Direct-connect memory components
By introducing buffer devices and unidirectional communication links into DRAM devices, efficient and deterministic data transmission between DRAM device stacks and the controller is achieved, solving the problems of low efficiency and latency uncertainty when DRAM devices are accessed concurrently by multiple memory devices, and improving the overall performance and reliability of the memory system.
Patent Information
- Application Number
- CN202580027802.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2024-04-16
- Filing Date
- 2025-04-10
- Publication Date
- 2026-11-10
AI Technical Summary
In the prior art, dynamic random access memory (DRAM) devices suffer from inefficiency and latency uncertainty during data transmission, especially when multiple memory devices are accessed concurrently, making it difficult to achieve efficient synchronous transmission of data and error detection information.
By stacking multiple DRAM devices and coupling them with a buffer device to form a memory component, the buffer device serializes/deserializes data using a unidirectional communication link and a deterministic protocol, and transmits data and error detection information in parallel, ensuring that commands and data are transmitted synchronously on the same link.
It enables efficient and deterministic data transfer between memory devices and the controller, reduces latency, improves the overall performance and reliability of the memory system, and simplifies data transfer scheduling for the controller.
Smart Images

Figure CN123076906A_ABST
Abstract
Description
Attached Figure Description
[0001] Figure 1 This is a block diagram of a memory system.
[0002] Figures 2A to 2B This is an isometric exploded view of the direct-connected memory component illustrated in the diagram.
[0003] Figure 3 This is an illustration of an example direct-connect memory assembly with a buffer die having a reduced area.
[0004] Figures 4A to 4B This is a timing diagram illustrating example communication with a direct-attached memory component.
[0005] Figures 5A to 5B This is a timing diagram illustrating example sequential communication with a direct-attached memory component.
[0006] Figure 6 This is a flowchart illustrating the method of operating a directly connected memory component.
[0007] Figure 7 This is a flowchart illustrating a method for operating a direct-connect memory component to access data and error detection information.
[0008] Figure 8 This is a flowchart illustrating a method for operating a directly connected memory component using a sequential command protocol.
[0009] Figure 9 This is a flowchart illustrating a method of using a sequential command protocol to operate the controller to transmit data to a direct-attached memory component.
[0010] Figure 10 This is a block diagram of the graphical processing system. Detailed Implementation
[0011] In an embodiment, multiple (e.g., four) stacks of dynamic random access memory (DRAM) devices are coupled to and attached to a buffer device to form a memory assembly. The buffer device interfaces between a controller (e.g., a host) and the memory device stacks, such that each memory device stack serves as a separate memory rank residing on the same memory channel. The buffer device also serializes / deserializes data transmitted with the memory device stacks, such that memory devices in the stack that are being accessed concurrently use a wider data word to transmit with the buffer device than the data word used to transmit with the controller. For example, the buffer device can transmit with the controller using an 8-bit data word, while each of the eight (eight) memory devices in the stack concurrently transmits 64 bits in parallel with the buffer device. Thus, in this example, the buffer device serializes 512 bits (64 bits multiplied by eight memory devices) to an 8-bit word, or deserializes it from an 8-bit word, for communication with the controller.
[0012] In this embodiment, the buffer device and controller use a unidirectional communication link for transmission. The buffer device and controller may use a deterministic protocol for transmission, which combines row functions, column functions, and addressing commands. In this embodiment, commands and data are transmitted on the same unidirectional link (command / data link). In this embodiment, at least read access commands can be pre-scheduled (or queued) to allow multiple consecutive read commands to be issued without delaying read commands to allow communication of write data on the unidirectional command / data link.
[0013] Figure 1 This is a block diagram of the memory system. Figure 1 In the memory system 100, there are a controller 120 and a component 101. The component 101 includes a buffer device 140 and memory device stacks 130a to 130d. Each of the memory device stacks 130a to 130d is at least partially stacked on the buffer device 140.
[0014] Controller 120 and memory devices 110aa to 110de can be integrated circuit type devices, such as those commonly referred to as "chips". The memory controller (such as controller 120) manages the flow of data to and from memory devices and / or memory components. The memory controller can be a separate, independent chip or integrated into another chip. For example, the memory controller can be included on a single die along with a microprocessor, or it can be included as part of a more complex integrated circuit system, such as a block of a system-on-a-chip (SoC).
[0015] Each memory device stack in memory device stacks 130a to 130d includes memory devices 110aa to 110ae, memory devices 110ba to 110be, memory devices 110ca to 110ce, and memory devices 110da to 110de, command / address (CA) interfaces 131a to 131d, data (DQ) interfaces 132a to 132d, and error detection and correction (EDC) information interfaces 133a to 133d. Each memory device 110aa to 110de includes CA interfaces 111aa to 111de and data interfaces 112aa to 112de. In an embodiment, each CA interface 111aa to 111de is operatively coupled to CA interfaces 131a to 131d of memory device stacks 130a to 130d, which include memory devices 111aa to 111de.
[0016] Buffer device 140 includes CA interfaces 141a to 141d, data interfaces 142a to 142d, and EDC information interfaces 143a to 143d, control circuitry 145, optional access sequencer circuitry 147, command and data receive interface (dRx) 146r, data transfer interface (dTx) 146t, EDC information receive interface (eRx) 147r, and EDC information transfer interface (eTx) 147t. Control circuitry 145 includes mode circuitry 145a (e.g., a register). CA interfaces 141a to 141d are operatively coupled to CA interfaces 131a to 131d of memory device stacks 130a to 130d, respectively. Data interfaces 142a to 142d are operatively coupled to data interfaces 132a to 132d of memory device stacks 130a to 130d, respectively. EDC interfaces 143a to 143d are operatively coupled to EDC interfaces 142a to 142d of memory device stacks 130a to 130d, respectively.
