Memory system performing hierarchical duty cycle regulator training and duty cycle regulator training method thereof

Through the hierarchical DCA training method, the DCA code offset size is gradually adjusted and the optimal DCA code combination is selected, which solves the signal integrity and time delay problems in semiconductor memory devices, and realizes efficient duty cycle adjustment and read margin.

CN119068932APending Publication Date: 2024-12-03SAMSUNG ELECTRONICS CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410658913.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2023-08-31
Filing Date
2024-05-27
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Existing semiconductor memory devices may cause excessive time delays when improving data signal integrity, and the application of duty cycle regulator (DCA) codes may fail if the command sequence does not meet even conditions.

Method used

Using the hierarchical DCA training method, by gradually adjusting the offset size of the DCA code, calculating the eye window size of the data signal and selecting the best DCA code combination, ensuring the duty cycle of the internal clock signal, reducing training time and increasing the reading margin .

Benefits of technology

Effectively reduces training time, provides high read margin, and avoids DCA code application failures due to the command sequence not meeting even conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119068932A_ABST
    Figure CN119068932A_ABST
Patent Text Reader

Abstract

A method of training a memory device is provided. In the first to third DCA training steps, a score for each of the first to third DCA code combinations is calculated based on an eye window size of the data signal, and the DCA code combination is selected based on a sum of an even eye window minimum and an odd eye window minimum of the data signal in response to an average occurring between the scores.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the priority of Korean Patent Application No. 10-2023-0071689 filed on June 2, 2023, in the Korean Intellectual Property Office, and Korean Patent Application No. 10-2023-0115410 filed on August 31, 2023, in the Korean Intellectual Property Office, the disclosures of which are incorporated herein by reference in their entirety. Background Art

[0003] Various example embodiments described herein relate to a semiconductor memory device, and more particularly, to a memory system including a memory device trained by adjusting a duty cycle adjuster (DCA) code and a DCA training method thereof.

[0004] Recently, various mobile devices and / or electronic devices, such as one or more of smart phones, desktop computers, laptop computers, tablet PCs, and wearable devices, are widely used. These electronic devices generally include semiconductor memory devices for storing data. As an example of a semiconductor memory device, a dynamic random access memory (DRAM) device, which is a volatile memory, uses charges stored in a capacitor to store data.

[0005] As the input / output speed of memory devices tends to increase, the signal integrity (SI) characteristics of data signals (DQ) input at high speed become relatively poor. One method to solve this signal integrity problem is a training method using a duty cycle adjuster (DCA). Through DCA training, the read margin of the data signal DQ or the data strobe signal (DQS) can be increased by adjusting the internal clock of the memory device using a DCA code.

[0006] However, excessive time delays may occur in setting an improved or optimal duty cycle through such DCA training. Additionally or alternatively, application of the DCA code may fail if the command sequence does not satisfy the even condition (command even gap violation). Summary of the invention

[0007] Various example embodiments provide a memory system and a DCA training method thereof that perform hierarchical DCA training, which may reduce training time and provide a high read margin.

[0008] According to some example embodiments, a method of training a memory device to adjust an eye window of a data signal according to a written duty cycle adjuster (DCA) code includes: performing a first DCA training step on the memory device using a first DCA code combination having a first offset size to adjust the duty cycle of each of internal clock signals QCLK, IBCLK, and QBCLK. Under the condition that the memory device is set according to a result of the first DCA training step, performing a second DCA training step on the memory device using a second DCA code combination having a second offset size, and performing a third DCA training step on the memory device using a third DCA code combination having a third offset size under the condition that the memory device is set according to a result of the second DCA training step. In the first to third DCA training steps, the method includes calculating a score of each of the first to third DCA code combinations based on the eye window size of the data signal, and in response to a tie occurring between the scores, selecting a DCA code combination based on a sum of an even eye window minimum value and an odd eye window minimum value of the data signal.

[0009] Alternatively or additionally, according to various example embodiments, a memory system configured to adjust the duty cycle of four internal clock signals by a duty cycle adjuster (DCA) code includes: a memory device configured to generate a data signal or a data strobe signal having a duty cycle that changes in response to a DCA code, and a memory controller configured to perform hierarchical DCA training on the memory device to select an improved DCA code and set the memory device using the improved DCA code. The memory controller is configured to perform a first DCA training step on the memory device using a first DCA code combination with a first offset size to adjust the duty cycle of each internal clock signal, and perform a second DCA training step on the memory device using a second DCA code combination with a second offset size under the condition that the memory device is set according to the result of the first DCA training step.

[0010] Alternatively or additionally, according to various example embodiments, a method of training a memory device includes: performing a first DCA training on the memory device using a first DCA code combination having a first offset size for adjusting a duty cycle of each of a plurality of internal clock signals, and performing a second DCA training on the memory device using a second DCA code combination having a second offset size under a condition that the memory device is set according to a result of the first DCA training. A maximum value of the first offset size is greater than a maximum value of the second offset size, and the method includes calculating a score based on an eye window size of a data signal corresponding to each of the first to second DCA code combinations in the first to second DCA training. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The above and other objects and features of example embodiments will become apparent by describing in detail embodiments of example embodiments with reference to the accompanying drawings.

[0012] Figure 1 is a block diagram showing the structure of a memory system according to various example embodiments.

[0013] Figure 2 It is a display Figure 1 A block diagram of a memory device.

[0014] Figure 3 is an exemplary illustration according to Figure 2 A timing diagram showing the change of the duty cycle of the data strobe signal regulated by the DCA code in the memory device.

[0015] Figure 4 is a table showing an example of a hierarchical DCA training method according to various example embodiments.

[0016] Figure 5 It is visually shown Figure 4 Diagram of the DCA code combinations at each step in hierarchical DCA training.

[0017] Figure 6 is a flow chart briefly illustrating a hierarchical DCA training process according to various example embodiments.

[0018] Figure 7 is shown in more detail Figure 6 Flow chart of operation S110.

[0019] Figure 8 is a table explaining tie-breaking algorithms corresponding to tie-breaking techniques according to various example embodiments.

[0020] Fig. 9 is shown in more detail Figure 6 Flow chart of operation S130.

[0021] Fig.10 is shown in more detail Figure 6 Flow chart of operation S150.

[0022] Fig.11 is a diagram showing the interaction of a host and a memory device in the first step of hierarchical DCA training.

[0023] Fig.12 is a flow chart briefly illustrating a hierarchical DCA training process according to various example embodiments.

[0024] Fig.13 It is visually shown Fig.12Plot of the DCA offset combinations for each step in the “3-2-1-1” hierarchical DCA training described in .

[0025] Fig.14 is a block diagram showing a memory system according to various example embodiments.

[0026] Fig.15 It is shown in Fig.14 Flowchart of hierarchical DCA training performed in a memory system of FIG.

[0027] Fig.16 is a block diagram illustrating a computing system 3000 according to various example embodiments.

[0028] Fig.17 is a block diagram illustrating a mobile system according to various example embodiments.

[0029] Fig.18 is a block diagram illustrating another embodiment of a memory system using hierarchical DCA training of various example embodiments.

[0030] Fig.19 is a diagram illustrating a memory system according to various example embodiments.

[0031] Fig. 20 is a cross-sectional diagram showing a memory system according to various example embodiments. DETAILED DESCRIPTION

[0032] It should be understood that both the foregoing general description and the following detailed description are exemplary and should be considered to provide additional description of the claimed invention. Reference numerals are indicated in detail in various embodiments, examples of which are indicated in the referenced drawings. Wherever possible, the same reference numerals are used in the specification and drawings to refer to the same or similar parts.

[0033] Figure 1 is a block diagram showing the structure of a memory system according to various example embodiments. Figure 1 , the memory system 1000 includes a host 1100 and a memory device 1300. The host 1100 performs duty cycle adjuster (DCA) training of the memory device 1300 to adjust the duty cycle (or duty ratio) of the data signal DQ and / or the data strobe signal DQS. The DCA training can be applied, for example, during boot and / or initialization of the memory system 1000 and / or during a test process, but the application timing of the DCA training is not limited to the description herein.

[0034] The host 1100 may write data to the memory device 1300 and / or may perform an access operation to read data stored in the memory device 1300. The host 1100 may generate clock signals (CKt, CKc), commands CMD, and addresses ADDR for writing data to the memory device 1300 and / or for reading data stored in the memory device 1300. The host 1100 may be or may include at least one of a chipset for controlling the memory device 1300, a system on chip SoC such as a mobile application processor AP, a CPU, a GPU, or a test device, or may be included therein.

[0035] For example, the host 1100 can adjust the duty cycle of the data signal DQ and / or the data strobe signal DQS output from the memory device 1300 through DCA training. The memory device 1300 generates a four-phase internal clock (ICLK, QCLK, IBCLK, and QBCLK), whose duty cycle is adjusted according to the DCA code. The size of the eye window EW of the data signal DQ of the memory device 1300 is adjusted by adjusting the duty cycle of the four phases (ICLK, QCLK, IBCLK, and QBCLK) of the internal clock. For example, through DCA code adjustment, the host 1100 can adjust the duty cycle of the internal clock (ICLK, QCLK, IBCLK, and QBCLK) of the memory device 1300. This indicates that the signal integrity (SI) characteristics of the data strobe signal DQS and the data signal DQ output from the memory device 1300 can be adjusted. Adjustment of the duty cycle of one or more of the four phases (ICLK, QCLK, IBCLK, and QBCLK) of the internal clock is a factor that determines the eye window size or signal integrity of the data signal DQ.

