instructions such as OpenGL instructions, a bidirectional connection to cache 113, and a bidirectional connection to graphics SIMD core 112. In response to receiving the high-level instructions, command processor issues low-level instructions for rendering, geometric processing, shading, and rasterization of data such as frame data, using cache 113 as temporary storage. In response to the graphics instructions, graphics SIMD core 112 executes the low-level instructions on large data sets in a massively parallel manner. Command processor 111 and cache 113 are used for temporary storage of input data and output (e.g., rendered and rasterized) data. Cache 113 also has a bidirectional connection to graphics SIMD core 112 and a bidirectional connection to memory controller 114.
[0010] The memory controller 114 has a first upstream port connected to the command processor 111, a second upstream port connected to the cache 113, a first downstream bidirectional port to the DDR PHY 117, and a second downstream bidirectional port to the GDDR PHY 118. As used herein, the "upstream" port is the side of the circuitry towards the data processor and away from the memory, and the "downstream" port is in the direction away from the data processor and towards the memory. The memory controller 114 controls the timing and ordering of data transfers to and from the DDR memory 130 and the GDDR memory 140. DDR and GDDR memories have asymmetric access, i.e., access to open pages in the memory is faster than access to closed pages. The memory controller 114 stores memory access commands and processes them out of order for efficiency, for example, by prioritizing access to open pages while adhering to certain quality of service targets.
[0011] The DDR PHY 117 has an upstream port connected to a first downstream port of the memory controller 114 and a downstream port bidirectionally connected to the DDR memory 130. The DDR PHY 117 meets all specified timing parameters of a version of the DDR memory 130, such as DDR version 5 (DDR5), and performs timing calibration operations at the direction of the memory controller 114. Similarly, the GDDR PHY 118 has an upstream port connected to a second downstream port of the memory controller 114 and a downstream port bidirectionally connected to the GDDR memory 140. The GDDR PHY 118 meets all specified timing parameters of a version of the GDDR memory 140, and performs timing calibration operations at the direction of the memory controller 114.
[0012] Interface timing to DDR memory 130 and GDDR memory 140 is susceptible to VT drift. Known techniques for compensating for VT drift center around periodic retraining of the links. However, retraining stops all activity in the system while the retraining is taking place, which can impair performance and cause jumps and stalls in graphics workloads, impairing the user experience.
[0013] To overcome the burden of periodic retraining, the present inventors have developed various methods to reduce system link sensitivity to VT-induced phase drift. The disclosed VT drift compensation methods reduce and in some cases eliminate the need for periodic high-speed link phase retraining. In an exemplary embodiment, the present technology is applied to a GDDR memory interface, but is not limited to only GDDR memory or only memory interfaces.
[0014] As shown in FIG. 1, the memory controller 114 includes a calibration controller 115 for performing basic link calibration and a compensation circuit 116 for compensating for VT drift without the need for frequent retraining, thus improving system performance and improving user experience.
[0015] The calibration controller 115 is a circuit that controls the calibration of timing parameters of the DDR PHY 117 and the GDDR PHY 118. At system startup, the link between the DDR PHY 117 and the DDR memory 130 requires training, and the link between the GDDR PHY 118 and the GDDR memory 140 is trained. Training generally involves determining the values of the reference voltages used by the memory and PHY to capture the input data, the timing relationship between the command clock and the data clock, and the timing relationship between the data and the clock at the transmitter so that the data can be reliably captured at the receiver. Techniques for performing these calibrations are well known and vary based on the version of DDR and GDDR. Additionally, a de facto industry standard for the interface between a memory controller and a memory PHY, known as the "DFI" standard, has been developed to specify the signaling and characteristics of the interface between the memory controller and the PHY. One of the features of the latest version of the DFI standard is the definition of certain low-level training features, where the entire calibration flow is directed by the memory controller, while most of the calibration functions are performed automatically by the PHY.
[0016] According to various embodiments disclosed herein, compensation circuit 116 leverages these capabilities of PHY circuits such as GDDR PHY 118 to adjust for VT drift without the need to perform recalibration operations using calibration controller 115 and GDDR PHY 118. Compensation circuit 116 calculates drift in timing parameters used to control delays in GDDR PHY 118. In one particular embodiment, compensation circuit 116 calculates the drift based on measured temperature changes, at least one voltage sensitivity coefficient and at least one temperature sensitivity coefficient, and compensates for timing changes based on these parameters by updating the delays of GDDR PHY 118.
