Memory control device
Through detection and converter optimization of BL length, the problem of low access efficiency during read/write switching in the BG compatible DRAM standard is solved, and more efficient DRAM access is achieved.
Patent Information
- Application Number
- CN202380028757.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2022-03-31
- Filing Date
- 2023-03-09
- Publication Date
- 2024-11-12
AI Technical Summary
In DRAM standards that have BG and are compatible with multiple BL lengths, it is difficult for the prior art to effectively access DRAM during the read/write switching process, and the selection of BL length is limited, affecting access efficiency.
The detector detects the switching between the read library group interleaving and the write library group interleaving in the memory access request, and is based on the number of memory access requests in the read or write library group interleaving and the timing information of the previous access request, Use a converter to convert the BL length to optimize the BL length selection.
In the DRAM standard with BG, DRAM access efficiency during read/write switching is improved, the appropriate selection of BL length is ensured, and the overall efficiency of memory access is improved.
Smart Images

Figure CN118946879A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to a memory control device. Background Art
[0002] Various measures have been proposed to more efficiently access DRAM (Dynamic Random Access Memory) (for example, see Patent Document 1 and Patent Document 2).
[0003] Reference List
[0004] Patent Literature
[0005] Patent Document 1: Japanese Unexamined Patent Application Publication No. 2021-157295
[0006] Patent Document 2: Japanese Unexamined Patent Application Publication No. 2006-252173 Summary of the invention
[0007] By the way, in existing DRAM technology, the larger the BL (burst length) length, the more efficient the access to DRAM. However, in DRAM standards such as LPDDR (Low Power Double Data Rate) or LPDDR5X that have BG (bank group) and are compatible with multiple BL lengths, access to DRAM is sometimes more efficient even when the BL length is short. Therefore, it is desirable to provide a memory control device that can access DRAM more efficiently in a DRAM standard that has BG and is compatible with multiple BL lengths.
[0008] A memory control device according to an embodiment of the present disclosure includes a detector and a converter. The detector detects switching between read bank group interleaving and write request or switching between write bank group interleaving and read request in multiple memory access requests received for memory access. The converter converts the BL length of the memory access request included in the read bank group interleaving or write bank group interleaving based on the number of memory access requests in the read bank group interleaving or write bank group interleaving and the timing information of the command corresponding to the memory access request immediately before the read bank group interleaving or write bank group interleaving.
[0009] In a memory control device according to an embodiment of the present disclosure, based on the number of memory access requests in read bank group interleaving or write bank group interleaving and the timing information of the command corresponding to the memory access request immediately before the read bank group interleaving or write bank group interleaving, the BL length of the memory access request included in the read bank group interleaving or write bank group interleaving is converted. This makes it possible to select the BL length in consideration of the access efficiency of the DRAM. BRIEF DESCRIPTION OF THE DRAWINGS
[0010] Figure 1is a diagram illustrating an embodiment of the operation of a command scheduler.
[0011] Figure 2 Is shown following Figure 1 An illustration of an embodiment of the following operation.
[0012] Figure 3 is a diagram illustrating an embodiment of the operation of a command scheduler.
[0013] Figure 4 is a diagram illustrating an embodiment of the operation of a command scheduler.
[0014] Figure 5 is a diagram showing a schematic configuration example of an information processing system including a memory control device according to one embodiment of the present disclosure.
[0015] Figure 6 is a diagram illustrating the concept of a FIFL memory.
[0016] Figure 7 is a diagram illustrating an embodiment of data stored in a FIFO memory.
[0017] Figure 8 is a diagram illustrating an embodiment of the operation of a RW switching detector.
[0018] Fig. 9 is a diagram illustrating conditions for determining whether to convert a BL length of a command included in 2BG interleaving.
[0019] Fig.10 It is shown Fig. 9 Graph of the conditions.
[0020] Fig.11 It is shown Fig. 9 Graph of the conditions.
[0021] Fig.12 It is shown Fig. 9 Graph of the conditions.
[0022] Fig.13 is a diagram showing an embodiment of the operation of the BL conversion determination unit.
[0023] Fig.14 is a diagram showing an embodiment of the operation of the BL conversion determination unit.
[0024] Fig.15 is a diagram showing an embodiment of the operation of the BL conversion determination unit.
[0025] Fig.16 is a diagram illustrating an embodiment of the operation of a BL converter.
[0026] Fig.17 is a diagram illustrating an embodiment of data stored in a FIFO memory.
[0027] Fig.18 is a diagram illustrating an embodiment of a sequence of memory access requests after conversion is performed by a BL converter.
[0028] Fig.19 is a diagram illustrating an embodiment of data stored in a FIFO memory.
[0029] Fig. 20 is a diagram illustrating an embodiment of a sequence of memory access requests after conversion is performed by a BL converter.
[0030] Fig.21 is a diagram illustrating an embodiment of an out-of-order execution process.
[0031] Fig. 22 is a diagram illustrating an embodiment of an out-of-order execution process.
[0032] Fig.23 is a diagram showing the operation of the command scheduler when a sequence of access requests changing from a read request to a write request is input.
[0033] Fig.24 is a diagram showing the operation of the command scheduler when a sequence of access requests changing from a write request to a read request is input.
[0034] Fig.25 It is shown Figure 5 FIG. 2 is a diagram of a modified example of the schematic configuration of an information processing system.
[0035] Fig.26 It is shown Figure 5 FIG. 2 is a diagram of a modified example of the schematic configuration of an information processing system.
[0036] Fig. 27 is a diagram showing conditions for determining whether to convert the BL length of a command included in 2BG interleaving.
[0037] Fig.28 It is shown Fig. 27 Graph of the conditions.
[0038] Fig.29 is a diagram illustrating an embodiment of data stored in a FIFO memory.
[0039] Fig.30 is a diagram showing an embodiment of a sequence of access requests after conversion is performed by a BL converter. DETAILED DESCRIPTION
[0040] Hereinafter, some embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. However, it should be noted that the embodiments described below are exemplary only and are not intended to exclude the application of various modifications or technologies not explicitly described below. Without departing from the spirit of the present technology, the present technology can be implemented by making various deformations (for example, by combining the embodiments). In addition, in the following description of the accompanying drawings, the same or similar parts are marked with the same or similar figure signs. The accompanying drawings are schematic and do not necessarily correspond to actual sizes or proportions. In some cases, the accompanying drawings may include parts that are different in their mutual size relationship or proportion.
[0041] <1. Challenges in the latest generation of standards>
[0042] As a memory system, synchronous DRAM (SDRAM: Synchronous Dynamic Random Access Memory) which is advantageous in terms of price, bus band, and capacity has been widely used. SDRAM is a DRAM that operates in synchronization with a clock signal, and generally includes a plurality of banks.
[0043] In an SDRAM having such a configuration, when the same bank is accessed continuously, the efficiency of accessing the SDRAM is significantly reduced. On the contrary, as a technology for improving the access efficiency to the DRAM, a technology for maintaining multiple access requests to the DRAM and outputting the access request to a bank different from the bank accessed last time among the maintained multiple access requests has been proposed.
[0044] In existing DRAM technology, the larger the BL (burst length) length, the higher the DRAM access efficiency. One of the reasons is that a larger BL length increases the amount of data to be transmitted for each use time of the address command line, thereby reducing the use time of the command address line. However, in the latest DRAM standards such as LPDDR5 or LPDDR5X, access to DRAM is sometimes more efficient even when the BL length is short. LPDDR5 and LPDDR5X are DRAM standards specified by JEDEC (Join Electron Device Engineering Council).
[0045] Figure 1 and Figure 2 Both show embodiments of the operation of the command scheduler when two bank groups are interleaved. Figure 2 Shows Figure 1 An example of the following operation. Figure 1 and Figure 2 The upper part of FIG. 1 shows an embodiment of the operation of the command scheduler when the BL length is BL16. Figure 1 and Figure 2 The lower part of FIG. 1 shows an embodiment of the operation of the command scheduler when the BL length is BL32. Figure 1 and Figure 2 The upper and lower parts of the diagram show commands output to the address command line, which are divided for each bank group and bank address, and the data output to the data line is represented in a different mode for each bank group and bank address. Note that the blackened part of the data line indicates that no data is output.