[0036] In particular, the host 1100 can adjust the clock deviation between the internal clock signals (ICLK, QCLK, IBCLK, and QBCLK) through hierarchical DCA training. By adjusting the clock deviation between the internal clock signals (ICLK, QCLK, IBCLK, and QBCLK), the eye window EW size of the data selection signal DQS or the data signal DQ can be increased (e.g., maximized). The hierarchical DCA training of the example embodiment refers to a technique for training the DCA code starting from a large offset and moving to a small offset.

[0037] For example, hierarchical DCA training can be performed through a process of three steps or four steps or more than four steps. Typically, the deviation of the first internal clock signal ICLK is fixed, so it may be impossible to adjust the duty cycle using the DCA code. Therefore, when hierarchical DCA training is performed in three steps, in the first step, a DCA code offset with an amplitude of "-4, 0, 4" is used to adjust the deviation of each of the second to fourth internal clock signals QCLK, IBCLK and QBCLK to perform the widest range of deviation adjustment. For example, DCA training is performed using a combination of internal clock signals QCLK, IBCLK and QBCLK, where each DCA code has an offset of "-4, 0 or 4". Then, the output data signal DQ is checked, and the DCA code combination corresponding to the maximum eye window is selected. Ultimately, the first step STEP_1 can be referred to as a rough training using the widest range of offsets. If a tie score situation occurs, an improved value or an optimal value can be selected by applying a tie-breaking algorithm, which may or may not be a randomization algorithm. According to the tie-breaking algorithm, a DCA code combination is selected based on the sum of the even-eye window minimum value and the odd-eye window minimum value. The tie-breaking algorithm will be explained in more detail with reference to the drawings described later.

[0038] After the memory device 1300 is provided with the improved or optimal DCA code selected in the first step STEP_1, the DCA training corresponding to the second step STEP_2 continues. In the second step STEP_2, by inputting the DCA code offset '-2, 0, 2' for each internal clock signal QCLK, IBCLK and QBCLK, the deviation adjustment is performed within a smaller range than the first step STEP_1. For example, the second step STEP_2 corresponds to the first fine (or finer) training that performs more accurate deviation adjustment based on the DCA code determined in the first step STEP_1. If a tie occurs in the second step STEP_2, a tie-breaking algorithm that can select an improved value or an optimal value can be applied. The tie-breaking algorithm used after the second step STEP_2 can be the same as or different from the tie-breaking algorithm used after the first step STEP_1.

[0039] After the memory device 1300 is provided with the improved or optimal DCA code selected in the second step STEP_2, the DCA training corresponding to the third step STEP_3 continues. In the third step STEP_3, the deviation adjustment within a smaller range than the second step STEP_2 is performed by inputting the DCA code offset '-1, 0, 1' for each internal clock signal QCLK, IBCLK and QBCLK. For example, the third step STEP_3 corresponds to the second fine training that performs more accurate deviation adjustment based on the DCA code determined in the second step STEP_2. If a tie occurs in the third step STEP_3, a tie-breaking algorithm that can select an improved or optimal value can be applied. The tie-breaking algorithm after the third step STEP_3 can be the same or different from the tie-breaking algorithm after the second step STEP_2 or after the first step STEP_1.

[0040] The three-step hierarchical DCA training performed by the host 1100 is not limited to the description here. The hierarchical DCA training can be applied in four or more steps. For example, four or more steps can be used to apply the hierarchical DCA training. Through the hierarchical DCA training, even if the command even gap that may be important in the magnitude of the even / odd unit interval is not observed, there is no possibility of the inter-phase DCA code being mixed or the possibility of the inter-phase DCA code being mixed is reduced. And a training mode such as a tie-breaking algorithm and a pseudo-random binary sequence PRBS used in the hierarchical DCA training is applied. Therefore, the eye window EW size of the data signal DQ can be improved with high accuracy, or ensured or more likely to be ensured.

[0041] The memory device 1300 may perform hierarchical DCA training under the control of the host 1100. The memory device 1300 generates a 4-phase internal clock 4P_CLK, whose duty cycle is adjusted according to the DCA code written by the host 1100. The 4-phase internal clock 4P_CLK may be an internal clock signal (ICLK, QCLK, IBCLK, and QBCLK), each of which has a 90° phase difference in this order. In some example embodiments, one internal clock signal (e.g., ICLK) may have a fixed duty cycle, while the duty cycles of the remaining three internal clock signals QCLK, IBCLK, and QBCLK may be adjusted using the DCA code. For this purpose, the memory device 1300 may include a duty cycle regulator 1350, a clock generation circuit 1360, and an input / output circuit 1370.

[0042] The duty cycle regulator 1350 generates a 4-phase internal clock signal 4P_CLK, whose duty cycle is adjusted according to the DCA code written by the host 1100. The 4-phase internal clock signal 4P_CLK can be generated based on the clock signal (CKt, CKc) or the write clock signal (WCKt, WCKc) provided from the host 1100. Generally, the first internal clock signal ICLK can be a fixed internal clock signal, and its duty cycle is not adjusted using the DCA code. On the other hand, the second internal clock signal QCLK has a phase difference of 90° with the first internal clock signal ICLK, and the third internal clock signal IBCLK has a phase difference of 180° with the first internal clock signal ICLK, and the fourth internal clock signal QBCLK has a phase difference of 270° with the first internal clock signal ICLK. The duty cycle regulator 1350 can adjust the duty cycle of the second internal clock signal to the fourth internal clock signal according to the written DCA code.

[0043] The clock generation circuit 1360 generates a clock signal and / or a data strobe signal DQS according to the 4-phase internal clock signal 4P_CLK provided from the duty cycle adjuster 1350. The clock generation circuit 1360 may increase or decrease the even clock duty cycle of the data strobe signal DQS according to the second internal clock QCLK having an increased or decreased duty cycle. Similarly, the clock generation circuit 1360 may increase or decrease the even clock duty cycle of the data strobe signal DQS according to the third internal clock IBCLK or the fourth internal clock QBCLK having an increased or decreased duty cycle.

[0044] The input / output circuit 1370 may output the read data Dout as a data signal DQ using a clock signal or a data strobe signal DQS provided by the clock generation circuit 1360. The data signal DQ is transmitted to the host 1100 in synchronization with the data strobe signal DQS generated by the clock generation circuit 1360.

[0045] Figure 2 It is a display Figure 1 A block diagram of a memory device. Figure 2 , the memory device 1300 includes a cell array 1310, a sense amplifier 1315, an address decoder 1320, a command decoder 1330, a mode register set 1340, a duty cycle regulator 1350, a clock generation circuit 1360, an input / output circuit 1370 and an input / output sense amplifier 1380.

[0046] The cell array 1310 includes a plurality of memory cells, such as, but not limited to, nonvolatile memory cells and / or volatile memory cells. In some example embodiments, the cell array 1310 may include redundant cells and / or dummy cells; example embodiments are not limited thereto. Write data received through the input / output sense amplifier 1380 may be written to the plurality of memory cells by the sense amplifier 1315, and the data written to the memory cells is selected by the command CMD and the address ADDR of the host 1100 and sensed by the sense amplifier 1315. The data sensed by the sense amplifier 1315 may be sent to the input / output sense amplifier 1380 and then to the input / output circuit 1370.

[0047] The address decoder 1320 may receive an address ADDR of a memory cell being accessed. When data is stored in and / or read from a memory cell, the address ADDR may be transmitted to the cell array 1310 as a row address and a column address through the address decoder 1320.

[0048] The command decoder 1330 may decode a command CMD input from the outside. The command decoder 1330 may access a mode register set (MRS, 1340) for setting various modes and operations of the memory device 1300. The command decoder 1330 may identify the attributes of the input command CMD by referring to an externally applied signal (such as but not limited to one or more of / RA, / CAS, / WE). In addition, the command decoder 1330 may write data into the mode register set 1340 according to the command CMD and address ADDR provided from the outside. For example, the DCA code provided for the DCA training of the present invention is written into the mode register set 1340 by the command decoder 1330.

[0049] The mode register group 1340 sets the internal mode register in response to the MRS command and the address ADDR for specifying the operation mode of the memory device 1300. The mode register group 1340 according to various example embodiments may include a DCA code register 1345, in which a DCA code for adjusting a duty cycle of a data signal DQ and / or a data strobe signal DQS is written. A mode register write MRW command may be used to write the DCA code into the DCA code register 1345.

[0050] The DCA code of each of the internal clock signals QCLK, IBCLK, and QBCLK for adjusting the duty cycle of each of the internal clock signals QCLK, IBCLK, and QBCLK is written into the DCA code register 1345. By writing the DCA code, the current duty cycle of each of the internal clock signals QCLK, IBCLK, and QBCLK is adjusted to one of the steps of "+7" to "-7" (15 steps). For example, the duty cycle of each of the internal clock signals QCLK, IBCLK, and QBCLK can be increased or decreased by writing the corresponding DCA code. Therefore, the even clock duty cycle and / or the odd clock duty cycle of the data signal DQ or the data strobe signal DQS is increased or decreased. Generally, the DCA code register 1345 is set so that the first internal clock signal ICLK cannot be adjusted, and the second to fourth internal clock signals QCLK, IBCLK, and QBCLK can be adjusted. However, the range of the adjustable internal clock signals is not limited to the disclosure herein.

[0051] The duty cycle regulator 1350 generates internal clock signals ICLK, QCLK, IBCLK, and QBCLK according to the DCA code in the write mode register group 1340. The duty cycle regulator 1350 generates the internal clock signals ICLK, QCLK, IBCLK, and QBCLK using externally provided clock signals (CK_t, CK_c) and / or write clock signals (WCK_t, WCK_c).