[0017] The GDDR memory 140 includes a set of mode registers 141 and a temperature sensor 142. The mode registers 141 provide a programming interface for controlling the operation of the GDDR memory 140 in a data processing system. As described further below, the mode registers 141 store at least one voltage sensitivity coefficient and at least one temperature sensitivity coefficient used in VT drift compensation. The GDDR memory 140 also includes a temperature sensor for measuring the temperature of the GDDR memory 140. In one form, the temperature sensor 142 provides temperature data to the compensation circuit 116 in the GPU 110 during an update operation that ensures that the compensation circuit 116 periodically receives updated temperature information.
[0018] The inventors have discovered that certain calibrated timing parameters are adjusted based solely on measured temperature and voltage differences without the need for performance-impacting recalibration during normal operation. Accordingly, the present disclosure describes various methods for reducing system link sensitivity to VT-induced phase drift. The disclosed VT drift compensation methods reduce and in some cases eliminate the need for periodic high-speed link phase retraining. The present disclosure is presented in the context of graphics DDR memory interfaces, but is not limited to only GDDR memory or only memory interfaces.
[0019] In some GDDR, version 6 (GDDR6) physical layer interface (PHY) systems, the voltage temperature (VT) drift of a parameter known as "WCK2DQI" VT drift direction and magnitude was successfully inferred by monitoring the VT phase drift of the error detection and correction (EDC) lane (WCK2DQO) relative to the PHY reference clock. As used herein, WSK2DQI means the write clock (WCK) to data input delay and WCK2DQO means the WCK to data output delay. The PHY reference clock was a branched clock source shared with the error detection and correction (EDC) lane. This basic relationship can be expressed as shown in Equation [1]. WCK2DQI_drift=WCK2DQO_drift * Alpha [1] where α is a scaling factor derived from hardware estimation.
[0020] Many products in use today leverage this WCK2DQI drift correlation to WCK2DQO phase drift, but this is not a perfect solution, does not work for all DRAM vendors and applications, and the method has some limitations or drawbacks. The inventors herein propose a method to better leverage drift tracking to reduce or eliminate the overhead of periodic training for high-speed link interfaces, including parameters in GDDR interfaces.
[0021] The simple model of temperature drift correlation expressed in equation [1] has two main limitations. First, equation [1] assumes that there is little or no process variation between DRAM devices. Second, equation [1] assumes that WCK2DQO VT drift scales symmetrically to WCK2DQI for both voltage and temperature sensitivity. In other words, the α scalar must be equivalent for both temperature and voltage, i.e., as expressed in equation (2). α_temp=α_volt=WCK2DQI_drift / WCK2DQO_drift [2]
[0022] The inventors have found that in practice, some DRAM devices do not have a symmetric correlation between WCK2DQO drift and WCK2DQI VT drift. As an example, Table I shows the VT drift coefficient of the write clock to DQ of one such DRAM device.
[0023] [Table 1]
[0024] In the table, ps stands for time in picoseconds, V stands for voltage in volts, and T C V represents temperature in degrees Celsius. DD represents the typical internal supply voltage of the memory at the worst-case processing corner, and V DDQ represents the typical input / output power supply voltage of the memory at the worst-case processing corner, and T C Note that represents the temperature at the worst case processing corner.
[0025] On the other hand, the measurements vary with the VT drift coefficient of the write clock for the DQ inputs from the same DRAM vendor, as shown in Table II below.
[0026] [Table 2]
[0027] As can be seen from Tables I and II above, equation (2) does not apply to this DRAM vendor. A variation of this particular example is described by the following equation: Formula (1): α_temp=0.7 / 1.1=0.636 α_volt=-30 / -180=0.166 α_avg=(α_temp+α_volt) / 2
[0028] Using this conventional technique to determine WCK2DQI VT drift based on WCK2DQO VT drift results in significant phase tracking error, as defined by equation (4). Phase tracking error = α_error * WCK2DQ_drift [4] where α_error=abs(α_temp±α_volt)*0.5 and WCK2DQ_drift is the total phase drift observed. The 0.5 multiplier used to derive α_error assumes that the asymmetry between the voltage and temperature α coefficients is averaged out.