[0046] It should be noted that Figure 1 and Figure 2 The various symbols in represent the following:
[0047] -ACT: Activate command
[0048] -READ16: Read command with BL16
[0049] -READ32: Read command with BL32
[0050] -PRE: Precharge command
[0051] in addition, Figure 1 and Figure 2 The various parameters in are defined as follows:
[0052] - Command clock = 800MHz
[0053] -tRRD=3nCK
[0054] -tFAW=15ns
[0055] -tRCD=15nCK
[0056] -tRCD_W=7nCK
[0057] -nRBTP=4nCK
[0058] -tRPpb=15nCK
[0059] -tWR=27nCK
[0060] -RL=18
[0061] (In the figure, the delay from read to DQ is indicated as RL=0.)
[0062] -WL=9
[0063] (In the figure, the delay from write to DQ is indicated as WL=0.)
[0064] -tRRD: shortest interval between ACT-ACT
[0065] -tFAW: There may be up to ACT cycles
[0066] -tRCD: shortest interval between ACT-REAAD or ACT-MASKEDWRITE
[0067] -tRCD_W: shortest interval between ACT-WRITE
[0068] -nRBTP: The shortest interval between READ-burst end-PRE
[0069] -tRPPb: shortest interval between PRE-ACT
[0070] -tWR: WRTE recovery time
[0071] from Figure 1 and Figure 2 It can be seen that in the first BG interleave, the data transmission using BL16 ends Δt1 earlier than BL32. However, in the second BG interleave, it can be seen that the data transmission using BL32 starts Δt2 (=1nCK) earlier than BL16. One reason for this is that when using BL16, two or more groups must be opened to perform BG interleave, and the shortest interval between the first two read commands is tRRD or greater. Therefore, it can be seen that as the BG interleave is repeated for the third, fourth, and so on, the start time of the BG interleave using BL32 becomes earlier in increments of 1nCK compared to BL16. In addition, in the fourth BG interleave, the data transmission waiting time reaches the equivalent level of BL16 and BL32. Therefore, it can be seen that, in general, BL32 has a higher access efficiency to DRAM than BL16.
[0072] Figure 3 and Figure 4 Each shows an embodiment of the operation of the command scheduler. Figure 3 and Figure 4 The upper part of FIG. 1 shows an embodiment of the command scheduler when the BL length is BL16. Figure 3 and Figure 4 The lower part of FIG. 1 shows an embodiment of the command scheduler when the BL length is BL32.
[0073] It should be noted that Figure 3 and Figure 4 The various symbols in refer to the following:
[0074] -ACT: Activate command
[0075] -READ16: Read command with BL16
[0076] -READ32: Read command with BL32
[0077] -PRE: Precharge command
[0078] -Write: Write command
[0079] The data lines of DRAM are shared by read and write. For this reason, a penalty time is required when switching from read to write or from write to read. Here, in BG mode, when BL / n_min of BL16 is 2, BL / n_min of BL32 is 6. Therefore, if Figure 3 and Figure 4 As shown, it can be seen that, according to the usage of the data line, the timing of the write command issued by BL16 is earlier than that of the write command issued by BL32. As such, it can be seen that in the latest generation standard, the access efficiency of BL16 to DRAM is sometimes more efficient than that of BL32.
[0080] From the above, it can be seen that BL32 is more efficient than BL16 in accessing DRAM without read / write switching, and BL16 is more efficient than BL32 in accessing DRAM with read / write switching. Therefore, the inventor of the present application proposes a technique for selecting BL length below, which improves the efficiency of accessing DRAM during read / write switching.
[0081] <2. Implementation Method>
[0082] [Configuration]
[0083] Figure 5 An example of a schematic configuration of an information processing system including a memory control device according to an embodiment of the present disclosure is shown. Figure 5 As shown, for example, the information processing system includes a plurality of initiators 10 , an adjustment unit 30 , a memory controller 40 , and a DRAM 50 .
[0084] (DRAM 50)
[0085] DRAM 50 is a DRAM that complies with LPDDR5 or LPDDR5X. Figure 5 As shown, for example, two bank groups BG0 and GB1 are assigned to the DRAM 50. For example, four banks Bank0, Bank1, Bank2, Bank3 are assigned to each bank group BG0, GB1. Note that the number of bank groups and banks in the DRAM 50 is not limited to the above embodiment.
[0086] LPDDR5 or LPDDR5X allows dynamic driving frequency to be changed. In addition, there are multiple bank modes in LPDDR5 or LPDDR5X, the bank mode is the use of bank configuration, and the specification that the bank mode changes according to the driving frequency has been standardized.
[0087] In the case where BL16 and BL32 are used differently in units of commands, and BL16 and BL32 are the minimum burst units of DRAM 50, the 16B (bank) mode will be used for low-speed driving frequency, and the BG (bank group) mode will be used for high-speed driving frequency. In the BG mode, for data transmission of BL32, data is transmitted in response to a command after a BL16 (1nCK) cycle. This transmission penalty on the data line is a rule unique to the BG mode. However, because the transmission data of another bank group can be included in this transmission penalty cycle, the transmission data can be continuously released on the data line.
[0088] (Starter 10)
[0089] The plurality of initiators 10 writes or reads data to or from the DRAM 50 via the adjustment unit 30 and the memory controller 40. The initiator 10 is, for example, a central processing unit (CPU: Central Processing Unit) or a functional block or the like.
[0090] Each initiator 10 issues a memory access request to write data to the DRAM 50 or read data from the DRAM 50, and outputs the memory access request to the adjustment unit 30. For example, the memory access request includes a logical address in a virtual storage area given to each initiator 10, a BL length as the length of the data to be accessed, identification information for identifying the initiator 10, and a transmission direction. The transmission direction mentioned here is an indication of whether the memory access request is a write request to write data or a read request to read data. According to the data output command from the adjustment unit 30, each initiator 10 outputs the write data to be written to the DRAM 50 to the adjustment unit 30. For example, each initiator 10 communicates with the adjustment unit 30 using a protocol defined by AMBA (Advanced Microcontroller Bus Architecture) (e.g., AXI (Advanced eXtensible Interface) protocol).
[0091] (Adjustment unit 30)
[0092] like Figure 5 As shown, for example, the adjustment unit 30 includes an arbitration section 31 , an RW switching detector 32 , a buffer 33 , a BL switching determination section 34 , and a BL converter 35 .
[0093] The arbitration section 31 converts the logical address included in the memory access request output from each initiator 10 into a physical address corresponding to the DRAM 50. The physical address mentioned here is an address indicating the bank, row, and column constituting the DRAM 50, and refers to a bank address, a row address, and a column address. By converting from the logical address to the physical address in this manner, the bank address, the row address, and the column address in the DRAM 50 are indicated in the converted memory access request.
[0094] The arbitration section 31 further performs arbitration (arbitration) based on the physical addresses indicated in the multiple memory access requests obtained from the multiple initiators 10. For example, in the case where the arbitration section 31 receives memory access requests from the respective initiators 10 at the same time, the arbitration section 31 suppresses the output of a memory access request having the same bank address as the bank address of the last memory access request output to the RW switching detector 32. That is, in the case where the arbitration section 31 receives multiple memory access requests at the same time, the arbitration section 31 outputs a memory access request having a bank address different from the last memory access request output to the RW switching detector 32 to the RW switching detector 32. In this way, the arbitration section 31 adjusts the order in which the multiple memory access requests are output to the RW switching detector 32 by using the interleaving technique. In the BG mode, the arbitration section 31 further adjusts the order in which the multiple memory access requests are output to the RW switching detector 32 by using the interleaving technique for each bank group. At this time, the arbitration section 31 adds the bank group identifier to the memory access request. When a write request is output to the RW switching detector 32 as a memory access request, the arbitration section 31 instructs the initiator 10 identified by the identification information indicated in the memory access request to output write data corresponding to the memory access request.