[0052] The clock generation circuit 1360 generates a clock signal and / or a data strobe signal DQS according to the internal clock signals ICLK, QCLK, IBCLK, and QBCLK provided from the duty cycle adjuster 1350. The clock generation circuit 1360 may increase or decrease the duty cycle of the even clock and / or odd clock of the data strobe signal DQS according to the internal clock signals ICLK, QCLK, IBCLK, and QBCLK that increase or decrease the duty cycle.

[0053] The input / output circuit 1370 outputs the read data Dout as the data signal DQ using the clock signal or the data strobe signal DQS provided by the clock generation circuit 1360. The data signal DQ is transmitted to the host 1100 in synchronization with the data strobe signal DQS generated by the clock generation circuit 1360.

[0054] As described above, the memory device 1300 of various example embodiments may be set to a DCA code that increases or maximizes the eye window EW of the data signal DQ through hierarchical DCA training.

[0055] Figure 3 is an exemplary illustration according to Figure 2The timing diagram of the change of the duty cycle of the data strobe signal regulated by the DCA code in the memory device. Figure 3 , shows the change of the data select signal DQS when the DCA code of the second internal clock signal QCLK is increased or decreased in the four phases of the internal clock signals (ICLK, QCLK, IBCLK and QBCLK).

[0056] The second internal clock signal QCLK has a 90° phase difference with the first internal clock signal ICLK. Assume that the DCA code corresponding to the currently written second internal clock signal QCLK is in the '0' step state. In addition, when the DCA code is increased by "+2" steps by writing the DCA code, the rising edge of the second internal clock signal QCLK retreats, as shown by the dotted circle 1351. The retreated rising edge of the second internal clock signal QCLK increases the size of the even duty cycle portion of the data select signal DQS. On the other hand, if the DCA code is reduced by "-2" steps by writing the DCA code, the rising edge of the second internal clock signal QCLK is advanced as shown by the dotted line 1352. The early rising edge of the second internal clock signal QCLK reduces the size of the even duty cycle portion of the data select signal DQS.

[0057] In the above, the effect of writing the DCA code to adjust the second internal clock signal QCLK among the internal clock signals ICLK, QCLK, IBCLK and QBCLK is briefly described. Similarly, the size of the duty cycle portion of the data selection signal DQS can be adjusted by adjusting the DCA code of the third to fourth internal clock signals IBCLK and QBCLK.

[0058] Figure 4 is a table showing a hierarchical DCA training method according to various example embodiments. Figure 4 , for hierarchical DCA training, a DCA scan is performed to adjust each of the internal clock signals QCLK, IBCLK, and QBCLK in three steps (STEP_1, STEP_2, and STEP_3). Here, the DCA scan refers to an input / output operation of measuring the eye window EW size of the data signal DQ output from the memory device 1300 while changing the DCA code. Here, the data pattern for DCA training may be a random pattern such as a pseudo-random binary sequence (PRBS); however, example embodiments are not limited thereto.

[0059] In the first step STEP_1, the host 1100 performs a rough DCA scan on each of the internal clock signals QCLK, IBCLK, and QBCLK. For example, in the first step STEP_1, the step size of the DCA code written (hereinafter referred to as DCA offset) can be any one of '-4', '0', and '4'. For example, for the DCA scan, the DCA offset of each of the internal clock signals QCLK, IBCLK, and QBCLK is divided into 27 offset combinations {(-4, -4, -4), (-4, -4, 0), ..., (4, 4, 4)}. Therefore, in the first step STEP_1, the eye window size of the input and output data signal DQ of the 27 DCA offsets of the memory device 1300 is measured. At this time, it can be assumed that the state of the DCA offset before the initial input of the DCA offset is (0, 0, 0), which is the default value of each of the internal clock signals QCLK, IBCLK, and QBCLK.

[0060] The eye window size of each of the 27 DCA offset combinations {(-4, -4, -4), (-4, -4, 0), ..., (4, 4, 4)} can be calculated according to the result of the rough DCA scan in the first step STEP_1. The minimum value minEW in the eye window of the four unit intervals (UI0, UI1, UI2, UI3) of the data signal DQ corresponding to each DCA offset combination can be assigned as the score of each DCA offset combination. At this time, among the 27 DCA offset combinations, any one DCA offset combination with the maximum score can be selected as the improved or optimal DCA offset combination in the first step STEP_1.

[0061] If there is a tie between the scores minEW of the 27 DCA offset combinations, a tie-breaking algorithm according to various example embodiments may be applied. For example, in the case of a tie, a DCA offset combination may be selected based on the sum of the even-eye window minimum value Even_minEW and the odd-eye window minimum value Odd_minEW in each output eye window. For example, in the case of a tie, a DCA offset combination having a larger sum of the even-eye window minimum value Even_minEW and the odd-eye window minimum value Odd_minEW is selected as an improved or optimal DCA offset combination. The sum of the even-eye window minimum value Even_minEW and the odd-eye window minimum value Odd_minEW may or may not be a weighted sum; example embodiments are not limited thereto. A tie processing algorithm will be defined as a tie-breaking algorithm. Then, the host 1100 writes the determined optimal DCA offset combination as a DCA code into the memory device 1300, and completes the first step STEP_1 of hierarchical DCA training.

[0062] In the second step STEP_2, the host 1100 performs a first FINE1 DCA scan with increased precision on each of the internal clock signals QCLK, IBCLK, and QBCLK based on the DCA code determined in the first step STEP_1. In some examples, the second step STEP_2 uses an offset '-2,0,2' that is more accurate than the DCA offset value of the first step STEP_1. For example, for the DCA scan in the second step STEP_2, 27 offset combinations {(-2,-2,-2), (-2,-2,0), ..., (2,2,2)} are used. In the second step STEP_2, an eye window size measurement of the 27 DCA offset input and output data signals DQ to the memory device 1300 will be performed.

[0063] The eye window size of each of the 27 DCA offset combinations {(-2,-2,-2), (-2,-2,0), …, (2,2,2)} will be calculated based on the results of the first fine FINE1 DCA scan in the second step STEP_2. In addition, the scores of the 27 DCA offset combinations are determined in the same manner as in the first step STEP_1. At this time, if there is a tie between the scores of the 27 DCA offset combinations, a tie-breaking algorithm can be applied. When an improved or optimal DCA offset combination is determined in the second step STEP_2, the host 1100 sets the memory device 1300 with the corresponding DCA code and completes the second step STEP_2 of the hierarchical DCA training. The tie-breaking algorithm can be based on the weighted sum of the minimum even and odd eye windows, which may or may not use the same weights as used in the tie-breaking algorithm performed after the first step STEP_1; example embodiments are not limited thereto.

[0064] In the third step STEP_3, the host 1100 performs a second fine FINE2 DCA scan on each of the internal clock signals QCLK, IBCLK, and QBCLK based on the DCA code determined in the second step STEP_2. That is, the third step STEP_3 uses an offset '-1, 0, 1' that is more accurate than the offset value of the DCA code of the second step STEP_2. For the second fine FINE2 DCA scan in the third step STEP_3, 27 offset combinations {(-1, -1, -1), (-1, -1, 0), ..., (1, 1, 1)} of each internal clock signal QCLK, IBCLK, and QBCLK are used.

[0065] The eye window size of each of the 27 DCA offset combinations {(-1,-1,-1), (-1,-1,0), ..., (1,1,1)} will be calculated based on the result of the second FINE2 DCA scan in the third step STEP_3. The scores of the 27 DCA offset combinations are determined in the same manner as in the first step STEP_1 or the second step STEP_2. When the improved or optimal DCA offset combination is determined in the third step STEP_3, the host 1100 sets the memory device 1300 with the corresponding DCA code and completes the third step STEP_3 of the hierarchical DCA training.

[0066] The above briefly explains the characteristics of hierarchical DCA training. The adjustment accuracy of the internal clock signals QCLK, IBCLK and QBCLK can be increased by gradually reducing the DCA offset size in each of the three stages in a "4-2-1" manner. The number of DCA codes written for DCA scanning in each of the three steps is 27 offset combinations. Therefore, for three-stage hierarchical DCA training, 81 (=27+27+27) DCA scans are performed. If training is performed using all DCA code combinations instead of hierarchical DCA training, DCA code combinations for the "15×15×15=3375" case must be tested or used. This indicates that the DCA training time according to various example embodiments can be significantly shortened. Alternatively or additionally, high-precision DCA training is possible using the tie-breaking algorithm of the example embodiments.

[0067] Figure 5 It is visually shown Figure 4 Figure 2. DCA code combinations for each step in the “4-2-1” hierarchical DCA training described in [1]. Figure 5 , the DCA offset combinations used in each of the three steps (STEP_1, STEP_2, and STEP_3) for hierarchical DCA training are arranged in the form of 3-dimensional coordinates.

[0068] The DCA offset value of each internal clock signal written for the DCA scan in the first step STEP_1 may be any one of '-4', '0', and '4'. In short, the DCA offset value of each of the internal clock signals QCLK, IBCLK, and QBCLK is mapped to any one of '-4', '0', and '4' on the three axes (Q, IB, QB). Therefore, the offset combinations may each correspond to 27 coordinate points. In order to perform the first step STEP_1 of hierarchical DCA training, the host 1100 writes a DCA code corresponding to each of the 27 offset combinations and detects the data margin.

[0069] The eye window size of each of the 27 DCA offset combinations {(-4, -4, -4), (-4, -4, 0), ..., (4, 4, 4)} according to the result of the rough DCA scan in the first step STEP_1 is detected. In addition, the minimum value minEW of the eye window size of the four unit intervals UI0, UI1, UI2 and UI3 of the data signal DQ corresponding to each DCA offset combination is calculated as the score of each DCA offset combination. At this time, among the 27 DCA offset combinations, the DCA offset combination with the maximum score can be selected as the improved or best offset combination in the first step STEP_1. If there is a tie between the scores (i.e., minEWs) of the 27 DCA offset combinations, a tie-breaking algorithm can be used. As shown in the figure, if (4, 0, 0) is selected as the improved or best DCA offset combination, the host 1100 will set the memory device 1300 with the corresponding DCA code.