[0029] Thus, for example, if a drift of 100 picoseconds (ps) is observed from WCK2DQO, this drift will result in a phase tracking error in WCK2DQI of 100ps*0.47 / 2, which results in an error of 23ps. This amount of phase tracking error is significant and limits the accuracy and therefore the usefulness of existing phase tracking techniques based on equation [2]. Furthermore, this amount was calculated without considering process mismatch terms for different DRAMs in the same vendor's product line.
[0030] The inventors of the present disclosure have developed a novel method and apparatus to overcome these above limitations, the causes of which are illustrated in a typical GDDR memory PHY to GDDR memory link and will now be described.
[0031] 2 is a block diagram of a GDDR PHY-DRAM link 200 of the data processing system 100 of FIG. 1 according to some embodiments. The GDDR PHY-DRAM link 200 includes the GPU 110 and a portion of the GDDR memory 140 communicating via a physical interface 260.
[0032] GPU 110 includes a phase locked loop (PLL) 210, a command and address ("C / A") circuit 220, a read clock circuit 230, a data circuit 240, and a write clock circuit 250. These circuits form part of GDDR PHY 118 of GPU 110.
[0033] The phase-locked loop 210 operates as a reference clock generating circuit and generates a clock signal “CK IN ” and an output.
[0034] C / A circuit 220 includes a delay element 221, a selector 222, and a transmit buffer 223 labeled "TX." Delay element 221 has an input connected to the output of PLL 210, an output, and has a variable delay controlled by an input not specifically shown in FIG. 2. The variable delay is determined at start-up by calibration controller 115 and adjusted during operation by compensation circuit 116 in accordance with the techniques described herein. Selector 222 has a first input for receiving a first command / address value, a second input for receiving a second command / address value, and a control input connected to the output of delay element 221. Transmitter 223 has an input connected to the output of selector 222 and an output connected to a corresponding integrated circuit terminal for providing a command / address signal labeled "C / A." C / A circuit 220 includes a set of individual buffers for each signal in the C / A signal group, configured similarly to the exemplary selector 222 and buffer 223 shown in FIG. 2, but it should be noted that only the exemplary C / A circuit 220 is shown.
[0035] Read clock circuit 230 includes a receive buffer 231 labeled "RX" and a selector 232. Receive buffer 231 has an input connected to a corresponding integrated circuit terminal for receiving a signal labeled "RCK", and an output.
[0036] Receive clock selector 232 has a first input connected to the output of PLL 210, a second input connected to the output of receive buffer 231, an output, and a control input for receiving a mode signal not shown in FIG. 2.
[0037] Data circuit 240 includes a receive buffer 241, a latch 242, delay elements 243 and 244, a serializer 245, and a transmit buffer 246. Receive buffer 241 has a first input connected to an integrated circuit terminal that receives a data signal generally labeled "DQ" and a second input connected to an integrated circuit terminal that receives a data signal generally labeled "V REF Latch 242 is a D-type latch having an input labeled "D" connected to the output of receive buffer 241, a clock input, and an output labeled "Q" for providing an output data signal. The interface between GDDR PHY 118 and GDDR memory 140 implements a four-level Pulse Amplitude Modulation data signaling system known as "PAM-4," which encodes two data bits into one of four nominal voltage levels. Thus, receive buffer 241 identifies which of the four levels is indicated by the input voltage and outputs two data bits representing a state accordingly. For example, receive buffer 241 may have V 100, V 200, V 300, V 400 defining four voltage ranges. REFBased on the three slicing levels, three comparators can be used to determine which range the received data signal falls into. Data circuit 240 includes a latch that latches two data bits and is replicated for each bit position. Delay element 243 has an input connected to the output of selector 232 and an output connected to the clock input of latch 242. Delay element 244 has an input connected to the output of PLL 210 and an output. Serializer 245 has inputs for receiving a first data value for a given bit position and a second data value for a given bit position corresponding to successive cycles of the burst, a control input connected to the output of delay element 244, and an output connected to a corresponding DR terminal. Each data byte of the data bus has a set of data circuits such as data circuit 240 for each bit of the byte. This replication allows different data bytes with different routing on the printed circuit board to have different delay values.
[0038] Write clock circuit 250 includes a delay element 251, a selector 252, and a transmit buffer 253. Delay element 251 has an input connected to the output of PLL 210 and an output. Selector 252 has a first input for receiving a first clock state signal, a second input for receiving a second clock voltage, a control input connected to the output of delay element 251, and an output. Transmit buffer 253 has an input connected to the output of selector 252, a first output connected to a corresponding integrated circuit terminal for providing a true write clock signal labeled "WCK_t" and a second output connected to a corresponding integrated circuit terminal for providing a complementary write clock signal labeled "WCK_c."