[0095] Each time the RW switching detector 32 receives a memory access request from the arbitration section 31, the RW switching detector 32 stores the received memory access request in the buffer 33. The buffer 33 includes, for example, a FIFO (First In First Out) memory 33A, such as Figure 6 As shown in . In the FIFO memory 33A, a plurality of memory access requests are stored in the order of storage. When a new memory access request is stored in the FIFO memory 33A, the memory access request having the earliest storage order is output from the FIFO memory 33A.
[0096] The FIFO memory 33A is associated with the bank group information bgint and the partition necessity information dev of each stored memory access request. Each time the FIFO memory 33A stores a new memory access request, the FIFO memory 33A outputs the memory access request with the earliest storage order in the FIFO memory 33A to the BL converter 35 together with the bank information bgint and the partition necessity information dev.
[0097] The bank group information bgint is a flag for determining whether a memory access request among a plurality of memory access requests stored in the FIFO memory 33A is a memory access request of BG interleaving immediately before the read / write switching. For example, when the bank group information bgint is "1", it indicates that the memory access request is a memory access request of BG interleaving immediately before the read / write switching. In addition, when the bank group information bgint is "0", it indicates that the memory access request is not a memory access request of BG interleaving immediately before the read / write switching.
[0098] The division necessity information dev is a flag for determining whether to convert the BL length of the memory access request of the BG interleaving included in the FIFO memory 33A from BL32 to BL16. For example, when the division necessity information dev is "1", it means that the BL length of the memory access request needs to be converted from BL32 to BL16. In addition, for example, when the division necessity information dev is "0", it means that the BL length of the memory access request does not need to be converted from BL32 to BL16.
[0099] The RW switching detector 32 detects a switch from a read BG interleave (read request) to a write request or a switch from a write BG interleave (write request) to a read request among a plurality of memory access requests stored in the buffer 33. For example, assuming that Figure 7 The plurality of memory access requests shown are stored in the FIFO memory 33A. At this time, the RW switching detector 32 detects switching between the read request and the write request in the memory access request (read request) having the fifth earliest storage order and the memory access request (write request) having the sixth earliest storage order.
[0100] Figure 8An embodiment of a process for detecting the above-mentioned switching in the RW switching detector 32 is shown. The RW switching detector 32 sets the parameter i representing the storage order to 0 (step S101). Next, the RW switching detector 32 determines whether the parameter i is less than the latest storage order N-1 (step S102). As a result, in the case where i<N-1 is satisfied (step S102; Y), the RW switching detector 32 compares the access request of the i-th storage order and the access request of the (i+1)-th storage order to determine whether the two access requests are read requests or write requests (step S103). As a result, in the case where both are read requests or write requests (step S103; Y), the RW switching detector 32 adds 1 to the parameter i (step S104) and returns to step S102. In the case where neither is a read request or a write request (step S103; N), the RW switching detector 32 determines that the switch from a read request to a write request or from a write request to a read request will be performed in a memory access request with the i-th storage order and a memory access request with the (i+1)-th storage order. Then, the RW switching detector 32 writes "1" of the i-th storage order into the library group information bgint corresponding to the memory access request, and enters the BL conversion necessity determination (step S105). At this time, the RW switching detector 32 outputs the parameter i as information about the above-mentioned switch (RW switching information) to the BL conversion determination unit 34. It should be noted that in the case where i<N-1 is not satisfied in step S102 (step S102; N), the RW switching detector 32 determines that there is no switch from a read request to a write request or from a write request to a read request, and ends the processing.
[0101] The BL conversion determination unit 34 and the BL converter 35 convert the BL length of the memory access request included in the 2BG interleaving IL-a (read BG group interleaving or write BG interleaving) based on the number of memory access requests in the 2BG interleaving IL-a immediately before the above-mentioned switching and the timing information of the command Cmd-a (first command) corresponding to the memory access request Ra immediately before the 2BG interleaving IL-a. It should be noted that "2BG interleaving" means interleaving two bank groups. At this time, the BL conversion determination unit 34 and the BL converter 35 are based on Fig. 9 Whether to convert the BL length of the memory access request included in the 2BG interleaved IL-a is determined based on the conditions described in
[0102] Fig. 9 Conditions for determining whether to convert the BL length of a memory access request included in 2BG interleaving IL-a are listed. Fig. 9Conditions for converting the BL length of the memory access request included in the 2BG interleaved IL-a from BL32 to BL16 are listed. In the case where any of the following two conditions is satisfied, the BL conversion determination section 34 and the BL converter 35 convert the BL length of the memory access request included in the 2BG interleaved IL-a from BL32 to BL16. It is to be noted that the "command Cmd-b" included in the following condition A is a command corresponding to the memory access request Rb immediately before the memory access request Ra.
[0103] -Condition A
[0104] The number of memory access requests in 2BG interleaved IL-a (the number of memory access requests when converted to the equivalent of BL16) is an even number × 2, and command Cmd-a is issued at any timing except 2nCK after command Cmd-b (second command) immediately before command Cmd-a.
[0105] -Condition B
[0106] The number of memory access requests in 2BG interleave IL-a (the number of memory access requests when converted to BL16 equivalent) is an odd number × 2, and command Cmd-a is issued at timing 2nCK after command Cmd-b.
[0107] Fig.10 (A) and Fig.10 (B) is a simplified diagram describing 2BG interleaving IL-a, command Cmd-a, and command Cmd-b. Fig.10 (A) illustrates the case where 2BG interleave IL-a is configured to interleave two bank groups BG2 and BG3 with four read commands using BL32. Fig.10 (B) illustrates a case where 2BG interleave IL-a is configured to interleave two read commands with two bank groups BG2 and BG3 and BL32.
[0108] exist Fig.10 In (A), command Cmd-a is a read command with BL32 for bank group BG1, and command Cmd-b is a read command with BL32 for bank group BG2. Fig.10 In (B), the command Cmd-a is a read command with BL32 for the bank group BG1, and the command Cmd-b is a read command with BL32 for the bank group BG0.
[0109] Fig.11 (A) shows that when multiple commands are Fig.10 The sequential arrangement shown in (A) is an example of the timing of outputting each command to the command access line CA. Fig.11 (B) shows that when each command Fig.11 (A) is an embodiment of the timing at which the transmission data is output to the data line DQ when the timing at which the command access line CA is output is output. Fig.11 In (A), command Cmd-a is issued 2nCK after command Cmd-b is issued.
[0110] Fig.12 (A) shows that Fig.10 (A) is an example of the timing of outputting each command to the command access line CA when a plurality of commands are sequentially arranged. Fig.12 (B) shows the Fig.12 (A) is an embodiment of the timing at which each command is output to the command access line CA and the transmission data is output to the data line DQ. Fig.12 In (A), command Cmd-a is issued 6 nCK after command Cmd-b is issued.
[0111] In the case where command Cmd-a is issued 2nCK after command Cmd-b is issued, it can be seen that the transmission data corresponding to command Cmd-a and the transmission data corresponding to command Cmd-b fill the data line DQ without any gap. However, in the case where command Cmd-a is issued 6nCK after command Cmd-b is issued, data is transmitted in response to command Cmd-a after the gap (1nCK) of BL16. This does not allow the transmission data corresponding to command Cmd-b to fill the gap, thus leaving the gap empty. The condition for generating such a gap corresponds to condition A.
[0112] Next, the operation of the BL switch determination unit 34 will be described. Fig.13 An example of a process of determining whether BL conversion is required in the BL conversion determination section 34 is shown. Fig.13 The FIFO memory 33A is illustrated in a state where all flags have been initialized.
[0113] The BL conversion determination unit 34 first copies the parameter i acquired from the RW switching detector 32 to the parameter j (step S201). Next, the BL conversion determination unit 34 determines whether the parameter j is 0 or more (step S202). In the case where the parameter j is 0 or more (step S202; Y), the BL conversion determination unit 34 accesses the memory access of the jth storage order stored in the FIFO memory 33A, and reads out the BG identifier included in the memory access request with the jth storage order. Then, the BL conversion determination unit 34 records the BG identifier and the number of BG identifiers (the number of BGs) included in the memory access request accessed from the start of execution of step S201 to the present in the internal memory (step S203). The BL conversion determination unit 34 also records the number of memory access requests that have been accessed from the start of execution of step S201 to the present in the internal memory. In the case where the parameter j has a negative value (step S202; N), the process proceeds to the determination on the generation of the division necessity information (step S204).