[0070] The offset value of the DCA code written for the DCA scan in the second step STEP_2 may be any one of '-2', '0', and '2'. The offset value of each of the internal clock signals QCLK, IBCLK, and QBCLK may be mapped to any one of '-2', '0', and '2' on the three axes (Q, IB, QB). The offset combination may correspond to 27 coordinate points having a magnitude of '-2', '2', or '0' in the three-dimensional coordinates of the three axes (Q, IB, QB).

[0071] The eye window size of each 27 offset combinations {(-2, -2, -2), (-2, -2, 0), ..., (2, 2, 2)} according to the result of the first FINE1 DCA scan applying the offset size "2" in the second step STEP_2 is calculated. In addition, the minimum value minEW of the eye window of the four unit intervals UI0, UI1, UI2 and UI3 of the data signal DQ corresponding to each offset combination is calculated as the score of each offset combination. At this time, among the 27 offset combinations, the DCA offset combination with the maximum score can be selected as the improved or best offset combination in the second step STEP_2. In addition, if there is a tie between the scores (i.e., minEWs) of the 27 DCA offset combinations, a tie-breaking algorithm can be used to select the improved or best offset combination. Let us assume that the improved or best offset combination is selected as (2, 2, -2). Next, the host 1100 writes the DCA code corresponding to the offset values ​​'2', '2' and '−2' to the memory device 1300 to set the duty cycle of each of the internal clock signals QCLK, IBCLK and QBCLK to complete the second step STEP_2 .

[0072] In the third step STEP_3, any one of '-1', '0' and '1' may be written as the offset value of the DCA code for the DCA scan. In other words, the offset value of each of the internal clock signals QCLK, IBCLK and QBCLK may be mapped to any one of '-1', '0' and '1' on the three axes (Q, IB, QB). The offset combination may correspond to 27 coordinate points having an amplitude of '-1', '1' or '0' in the three-dimensional coordinates of the three axes (Q, IB, QB). Therefore, the host 1100 will perform a DCA scan corresponding to each of the 27 offset combinations to continue the third step STEP_3 of the hierarchical DCA training.

[0073] The eye window size of each 27 offset combinations {(-1, -1, -1), (-1, -1, 0), ..., (1, 1, 1)} according to the result of the second FINE2 DCA scan applying the offset size '1' in the third step STEP_3 is calculated. In addition, the minimum value minEW of the eye window of the four unit intervals UI0, UI1, UI2 and UI3 of the data signal DQ corresponding to each offset combination is calculated as the score of each offset combination. At this time, among the 27 offset combinations, any one DCA offset combination with the maximum score can be selected as the improved or best offset combination in the third step STEP_3. In addition, if there is a tie between the scores (i.e., minEW) of the 27 DCA offset combinations, a tie-breaking algorithm can be used to select the improved or best offset combination. Let us assume that the improved or best offset combination is selected as (0, 1, -1). Next, the host 1100 writes the DCA codes corresponding to the offset values ​​'0', '1' and '-1' to the memory device 1300 to complete the third step STEP_3. When the improved or optimal DCA code is determined through the DCA scan of the third step STEP_3 and the memory device 1300 is set using the determined optimal DCA code, the hierarchical DCA training of the present invention is completed.

[0074] The gradual reduction process of the DCA code used in hierarchical DCA training is briefly described above. By gradually reducing the offset size to a '4-2-1' pattern in each of the three stages, the adjustment accuracy of the duty cycle of the internal clock signals QCLK, IBCLK and QBCLK can be increased.

[0075] Figure 6 is a flow chart briefly illustrating the hierarchical DCA training process of the present invention. Figure 6 , the memory system 1000 of the present invention (see Figure 1 ) performs hierarchical DCA training on the internal clock signals QCLK, IBCLK and QBCLK sequentially.

[0076] In operation S110, the host 1100 performs the first step STEP_1 of the hierarchical DCA training. The host 1100 may set the DCA offset to any one of '-4', '0', and '4' for the coarse DCA scan in the first step STEP_1. The internal clock signals QCLK, IBCLK, and QBCLK may have 27 offset combinations. The host 1100 will perform a DCA scan using a DCA code corresponding to each of the 27 offset combinations to continue the first step STEP_1 of the hierarchical DCA training. The host 1100 generates a score from the eye window EW size for each of the 27 offset combinations. In addition, if there is a tie in the scores of the 27 DCA offset combinations, the host 1100 may use a tie-breaking algorithm to select an improved or optimal offset combination. Once the improved or optimal offset combination is determined, the host 1100 writes the DCA code corresponding to the improved or optimal offset combination into the memory device 1300.

[0077] In operation S130, the host 1100 performs the second step STEP_2 of the hierarchical DCA training. The host 1100 performs the second step STEP_2 of the hierarchical DCA training based on the improved or optimal DCA code generated as a result of the setting of the first step STEP_1. For the DCA scan in the second step STEP_2, the host 1100 may set the DCA offset to any one of "-2", "0" and "2". The host 1100 will perform the DCA scan using the DCA code corresponding to each of the 27 offset combinations to continue the second step STEP_2 of the hierarchical DCA training. In addition, as a result of the DCA scan, the host 1100 calculates the eye window EW size of each of the 27 offset combinations. The host 1100 generates a score from the eye window EW size for each of the 27 offset combinations. The score generation method is the same as the method applied to the first step STEP_1. In addition, if there is a tie in the scores of the 27 DCA offset combinations, the host 1100 may use a tie-breaking algorithm to select an improved or optimal offset combination. Once the improved or optimal offset combination is determined, the host 1100 writes the DCA code corresponding to the selected optimal offset combination into the memory device 1300. Then, the duty cycle of each of the internal clock signals QCLK, IBCLK and QBCLK in the second step STEP_2 is set to an optimal value.

[0078] In operation S150, the host 1100 performs the third step STEP_3 of the hierarchical DCA training. The host 1100 performs the third step STEP_3 of the hierarchical DCA training based on the improved or optimal DCA code generated as a result of the setting of the second step STEP_2. For the DCA scan in the third step STEP_3, the host 1100 may set the DCA offset to one of '-1', '0', and '1'. As a result of the DCA scan, the host 1100 calculates the eye window EW size for each of the 27 offset combinations. The host 1100 generates a score from the eye window EW size for each of the 27 offset combinations. The score generation method is the same as the score generation method applied in the first step STEP_1 and the second step STEP_2. In addition, if there is a tie in the scores of the 27 DCA offset combinations, the host 1100 may use a tie-breaking algorithm to select an improved or optimal offset combination. Once the improved or optimal offset combination is determined, the host 1100 writes the DCA code corresponding to the selected optimal offset combination into the memory device 1300. Then, the duty ratio of each of the internal clock signals QCLK, IBCLK, and QBCLK in the third step STEP_3 may be set to an optimal value.

[0079] Above, a hierarchical DCA training method for internal clock signals according to various example embodiments has been described. By gradually reducing the DCA offset size in three steps (such as "4-2-1"), the adjustment accuracy of the duty cycle of the internal clock signals QCLK, IBCLK, and QBCLK can be increased. In addition, if a tie occurs between the scores of each step of the hierarchical DCA training, a tie-breaking algorithm of the present invention can be used to perform high-precision DCA code selection.

[0080] Figure 7 is shown in more detail Figure 6 Flow chart of operation S110. Figure 7 , host 1100 (see Figure 1 )Perform a coarse DCA scan to perform the first step STEP_1 of hierarchical DCA training and determine the optimal DCA code.

[0081] In operation S111, the host 1100 writes a DCA code corresponding to one of 27 offset combinations of the internal clock signals QCLK, IBCLK, and QBCLK to the memory device 1300 for DCA scanning. Here, the host 1100 uses relatively large offsets (e.g., "-4", "0", and "4") to determine the duty ratios of the internal clock signals QCLK, IBCLK, and QBCLK.

[0082] In operation S112, the memory device 1300 is adjusted according to the DCA code written in operation S111 (see Figure 1 ) of the internal clock signals QCLK, IBCLK, and QBCLK. Then, the size of the eye window EW of the data signal DQ is adjusted according to the duty ratio of the internal clock signals QCLK, IBCLK, and QBCLK after the adjustment. As a result, the memory device 1300 outputs a data pattern corresponding to the DCA code written in operation S111 through the data signal DQ. At this time, the data pattern used can be, for example, a random sequence, such as a pseudo-random binary sequence PRBS. Then, the host 1100 can measure the data margin or eye window size of the output data pattern.

[0083] In operation S113, the method may determine whether the DCA code combination applied in operations S111 and S112 is the last combination (or case) among all the combinations. When all DCA scans of each of the 27 DCA code combinations are completed, the process moves to operation S115. On the other hand, if there is a combination that is not used in the DCA scan among the 27 DCA code combinations, the process moves to operation S114.

[0084] In operation S114, the host 1100 selects a DCA offset combination that is not applied in the DCA scan of the first step STEP_1. Thereafter, the process moves to operation S111, where a DCA scan using the selected DCA offset combination will be performed.

[0085] In operation S115, the host 1100 calculates a score using an output pattern or an eye window of the data signal DQ corresponding to the 27 DCA code combinations. For example, the minimum value minEW in the eye window of four unit intervals (UI0, UI1, UI2, and UI3) of the data signal DQ may be calculated as a score for each offset combination.