[0039] The GDDR memory 140 generally includes a write clock receiver 270, a command / address receiver 280, and a data path transceiver 290. The write clock receiver 270 includes a receive buffer 271, a buffer 272, a divider 273, a buffer / tree 274, and a divider 275. The receive buffer 271 has a first input connected to an integrated circuit terminal of the GDDR memory 140 that receives the WCK_t signal, a second input connected to an integrated circuit terminal of the GDDR memory 140 that receives the WCK_c signal, and an output. In the example shown in FIG. 2, the output of the receive buffer 271 is a clock signal having a nominal frequency of 8 GHz. The buffer 272 has an input connected to the output of the receive buffer 271 and an output. The divider 273 has an input connected to the output of the buffer 272 and an output for providing a divided clock having a nominal frequency of 4 GHz. The divider 275 has an input connected to the output of the buffer / tree 274 and an output for providing a clock signal labeled "CK4" having a nominal frequency of 2 GHz.
[0040] The command / address receiver 280 includes a receive buffer 281 and a slicer 282. The receive buffer 281 has a first input connected to the corresponding integrated circuit terminal of the GDDR memory 140 that receives the C / A signal, a second input for receiving V REF and an output. The C / A input signal is received as a normal binary signal having two logic state levels and is considered non-return-to-zero (NRZ) signal encoding. The slicer 282 has a set of two data latches each having a D input connected to the output of the receive buffer 281, a clock input for receiving the corresponding output of the output of the divider 275, and a Q output for providing the corresponding C / A signal.
[0041] The data path transceiver 290 includes a serializer 291, a transmitter 292, a serializer 293, a transmitter 294, a receive buffer 295, and a slicer 296. The serializer 291 has an input for receiving a first read clock level, a second input for receiving a second read clock level, a select input connected to an output of the buffer / tree 274, and an output. The transmitter 292 has an input connected to an output of the serializer 293, and an output connected to an RCK terminal of the GDDR memory 140. The serializer 293 has an input for receiving a first read data value, a second input for receiving a second data value, a select input connected to an output of the buffer / tree 274, and an output. The transmitter 294 has an input connected to an output of the serializer 293, and an output connected to a corresponding DQ terminal of the GDDR memory 140. The receive buffer 295 has a first input connected to a corresponding DQ terminal of the GDDR memory 140, and a second input connected to a V REF Slicer 296 has a set of four data latches, each having a D input connected to the output of receive buffer 295, a clock input connected to the output of buffer / tree 274, and a Q output for providing a corresponding DQ signal.
[0042] The interface 260 includes a set of physical connections routed from the bond pads of the GPU 110 die, through the package impedance to the package terminals, through traces on the printed circuit board to the package terminals of the GDDR memory 140, and back through the package impedance to the bond pads of the GDDR memory 140 die.
[0043] During operation, the data processing system can be used as a graphics card or accelerator due to the high bandwidth graphics processing performed by the graphics SIMD cores 112. The host CPU 120, running an operating system or application programs, sends graphics processing commands to the GPU 110 through the DDR memory 130, which serves as a unified memory for the GPU 110 and the host CPU 120. It may send commands to the GPU interface, for example, using OpenGL commands or through any other host CPU. OpenGL is a cross-language, cross-platform application programming interface for rendering 2D and 3D vector graphics developed by the Khronos Group. The host CPU 120 interacts with the GPU 110 using an application programming interface (API) to provide hardware accelerated rendering.
[0044] Data processing system 100 uses two types of memory. The first type of memory is DDR memory 130, which is accessible by both GPU 110 and host CPU 120. As part of the high performance of graphics SIMD core 112, GPU 110 uses high speed graphics double data rate (GDDR) memory.
[0045] In high-speed DDR memories, read or write data may have variable transmission path delays that vary with respect to the clock signal used to latch the data elements. In addition, the JEDEC committee specifies that the processor calibrates the link to implement a set of data element delays between the GPU 110 and the GDDR memory 140 so that the data elements can be properly transferred between the data processor and the memory. The various signal processing path lengths inject skew into the system such that as VT changes during operation, drifts in the various signal paths do not track each other and the simple temperature scaling adjustment shown in equation [2] does not produce an accurate compensated calibration value. This characteristic is now described.