[0114] After executing step S203, the BL conversion determination unit 34 determines whether the number of BGs recorded in the internal memory is 2 or more (step S205). In the case where the number of BGs recorded in the memory is 2 or more (step S205; Y), the BL conversion determination unit 34 proceeds to the determination on the generation of the division necessity information (step S204). In the case where the number of BGs recorded in the memory is 2 or less (step S205; N), the BL conversion determination unit 34 determines whether the identifiers of the two most recently recorded BGs among the BG identifiers recorded in the internal memory are equal (step S206).
[0115] In the case where the identifiers of the two most recently recorded BGs are equal to each other (step S206; Y), the BL conversion determination unit 34 determines that the last accessed memory access request is not a valid request and ends the present process. In the case where the identifiers of the two most recently recorded BGs are different from each other (step S206; N), the BL conversion determination unit 34 determines that the last accessed memory access request is a valid request. Then, the BL conversion determination unit writes "1" to the library group information bgint corresponding to the memory access request in the j-th storage order stored in the FIFO memory 33A. The BL conversion determination unit 34 further subtracts 1 from the parameter j (step S207) and returns to step S202.
[0116] Fig.14 An example of a process for determining generation of division necessity information in the BL conversion determination section 34 (step S204) is shown. Fig.14 An example is given of a FIFO memory 33A in which the following data have been written: Fig.13 Flags after processing as described in .
[0117] The BL conversion determination section 34 determines whether the parameter j is 0 (step S301). In the case where the parameter j is 0 (step S301; Y), the BL conversion determination section 34 proceeds to the generation of the division necessity information (step S302). In the case where the parameter j is not 0 (step S301; N), the BL conversion determination section 34 ends the determination on the generation of the division necessity information.
[0118] Fig.15 An example of a process of generating division necessity information in the BL conversion determination section 34 is shown. Fig.15 The embodiment of the FIFO memory 33A is marked as follows Fig.15 The processing described in has since been written into it.
[0119] The BL conversion determination unit 34 determines whether the number of memory access requests recorded in the internal memory is an even number × 2 (step S401). The above-mentioned "number of memory access requests recorded in the internal memory" is the number of memory access requests of BG interleaved IL-a (the number of memory access requests when converted to BL16 is equal). As a result, when the number of memory access requests recorded in the internal memory is an even number × 2 (step S401; Y), the BL conversion determination unit 34 determines whether the command Cmd-a is offset from the command Cmd-b by 2nCK based on the release time information obtained from the command scheduler 41 (step S402). Specifically, it is assumed that the release timing (release time) of the command Cmd-b and the release timing (release time) of the command Cmd-a have been obtained from the release timing information obtained from the command scheduler 41. At this time, the BL conversion determination unit 34 can determine whether the command Cmd-a is offset from the command Cmd-b by 2nCK from the release time.
[0120] As a result, when command Cmd-a is not offset from command Cmd-b by 2nCK (step S402; Y), the BL conversion determination unit 34 writes "1" into the division necessity information dev corresponding to the memory access request of BG interleave IL-a in the FIFO memory 33A (step S403).
[0121] On the contrary, in the case where the number of memory access requests of BG interleaved IL-a (the number of memory access requests converted to BL16 equivalent) is an odd number × 2 (step S401; N), the BL conversion determination section 34 determines whether the command Cmd-a is offset from the command Cmd-b by 2nCK based on the issuance timing information acquired from the command scheduler 41 (step S404). Specifically, it is assumed that the issuance timing (issuance time) of the command Cmd-b and the issuance timing (issuance time) of the command Cmd-a have been obtained from the issuance timing information acquired from the command scheduler 41. At this time, the BL conversion determination section 34 can determine whether the command Cmd-a is offset from the command Cmd-b by 2nCK from the issuance time.
[0122] As a result, in the case where the command Cmd-a is offset from the command Cmd-b by 2nCK (step S404; Y), the BL conversion determination unit 34 writes "1" to the division necessity information dev corresponding to the memory access request of the BG interleaved IL-a in the FIFO memory 33A (step S403). In the case where the memory access request Ra is not offset from the memory access request Rb by 2nCK in step S402 (step S404; N), or in the case where the memory access request Ra is offset from the memory access request Rb by 2nCK in step S402 (step S402; Y), the BL conversion determination unit 34 writes "0" to the division necessity information dev corresponding to the memory access request of the BG interleaved IL-a in the FIFO memory 33A (step S405).
[0123] Next, the operation of the BL converter 35 will be described. Fig.16 An embodiment of the BL conversion process in the BL converter 35 is shown. The FIFO memory 33A outputs the memory access request with the earliest storage order in the FIFO memory 33A to the BL converter 35 together with the division necessity information dev each time a new memory access request is stored. When the memory access request and the division necessity information dev are input from the FIFO memory 33A, the BL converter 35 determines whether the division necessity information dev is "1" (step S501). As a result, in the case where the division necessity information dev is "1" (step S501; Y), the BL converter 35 converts the BL length of the memory access request from BL32 to BL16 (step S502). Specifically, the BL converter 35 divides the memory access request with BL32 into two memory access requests with BL16. In the case where the division necessity information dev is "0" (step S501; N), the BL converter 35 maintains the BL length of the memory access request as BL32 without change (step S503).
[0124] Next, a specific method of converting the BL length of the memory access request of BG interleave IL-a will be described.
[0125] Fig.17 FIG. 3 shows an embodiment of the FIFO memory 33A after executing the flag writing process when the condition A is satisfied. Fig.17 As shown in FIG. 1 , the output of the BL converter 35 when a plurality of memory access requests are sequentially input to the FIFO memory 33A will be described.
[0126] When a new memory access request is input to the FIFO memory 33A, the memory access request (read request (BL32, BG2) = memory access request Rb) having the earliest storage order in the FIFO memory 33A and the division necessity information dev are output to the BL converter 35. When the read request (BL32, BG2) and the division necessity information dev are input from the FIFO memory 33A, the BL converter 35 directly outputs the read request (BL32, BG2) to the memory controller 40 since dev = 0.
[0127] When a memory access request is further input to the FIFO memory 33A, the memory request (read request (BL32, BG1) = memory access request Ra) having the earliest storage order in the FIFO memory 33A and the division necessity information dev are output to the BL converter 35. When the read request (BL32, BG1) and the division necessity information dev are input from the FIFO memory 33A, the BL converter 35 directly outputs the read request (BL32, BG1) to the memory controller 40 since dev = 0.
[0128] When a memory access request is further input to the FIFO memory 33A, a memory request having the earliest storage order in the FIFO memory 33A (a lead read request of 2BG interleaved IL-a (BL32, BG2)) and the division necessity information dev are output to the BL converter 35. When a read request (BL32, BG2) and the division necessity information dev are input from the FIFO memory 33A, the BL converter 35 divides the read request (BL32, BG2) into two read requests (BL16, BG2), and outputs the read request to the memory controller 40 because dev=1.
[0129] When a memory access request is further input to the FIFO memory 33A, the memory request with the earliest storage order in the FIFO memory 33A (the second read request of 2BG interleaved IL-a (BL32, BG3)) and the division necessity information dev are output to the BL converter 35. When the read request (BL32, BG3) and the division necessity information dev are input from the FIFO memory 33A, the BL converter 35 divides the read request (BL32, BG3) into two read requests (BL16, BG3), and since dev=1, the read request is output to the memory controller 40. Thereafter, the BL converter 35 outputs the memory access request to the memory controller 40 in the same manner as described above. As a result, for example, Fig.18 The memory access request sequence shown is input to the memory controller 40. Fig.18 , a plurality of memory access requests input to the memory controller 40 are arranged in order from the bottom.