[0086] In operation S116, the host 1100 determines whether there is a tie between the scores of each offset combination. For example, there may be scores of the same value in the scores of each offset combination. If there is no tie score, the process moves to operation S118. On the other hand, if there are scores of the same value in the scores of each offset combination, the process moves to operation S117.

[0087] In operation S117, if there is a tie between the scores of the DCA offset combinations, the host 1100 uses a tie-breaking algorithm to select an improved or optimal offset combination. That is, a DCA offset combination having a maximum sum of the even-eye window minimum value Even_minEW and the odd-eye window minimum value Odd_minEW for each tie case may be selected as an improved or optimal DCA offset combination.

[0088] In operation S118, the host 1100 may select the DCA offset combination with the maximum score as the improved or best DCA offset combination. Alternatively, if there is a tie score, the DCA offset combination selected using the tie-breaking algorithm may be selected as the improved or best DCA offset combination.

[0089] In operation S119, the host 1100 writes the DCA code corresponding to the determined optimal offset combination into the memory device 1300. Then, the first step STEP_1 of the hierarchical DCA training is completed.

[0090] The above briefly describes the coarse DCA scanning process of the first step STEP_1 for performing hierarchical DCA training. Through the coarse DCA scanning, the duty cycle of the internal clock signals QCLK, IBCLK, and QBCLK of the memory device 1300 is improved or optimized within a large offset value range (e.g., "-4", "0", and "4").

[0091] Figure 8 is a table explaining the tie-breaking algorithm corresponding to the tie-breaking technique of the present invention. Figure 8 , the host 1100 performs a DCA scan on the memory device 1300 and calculates a score for each of the 27 DCA code combinations. In addition, when a tie occurs, the host 1100 calculates Sum (even / odd), which is the sum of the even-eye window minimum value Even_minEW and the odd-eye window minimum value Odd_minEW. In addition, the DCA offset combination with the largest Sum (even / odd) is selected as an improved or optimal offset combination. Although the example embodiment describes calculating the sum of the even-eye window and the odd-eye window, the example embodiment is not limited to this. For example, in some examples, for the tie-breaking algorithm, the average value and / or product and / or geometric mean and / or harmonic mean of the even-eye window and the odd-eye window may be calculated, and / or another function of the even-eye window and the odd-eye window may be calculated.

[0092] First, for each DCA code combination, the minimum value minEW in the eye window of the four unit intervals UI0, UI1, UI2 and UI3 of the data signal DQ is calculated as a score. For example, the minimum eye window value minEW of the DCA code combination (-4, 0, 4) corresponding to the DCA index '5' will be detected as '35'. In this way, the host 1100 calculates the scores of 27 DCA offset combinations. Next, the host 1100 will select the DCA code combination corresponding to the maximum value of the 27 scores as the improved or best offset combination. However, as shown in the figure, two cases (A, B) are detected to have the same score "40". Therefore, a tie-breaking algorithm must be applied to handle the tie.

[0093] In order to handle the tie, the host 1100 may, for example, add the even-eye window minimum value Even_minEW and the odd-eye window minimum value Odd_minEW for each tie case. The minimum even-eye window Even_minEW and the minimum odd-eye window Odd_minEW in the DCA code combination of the DCA index '13' are detected as 40 and 41, respectively. On the other hand, the minimum even-eye window Even_minEW and the minimum odd-eye window Odd_minEW in the DCA code combination of the DCA index '22' are detected as 40 and 43, respectively. Therefore, 'Sum (even / odd)', which is the sum of the even-eye window minimum value Even_minEW and the odd-eye window minimum value Odd_minEW, are expressed as '81' and '83', respectively. Therefore, the B case with a larger 'Sum (even / odd)' may be selected as an improved or optimal offset combination.

[0094] Fig. 9 is shown in more detail Figure 6 Flow chart of operation S130. Fig. 9 , host 1100 (see Figure 1 ) performs a DCA scan to perform the second step STEP_2 of hierarchical DCA training and determine the optimal DCA offset.

[0095] In operation S131, the host 1100 writes a DCA code corresponding to one of 27 offset combinations of the internal clock signals QCLK, IBCLK, and QBCLK for a first fine DCA scan to the memory device 1300. Here, the host 1100 sets the duty ratios of the internal clock signals QCLK, IBCLK, and QBCLK using offset values ​​(e.g., "-2", "0", and "2") that are smaller than the offset size in the first step STEP_1.

[0096] In operation S132, the duty ratios of the internal clock signals QCLK, IBCLK, and QBCLK of the memory device 1300 will be adjusted according to the DCA code written in operation S131. The memory device 1300 will output a data pattern corresponding to the DCA code written in operation S131 through the data signal DQ. Then, the host 1100 may measure the data margin or eye window size of the output data pattern.

[0097] In operation S133, it is determined whether the DCA offset combination applied in operations S131 and S132 is the last combination (or case) among all the combinations. When all DCA scans of each of the 27 DCA code combinations are completed, the process moves to operation S135. On the other hand, if there is a combination that is not used in the DCA scan among the 27 DCA code combinations, the process moves to operation S134.

[0098] In operation S134, the host 1100 selects a DCA code combination that is not applied in the DCA scan of the second step STEP_2. Thereafter, the process proceeds to operation S131, where a DCA scan using the selected DCA code combination will be performed.

[0099] In operation S135, the host 1100 calculates the score using the output mode or eye window size of the data signal DQ corresponding to the 27 DCA code combinations. For example, the minimum value minEW in the eye window of the four unit intervals UI0, UI1, UI2, and UI3 of the data signal DQ can be calculated as the score of each offset combination.

[0100] In operation S136, the host 1100 determines whether there is a tie between the scores of each offset combination. For example, there may be scores of the same value in the scores of each offset combination. If there is no tie score, the process moves to operation S138. On the other hand, if there are scores of the same value in the scores of each offset combination, the process moves to operation S137.

[0101] In operation S137 , if there is a tie in the scores of the DCA offset combinations, the host 1100 uses a tie-breaking algorithm to select an improved or optimal offset combination.

[0102] In operation S138, the host 1100 may select the DCA offset combination having the maximum score corresponding to each of the generated DCA offset combinations as an improved or optimal DCA offset combination. Alternatively, if a tie occurs, the host 1100 may use a tie-breaking algorithm to select the selected DCA offset combination as an improved or optimal DCA offset combination.

[0103] In operation S139, the host 1100 writes the DCA code corresponding to the determined optimal offset combination into the memory device 1300. Then, the second step STEP_2 of the hierarchical DCA training is completed.

[0104] In the above, the first fine DCA scan performed in the second step STEP_2 of the hierarchical DCA training is briefly described.

[0105] Fig.10 is shown in more detail Figure 6 Flow chart of operation S150. Fig.10 , host 1100 (see Figure 1 )Perform a DCA scan to perform the third step STEP_3 of hierarchical DCA training and determine the optimal DCA offset.

[0106] In operation S151, the host 1100 writes a DCA code corresponding to one of the 27 offset combinations of the internal clock signals QCLK, IBCLK, and QBCLK to the memory device 1300 for the second fine DCA scan. Here, the host 1100 sets the duty cycle of the internal clock signals QCLK, IBCLK, and QBCLK using an offset value (e.g., '-1', '0', and '1') that is smaller than the offset size in the second step STEP_2.

[0107] In operation S152, the memory device 1300 (see Figure 1 ) will adjust the duty ratios of the internal clock signals QCLK, IBCLK, and QBCLK according to the DCA code written in operation S151. The memory device 1300 will output a data pattern corresponding to the DCA code written in operation S151 through the data signal DQ. Then, the host 1100 may measure the data margin or eye window size of the output data pattern.

[0108] In operation S153, it is determined whether the DCA code combination applied in operations S151 and S152 is the last combination (or case) among all the combinations. When all DCA scans of each of the 27 DCA code combinations are completed, the process moves to operation S155. On the other hand, if there is a combination that is not used in the DCA scan among the 27 DCA code combinations, the process moves to operation S154.

[0109] In operation S154, the host 1100 selects a DCA offset combination that is not applied in the second fine DCA scan of the third step STEP_3. Thereafter, the process moves to operation S151, where a DCA scan using the selected DCA offset combination will be performed.

[0110] In operation S155, the host 1100 calculates a score using an output pattern or an eye window of the data signal DQ corresponding to the 27 DCA code combinations.

[0111] In operation S156, the host 1100 determines whether there is a tie between the scores of each offset combination. If there is no tie score, the process moves to operation S158. On the other hand, if there are scores of the same value in the scores of each offset combination, the process moves to operation S157.

[0112] In operation S157, the host 1100 uses a tie-breaking algorithm to select an improved or optimal offset combination. That is, a DCA offset combination having a maximum sum of a minimum value of the even-eye pattern Even_minEW and a minimum value of the odd-eye pattern Odd_minEW in each tie case may be selected as an improved or optimal DCA offset combination.

[0113] In operation S158, the host 1100 may select the DCA offset combination having the maximum score corresponding to each of the generated DCA offset combinations as an improved or optimal DCA offset combination. Alternatively, if a tie occurs, the host 1100 may use a tie-breaking algorithm to select the selected DCA offset combination as an improved or optimal DCA offset combination.

[0114] In operation S159, the host 1100 writes the DCA code corresponding to the determined optimal offset combination into the memory device 1300. Then, the third step STEP_3 of the hierarchical DCA training is completed.

[0115] In the above, the second fine DCA scan process corresponding to the third step STEP_3 of the hierarchical DCA training is briefly described. Through DCA scanning, the duty ratios of the internal clock signals QCLK, IBCLK, and QBCLK of the memory device 1300 are improved or optimized to a range of small offset values ​​(e.g., ', -1', '0', and '1').