[0017] Controller 120 includes a control circuit system 125, a command and data transmission interface (dTx) 126t, a data receiving interface (dRx) 126r, an EDC information transmission interface (eTx) 127t, and an EDC information receiving interface (eRx) 127r. The data transmission interface 126t of controller 120 is operatively coupled to the data receiving interface 146r of buffer device 140 of component 101. The data receiving interface 126r of controller 120 is operatively coupled to the data transmission interface 146t of buffer device 140 of component 101. The EDC transmission interface 127t of controller 120 is operatively coupled to the EDC information receiving interface 147r of buffer device 140 of component 101. The EDC information receiving interface 127r of controller 120 is operatively coupled to the EDC information transmission interface 147t of buffer device 140 of component 101. Therefore, it should be understood that in the embodiments, the controller 120 is operatively coupled to the component 101 via a one-way communication link.
[0018] In an embodiment, each of the memory device stacks 130a to 130d comprises ten (10) memory devices stacked on top of each other. However, other numbers of devices (e.g., 12, 14, 20, etc.) are also contemplated. In an embodiment, eight of the ten memory devices in each of the memory device stacks 130a to 130d are used to store data, and two memory devices in each of the memory device stacks 130a to 130d are used to store error detection and correction (EDC) information. For example, memory device stack 130a may include ten memory devices 110aa to 110ae, wherein memory devices 110ac and 110ae are used to store EDC information, and the remaining memory devices in memory device stack 130a are used to store data.
[0019] The memory devices storing data in the memory device stacks 130a to 130d are operatively coupled to data interfaces 142a to 142d, and data is transmitted via data interfaces 142a to 142d respectively. The memory devices storing EDC information in the memory device stacks 130a to 130d are operatively coupled to EDC interfaces 143a to 143d, and EDC information is transmitted via EDC interfaces 143a to 143d respectively.
[0020] In one embodiment, EDC information stored by memory devices (e.g., memory devices 110aa and 110ae) in a memory device stack (e.g., memory device stack 130a) is associated with data stored by other memory devices in the stack (i.e., protecting that data). In another embodiment, EDC information stored by memory devices (e.g., memory devices 110a and 110e) in a memory device stack is associated with data stored in a different memory device stack (e.g., memory device stack 130b) (i.e., protecting that data). In this way, “die failure” level EDC protection can be achieved. “Die failure” refers to the ability to detect and correct errors that affect all bits transmitted by a single die in an accessed DRAM stack. Typical EDC codes require twice the number of bits (symbols) to detect and correct the number of bits corresponding to one symbol; (e.g., if each die access provides 64 bits of data (1 symbol), more than 128 EDC bits (2 symbols) are needed to perform “die failure” correction.)
[0021] In this embodiment, memory device stacks 130a to 130d each serve as separate rank. Therefore, component 101 can be used as a complete memory channel and appears as such to controller 120. Commands and addresses transmitted by controller 120 to buffer device 140 (i.e., component 101) via data transfer interface 126t and data receive interface 146r can conform to a deterministic access protocol. To access one of the memory stacks 130a to 130d, controller 120 can transmit a command packet via data transfer interface 126t and data receive interface 146r, which combines row operations, column operations, and associated row and column addressing. An example command packet format, and the associated EDC transmissions protecting the command packet from errors, are illustrated in Table 1.
[0022] Based on commands received from controller 120 via data receive interface 146r, the buffer device accesses memory device stacks 130a to 130d. For example, in response to commands from controller 120, buffer device 140 (and particularly control circuitry 145) can translate command packets into command and address transactions (e.g., precharge, activate, read, write, refresh, mode register set, etc.) and transmit those command and address transactions to the addressed memory stacks (ranks) 130a to 130d, which are compatible with the CA interfaces 111aa to 111de of memory devices 110aa to 110de. Based on these command and address transactions, data is transferred in parallel between the individual memory devices (e.g., memory devices 110aa to 110ae) in the addressed memory device stack (e.g., memory device stack 130a) and the buffer device. Similarly, in response to the command, the buffer device 140 serializes or deserializes the data transmitted with the accessed memory device stacks 130a to 130d, as appropriate.
[0023] For example, in response to a read command from controller 120 to memory device stack 130a, buffer device 140 may issue a sequence of read command commands (e.g., ACT, RD) to memory devices 110aa to 110ae of memory device stack 130a via CA interface 141a and CA interface 131a. In response, each of the accessed memory devices 110aa to 110ae of memory device stack 130a provides data read from its memory array(s) in parallel to buffer device 140 via data interface 132a and data interface 142a. The buffer device serializes the received data from memory devices 110aa to 110ae into packets with a width smaller than the width used to transmit the data between memory device stack 130a and buffer device 140 (i.e., fewer bits, more transmission). The buffer then transmits the packet to controller 120. Example packet formats and associated EDC transmissions for protecting command packets from errors are illustrated in Table 2.