[0130] Fig.19 FIG. 2 shows an embodiment of the FIFO memory 33A after executing the flag writing process when the condition B is satisfied. Fig.19 The output from the BL converter 35 when a plurality of commands are sequentially input to the FIFO memory 33A as described in FIG.
[0131] When a new memory access request is input to the FIFO memory 33A, the memory access request (read request (BL32, BG1) = memory access request Rb) having the earliest storage order in the FIFO memory 33A and the division necessity information dev are output to the BL converter 35. When the read request (BL32, BG1) and the division necessity information dev are input from the FIFO memory 33A, the BL converter 35 directly outputs the read request (BL32, BG1) to the memory controller 40 since dev=0.
[0132] When a memory access request is further input to the FIFO memory 33A, a memory request having the earliest storage order in the FIFO memory 33A (leading read request (BL32, BG0) = memory access request Ra) and the division necessity information dev are output to the BL converter 35. When the read request (BL32, BG0) and the division necessity information dev are input from the FIFO memory 33A, the BL converter 35 directly outputs the read request (BL32, BG0) to the memory controller 40 because dev = 0.
[0133] When a memory access request is further input to the FIFO memory 33A, a memory request having the earliest storage order in the FIFO memory 33A (a lead read request of 2BG interleaved IL-a (BL32, BG2)) and the division necessity information dev are output to the BL converter 35. When a read request (BL32, BG2) and the division necessity information dev are input from the FIFO memory 33A, the BL converter 35 divides the read request (BL32, BG2) into two read requests (BL16, BG2), and outputs the read request to the memory controller 40 because dev=1.
[0134] When a memory access request is further input to the FIFO memory 33A, the memory request (the second read request of 2BG interleaved IL-a (BL32, BG3)) with the earliest storage order in the FIFO memory 33A and the division necessity information dev are output to the BL converter 35. When the read request (BL32, BG3) and the division necessity information dev are input from the FIFO memory 33A, the BL converter 35 divides the read request (BL32, BG3) into two read requests (BL16, BG3), and since dev=1, the read request is output to the memory controller 40. Thereafter, the BL converter 35 outputs the memory access request to the memory controller 40 in the same manner as described above. As a result, for example, Fig. 20 The memory access request sequence shown is input to the memory controller 40. Fig. 20 , a plurality of memory access requests input to the memory controller 40 are arranged in order from the bottom.
[0135] Next, a description will be given of the memory controller 40. Figure 5 As shown, for example, the memory controller 40 includes a command scheduler 41 and a LPDDR-PHY (hereinafter referred to as “physical layer”) 42 .
[0136] The command scheduler 41 issues a command to the DRAM 50 based on the memory access request input from the adjustment unit 30. At this time, the command scheduler 41 generates issue timing information for the command to be issued, and outputs the command to the physical layer 42 according to the generated issue timing information. The command scheduler 41 outputs the generated issue timing information to the BL transition determination section 34.
[0137] The command scheduler 41 issues commands to a plurality of bank groups in the DRAM 50, for example, using a bank group interleaving technique for accessing bank groups in parallel. The memory controller 40 outputs the data written in the internal buffer to the DRAM 50 in synchronization with the command issuance. The memory controller 40 reads out the read data from the DRAM 50 in synchronization with the command issuance and stores the read data in the internal buffer.
[0138] The physical layer 42 outputs commands provided in synchronization with the operation clock of the memory controller 40 and the write data stored in the internal buffer based on the memory clock of the DRAM 50. The physical layer 42 also stores the data read out in synchronization with the memory clock in the DRAM 50 in the internal buffer, in synchronization with the operation clock of the memory controller 40.
[0139] The command scheduler 41 has a function of changing the access order (out-of-order execution function) in order to improve the efficiency of memory access. Hereinafter, a description will be given of the out-of-order execution function.
[0140] Fig.21 An embodiment of the out-of-order execution process is shown. First, the command scheduler determines whether the excess flag is set (step S501). As a result, in the case where the excess flag is set (step S501; Y), the command scheduler 41 clears the excess flag (step S502). At this time, the command scheduler 41 decides the time to issue the command corresponding to (b), and sets the command corresponding to (b) as the command corresponding to (a) (steps S502 and S503).
[0141] In the case where the excess flag is not set (step S501; N), the command scheduler 41 determines whether the BL length of the command corresponding to (b) is BL32 (step S505). As a result, in the case where the BL length of the command corresponding to (b) is BL32 (step S505; Y), the command scheduler 41 determines the time to issue the command corresponding to (b), and sets the command corresponding to (b) to the command corresponding to (a) (steps S506 and S507).
[0142] In the case where the BL length of the command corresponding to (b) is not BL32 (step S505; N), the command scheduler 41 determines whether the command corresponding to (a) and the command corresponding to (b) are both read commands or write commands (step S508). As a result, in the case where the command corresponding to (a) and the command corresponding to (b) are not read commands and write commands to each other (step S508; N), the command scheduler 41 determines the time to issue the command corresponding to (b), and sets the command corresponding to (b) to the command corresponding to (a) (steps S509 and S510).
[0143] In the case where the command corresponding to (a) and the command corresponding to (b) are both read commands or write commands (step S508; Y), the command scheduler 41 determines whether the BG of the command corresponding to (a) is equal to the BG of the command corresponding to (b) (step S511). As a result, in the case where the BG of the command corresponding to (a) and the BG of the command corresponding to (b) are different (step S511; N), the command scheduler 41 determines the time to issue the command corresponding to (b), and sets the command corresponding to (b) to the command corresponding to (a) (steps S512 and S513).
[0144] When the BG of the command corresponding to (a) and the BG of the command corresponding to (b) are equal (step S511; Y), the command scheduler sets the excess flag (step S514). Subsequently, the command scheduler 41 determines the time to issue the command corresponding to (c), and sets the command corresponding to (c) as the command corresponding to (a) (steps S515 and S516).
[0145] Fig. 22 An embodiment of a command sequence generated by the command scheduler 41 is shown. Fig. 22 Each time in represents a release time decided on the command corresponding to (b), and the command corresponding to (c) is the next earliest command after the command corresponding to (b).
[0146] like Fig. 22 As shown in (A), the command scheduler 41 assumes that the exceed flag is not set, the BL length of the command corresponding to (b) is BL16, the command corresponding to (a) and the command corresponding to (b) are both read commands, and the BG of the command corresponding to (a) is different from the BG of the command corresponding to (b) (steps S501; N, 505; N, S508; Y and S511; N). At this time, the command scheduler 41 sets the release time of the command corresponding to (b) to t1 (step S512). In addition, the command scheduler 41 outputs the command corresponding to (a) to the physical layer 42, and sets the command corresponding to (b) to the command corresponding to (a) (step S513).
[0147] Next, if Fig. 22As shown in (B), the command scheduler 41 assumes that the exceed flag is not set, the BL length of the command corresponding to (b) is BL16, the command corresponding to (a) and the command corresponding to (b) are both read commands, and the BG of the command corresponding to (a) is equal to the BG of the command corresponding to (b) (steps S501; N, S505; N, S508; Y and S511; Y). At this time, the command scheduler 41 sets the exceed flag and sets the release time of the command corresponding to (c) to t2 (steps S514 and S515). In addition, the command scheduler 41 outputs the command corresponding to (a) to the physical layer 42, and sets the command corresponding to (c) to the command corresponding to (a) (step S516).
[0148] Next, if Fig. 22 As shown in (C), in the case where the excess flag is set (step S501; Y), the command scheduler 41 clears the excess flag (step S502). At this time, the command scheduler 41 sets the time to issue the command corresponding to (b) to t3 (step S503). In addition, the command scheduler 41 outputs the command corresponding to (a) to the physical layer 42, and sets the command corresponding to (b) as the command corresponding to (a) (step S504).