[0116] Fig.11 is a diagram showing the mutual operation of the host and the memory device in the first step STEP_1 of the hierarchical DCA training. Fig.11 , host 1100 (see Figure 1 )Perform a coarse DCA scan to perform the first step STEP_1 of hierarchical DCA training and determine the optimal DCA code.

[0117] In operation S121, the host 1100 performs a DCA scan on the memory device 1300 using a DCA code corresponding to one of the coarse offset combinations. To this end, in operation S121a, the host 1100 writes a DCA code corresponding to one of the offset combinations of the internal clock signals QCLK, IBCLK, and QBCLK into the memory device 1300. At this time, the DCA code combination written to the memory device 1300 consists of, for example, at least one of '-4', '0', and '4'. And in operation S121b, the host 1100 receives a data pattern from the memory device 1300. Next, in operation S121c, the host 1100 can use the data pattern output from the memory device 1300 to measure the eye window size.

[0118] Through operations S122 and S123, the host 1100 performs a DCA scan using 27 DCA offset combinations. The processes of operations S122 and S123 are substantially the same as operation S121, except that the DCA code combinations are different. In operations S121 to S123, the host 1100 performs a DCA scan on all 27 DCA code combinations and measures the window size for each case.

[0119] In operation S124, the host 1100 calculates a score corresponding to each of the 27 DCA code combinations according to the data pattern output from the memory device 1300. For example, the minimum value minEW in the eye window of the four unit intervals UI0, UI1, UI2, and UI3 of the data signal DQ may be calculated as the score of each offset combination.

[0120] In operation S125, the host 1100 determines whether there is a tie between the scores of each DCA code combination. If there is no tie score, the process moves to operation S127. On the other hand, if there are scores of the same value in the scores, the process moves to operation S126.

[0121] In operation S126 , the host 1100 uses a tie-breaking algorithm to select an improved or optimal offset combination.

[0122] In operation S127, the host 1100 may select the DCA code combination with the maximum score corresponding to each DCA offset combination as an improved or optimal DCA offset combination. Alternatively, if a tie occurs, the host 1100 may use a tie-breaking algorithm to select the selected DCA offset combination as an improved or optimal DCA offset combination.

[0123] In operation S128, the host 1100 writes the DCA code corresponding to the determined optimal offset combination into the memory device 1300. Then, the first step STEP_1 of the hierarchical DCA training is completed.

[0124] Above, the coarse DCA scanning process for performing the first step STEP_1 of the hierarchical DCA training is briefly described. Through the coarse DCA scanning, the duty ratios of the internal clock signals QCLK, IBCLK, and QBCLK of the memory device 1300 are improved or optimized to large offset values ​​(e.g., "-4", "0", and "4"). The second step STEP_2 and the third step STEP_3 of the hierarchical DCA training will be performed through the same process except that the offset size is different.

[0125] Fig.12 FIG. 4 is a flowchart briefly illustrating a hierarchical DCA training process according to another embodiment of the present invention. Fig.12 , the memory system 1000 of the present invention (see Figure 1 ) can perform a four-step “3-2-1-1” hierarchical DCA training on the internal clock signals QCLK, IBCLK and QBCLK.

[0126] In operation S210, the host 1100 performs the first step STEP_1 of hierarchical DCA training. The host 1100 may set the DCA offset value to one of '-3', '0', and '3' for the DCA scan in the first step STEP_1. Therefore, 27 offset combinations of the internal clock signals QCLK, IBCLK, and QBCLK may be formed. The host 1100 will perform a DCA scan using a DCA code corresponding to each of the 27 offset combinations to proceed with the first step STEP_1. And as a result of the DCA scan, the host 1100 calculates the eye window EW size for each of the 27 offset combinations. The host 1100 generates a score from the eye window EW size for each of the 27 offset combinations. And among the 27 offset combinations, the DCA offset combination with the maximum score is selected as the improved or best offset combination in the first step STEP_1. In addition, if there is a tie between the scores (i.e., minEW) of the 27 DCA offset combinations, the host 1100 can use a tie-breaking algorithm to select an improved or optimal offset combination. Once the improved or optimal offset combination is determined, the host 1100 writes the DCA code corresponding to the improved or optimal offset combination into the memory device 1300.

[0127] In operation S230, the host 1100 performs the second step STEP_2 of the hierarchical DCA training. The host 1100 performs the second step STEP_2 of the hierarchical DCA training using the improved or optimal DCA code generated as a result of the setting of the first step STEP_1. For the DCA scan in the second step STEP_2, the host 1100 can set the offset of the DCA code to one of '-2', '0' and '2'. The host 1100 will perform a DCA scan using the DCA code corresponding to each of the 27 offset combinations to continue the second step STEP_2 of the hierarchical DCA training. The host 1100 generates a score from the eye window EW size for each of the 27 offset combinations. The score generation method is the same as the method applied to the first step STEP_1. In addition, if there is a tie in the scores of the 27 DCA offset combinations, the host 1100 can use a tie-breaking algorithm to select an improved or optimal offset combination. Once an improved or optimal offset combination is determined, the host 1100 writes the DCA code corresponding to the selected optimal offset combination into the memory device 1300 .

[0128] In operation S250, the host 1100 performs the third step STEP_3 of the hierarchical DCA training. The host 1100 proceeds to the third step STEP_3, in which an improved or optimal DCA code is generated as a result of the setting of the second step STEP_2. For the DCA scan in the third step STEP_3, the host 1100 may set the offset of the DCA code to one of '-1', '0', and '1'. The host 1100 will perform a DCA scan using the DCA code corresponding to each of the 27 offset combinations to proceed to the third step STEP_3 of the hierarchical DCA training. The host 1100 generates a score from the eye window EW size for each of the 27 offset combinations. The score generation method is the same as the score generation method applied in the first step STEP_1 or the second step STEP_2. Additionally or alternatively, if there is a tie among the scores of the 27 DCA offset combinations, the host 1100 may use a tie-breaking algorithm to select an improved or optimal offset combination. Once the improved or optimal offset combination is determined, the host 1100 writes the DCA code corresponding to the selected optimal offset combination into the memory device 1300. Then, the duty cycle of each of the internal clock signals QCLK, IBCLK, and QBCLK in the third step STEP_3 may be set to an optimal value.

[0129] In operation S270, the host 1100 performs the fourth step STEP_4 of the hierarchical DCA training. The host 1100 proceeds to the fourth step STEP_4, where an improved or optimal DCA code is generated as a result of the setting of the third step STEP_3. In the fourth step STEP_4, as in the third step STEP_3, the offsets of the DCA code are '-1', '0' and '1'. The host 1100 will perform a DCA scan using the DCA code corresponding to each of the 27 offset combinations to proceed to the fourth step STEP_4 of the hierarchical DCA training. And as a result of the DCA scan, the host 1100 calculates the eye window EW size for each of the 27 offset combinations. The host 1100 generates a score from the eye window EW size for each of the 27 offset combinations. The score generation method is the same as the score generation method applied in the first step STEP_1, the second step STEP_2 or the third step STEP_3. In addition, if there is a tie in the scores of the 27 DCA offset combinations, the host 1100 may use a tie-breaking algorithm to select an improved or optimal offset combination. Once the improved or optimal offset combination is determined, the host 1100 writes the DCA code corresponding to the selected optimal offset combination into the memory device 1300. Then, the duty cycle of each of the internal clock signals QCLK, IBCLK, and QBCLK in the fourth step STEP_4 may be set to an optimal value.

[0130] Above, a hierarchical DCA training method for internal clock signals according to various example embodiments has been described. By gradually reducing the offset size to '3-2-1-1' in each of the four steps, the adjustment accuracy of the duty cycle of the internal clock signals QCLK, IBCLK, and QBCLK can be increased. Alternatively or additionally, if a tie occurs between the scores of each step of the hierarchical DCA training, a tie-breaking algorithm of the present invention can be used to perform high-precision DCA code selection.

[0131] Fig.13 It is visually shown Fig.12 Figure 2. DCA offset combinations for each step in the “3-2-1-1” hierarchical DCA training described in . Fig.13 , shows the DCA offset combination used in each of the four steps (STEP_1, STEP_2, STEP_3 and STEP_4) for hierarchical DCA training in the form of 3D coordinates.

[0132] The offset value of the DCA code written for the DCA scan in the first step STEP_1 can be any one of '-3', '0' and '3'. That is, the offset value of each of the internal clock signals QCLK, IBCLK and QBCLK is mapped to any one of '-3', '0' and '3' on the three axes (Q, IB, QB). In short, the offset combination can correspond to 27 coordinate points. Therefore, the host 1100 will perform a DCA scan using each of the 27 offset combinations to continue the first step STEP_1 of the hierarchical DCA training.

[0133] The eye window size of each of the 27 offset combinations {(-3, -3, -3), (-3, -3, 0), ..., (3, 3, 3)} is calculated based on the result of the DCA scan applying the offset size "3" in the first step STEP_1. And the minimum value minEW of the eye window of the four unit intervals UI0, UI1, UI2 and UI3 of the data signal DQ corresponding to each offset combination is calculated as the score of each offset combination. At this time, among the 27 offset combinations, any one DCA offset combination with the maximum score can be selected as the improved or best offset combination in the first step STEP_1. In addition, if there is a tie between the scores of the 27 DCA offset combinations, a tie-breaking algorithm can be used to select the improved or best offset combination. For example, assume that (3, 0, 0) is selected as the improved or best offset combination. Next, the host 1100 writes the memory device 1300 with the DCA code corresponding to offsets '3', '0', and '0' to set the duty cycle of each of the internal clock signals QCLK, IBCLK, and QBCLK.