[0024] For another example, consider component 101, wherein: (1) each memory device stack 130a to 130d comprises ten (10) memory devices 110aa to 110de; (2) eight memory devices in each memory device stack 130a to 130d are configured or assigned to store data; (3) two memory devices in each memory device stack 130a to 130d are configured or assigned to store EDC information; and (4) data transmission interface 126t, data receiving interface 126r, data transmission interface 146t, and data receiving interface 146r are all oct-bit wide (i.e., each transmitted via eight data links). —For example, eight single-ended signal wires or sixteen differential signal wires); (5) EDC information transmission interface 127t, EDC information receiving interface 127r, EDC information transmission interface 147t and EDC information receiving interface 147r are all two bits wide (i.e., each is transmitted via two data links—for example, two single-ended signal wires or four differential signal wires); and (6) each memory device 110aa to 110de transmits bidirectionally and in parallel with buffer device 140, transmitting 64 bits per communication (i.e., each is transmitted via 64 data links—for example, 64 single-ended signal wires or 128 differential signal wires). In this example, buffer device 140 transmits 512 bits of data from / to the accessed stack in each communication transmission, and transmits 128 bits of EDC information in each communication transmission. For reads, the buffer device serializes data from the accessed memory device stack into an 8-bit wide pass (i.e., the number of data links) and serializes the EDC from the accessed memory device stack for transmission to controller 120. For writes, the buffer device deserializes future 8-bit data transmitted from the controller into a 512-bit wide pass to the accessed memory device stack, and deserializes future 2-bit EDC information transmitted from the controller into a 128-bit wide pass to the accessed memory device stack. To summarize this example, if the size of the data block transmitted with memory stacks 130a to 130d is N (e.g., N = 512) bits, and the size of the parallel transmission unit transmitted with the controller is M (e.g., M = 8) bits, then the data burst with controller 120 has P parallel transmission units, where P equals N divided by M (e.g., P = 64 = 512 / 8), and N, M, and P are positive integers greater than two.
[0025] As discussed herein, in response to the same command to component 101, EDC information and data can be accessed using different memory device stacks. In this embodiment, for a read, the buffer device serializes data from the addressed memory device stack into an 8-bit wide pass (i.e., the number of data links), and serializes EDC from different (e.g., hardwired, command-specified, and / or mode-specified) memory device stacks for communication to controller 120. For a write, the buffer device deserializes future 8-bit data passed from the controller into a 512-bit wide pass to the addressed memory device stack, and deserializes future 2-bit EDC information passed from the controller into a 128-bit wide pass to a different memory device stack.
[0026] Figures 2A to 2B This illustration shows an isometric exploded view of an example direct-connect memory component. Figure 2A In the diagram, component 200 is illustrated with memory device stacks 230a to 230d arranged on buffer device 240. Figure 2B In the diagram, component 201 is also illustrated with memory device stacks 230a to 230d arranged on buffer device 240. However, in Figure 2B In this context, memory devices 210ac and 210ae of memory device stack 230a, memory devices 210bc and 210be of memory device stack 230b, memory devices 210cc and 210ce of memory device stack 230c, and memory devices 210dc and 210de of memory device stack 230d are designated or configured to store EDC information. Figure 2B The memory devices are illustrated by the shading on the visible side of the memory devices.
[0027] Figure 3 This is an illustration of an example direct-connect memory assembly with a buffer die having a reduced area. Figure 3 In the diagram, memory device stacks 330a to 330d are illustrated only by their outlines (e.g., wireframes) to allow for visibility. Figure 3In the illustration, component 300 is shown as a memory device stack 330a to 330c partially disposed on substrate 301 and partially disposed on buffer device 340. Specifically, each memory device stack in memory device stacks 330a to 330d is disposed in a different quadrant (above) of buffer device 340. In this way, buffer device 340 can be smaller (i.e., with less area) than buffer device 240 and still form connections (e.g., through-silicon vias - TSVs) with memory device stacks 330a to 330c (and the memory devices therein).
[0028] Figures 4A to 4B This is a timing diagram illustrating example communication with a direct-attached memory component. Figures 4A to 4B In the diagram, example communication between a controller (e.g., controller 120) and a component (e.g., component 101) is illustrated from the perspective of the controller's unidirectional data transmission and data reception interfaces (dTx, dRx). Figures 4A to 4B In this process, on the controller's data transmission interface (or link), the controller transmits a first read command (RD1). In response to the first read command, the component transmits first read data (RDQ1) to the controller's data receiving interface on its unidirectional data transmission link, after a predetermined (i.e., consistent) delay period. This is in... Figure 4B The diagram is illustrated by arrow 401, which extends from the transmission of RD1 to the reception of RDQ1.
[0029] Following the first read command (RD1), the controller transmits the first write command (WR1). On the controller's data transmission interface, the first portion of the first write data (WDQ1a) immediately follows the first write command. This is... Figure 4B The diagram is illustrated by arrow 402, which extends the transmission from WR1 to WDQ1a. Following the transmission of WDQ1a, the controller transmits a second read command (RD2). On the controller's data transmission interface, the second read command RD2 is immediately followed by the second portion (WDQ1b) of the first write data associated with WR1. This is... Figure 4B The diagram is illustrated by arrow 403, which extends the transmission from WR1 to WDQ1b. In response to the second read command RD2, immediately following RDQ2 (i.e., no unused transmission cycles) and after a predetermined (i.e., consistent) delay period from when RD2 is transmitted, the component transmits the second read data (RDQ2) to the controller's data receive interface on its unidirectional data transmission link. This is in... Figure 4B The diagram is illustrated by arrow 404, which extends from the transmission of RD2 to the reception of RDQ2.
[0030] Following the transmission of RD2, the controller transmits a refresh command (REF). This refresh command can be, for example, a single memory bank refresh for a precharged memory bank in the currently accessed rank (memory device stack). In another example, the refresh command can be a refresh for all memory banks in a rank that is not currently accessed.