[0149] Next, if Fig. 22 As shown in (D), the command scheduler assumes that the exceed flag is not set, the BL length of the command corresponding to (b) is BL16, the command corresponding to (a) and the command corresponding to (b) are both read commands, and the BG of the command corresponding to (a) is different from the BG of the command corresponding to (b) (steps S501; N, 505; N, S508; Y and S511; N). At this time, the command scheduler 41 sets the time to issue the command corresponding to (b) to t4 (step S512). In addition, the command scheduler 41 outputs the command corresponding to (a) to the physical layer 42, and sets the command corresponding to (b) to the command corresponding to (a) (step S513).
[0150] Next, if Fig. 22 As shown in (E), the command scheduler 41 assumes that the exceed flag is not set, the BL length of the command corresponding to (b) is BL16, the command corresponding to (a) and the command corresponding to (b) are both read commands, and the BG of the command corresponding to (a) is different from the BG of the command corresponding to (b) (steps S501; N, 505; N, S508; Y and S511; N). At this time, the command scheduler 41 sets the time to issue the command corresponding to (b) to t5 (step S512). In addition, the command scheduler 41 outputs the command corresponding to (a) to the physical layer 42, and sets the command corresponding to (b) to the command corresponding to (a) (step S513).
[0151] Next, if Fig. 22 As shown in (F), the command scheduler 41 assumes that the exceed flag is not set, the BL length of the command corresponding to (b) is BL16, the command corresponding to (a) and the command corresponding to (b) are both read commands, and the BG of the command corresponding to (a) is equal to the BG of the command corresponding to (b) (steps S501; N, S505; N, S508; Y and S511; Y). At this time, the command scheduler 41 sets the exceed flag and sets the release time of the command corresponding to (c) to t6 (steps S514 and S515). In addition, the command scheduler 41 outputs the command corresponding to (a) to the physical layer 42, and sets the command corresponding to (c) to the command corresponding to (a) (step S516).
[0152] Next, if Fig. 22 As shown in (G), the command scheduler assumes that the excess flag is set (step S501; Y). At this time, the command scheduler 41 clears the excess flag and sets the issuance time of the command corresponding to (b) to t7 (steps S502 and S503). In addition, the command scheduler 41 outputs the command corresponding to (a) to the physical layer 42, and sets the command corresponding to (b) as the command corresponding to (a) (step S504).
[0153] Next, if Fig. 22 As shown in (H), the command scheduler 41 assumes that the exceed flag is not set, the BL length of the command corresponding to (b) is BL16, the command corresponding to (a) and the command corresponding to (b) are both read commands, and the BG of the command corresponding to (a) is different from the BG of the command corresponding to (b) (steps S501; N, 505; N, S508; Y and S511; N). At this time, the command scheduler 41 sets the time of issuance of the command corresponding to (b) to t8 (step S512). In addition, the command scheduler 41 outputs the command corresponding to (a) to the physical layer 42, and sets the command corresponding to (b) to the command corresponding to (a) (step S513).
[0154] Next, if Fig. 22 As shown in (I), the command scheduler 41 assumes that the excess flag is not set and the BL length of the command corresponding to (b) is BL32 (steps S501; N and 505; Y). At this time, the command scheduler 41 sets the time to issue the command corresponding to (b) to t9 (step S506). In addition, the command scheduler 41 outputs the command corresponding to (a) to the physical layer 42, and sets the command corresponding to (b) to the command corresponding to (a) (step S507).
[0155] In this way, the command scheduler 41 executes out of order.
[0156] Fig.23 An embodiment of the operation of the command scheduler 41 is shown when the input of a command sequence changes from a read command to a write command. Fig.23 The upper part of FIG. 4 shows an example of the operation of the command scheduler 41 when the BL length is BL32. Fig.23 The lower part shows the Fig.23 The upper part of the command scheduler 41 shows an embodiment of the operation of converting the BL length of multiple read commands (BL32, BG3) of a specific BG interleave from BL32 to BL16. It should be noted that the specific BG interleave represents a BG interleave composed of multiple commands, in which "1" is set as the division necessity information dev.
[0157] exist Fig.23 In the upper and lower parts of the diagram, commands output to the address command line are shown, divided for each library group and library address, and data output to the data line is represented in different modes for each library group and library address. It should be noted that the blocked part of the data line indicates no data output.
[0158] from Fig.23 As can be seen in FIG. 1 , the conversion of the BL length from BL32 to BL16 for multiple read commands of a specific BG interleaving makes it possible to reduce the penalty required for read / write switching.
[0159] Fig.24 An embodiment of the operation of the command scheduler 41 is shown when the input is a command sequence that changes from a write command to a read command. Fig.24 The upper part of FIG. 4 shows an example of the operation of the command scheduler 41 when the BL length is BL32. Fig.24 The lower part shows the Fig.24 The upper part of the command scheduler 41 shows an embodiment of the operation of converting the BL length of multiple read commands (BL32, BG3) of a specific BG interleave from BL32 to BL16. It should be noted that the specific BG interleave represents a BG interleave composed of multiple commands, in which "1" is set as the division necessity information dev.
[0160] exist Fig.24 In the upper and lower parts of the diagram, commands output to the address command line are shown, which are divided by each library group and library address, and the data output to the data line is represented in different modes for each library group and library address. It should be noted that the blocked part of the data line indicates that there is no data output.
[0161] from Fig.24As can be seen in FIG. 1 , the conversion of the BL length from BL32 to BL16 for multiple commands interleaved with a particular BG makes it possible to reduce the penalty required for read / write switching.
[0162] [Effect]
[0163] Next, a description will be given of the effects of the information processing system according to the present embodiment.
[0164] In the present embodiment, the BL length of the memory access request included in the read BG interleaving or write BG interleaving is converted according to the number of memory access requests in the read BG interleaving or write BG interleaving and the timing information of the command corresponding to the memory access request immediately before the read bank group interleaving or write bank group interleaving. This makes it possible to select the BL length in consideration of the efficiency of access to the DRAM 50. Therefore, in a DRAM standard having a BG and being compatible with a plurality of BL lengths, such as LPDDR5 or LPDDR5X, the DRAM 50 can be accessed more efficiently.
[0165] In the present embodiment, when any one of the above-mentioned conditions A and B is satisfied, the BL length of the memory access request included in the read BG interleaving or the write BG interleaving is converted from BL32 to BL16. This makes it possible to select the BL length in consideration of the efficiency of access to the DRAM 50. Therefore, in a DRAM standard such as LPDDR5 or LPDDR5X that has a BG and is compatible with a plurality of BL lengths, the DRAM 50 can be accessed more efficiently.
[0166] In the present embodiment, among the plurality of memory access requests, it is determined whether there is a memory access request corresponding to read BG interleaving or write BG interleaving, and a flag corresponding to the determination result is written as bank group information to the FIFO memory 33A of the buffer 33. This makes it possible to select the BL length using the FIFO memory 33A in consideration of the efficiency of access to the DRAM 50. Therefore, in a DRAM standard having a BG and being compatible with a plurality of BL lengths, such as LPDDR5 or LPDDR5X, the DRAM 50 can be accessed more efficiently.
[0167] In the present embodiment, whether to convert the BL length of the memory access request of a plurality of flags is determined based on the issuance timing information acquired from the command scheduler 41, and the flag according to the determination result is written as the division necessity information dev to the FIFO memory 33A of the buffer 33. This makes it possible to select the BL length using the FIFO memory 33A in consideration of the efficiency of access to the DRAM 50. Therefore, in a DRAM standard having a BG and being compatible with a plurality of BL lengths, such as LPDDR5 or LPDDR5X, the DRAM 50 can be accessed more efficiently.
[0168] <3. Modifications>
[0169] Hereinafter, a description will be given of a modification of the information processing system according to the above-described embodiment. In the following modification, the same configuration as that of the above-described embodiment will be described using the same reference numerals.
[0170] [Variation A]
[0171] In the above-described embodiment and the modified examples of the above-described embodiment, as Fig.25 As shown, for example, a memory controller 60 may be provided instead of the adjustment unit 30 and the memory controller 40. The memory controller 60 includes a command scheduler 43 and a physical layer 42. The command scheduler 43 includes, for example, a buffer 43a, an RW switching detector 32, a BL converter 35, a BL switching determination section 34, and a command issuing section 43b.