[0134] The DCA offset written for the DCA scan in the second step STEP_2 may be any one of '-2', '0', and '2'. That is, the offset value of each of the internal clock signals QCLK, IBCLK, and QBCLK may be mapped to any one of '-2', '0', and '2' on the three axes (Q, IB, QB). The offset combination may correspond to 27 coordinate points having a magnitude of '-2', '2', or '0' in the three-dimensional coordinates of the three axes (Q, IB, QB). Therefore, the host 1100 will perform a DCA scan corresponding to each of the 27 offset combinations to continue the second step STEP_2 of the hierarchical DCA training.

[0135] The eye window size of each of the 27 offset combinations {(-2, -2, -2), (-2, -2, 0), ..., (2, 2, 2)} is calculated based on the result of the DCA scan applying the offset size "2" in the second step STEP_2. And the minimum value minEW of the eye window of the four unit intervals UI0, UI1, UI2 and UI3 of the data signal DQ corresponding to each offset combination is calculated as the score of each offset combination. For example, assume that the improved or optimal offset combination is selected as (2, 2, -2). Next, the host 1100 sends the DCA codes corresponding to offsets '2', '2' and '-2' to the memory device 1300 to set the duty cycle of each of the internal clock signals QCLK, IBCLK and QBCLK to complete the second step STEP_2.

[0136] In the third step STEP_3, any one of '-1', '0' and '1' may be written as the offset value of the DCA code for DCA scanning. That is, the offset of each of the internal clock signals QCLK, IBCLK and QBCLK may be mapped to any one of '-1', '0' and '1' on the three axes (Q, IB, QB). The offset combination may correspond to 27 coordinate points having an amplitude of '-1', '1' or '0' in the three-dimensional coordinates of the three axes (Q, IB, QB). Therefore, the host 1100 will perform a DCA scan corresponding to each of the 27 offset combinations to continue the third step STEP_3 of the hierarchical DCA training.

[0137] The eye window size for each of the 27 offset combinations {(-1, -1, -1), (-1, -1, 0), ..., (1, 1, 1)} is calculated based on the result of applying the DCA scan of the offset size '1' in the third step STEP_3. And the minimum value minEW of the eye window of the four unit intervals UI0, UI1, UI2 and UI3 of the data signal DQ corresponding to each offset combination is calculated as the score of each offset combination. At this time, among the 27 offset combinations, any one DCA offset combination with the maximum score can be selected as the improved or best offset combination in the third step STEP_3. In addition, if there is a tie between the scores of the 27 DCA offset combinations, a tie-breaking algorithm can be used to select an improved or best offset combination. For example, assume that the improved or best offset combination is selected as (0, 1, -1). Next, the host 1100 writes the DCA codes corresponding to the offset values ​​'0', '1' and '-1' to the memory device 1300 to complete the third step STEP_3.

[0138] In the fourth step STEP_4, the DCA scan is performed again using the same offset size as the third step STEP_3. The fourth step STEP_4 can be considered as a DCA scan operation substantially the same as the third step STEP_3, except that the memory device 1300 is set to the DCA code determined in the third step STEP_3. Assume that as a result of the DCA scan, an improved or optimal offset combination (-1, 1, 0) is selected. Then, the host 1100 sends the DCA code corresponding to the offset values ​​'-1', '1' and '0' to the memory device 1300 to complete the fourth step STEP_4. When the improved or optimal DCA code is determined by the DCA scan of the fourth step STEP_4 and the memory device 1300 is set using the determined optimal DCA code, the hierarchical DCA training of the present invention is completed.

[0139] The above briefly describes the gradual reduction process of the DCA code used in the hierarchical DCA training. By reducing the offset size in each of the four stages in a '3-2-1-1' manner, the adjustment accuracy of the duty cycle of the internal clock signals QCLK, IBCLK and QBCLK can be increased. In addition, if a tie occurs between the scores of each step of the hierarchical DCA training, a high-precision DCA code selection can be performed using the tie-breaking algorithm of the present invention.

[0140] Fig.14 is a block diagram showing a memory system according to another embodiment of the present invention. Fig.14 , the memory system 2000 includes a host 2100 and a memory device 2300. The host 2100 performs DCA training with reference to DCA step information DCA_SI provided by the memory device 2300.

[0141] The host 2100 reads the DCA step information DCA_SI from the DCA step register 2390 of the memory device 2300 before performing DCA training. The DCA step information DCA_SI may include information on whether to continue the hierarchical DCA training method in a "4-2-1" manner or in a "3-2-1-1" manner. Alternatively, the DCA step information DCA_SI may store the final DCA code obtained as a result of the hierarchical DCA training performed in the test phase in the memory device 2300. In this case, the host 2100 may write the final DCA code to the memory device 2300 without separate DCA training.

[0142] The memory device 2300 may include a duty cycle regulator 2350, a clock generation circuit 2360, an input / output circuit 2370, and a DCA step register 2390. The memory device 2300 is substantially similar to Figure 1 The memory device 1300 is the same as the memory device 1300 except that it includes the DCA step register 2390. Therefore, a detailed description of the memory device 2300 will be omitted.

[0143] Fig.15 It is shown in Fig.14 Flowchart of hierarchical DCA training performed in a memory system of FIG. Fig.15 , host 2100 (see Fig.14 ) can be in the slave memory device 2300 (see Fig.14 )After reading the DCA step information DCA_SI, select the hierarchical DCA training method.

[0144] In operation S310, the host 2100 reads DCA step information DCA_SI from the DCA step register 2390 of the memory device 2300. The DCA step information DCA_SI includes information on a hierarchical DCA training method of the memory device 2300.

[0145] In operation S320, the host 2100 checks the hierarchical DCA training method using the DCA step information DCA_SI. If the hierarchical DCA training method stored in the DCA step information DCA_SI is 3 steps or '4-2-1' mode, the process moves to operation S330. On the other hand, if the hierarchical DCA training method indicated in the DCA step information DCA_SI is 4 steps or '3-2-1-1' mode, the process moves to operation S340.

[0146] In operation S330, the host 2100 performs 3-step hierarchical DCA training on the memory device 2300. For example, the host 2100 will perform hierarchical DCA training in a "4-2-1" manner, which reduces the input size of the DCA code in three steps.

[0147] In operation S340, the host 2100 performs 4-step hierarchical DCA training on the memory device 2300. For example, the host 2100 will apply a "3-2-1-1" style of hierarchical DCA training to the memory device 2300, which reduces the input size of the DCA code to 4 steps.

[0148] In operation S350, the host 2100 uses the improved or optimal DCA code determined by the 3-step or 4-step hierarchical DCA training to set up the memory device 2300. Once the setting of the memory device 1300 is completed, the hierarchical DCA training of the present invention is completed.

[0149] Fig.16 3 is a block diagram showing a computing system 3000 according to another embodiment of the present invention. Fig.16 , the computing system 3000 may include a processor 3050 , a chipset 3100 , an input / output controller 3200 , and a memory device 3300 .

[0150] The processor 3050 executes software (applications, operating systems, device drivers, etc.) running on the computing system 3000. The processor 3050 will execute an operating system (OS, not shown) loaded into the memory device 2300. The processor 3050 will execute various application programs to be run based on the operating system.

[0151] The chipset 3100 controls connections between the processor 3050, the memory device 3300, and the input / output controller 3200. The chipset 3100 may include a north bridge for controlling the processor 3050 and the memory device 3300 and / or a south bridge for controlling the input / output controller 3200. Alternatively, the chipset 3100 may be provided without distinguishing between the functions of the north bridge and the south bridge.

[0152] In particular, the chipset 3100 may apply the hierarchical DCA training of the present invention to the memory device 3300. To this end, the chipset 3100 includes a memory controller 3150 that performs DCA training. The memory controller 3150 may increase the duty cycle adjustment accuracy by applying 3-step or 4-step hierarchical DCA training to the memory device 3300. In addition, if a tie occurs between the scores at each step of the hierarchical DCA training, a tie-breaking algorithm using various example embodiments may be used to perform high-precision DCA code selection. Here, it will be well understood that the functions of the chipset 3150 or the functions or configurations of the memory controller 3150 may be integrated into the processor 3050.

[0153] The input / output controller 3200 controls the data flow between the processor 3050 and one or more input / output interfaces (e.g., wired and wireless network interfaces) and I / O devices. For example, the input / output controller 3200 may control the data flow between the processor 3050 and at least one of a keyboard, a mouse, a trackball, a pen, a touch screen, a display, a monitor, a speaker, and a printer. The memory device 3300 may be connected to the processor 3050. Figure 1 The memory device 1300 or Fig.14 The memory device 2300 operates substantially the same.

[0154] Fig.17 is a block diagram illustrating a mobile system according to various example embodiments. Fig.17 , the computing system 4000 may include an application processor 4100 and a memory device 4300. Here, the memory device 4300 and Figure 1 The memory device 1300 or Fig.14 The memory device 4300 is substantially the same as the memory device 2300. In this embodiment, the memory device 4300 may be implemented as a low power DDR (LPDDR) memory improved or optimized for a mobile environment.

[0155] The application processor 4100 may apply the hierarchical DCA training of various example embodiments to the memory device 4300. The application processor 4100 may increase the duty cycle adjustment accuracy by applying the 3-step or 4-step hierarchical DCA training to the memory device 4300. In addition, if a tie occurs between the scores of each step of the hierarchical DCA training, high-precision DCA code selection may be performed using the tie-breaking algorithm of the present invention.

[0156] Fig.18 is a block diagram illustrating an embodiment of a memory system using hierarchical DCA training. Fig.18 , the memory system 5000 may include an interface (I / F) 5050 , a memory controller 5100 , and a plurality of DRAMs 5300 .