[0031] Following the transmission of REF, the controller transmits a third read command (RD3). On the controller's data transmission interface, the second write command (WR2) immediately follows the third read command RD3. On the controller's data transmission interface, the first portion of the second write data (WDQ2a) immediately follows the second write command. This... Figure 4B The diagram is illustrated by arrow 406, which extends the transmission from WR2 to WDQ2a. Following the transmission of WDQ2a, the controller transmits the fourth read command (RD4). On the controller's data transmission interface, the fourth read command RD4 is immediately followed by the second portion (WDQ2b) of the second write data associated with WR2. This is... Figure 4B The diagram is illustrated by arrow 407, which extends the transmission from WR2 to WDQ2b. In response to the third read command, immediately following RDQ2 (i.e., without any unused transmission cycles) and after a predetermined (i.e., consistent) delay period from when RD3 is transmitted, the component transmits the third read data (RDQ3) to the controller's data receive interface on its unidirectional data transmission link. This is in... Figure 4B The diagram is illustrated by arrow 405, which extends from the transmission of RD2 to the reception of RDQ2.
[0032] Figures 5A to 5B This is a timing diagram illustrating example sequential communication with a direct-attached memory component. Figures 5A to 5B In the diagram, example communication between a controller (e.g., controller 120) and a component (e.g., component 101) is illustrated from the perspective of the controller's unidirectional data transmission and data reception interfaces (dTx, dRx). Figures 5A to 5B In the process, on the controller's data transmission interface (or link), the controller successively transmits the first read command (RD1), the second read command (RD2), the third read command (RD3), the fourth read command (RD4), the refresh command (REF), and the first write command (WR1). In response to commands RD1 to RD4, the component immediately connects the first read data (RDQ1), the second read data (RDQ2), the third read data (RDQ3), and the fourth read data (RDQ4) to its unidirectional data transmission link. Figures 5A to 5B(Not shown) The data is transmitted to the controller's data receiving interface. Note that because commands RD1 to RD4 take fewer bus cycles to transmit than the associated read data RDQ1 to RDQ4, the close succession of RD1 to RD4 and RDQ1 to RDQ4 results in a non-uniform delay between the transmission of each command in RD1 to RD4 and the transmission of the associated data RDQ1 to RDQ4. These delays occur in... Figure 5B The diagram is illustrated by arrow 501 extending from the transmission of RD1 to the reception of RDQ1, arrow 502 extending from the transmission of RD2 to the reception of RDQ2, and arrow 503 extending from the transmission of RD3 to the reception of RDQ3.
[0033] Immediately following the first write command WR1, and on the controller's data transfer interface, the data associated with the first write command (WDQ1) is transmitted. This in Figure 5B The diagram is illustrated by arrow 504, which extends from the transmission of WR1 to the transmission of WDQ1 immediately following WR1. Immediately following the transmission of WDQ1, a second write command (WR2) is transmitted by the controller. Following the second write command WR2, and on the controller's data transfer interface, the data associated with the second write command (WDQ2) is transmitted. This is... Figure 5B The diagram is illustrated by arrow 505, which extends from the transmission of WR2 to the transmission of WDQ2 immediately following WR2. Therefore, it should be obvious that... Figures 4A to 4B Compared to the command sequence illustrated in the figure, Figures 5A to 5B The ordering or queuing of commands illustrated in the figure (e.g., by sequencer circuitry 147) results in non-consistent latency between read commands and their associated data, allowing write data to be sent in single blocks (i.e., not split into multiple parts) and simplifying command scheduling on the controller's data transfer interface. Sequencer 147 includes a state machine that schedules access commands received from controller 120 to DRAM stacks 130a through 130d based on DRAM timing parameters. In this example, sequencer 147 schedules received read commands and addresses to the appropriate DRAM rank based on DRAM read timing requirements (e.g., tCCD) to ensure read data becomes available at the correct time and to avoid DRAM timing conflicts, as DRAM responses to read requests have a consistent latency. The sequencer also ensures that component 101 executes all received commands in the order they are received. In other words, sequencer 147 ensures that communication between buffer 140 and DRAM stacks 130a through 130d remains deterministic with consistent latency. In this approach, even if the command response latency at the interface to the controller is inconsistent, the access response for each command remains deterministic for the controller 120.
[0034] Figure 6 This is a flowchart illustrating a method for operating a direct-connect memory component. Figure 6 One or more steps illustrated herein can be performed, for example, by memory system 100 and / or its components. A first command (602) to be executed by a first stack of the multiple memory integrated circuit device stacks is received by a component including a buffer integrated circuit and a plurality of memory integrated circuit device stacks arranged at least partially on the buffer integrated circuit. For example, component 101 (and, in particular, buffer device 140) can receive a command from controller 120 to access memory device stack 130a.
[0035] A first data block is transmitted in parallel with the buffer integrated circuit based on a first command from the first stack (604). For example, based on the first command, the buffer device 140 can control the memory devices 110aa to 110ae of the memory device stack 130a to transmit data blocks (e.g., 640 bits of read / write data) in parallel with each memory device 110aa to 110ae using a wide data transfer width (e.g., 64 bits of data are transmitted in parallel with each memory device 110aa to 110ae, achieving a total of 640 bits transmitted in parallel with the buffer device 140 for each transfer). The first data block is transmitted in the form of a first data burst by the buffer integrated circuit based on the first command and the controller transmitting the first command, wherein the first data burst includes a first plurality of data sub-blocks, and the first plurality of data sub-blocks are transmitted with the controller using parallel transfer units smaller than the first data block (606). For example, buffer device 140 can serialize / deserialize data blocks (e.g., 640-bit blocks) from / to memory device stack 130a and transmit those data blocks with controller 120 using smaller parallel transfer units (e.g., 8-bit, 10-bit, etc.).