[0046] Figure 3 is a block diagram of an annotated GDDR PHY-DRAM link 300, which corresponds to the GDDR PHY-DRAM link 200 of Figure 2. The GDDR PHY-DRAM link 300 is annotated to show signal paths that account for specific timing differences as VT changes.
[0047] Timing path 310 shows the path of the write clock formed by differential signals WCK_t and WCK_c relative to the capture of input (write) data at slicer 296. Timing path 310 shows the flow of the received write clock through the DRAM package, receive buffer 271, buffer 272, divider 273, and buffer / tree 274 before reaching the clock input of slicer 296. Timing path 350 shows the path of the data input signal during a write cycle, showing the received data flowing through the DRAM package impedance and receive buffer 283 to the input of slicer 284. Timing path 310 passes through more circuitry than timing path 350, and changes in VT have a larger effect than changes in timing path 350. These path delays affect a timing parameter known as WCK2DQI.
[0048] Timing path 320 shows the path of the write clock to the output of the read clock RCK. Timing path 320 shows that the received write clock flows through the DRAM package resistor receive buffer 271, buffer 272, divider 273, buffer / tree 274, divider 275, serializer 291, transmit buffer 292, and package impedance to form the read clock. This path delay determines a timing parameter known as WCK2RCK.
[0049] Timing path 330 shows the path of the write clock relative to the capture of the command / address signal at slicer 296. Timing path 320 shows the path of the received write clock through the DRAM package impedance, receive buffer 271, buffer 272, divider 273, buffer / tree 274, and divider 275 before reaching the clock input of slicer 282. Timing path 340 shows the path of the C / A input signal during a command cycle, showing the received data flowing through the DRAM package and receive buffer 281 to the input of slicer 282. This path affects a timing parameter known as WCK2CA. Timing path 330 passes through more circuitry than timing path 340, and changes in VT have a larger effect than changes in timing path 340. These path delays affect the timing parameter known as WCK2CA.
[0050] These representative circuit diagrams show that VT drift affects each of these paths differently. For example, the propagation time through the package routing paths is affected by temperature but not by the memory's supply voltage, whereas the propagation time through the active circuits is affected by supply voltage as well as temperature.
[0051] FIG. 4 is a timing diagram 400 useful for understanding how to capture the WCK2RCK_drift parameter without impacting system performance or latency. In the timing diagram 400, the horizontal axis represents time in picoseconds (ps) and the vertical axis represents the amplitude of several signals in volts (V). The timing diagram 400 shows the waveform of a differential clock signal formed by a true clock signal CK_t and a complementary clock signal CK_c. The differential clock signal is used to latch command and address signals, labeled "CMD" (not shown in FIG. 4), into the GDDR memory 140. To ensure that the commands are captured reliably at the memory, the calibration controller 115 of FIG. 1 previously performed command / address training to determine the amount of delay between the two signal groups, which is applied by the GDDR PHY 118 so that the CMD and address signals arrive at the input to the GDDR memory 140 near the center of the data "eye" with the appropriate setup and hold times relative to the transitions in the CK_t and CK_c signals. Thus, a precharge all command PREALL is latched by the GDDR memory 140 at a time labeled "T0", a refresh all banks (REFab) command is latched at a time labeled "Ta0", and a write training command (WRTR) is latched at a time labeled "Tb0". In response to the WRTR command, the GDDR memory 140 provides read data that can be compared to the expected read data, and the GDDR PHY 118 can gradually change the delay of the delay element 251 until the expected read data is returned from the GDDR memory 140 to the DQ pins to define the current WCK2RCK_drift. Thus, the calibration controller 115 can perform incremental write training to find the WCK2RCK_drift parameter during one or more refresh all banks periods, but the GDDR memory 140 cannot perform any read or write operations.
[0052] FIG. 5 is a timing diagram 500 useful for understanding how to read memory temperature without affecting system performance or command latency. In timing diagram 500, the horizontal axis represents time in picoseconds (ps) and the vertical axis represents the amplitude of several signals in volts (V). Timing diagram 500 shows the waveform of a differential clock signal formed by a true clock signal CK_t and a complementary clock signal CK_c. The differential clock signal is used to latch command and address signals, labeled "CMD" signals (not shown in FIG. 5), into GDDR memory 140. To ensure that commands are reliably captured in the memory, the calibration controller 115 of FIG. 1 previously performed command / address training to determine the amount of delay between the two signal groups, which is applied by GDDR PHY 118 so that the CMD and address signals arrive at the input to GDDR memory 140 near the center of the data "eye" with the appropriate setup and hold times relative to the transitions in the CK_t and CK_c signals. Thus, a precharge all command PREALL is latched by the GDDR memory 140 at a time labeled "T0," a refresh all banks (REFab) command is latched at a time labeled "Ta0," and a mode register set command (MRS) is latched at a time labeled "Tb0." This MRS command reads the mode register which holds the temperature value of the memory.