[0157] The hierarchical DCA training technology of the present invention may be applied within the memory system 5000. That is, the memory controller 5100 may include a hierarchical DCA training logic 5150 capable of performing the hierarchical DCA training of the present invention. For example, the memory system 5000 may be implemented with a Computer Express Link (CXL) DRAM. In this case, the memory controller 5100 may ensure the signal integrity of the plurality of DRAMs 5300 by performing hierarchical DCA training. Alternatively or additionally, a method of using a DCA code obtained as a result of training as a DCA code during initialization or booting without additional training or with a reduced probability of using additional training may be applied to the memory system 5000.

[0158] Fig.19 is a diagram illustrating a memory system according to various example embodiments. Fig.19 , shows an accelerated double-sided memory module (6000, accelerated DIMM: hereinafter referred to as AxDIMM) as an example of a memory system equipped with an artificial intelligence engine. AxDIMM 6000 may include a plurality of DRAM chips (6110 to 6180), an AxDIMM buffer 6200, and an FPGA 6300.

[0159] When the AxDIMM 6000 is booted, firmware or software that performs the hierarchical DCA training of the present invention may be loaded into the AxDIMM buffer 6200. Alternatively, as shown, the hierarchical DCA training of the present invention may be applied to multiple DRAM chips 6110 to 6180 via an FPGA 6300 including the hierarchical DCA training logic 6350 of the present invention.

[0160] Fig. 20 is a cross-sectional diagram showing a memory system according to various example embodiments. Fig. 20 , a memory system 7000 implemented as a stacked semiconductor device includes a PCB substrate 7100 , an interposer 7150 , a processor 7200 , a logic die 7300 , and a plurality of DRAM dies 7410 , 7420 , 7430 , and 7440 .

[0161] The memory system 7000 uses an interposer 7150 to connect the high bandwidth memory 7400 and the processor 7200. The interposer 7150 is disposed on the upper portion of the PCB board 7100 and is electrically connected to the PCB board 7100 through a flip chip bump FB. The interposer 7150 can connect the logic die 7300 and the processor 7200. The interposer 7150 connects the physical layer 7350 of the logic die 7300 and the physical layer 7250 of the processor 7200, and can provide a physical path formed using a conductive material. Therefore, the logic die 7300 and the processor 7200 can send and receive signals to each other through the interposer 7150.

[0162] In some example embodiments, the processor 7200 may write a DCA code to adjust the duty cycle of the four orthogonal internal clock signals ICLK, QCLK, IBCLK, and QBCLK of the high bandwidth memory 7400. The processor 7200 may apply hierarchical DCA training to the internal clock signals. In addition, if a tie occurs between the scores of each step of the hierarchical DCA training, a high-precision DCA code selection may be performed using the tie-breaking algorithm of the present invention.

[0163] Any of the elements and / or functional blocks disclosed above may include or be implemented in a processing circuit, such as hardware including a logic circuit; a hardware / software combination, such as a processor that executes software; or a combination thereof. For example, the processing circuit may more specifically include, but is not limited to, a central processing unit (CPU), an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), a system on a chip (SoC), a programmable logic unit, a microprocessor, an application specific integrated circuit (ASIC), etc. The processing circuit may include an electrical component such as at least one of a transistor, a resistor, a capacitor, etc. The processing circuit may include an electrical component such as a logic gate, which includes at least one of the following: an AND gate, an OR gate, a NAND gate, a NOT gate, etc.

[0164] In addition, any or all elements described with reference to the various figures may communicate with any or all other elements described with reference to the various figures. For example, in some figures, any element may communicate with any or all other elements in a one-way and / or two-way and / or broadcast manner to transmit and / or exchange and / or receive information such as, but not limited to, data and / or commands via a bus such as a wireless and / or wired bus (not shown) in a serial and / or parallel manner. The information may be in various formats encoded, such as analog and / or digital formats.

[0165] The above are specific example embodiments for carrying out the inventive concept. In addition to the above example embodiments, the inventive concept may include simple design changes and / or easily changeable embodiments. Alternatively or additionally, the inventive concept may include techniques that can be easily modified and implemented using the example embodiments. Therefore, the scope of the inventive concept should not be limited to the above example embodiments, and should be defined by the claims of the example embodiments and the equivalents of the claims and the claims described later. In addition, the example embodiments are not necessarily mutually exclusive. For example, some example embodiments may include one or more features described with reference to one or more figures, and may also include one or more other features described with reference to one or more other figures.

Claims

1. A method for training a memory device to adjust the eye window of a data signal according to a written duty cycle adjuster (DCA) code, the method include: performing a first DCA training step on the memory device using a first DCA code combination having a first offset size, the first DCA training adjusting a duty cycle of each of the internal clock signals QCLK, IBCLK, and QBCLK; performing a second DCA training step on the memory device using a second DCA code combination having a second offset size under the condition that the memory device is set according to the result of the first DCA training step; and performing a third DCA training step on the memory device using a third DCA code combination having a third offset size under the condition that the memory device is set according to the result of the second DCA training step, Wherein, in the first DCA training step to the third DCA training step, the method includes calculating a score of each of the first DCA code combination to the third DCA code combination based on an eye window size of the data signal, and in response to a tie between the scores, the method includes selecting the DCA code combination based on the sum of an even eye window minimum value and an odd eye window minimum value of the data signal.

2. The method according to claim 1, in, The first DCA training step comprises: writing each of the first DCA code combinations into the memory device, and measuring an eye window size of a data signal from the memory device corresponding to each of the first DCA code combinations; and A score for each of the first DCA code combinations is calculated based on the eye window size. 3 . The method of claim 2 , wherein the score corresponds to a minimum value among eye window sizes of four consecutive unit intervals of the data signal.

4. The method of claim 1, wherein the first offset size is at least one of '-4', '0', and '4', the second offset size is at least one of '-2', '0', and '2', and the third offset size is at least one of '-1', '0', and '1'.

5. The method according to claim 1, further comprising: include: A fourth DCA training step is performed on the memory device using a fourth DCA code combination having a fourth offset size.

6. The method according to claim 5, wherein the first offset size corresponds to at least one of '-3', 0' and '3', the second offset size corresponds to at least one of '-2', '0' and '2', the third offset size corresponds to at least one of '-1', '0' and '1', and the fourth offset size corresponds to at least one of '-1', '0' and '1'.

7. The method according to claim 1, in, Each of the first DCA code combination, the second DCA code combination, and the third DCA code combination includes 27 cases.

8. The method according to claim 1, in, The data signals in the first DCA training step to the third DCA training step are output based on a pseudo random binary sequence PRBS.

9. A memory system configured to adjust the duty cycle of four internal clock signals by a duty cycle adjuster (DCA) code, include: a memory device configured to generate a data signal or a data strobe signal having a duty cycle that changes in response to a DCA code; and a memory controller configured to perform hierarchical DCA training on the memory device to select an improved DCA code and to configure the memory device using the improved DCA code, The memory controller is configured to perform a first DCA training step on the memory device using a first DCA code combination having a first offset size to adjust the duty cycle of each of the internal clock signals, and to perform a second DCA training step on the memory device using a second DCA code combination having a second offset size under the condition that the memory device is set according to the result of the first DCA training step.

10. The memory system according to claim 9, in, The memory controller is configured to perform a third DCA training step on the memory device using a third DCA code combination having a third offset size under the condition that the memory device is set according to a result of the second DCA training step.

11. The memory system according to claim 10, in, The memory controller is configured to calculate a score for each of the first to third DCA code combinations based on an eye window size of the data signal in the first to third DCA training steps.

12. The memory system according to claim 11, in, In response to a tie occurring between the scores, the memory controller is configured to select a modified DCA code combination based on a sum of a minimum even eye window value and a minimum odd eye window value of the data signal.

13. The memory system of claim 10, wherein the first offset size corresponds to at least one of '-4', '0', and '4', the second offset size corresponds to at least one of '-2', '0', and '2', and the third offset size corresponds to at least one of '-1', '0', and '1'.

14. The memory system according to claim 10, in, The memory controller is configured to perform a fourth DCA training step on the memory device using a fourth DCA code combination having a fourth offset size under the condition that the memory device is set according to a result of the third DCA training step.

15. The memory system of claim 14, wherein the first offset size corresponds to at least one of '-3', '0', and '3', the second offset size corresponds to at least one of '-2', '0', and '2', the third offset size corresponds to at least one of '-1', '0', and '1', and the fourth offset size corresponds to at least one of '-1', '0', and '1'.

16. The memory system according to claim 12, in, The memory device includes a DCA step register configured to store information of a number of steps in the hierarchical DCA training.

17. The memory system according to claim 16, in, The memory controller is configured to read the DCA step register to determine a number of steps for hierarchical DCA training.

18. A method of training a memory device, include: performing a first duty cycle adjuster (DCA) training on the memory device using a first DCA code combination having a first offset size to adjust a duty cycle of each of a plurality of internal clock signals; as well as performing a second DCA training on the memory device using a second DCA code combination having a second offset size under the condition that the memory device is set according to the result of the first DCA training, wherein a maximum value of the first offset size is greater than a maximum value of the second offset size, and the method includes calculating a score based on an eye window size of a data signal corresponding to each of a first DCA code combination to a second DCA code combination in a first DCA training to a second DCA training.

19. The method of claim 18, wherein in response to a tie occurring between the scores, the method comprises selecting a modified DCA code combination based on a sum of an even eye window minimum value and an odd eye window minimum value of the data signal.

20. The method according to claim 18, in, The method includes outputting the data signal from the memory device based on a pseudo-random binary sequence (PRBS) in the first DCA training to the second DCA training.

Citation Information

Patent Citations

  • Method and apparatus for image processing

    KR1020230071689A

  • Light emitting device including organometallic compound, electronic apparatus including the light emitting device and the organometallic compound

    KR1020230115410A