[0036] The component receives a second command (608) executed by a second stack of multiple memory integrated circuit device stacks. For example, component 101 (and, in particular, buffer device 140) may receive a command from controller 120 to access memory device stack 130b.
[0037] The second data block is transmitted in parallel with the buffer integrated circuit based on the second command from the second stack (610). For example, based on the second command, the buffer device 140 can control the memory devices 110ba to 110be of the memory device stack 130b to transmit data blocks (e.g., 640 bits of read / write data) in parallel with each memory device 110ba to 110be using a wide data transfer width (e.g., 64 bits are transmitted in parallel with each memory device 110ba to 110be, achieving a total of 640 bits transmitted in parallel with the buffer device 140 for each transfer). The second data block is transmitted in the form of a second data burst by the buffer integrated circuit based on the first command and the controller transmitting the first and second commands, wherein the second data burst includes a second plurality of data sub-blocks, which are transmitted with the controller using a parallel transfer unit smaller than the first data block (612). For example, buffer device 140 can serialize / deserialize data blocks (e.g., 640-bit blocks) from / to memory device stack 130b and transmit those data blocks with controller 120 using smaller parallel transfer units (e.g., 8-bit, 10-bit, etc.).
[0038] Figure 7 This is a flowchart illustrating a method for operating a direct-access memory component to access data and error detection information. Figure 7 One or more steps illustrated herein can be performed by, for example, memory system 100 and / or its components. A first access command (702) to be executed by a component comprising a buffer integrated circuit and a stack of multiple memory integrated circuit devices at least partially disposed on the buffer integrated circuit is received. For example, component 101 (and, in particular, buffer device 140) can receive the access command to be executed by component 101 from controller 120.
[0039] A first data block (704) is transmitted in parallel with the buffer integrated circuit from the first stack based on a first command. For example, based on the first command, the buffer device 140 can control a first subset (e.g., eight devices) of the memory device stack 130a to transmit data blocks (e.g., 512 bits of read / write data) in parallel with each memory device in the first subset using a wide data transfer width (e.g., each of the eight memory devices in the first subset transmits 64 bits in parallel, achieving a total of 512 bits transmitted in parallel with the buffer device 140 per transfer). A first error detection information block (706) is transmitted in parallel with the buffer integrated circuit from the second stack based on the first command. For example, based on the first command, the buffer device 140 can control a second subset (e.g., two devices) of the memory device stack 130b to transmit EDC information data blocks (e.g., 128 bits of EDC read / write data) in parallel with each memory device in the second subset using a wide data transfer width (e.g., each of the two memory devices in the second subset transmits 64 bits in parallel, achieving a total of 128 bits transmitted in parallel with the buffer device 140 for each transfer).
[0040] Based on a first command, the buffer integrated circuit transmits a first data block and a first error detection information block in the form of a first data burst with the controller that transmitted the first command. The first data burst includes a first plurality of data sub-blocks, which are transmitted with the controller using a parallel transfer unit smaller than the first data block (708). For example, the buffer device 140 can serialize / deserialize data blocks (e.g., 512-bit blocks) from / to memory device stack 130a and transmit those data blocks with the controller 120 using a smaller parallel transfer unit (e.g., 8 bits). The buffer device 140 can also, for example, serialize / deserialize EDC information data blocks (e.g., 128-bit blocks) from / to memory device stack 130b and transmit those EDC information data blocks with the controller 120 using a smaller parallel transfer unit (e.g., 2 bits).
[0041] Figure 8 This is a flowchart illustrating a method for operating a directly connected memory component using a sequential command protocol. Figure 8 One or more steps illustrated herein may be performed by, for example, memory system 100 and / or its components. Multiple commands are received from the controller (802). For example, controller 120 may transmit and component 101 may receive: a first read command (RD1), a second read command (RD2), a third read command (RD3), and a fourth read command (RD4) in succession.
[0042] Based on multiple read commands, data associated with a corresponding read command among the multiple read commands is transmitted (804) in a manner with a non-uniform delay between the reception of the corresponding read command among the multiple read commands and the transmission of data associated with the corresponding read command among the multiple read commands. For example, the component may, on its unidirectional data transmission link, transmit first read data (RDQ1), second read data (RDQ2), and third read data (RDQ3) immediately after each read command RD1 to RD3 is transmitted. Because, for example, commands RD1 to RD3 take fewer bus cycles to transmit than the associated read data RDQ1 to RDQ3, the close succession of RD1 to RD3 and the close succession of RDQ1 to RDQ3 results in a non-uniform delay between the transmission of each command among the RD1 to RD3 commands and the transmission of the associated data RDQ1 to RDQ3.
[0043] Figure 9 This is a flowchart illustrating a method for using a sequential command protocol to operate the controller to transmit data to a direct-attached memory component. Figure 9 One or more steps shown can be performed by, for example, memory system 100 and / or its components. Multiple read commands are transmitted to a component including a buffer integrated circuit and a stack of memory integrated circuit devices (902) at least partially disposed on the buffer integrated circuit. For example, controller 120 can transmit and component 101 can receive a first read command (RD1), a second read command (RD2), a third read command (RD3), and a fourth read command (RD4) in close succession.