[0053] For example, a mode register set command is a command that writes to a specific bit of a specific mode register of the GDDR4 memory 140 to invoke a temperature read operation. The GDDR memory 140 provides a temperature reading derived from a temperature sensor 142 on DQ pins 7:0. The GDDR memory 140 holds the DQ pins stable for an extended period of time to allow a temperature reading to be taken prior to an initial timing calibration. In the illustrated embodiment, the GDDR memory 140 waits for a maximum time t following Tb0. WRIDON The GDDR memory 140 provides a binary temperature readout within at least a time t MRD3, driving the binary temperature readout on DQ[7:0] until receipt of an MRS command to disable the binary temperature readout at time Tc2, which may also be provided to the calibration controller 115. Thus, the calibration controller 115 can perform temperature reads to find the DRAM_deltaTemp parameter during one refresh all-banks period, during which the GDDR memory 140 cannot perform any pending read or write operations.
[0054] According to some embodiments, the calibration controller 115 in the memory controller 114 has the flexibility to leverage drift tracking information from WCK2RCK in combination with other VT compensation methods. For example, if the phase drift registered from WCK2RCK_drift exceeds a threshold, the calibration controller 115 can optionally trigger a full write / read / CA calibration. This full calibration can be used to update the VT sensitivity coefficients. To facilitate this technique, the calibration controller 115 can extract multiple voltage and temperature drift magnitudes throughout device operation and update one or more offsets to better predict future VT behavior.
[0055] To fully utilize the VT sensitivity information stored in mode register 141, a tolerance is set along with a maximum drift threshold that will cause the UMC to issue a full link retrain when necessary. An exemplary set of parameters that can be used for this process are shown in Tables III-V.
[0056] Table III corresponds to the GDDR DRAM reference RX (write) operation at a transfer rate of 32 Gbps.
[0057] [Table 3]
[0058] Table IV corresponds to a GDDR DRAM reference TX (read) operation at a transfer rate of 32 Gbps.
[0059] [Table 4]
[0060] Table V corresponds to the GDDR DRAM C / A timing reference operation at a transfer rate of 32 Gbps.
[0061] [Table 5]
[0062] The data processing system described herein, or portions thereof, may be embodied in one or more integrated circuits, any of which may be described or represented by a computer-accessible data structure in the form of a database or other data structure that may be read by a program and used directly or indirectly to manufacture the integrated circuit. For example, the data structure may be a behavioral level description or a register transfer level (RTL) description of the hardware functionality in a high level design language (HDL) such as Verilog or VHDL. The description may be read by a synthesis tool that may synthesize the description to generate a netlist that includes a list of gates from a synthesis library. The netlist includes a set of gates that also represent the functionality of the hardware that comprises the integrated circuit. The netlist may then be placed and routed to generate a data set that describes the geometric shapes to be applied to a mask. The mask may then be used in various semiconductor manufacturing processes to manufacture the integrated circuit. Alternatively, the database on the computer-accessible storage medium may be a netlist (with or without a synthesis library) or a data set, or Graphic Data System (GDS) II data, if desired.
[0063] Although specific embodiments have been described, various modifications to these embodiments will be apparent to those skilled in the art. For example, some embodiments have been described with reference to graphics double data rate (GDDR) DRAM, but may also be applied to other memory types, such as non-graphics DDR memory, high bandwidth memory (HBM), etc. Furthermore, while the embodiments have been described with reference to a data processing system having a separate GPU for very high performance graphics operations, they may also be applied to data processing systems having an accelerated processing unit (APU) in which a CPU and a GPU are integrated together on a single integrated circuit chip. The use of differential or single-ended signaling, and NRZ data signaling or PAM-4 signaling may vary in different embodiments.
[0064] Therefore, the appended claims are intended to cover all modifications of the disclosed embodiments that fall within the scope of the disclosed embodiments.