[0172] The buffer 43a includes, for example, a FIFO memory 33A. Each time the buffer 43a receives a memory access request from the arbitration section 31, the buffer 43a stores the received memory access request in the FIFO memory 33A. The BL converter 35 outputs the memory access request output from the FIFO memory 33A to the command issuing section 43b. The BL converter 35 outputs the memory access request output from the FIFO memory 33A to the command issuing section 43b.
[0173] The command issuing section 43b issues a command to the DRAM 50 based on the memory access request input from the BL converter 35. The command issuing section 43b issues a command to a plurality of bank groups in the DRAM 50, for example, using a bank group interleaving technique for parallel access to bank groups. The command issuing section 43b outputs the data written to the internal buffer to the DRAM 50 in synchronization with the command issuance. The command issuing section 43b reads out the read data from the DRAM 50 in synchronization with the command issuance, and stores the read data in the internal buffer.
[0174] In this modification, the function of the adjustment unit 30 is built into the memory controller 60. As configured in this manner, a command buffer within the memory controller 60 can be used as the buffer 43a.
[0175] It should be noted that in this modification, the command issuing section 43b may output information related to the use status of the DQ pin managed in the memory controller 60 to the BL transition determination section 34. At this time, the BL transition determination section 34 determines whether the DQ pin is used as described above based on the information related to the use status of the DQ pin. Fig.12(B) becomes empty (i.e., whether the aforementioned condition A is satisfied). As a result, in the case where the BL conversion determination unit 34 determines that the DQ pin becomes empty (i.e., the aforementioned condition A is satisfied), the BL length of the command included in the BG interleaving IL-a is converted from BL32 to BL16. Even in this case, the BL length can be selected in consideration of the efficiency of accessing the DRAM 50. Therefore, in a DRAM standard having a BG and being compatible with a plurality of BL lengths, such as LPDDR5 or LPDDR5X, the DRAM 50 can be accessed more efficiently.
[0176] [Variation Example B]
[0177] In the above-described embodiment and modification A, the BL length included in the memory access input from each initiator 10 is always BL32. However, in the above-described embodiment and modification A, the BL length included in a part of the memory access request input from each initiator 10 may be, for example, Fig.26 BL16 shown in.
[0178] Fig. 27 Conditions for determining whether to convert the BL length included in the 2BG interleave IL-a are shown. Fig. 27 Conditions for converting the BL length of the command included in the 2BG interleaved IL-a from BL16 to BL32 are listed. When neither of the following two conditions is satisfied, the BL conversion determination section 34 and the BL converter 35 convert the BL length of the command included in the 2BG interleaved IL-a from BL16 to BL32.
[0179] -Condition A
[0180] The number of memory access requests in 2BG interleaved IL-a (the number of memory access requests when converted to the BL16 equivalent) is an even number × 2, and command Cmd-a is issued at any timing except 2nCK after command Cmd-b (second command) immediately before command Cmd-a.
[0181] -Condition B
[0182] The number of memory access requests in 2BG interleave IL-a (the number of memory access requests when converted to BL16 equivalent) is an odd number × 2, and command Cmd-a is issued at timing 2nCK after command Cmd-b.
[0183] Fig.28 (A) and Fig.28 (B) is a simplified diagram describing 2BG interleaving IL-a, command Cmd-a, and command Cmd-b. Fig.28(A) illustrates a case where 2BG interleave IL-a is configured to interleave two bank groups BG2 and BG3 having eight read requests of BL16. Fig.29 (B) illustrates a case where 2BG interleave IL-a is configured to interleave two bank groups BG2 and BG3 having four read commands of BL16.
[0184] exist Fig.28 In (A), the command Cmd-a immediately before the 2BG interleave IL-a is a read command with BL32 for bank group BG1, and the command Cmd-b immediately before the command Cmd-a is a read command with BL32 for bank group BG2. Fig.29 In (B), the command Cmd-a just before the 2BG interleave IL-a is a read command with BL32 for the bank group BG1, and the command Cmd-b just before the command Cmd-a is a read command with BL32 for the bank group BG0.
[0185] In this modification, when both the above condition A and the above condition B are not satisfied, the BL length of the memory access request included in the read BG interleaving or the write BG interleaving is converted from BL16 to BL32. This makes it possible to select the BL length in consideration of the efficiency of access to the DRAM 50. Therefore, in a DRAM standard having a BG and being compatible with a plurality of BL lengths, such as LPDDR5 or LPDDR5X, the DRAM 50 can be accessed more efficiently.
[0186] [Variant C]
[0187] In the aforementioned embodiment and modifications A and B, the command scheduler 41 may interleave three library groups. At this time, the BL conversion determination unit 34 and the BL converter 35 convert the BL length of the memory access request included in the 3BG interleaving IL-b (read BG group interleaving or write BG interleaving) based on the number of memory access requests in the 3BG interleaving IL-b immediately before the read / write switching and the timing information of the command Cmd-a (first command) corresponding to the memory access request Ra immediately before the 3BG interleaving IL-b. It should be noted that "3BG interleaving" means interleaving three library groups. At this time, the BL conversion determination unit 34 and the BL converter 35 may determine whether to convert the BL length of the memory access request included in the 3BG interleaving IL-b based on the condition A or condition B listed below.
[0188] -Condition A
[0189] The number of memory access requests in 3BG interleaving IL-a (the number of memory access requests when converted to BL16 equivalent) is an even number × 2, and command Cmd-a is issued at any timing except 2nCK after command Cmd-b (second command) immediately before command Cmd-a.
[0190] -Condition B
[0191] The number of memory access requests in 3BG interleave IL-a (the number of memory access requests when converted to BL16 equivalent) is an odd number × 2, and command Cmd-a is issued at timing 2nCK after command Cmd-b.
[0192] At this time, for example, Fig.29 The FIFO memory 33A described in is stored in the buffer 33. Fig.29 FIG. 1 shows an embodiment of the FIFO memory 33A after executing the flag writing process when the condition A is satisfied. In the following, a description will be given of the operation when multiple memory access requests are made such as Fig.29 1 and 2. Description of the output from the BL converter 35 when being sequentially input to the FIFO memory 33A.
[0193] When a new memory access request is input to the FIFO memory 33A, the memory access request (read request (BL32, BG1) = memory access request Rb) having the earliest storage order in the FIFO memory 33A and the division necessity information dev are output to the BL converter 35. When the read request (BL32, BG1) and the division necessity information dev are input from the FIFO memory 33A, the BL converter 35 directly outputs the read request (BL32, BG1) to the memory controller 40 since dev=0.
[0194] When a memory access request is further input to the FIFO memory 33A, a memory request having the earliest storage order in the FIFO memory 33A (leading read request (BL32, BG0) = memory access request Ra) and the division necessity information dev are output to the BL converter 35. When the read request (BL32, BG0) and the division necessity information dev are input from the FIFO memory 33A, the BL converter 35 directly outputs the read request (BL32, BG0) to the memory controller 40 because dev = 0.
[0195] When a memory access request is further input to the FIFO memory 33A, a memory request having the earliest storage order in the FIFO memory 33A (a lead read request of 3BG interleaved IL-b (BL32, BG1)) and the division necessity information dev are output to the BL converter 35. When a read request (BL32, BG1) and the division necessity information dev are input from the FIFO memory 33A, the BL converter 35 divides the read request (BL32, BG1) into two read requests (BL16, BG1), and outputs the read request to the memory controller 40 because dev=1.
[0196] When a memory access request is further input to the FIFO memory 33A, the memory request (the second read request of 3BG interleaved IL-b (BL32, BG3)) with the earliest storage order in the FIFO memory 33A and the division necessity information dev are output to the BL converter 35. When the read request (BL32, BG3) and the division necessity information dev are input from the FIFO memory 33A, the BL converter 35 divides the read request (BL32, BG3) into two read requests (BL16, BG3), and since dev=1, the read request is output to the memory controller 40. Thereafter, the BL converter 35 outputs a command to the memory controller 40 in the same manner as described above. As a result, for example, Fig.30 The sequence of memory access requests shown is input to the memory controller 40. Fig.30 , a plurality of memory access requests input to the memory controller 40 are arranged in order from the bottom.