[0044] Based on multiple read commands, corresponding multiple inconsistencies are determined between the transmission of the corresponding read command among the multiple read commands and the receipt from the component of data associated with the corresponding read command among the multiple read commands (904). For example, controller 120 can determine the corresponding inconsistencies when RD1 to RDQ1 will be received, the corresponding inconsistencies when RD2 to RDQ2 will be received, and the corresponding inconsistencies when RD3 to RDQ3 will be received, based on the information that read commands RD1 to RD3 are transmitted (or will be transmitted) closely using these information (and optionally other data information, such as refresh command transmission time, etc.) and the information that commands RD1 to RD3 are transmitted using fewer bus cycles than the associated read data RDQ1 to RDQ3.
[0045] Based on the corresponding multiple non-consistent delays, data associated with the corresponding read command among multiple read commands is received (906). For example, based on the non-consistent delay determined when RD1 to RDQ1 will be received, the non-consistent delay determined when RD2 to RDQ2 will be received, and the non-consistent delay determined when RD3 to RDQ3 will be received, the controller 120 can receive RDQ1 to RDQ3.
[0046] Although CA and data transmission between component 101 and controller 120 are performed on a unidirectional link in the above embodiment, it is also possible to implement the present invention using a bidirectional link between controller 120 and component 101.
[0047] The methods, systems, and devices described above can be implemented in or stored in a computer system. The methods described above can also be stored on a non-transitory computer-readable medium. The devices, circuits, and systems described herein can be implemented using computer-aided design tools available in the art and are embodied in computer-readable documents containing software descriptions of such circuits. This includes, but is not limited to, one or more component memory systems 100 and / or components thereof. These software descriptions can be behavioral, register-pass, logic unit-level, transistor-level, and layout geometry-level descriptions. Furthermore, the software descriptions can be stored on a storage medium or transmitted by a carrier wave.
[0048] Such descriptions can be implemented in data formats including, but not limited to: formats supporting behavioral languages like C; formats supporting register-pass-level (RTL) languages like Verilog and VHDL; formats supporting geometric description languages (such as GDSII, GDSIII, GDSIV, CIF, and MEBES); and other suitable formats and languages. Furthermore, the transfer of such files on machine-readable media can be accomplished electronically over various media on the Internet, or, for example, via email. Note that physical files can be implemented on machine-readable media such as 4mm magnetic tape, 8mm magnetic tape, 3-1 / 2-inch floppy disk media, CDs, DVDs, etc.
[0049] Figure 10 This is a block diagram illustrating one embodiment of a processing system 1000 for including, processing, or generating representations of circuit components 1020. The processing system 1000 includes one or more processors 1002, a memory 1004, and one or more communication devices 1006. The processors 1002, memory 1004, and communication devices 1006 are transmitted using wired and / or wireless connections 1008 of any suitable type, number, and / or configuration.
[0050] Processor 1002 executes instructions of one or more processes 1012 stored in memory 1004 to process and / or generate circuit components 1020 in response to user input 1014 and parameters 1016. Process 1012 may be any suitable electronic design automation (EDA) tool or part of such a tool, used to design, simulate, analyze, and / or verify electronic circuit systems and / or generate photomasks for electronic circuit systems. As illustrated in the figures, representation 1020 includes data describing all or part of the memory system 100 and / or its components.
[0051] The representation 1020 may include one or more of the following: behavioral level, register-pass level, logic unit level, transistor level, and layout geometry level description. Furthermore, the representation 1020 may be stored on a storage medium or transmitted by a carrier wave.
[0052] The data formats in which 1020 can be implemented include, but are not limited to: formats supporting behavioral languages like C, formats supporting register-pass-level (RTL) languages like Verilog and VHDL, formats supporting geometry description languages such as GDSII, GDSIII, GDSIV, CIF, and MEBES, and other suitable formats and languages. Furthermore, the data transfer of such documents on machine-readable media can be accomplished electronically over various media on the Internet, or, for example, via email.
[0053] User input 1014 may include input parameters from a keyboard, mouse, voice recognition interface, microphone and speaker, graphics display, touchscreen, or other type of user interface device. The user interface may be distributed across multiple interface devices. Parameter 1016 may include specifications and / or characteristics input to help define the representation 1020. For example, parameter 1016 may include information defining the device type (e.g., NFET, PFET, etc.), topology (e.g., block diagram, circuit description, schematic, etc.), and / or device description (e.g., device attributes, device size, power supply voltage, simulation temperature, simulation model, etc.).
[0054] The memory 1004 includes any suitable type, number, and / or configuration of nontransitory computer-readable storage media, which stores process 1012, user input 1014, parameters 1016, and circuit components 1020.
[0055] Communication device 1006 includes wired and / or wireless devices of any suitable type, number, and / or configuration, which transmit information from processing system 1000 to another processing or storage system (not shown) and / or receive information from another processing or storage system (not shown). For example, communication device 1006 can transmit circuit component 1020 to another system. Communication device 1006 can receive process 1012, user input 1014, parameters 1016, and / or circuit component 1020, and store process 1012, user input 1014, parameters 1016, and / or circuit component 1020 in memory 1004.
[0056] The implementation methods discussed herein include, but are not limited to, the following examples:
[0057] Example 1: A memory device includes: a buffer device including an external memory channel interface for transmitting commands, addresses, and data with a controller; and a plurality of memory device stacks arranged with and coupled to the buffer device, each of the plurality of memory device stacks including a corresponding plurality of memory devices, each of the corresponding plurality of memory devices in the corresponding memory device stack serving as a rank of a memory device, the rank of the memory device being concurrently accessed in response to commands, addresses, and data transmitted via the buffer device.
[0058] Example 2: The memory device according to Example 1, wherein data bursts transmitted based on access via the external memory channel interface are to be transmitted in parallel with each memory device stack.
[0059] Example 3: The memory device according to Example 1, wherein the external memory channel interface includes a plurality of unidirectional receive links and a plurality of unidirectional transmit links.