Claims
1. 1. A data processing system comprising a data processor coupled to a memory, The data processor includes: a reference clock generation circuit for providing a reference clock signal; a first delay circuit for delaying the reference clock signal by a first amount to provide command and address signals; a second delay circuit for delaying the reference clock signal by a second amount to provide a read data signal; a calibration circuit for determining current values of the first and second quantities; a compensation circuit for calculating drift in the first and second quantities based on the measured temperature change, at least one voltage sensitivity coefficient and at least one temperature sensitivity coefficient, and for updating the first and second quantities according to the drift. Data processing system.
2. the compensation circuit calculates drift in the first amount and the second amount based on a timing drift between a write clock signal provided to the memory and a read clock signal received from the memory.
2. The data processing system of claim 1.
3. The calibration circuit includes: a calibration controller that issues memory operations to determine said current values; a memory physical layer interface circuit (PHY) coupled to the calibration controller for providing signals to the memory in response to the memory operations; 2. The data processing system of claim 1.
4. The memory includes: a mode register for storing said at least one voltage sensitivity coefficient and said at least one temperature sensitivity coefficient; 2. The data processing system of claim 1.
5. The memory includes: a temperature sensor for measuring a temperature of the memory; the memory providing the temperature to the data processor during a given operation; 2. The data processing system of claim 1.
6. The predetermined operation includes a refresh operation.
6. The data processing system of claim 5.
7. the compensation circuit causes the calibration circuit to execute a full link retraining sequence in response to drift of either the first amount or the second amount exceeding a respective tolerance threshold.
2. The data processing system of claim 1.
8. A data processor configured to be coupled to a memory, comprising: a reference clock generation circuit for providing a reference clock signal; a first delay circuit for delaying the reference clock signal by a first amount to provide command and address signals; a second delay circuit for delaying the reference clock signal by a second amount to provide a read data signal; a calibration circuit for determining current values of the first and second quantities; a compensation circuit for calculating drift in the first and second quantities based on the measured temperature change, at least one voltage sensitivity coefficient and at least one temperature sensitivity coefficient, and for updating the first and second quantities according to the drift. Data processor.
9. the compensation circuit calculates drift in the first amount and the second amount based on a timing drift between a write clock signal provided to the memory and a read clock signal received from the memory.
9. The data processor of claim 8.
10. The calibration circuit includes: a calibration controller that issues memory operations to determine said current values; a memory physical layer interface circuit (PHY) coupled to the calibration controller for providing signals to the memory in response to the memory operations; 9. The data processor of claim 8.
11. the compensation circuit is configured to read the at least one voltage sensitivity coefficient and the at least one temperature sensitivity coefficient from at least one mode register of the memory.
9. The data processor of claim 8.
12. the compensation circuit causes the calibration circuit to execute a full link retraining sequence in response to drift of either the first amount or the second amount exceeding a respective tolerance threshold.
9. The data processor of claim 8.
13. 1. A method for updating timing values for accesses to a memory by a data processor to compensate for voltage temperature (VT) drift during operation without retraining a link, comprising: generating a reference clock signal; delaying the reference clock signal by a first amount using a first delay circuit to provide command and address signals; delaying the reference clock signal by a second amount using a second delay circuit to provide a read data signal; determining current values of the first quantity and the second quantity using a calibration circuit; and calculating, using a compensation circuit, drift in the first quantity and the second quantity based on the measured temperature change, at least one voltage sensitivity coefficient, and at least one temperature sensitivity coefficient. method.
14. calculating drift in the first amount and the second amount based on timing drift between a write clock signal provided to the memory and a read clock signal received from the memory.
14. The method of claim 13.
15. measuring a timing drift between a write clock signal provided to the memory and a read clock signal received from the memory during a read cycle; The method of claim 14.
16. measuring a timing drift between a write clock signal provided to the memory and a read clock signal received from the memory during a refresh period; 15. The method of claim 14.
17. measuring timing drift between a write clock signal provided to the memory and the read clock signal received by setting a mode register of the memory to place the read clock signal in a continuous toggle mode; 15. The method of claim 14.
18. measuring a temperature of the memory; and providing said temperature to said data processor during a predetermined operation.
14. The method of claim 13.
19. Providing the temperature to the data processor during the predetermined operation includes: providing said temperature to said data processor during a refresh operation of said memory; 20. The method of claim 18.
20. in response to drift of either the first amount or the second amount exceeding a respective tolerance threshold, causing the calibration circuit to perform a full link retrain sequence.
14. The method of claim 13.