[0197] In this modification, based on the number of memory access requests of 3BG interleaved IL-b immediately before the read / write switching and the timing information of the command Cmd-a corresponding to the memory access request immediately before the 3BG interleaved IL-b, the BL length of the memory access request included in the 3BG interleaved IL-b is converted. Even in this case, the BL length can be selected in consideration of the efficiency of accessing the DRAM 50. Therefore, in a DRAM standard having a BG and being compatible with a plurality of BL lengths, such as LPDDR5 or LPDDR5X, the DRAM 50 can be accessed more efficiently.
[0198] It should be noted that in this modification, the command scheduler 41 can interleave M library groups. At this time, the BL conversion determination unit 34 and the BL converter 35 can convert the BL length of the memory access request included in the MBG interleaved IL-b (M≥4) according to the number of memory access requests before the read / write switching is about to be performed, and the timing information of the command Cmd-a (first command) corresponding to the memory access request Ra before the MBG interleaved IL-b. It should be noted that "MBG interleaving" refers to interleaving M library groups. At this time, the BL conversion determination unit 34 and the BL converter 35 can determine whether to convert the BL length of the memory access request included in the 3BG interleaved IL-b based on the condition A or condition B listed below.
[0199] -Condition A
[0200] The number of memory access requests in MBG interleaved IL-a (the number of memory access requests when converted to BL16) is an even number × 2, and command Cmd-a is issued at any timing except 2nCK after command Cmd-b (second command) immediately before command Cmd-a.
[0201] -Condition B
[0202] The number of memory access requests in MBG interleaving IL-a (the number of memory access requests when converted to BL16) is an odd number × 2, and command Cmd-a is issued at timing 2nCK after command Cmd-b.
[0203] Even in this case, the BL length may be selected in consideration of the efficiency of accessing the DRAM 50. Therefore, in a DRAM standard having a BG and being compatible with a plurality of BL lengths, such as LPDDR5 or LPDDR5X, the DRAM 50 may be accessed more efficiently.
[0204] As described above, the present technology has been described through some embodiments and variations of the embodiments. However, the present disclosure is not limited to the above-mentioned embodiments, etc., and various variations are possible. It should be noted that the effects described herein are only examples. The effects of the present disclosure are not limited to the effects described herein. The present disclosure may have any effects other than the effects described herein.
[0205] Furthermore, the present disclosure can adopt, for example, the following configurations. (1)
[0207] A memory control device, comprising:
[0208] a detector that detects, among a plurality of memory access requests received for memory access, a switch between a read bank group interleaving and a write request or a switch between a write bank group interleaving and a read request; and
[0209] A converter converts BL lengths of memory access requests included in the read bank group interleaving or the write bank group interleaving based on the number of memory access requests in the read bank group interleaving or the write bank group interleaving and timing information of a first command corresponding to a memory access request immediately before the read bank group interleaving or the write bank group interleaving. (2)
[0211] The memory control device according to (1), wherein the converter converts the BL length of the memory access request included in the read bank group interleave or the write bank group interleave from BL32 to BL16 when one of the following two conditions is satisfied:
[0212] Condition A: the number of memory access requests when converted to equivalent to BL16 in read bank group interleaving or write bank group interleaving is an even number × 2, and the first command is issued at any timing except the timing 2nCK after the second command immediately before the first command; and
[0213] Condition B: The number of memory access requests when converted to equivalent to BL16 in read bank group interleave or write bank group interleave is an odd number × 2, and the first command is issued at timing 2nCK after the second command. (3)
[0215] The memory control device according to (1), wherein the converter converts the BL length of a command included in a read bank group interleave or a write bank group interleave from BL16 to BL32 when neither of the following two conditions is satisfied:
[0216] Condition A: the number of memory access requests when converted to equivalent to BL16 in read bank group interleaving or write bank group interleaving is an even number × 2, and the first command is issued at any timing except the timing 2nCK after the second command immediately before the first command; and
[0217] Condition B: The number of memory access requests when converted to equivalent to BL16 in read bank group interleave or write bank group interleave is an odd number × 2, and the first command is issued at timing 2nCK after the second command. (4)
[0219] The memory control device according to (1) or (2), further comprising:
[0220] a storage unit that stores, for each command, bank group information and division necessity information associated with a plurality of memory access requests; and
[0221] The determination unit determines whether there is a memory access request corresponding to the read bank group interleave or the write bank group interleave among the plurality of memory access requests, and writes a flag as bank group information according to a result of the determination. (5)
[0223] The memory control device according to (4), wherein the determination section determines whether to convert BL lengths of a plurality of memory access requests to which the flag is set based on the timing information, and in the storage unit, writes the flag as the division necessity information according to a result of the determination.
[0224] In a memory control device according to one embodiment of the present disclosure, the BL length of the memory access request included in the read bank group interleaving or the write bank group interleaving is converted based on the number of memory access requests in the read bank group interleaving or the write bank group interleaving and the timing information of the command corresponding to the memory access request immediately before the read bank group interleaving or the write bank group interleaving. This makes it possible to select the BL length in consideration of the access efficiency of the DRAM. Therefore, in a DRAM standard having a BG and being compatible with multiple BL lengths (such as LPDDR5 or LPDDR5X), the DRAM can be accessed more efficiently. It should be noted that the effects of the present disclosure are not necessarily limited to the effects described herein, and may be any of the effects described herein.
[0225] This application claims priority based on Japanese Patent Application No. 2022-059698 filed with the Japan Patent Office on March 31, 2022, the entire contents of which are incorporated herein by reference.
[0226] It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and alterations may occur depending on design requirements and other factors insofar as they are within the scope of the appended claims or the equivalents thereof.
Claims
1. A memory control device, comprising: a detector for detecting, among the received plurality of memory access requests for memory access, a switch between a read bank group interleaving and a write request or a switch between a write bank group interleaving and a read request; as well as A converter converts BL lengths of the memory access requests included in the read bank group interleaving or the write bank group interleaving based on the number of memory access requests in the read bank group interleaving or the write bank group interleaving and timing information of a first command corresponding to a memory access request immediately before the read bank group interleaving or the write bank group interleaving.
2. The memory control device according to claim 1, wherein: The converter converts the BL length of a memory access request included in the read bank group interleaving or the write bank group interleaving from BL32 to BL16 if one of the following two conditions is satisfied: Condition A: the number of memory access requests converted to BL16 equivalent in the read bank group interleaving or the write bank group interleaving is an even number × 2, and the first command is issued at any timing except the timing 2nCK after the second command immediately before the first command; as well as Condition B: The number of memory access requests converted to the equivalent of BL16 in the read bank group interleave or the write bank group interleave is an odd number × 2, and the first command is issued at timing 2nCK after the second command.
3. The memory control device according to claim 1, wherein: The converter converts the BL length of the command included in the read bank group interleaving or the write bank group interleaving from BL16 to BL32 when neither of the following two conditions is satisfied: Condition A: the number of memory access requests converted to BL16 equivalent in the read bank group interleaving or the write bank group interleaving is an even number × 2, and the first command is issued at any timing except the timing 2nCK after the second command immediately before the first command; as well as Condition B: The number of memory access requests converted to the equivalent of BL16 in the read bank group interleave or the write bank group interleave is an odd number × 2, and the first command is issued at timing 2nCK after the second command.
4. The memory control device according to claim 1, further comprising: a storage unit storing, for each command, bank group information and division necessity information associated with a plurality of the memory access requests; as well as The determination unit determines whether there is a memory access request corresponding to the read bank group interleave or the write bank group interleave among the plurality of memory access requests, and writes a flag as the bank group information according to a result of the determination.
5. The memory control device according to claim 4, wherein: The determination section determines whether to convert the BL lengths of the plurality of memory access requests to which the flag is set based on the timing information, and writes the flag as the division necessity information in the storage unit according to a result of the determination.