[0060] Example 4: The memory device according to Example 3, wherein the plurality of unidirectional receive links include a plurality of error detection and correction (EDC) information links for receiving EDC information, and wherein the plurality of unidirectional transmit links include a plurality of EDC information links for transmitting EDC information.
[0061] Example 5: The memory device according to Example 1 further includes: an access sequencing circuit system for scheduling communication via the plurality of unidirectional transmission links in a manner having a non-constant delay relative to commands received via the plurality of unidirectional receive links.
[0062] Example 6: The memory device according to Example 1, wherein data and associated error detection and correction (EDC) information are to be accessed using different memory device stacks among the plurality of memory device stacks.
[0063] Example 7: The memory device according to Example 1, wherein a first region of the first surface of each of the plurality of memory device stacks is respectively arranged on each of the plurality of portions of the second surface of the buffer device.
[0064] Example 8: The memory device according to Example 7, wherein the total area of the first surface of the plurality of memory devices stacked is greater than the area of the second surface of the buffer device.
[0065] Example 9: A memory device comprising: a first memory device stack including a first plurality of memory devices configured as a first rank of memory devices to be accessed concurrently; a second memory device stack including a second plurality of memory devices configured as a second rank of memory devices to be accessed concurrently; and a buffer device disposed on and coupled to the first and second memory device stacks, the buffer device including an external memory channel interface for transmitting a plurality of commands, a plurality of addresses, and a plurality of data bursts to a controller, wherein a first data block corresponding to a first data burst of the plurality of data bursts is to be transmitted in parallel with the first memory device stack, and a second data block corresponding to a second data burst of the plurality of data bursts is to be transmitted in parallel with the second memory device stack.
[0066] Example 10: The memory device according to Example 9, wherein the external memory channel interface includes a first plurality of unidirectional receive links for transmitting a first command from the controller to the buffer device.
[0067] Example 11: The memory device according to Example 10, wherein the first plurality of unidirectional receive links are configured to receive the first data burst based on the first command, and the buffer device is configured to transfer the first data block to the first memory device stack based on the first command.
[0068] Example 12: The memory device according to Example 11, wherein the external memory channel interface further includes a first plurality of unidirectional transmission links, the buffer device being configured to: access the second data block from the second memory device stack based on a second command transmitted via the first plurality of unidirectional receive links, and transmit the second data burst to the controller via the first plurality of unidirectional transmission links.
[0069] Example 13: The memory device according to Example 12, wherein the external memory channel further includes a second plurality of unidirectional receive links for receiving error detection and correction (EDC) information from the controller, and wherein the external memory channel further includes a second plurality of unidirectional transmit links for transmitting EDC information to the controller.
[0070] Example 14: The memory device according to Example 13 further includes:
[0071] Access sequencing circuitry system for scheduling communication via the first plurality of unidirectional transmission links in a manner with a non-constant delay relative to commands received via the first plurality of unidirectional receive links.
[0072] Example 15: The memory device according to Example 9, wherein the first data block is associated with first error detection and correction (EDC) information accessed using the second memory device stack.
[0073] Example 16: The memory device according to Example 15, wherein the second data block is associated with second EDC information accessed using the first memory device stack.
[0074] Example 17: A method comprising: receiving, by a component including a buffer integrated circuit and a plurality of memory integrated circuit device stacks at least partially disposed on the buffer integrated circuit, a first command to be executed by a first stack of the plurality of memory integrated circuit device stacks; transmitting, by the first stack, in parallel with the buffer integrated circuit based on the first command; transmitting, by the buffer integrated circuit, in the form of a first data burst with a controller transmitting the first command based on the first command, the first data burst including a first plurality of data sub-blocks, the first plurality of data sub-blocks being transmitted with the controller using a parallel transmission unit smaller than the first data block; receiving, by the component, a second command to be executed by a second stack of the plurality of memory integrated circuit device stacks; transmitting, by the second stack, in parallel with the buffer integrated circuit based on the second command; and transmitting, by the buffer integrated circuit, in the form of a second data burst with the controller transmitting the first command and the second command based on the second command, the second data burst including a second plurality of data sub-blocks, the second plurality of data sub-blocks being transmitted with the controller using a parallel transmission unit smaller than the first data block.
[0075] Example 18: The method according to Example 17 further includes: transmitting first error detection and correction (EDC) information to the buffer integrated circuit by the second stack based on the first command; and transmitting the first EDC information to the controller by the buffer integrated circuit based on the first command.
[0076] Example 19: The method according to Example 17 further includes: receiving a plurality of read commands from the controller; and, based on a corresponding read command among the plurality of read commands, transmitting the data associated with the corresponding read command among the plurality of read commands in a manner with a non-consistent delay between the reception of the corresponding read command among the plurality of read commands and the data associated with the corresponding read command among the plurality of read commands.
[0077] Example 20: According to the method described in Example 17, the size of the first data block and the second data block is N bits, the size of the parallel transfer unit is M bits, the first data burst and the second data burst have P parallel transfer units, where P equals N divided by M, and N, M and P are positive integers greater than two.
[0078] The foregoing description of the invention has been presented for illustrative and descriptive purposes. It is not intended to be exhaustive or to limit the invention to the precise forms disclosed, and other modifications and variations may be possible in light of the foregoing teachings. This embodiment has been chosen and described to best explain the principles of the invention and its practical application, thereby enabling others skilled in the art to best utilize the invention in various embodiments and modifications suitable for the particular intended use. The appended claims are intended to be construed as including other alternative embodiments of the invention, unless limited by prior art.
Claims
1. A memory device, comprising: A buffer device, the buffer device including an external memory channel interface for transmitting commands, addresses, and data with the controller; as well as Multiple memory device stacks are arranged with and coupled to the buffer device, each memory device stack in the multiple memory device stacks includes a corresponding plurality of memory devices, each of the corresponding plurality of memory devices in the corresponding memory device stack serving as a rank of a memory device, the rank of the memory device being concurrently accessed in response to commands, addresses and data transmitted via the buffer device.
2. The memory device of claim 1, wherein data bursts transmitted based on access via the external memory channel interface are to be transmitted in parallel with each memory device stack.
3. The memory device according to claim 1, wherein the external memory channel interface includes a plurality of unidirectional receive links and a plurality of unidirectional transmit links.
4. The memory device of claim 3, wherein the plurality of unidirectional receiving links include a plurality of error detection and correction (EDC) information links for receiving EDC information, and wherein the plurality of unidirectional transmitting links include a plurality of EDC information links for transmitting EDC information.
5. The memory device according to claim 1, further comprising: Access sequencing circuitry system for scheduling communication via the plurality of unidirectional transmission links in a manner with a non-constant delay relative to commands received via the plurality of unidirectional receive links.
6. The memory device of claim 1, wherein the data and associated error detection and correction (EDC) information are to be accessed using different memory device stacks among the plurality of memory device stacks.
7. The memory device of claim 1, wherein a first region of the first surface of each of the plurality of memory device stacks is respectively disposed on each of a plurality of portions of the second surface of the buffer device.
8. The memory device of claim 7, wherein the total area of the first surface of the plurality of memory devices stacked is greater than the area of the second surface of the buffer device.
9. A memory device, comprising: A first memory device stack, the first memory device stack including a first plurality of memory devices, the first plurality of memory devices being configured as a first rank of memory devices to be accessed concurrently; A second memory device stack, the second memory device stack including a second plurality of memory devices, the second plurality of memory devices being configured as a second rank of memory devices to be accessed concurrently; as well as A buffer device is disposed on and coupled to the first memory device stack and the second memory device stack. The buffer device includes an external memory channel interface for transmitting multiple commands, multiple addresses, and multiple data bursts with the controller. A first data block corresponding to a first data burst of the multiple data bursts is to be transmitted in parallel with the first memory device stack, and a second data block corresponding to a second data burst of the multiple data bursts is to be transmitted in parallel with the second memory device stack.
10. The memory device of claim 9, wherein the external memory channel interface includes a first plurality of unidirectional receive links for transmitting a first command from the controller to the buffer device.
11. The memory device of claim 10, wherein the first plurality of unidirectional receive links are configured to receive the first data burst based on the first command, and the buffer device is configured to transfer the first data block to the first memory device stack based on the first command.
12. The memory device of claim 11, wherein the external memory channel interface further comprises a first plurality of unidirectional transmission links, the buffer device being configured to: access the second data block from the second memory device stack based on a second command transmitted via the first plurality of unidirectional receive links, and transmit the second data burst to the controller via the first plurality of unidirectional transmission links.
13. The memory device of claim 12, wherein the external memory channel further includes a second plurality of unidirectional receive links for receiving error detection and correction (EDC) information from the controller, and wherein the external memory channel further includes a second plurality of unidirectional transmit links for transmitting EDC information to the controller.
14. The memory device of claim 13, further comprising: Access sequencing circuitry system for scheduling communication via the first plurality of unidirectional transmission links in a manner with a non-constant delay relative to commands received via the first plurality of unidirectional receive links.
15. The memory device of claim 9, wherein the first data block is associated with first error detection and correction (EDC) information accessed using the second memory device stack.
16. The memory device of claim 15, wherein the second data block is associated with second EDC information accessed using the first memory device stack.
17. A method comprising: A component comprising a buffer integrated circuit and a stack of multiple memory integrated circuit devices at least partially disposed on the buffer integrated circuit receives a first command to be executed by a first stack of the multiple memory integrated circuit device stacks; The first data block is transmitted in parallel with the buffer integrated circuit based on the first command from the first stack; The buffer integrated circuit transmits the first data block in the form of a first data burst based on the first command and the controller that transmits the first command. The first data burst includes a first plurality of data sub-blocks, and the first plurality of data sub-blocks are transmitted with the controller using a parallel transmission unit smaller than the first data block. The component receives a second command to be executed by a second stack in the plurality of memory integrated circuit device stacks; The second stack transmits a second data block in parallel with the buffer integrated circuit based on the second command; as well as Based on the second command, the buffer integrated circuit transmits the second data block in the form of a second data burst with the controller that transmits the first command and the second command. The second data burst includes a second plurality of data sub-blocks, which are transmitted with the controller using the parallel transmission unit that is smaller than the first data block.
18. The method of claim 17, further comprising: The second stack transmits first error detection and correction (EDC) information to the buffer integrated circuit based on the first command; as well as The buffer integrated circuit transmits the first EDC information to the controller based on the first command.
19. The method of claim 17, further comprising: Receive multiple read commands from the controller; as well as Based on the corresponding read command among the plurality of read commands, the data associated with the corresponding read command among the plurality of read commands is transmitted in a manner with a non-consistent delay between the receipt of the corresponding read command among the plurality of read commands and the data associated with the corresponding read command among the plurality of read commands.
20. The method of claim 17, wherein the size of the first data block and the second data block is N bits, the size of the parallel transfer unit is M bits, the first data burst and the second data burst have P parallel transfer units, wherein P equals N divided by M, and N, M and P are positive integers greater than two.
