Multi-channel dynamic random access memory
The MCDRAM addresses bandwidth and power efficiency limitations by employing a dual-mode operation and hybrid bonding, achieving doubled data rates and reduced power consumption with enhanced system performance and capacity.
Patent Information
- Application Number
- PCT/US2026/025673
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-06-04
- Filing Date
- 2026-04-28
- Publication Date
- 2026-11-12
AI Technical Summary
Existing DRAM technologies face limitations in bandwidth, signal integrity, and power efficiency at higher data rates, necessitating the development of alternative memory device and module architectures.
A multi-channel DRAM (MCDRAM) that can operate in two-channel (2CH) and one-channel (1CH) modes, featuring hybrid bonding technology, interleaved data access, and shared memory spaces, enabling increased data bandwidth and capacity without additional buffers, while maintaining compatibility with existing systems.
The MCDRAM achieves doubled data rates, improved bank availability, and reduced power consumption, offering superior bandwidth and capacity with seamless integration and socket compatibility, thus enhancing overall system performance and efficiency.
Smart Images

Figure US2026025673_12112026_PF_FP_ABST
Abstract
Description
Attorney Docket No.: 27170.1116 (L1074PCT)MULTI-CHANNEL DYNAMIC RANDOM ACCESS MEMORYBACKGROUND
[0001] Various computing systems use dynamic random access memory (DRAM) to store data. A DRAM device (also referred to as DRAM chip) is a semiconductor memory integrated circuit that stores data in capacitive cells organized in a grid of rows and columns. The capacitive cells need periodic refreshing to maintain data, and are typically organized into ranks accessed through one or more memory channels. Some computing systems employ these DRAM devices in one or more dual in-line memory modules (DIMMs) to provide volatile system memory. A DIMM includes one or more DRAM devices mounted on a printed circuit board and interfaces with a memory controller via a parallel electrical bus. In some systems, DIMMs may include a register (e.g., in registered DIMMs) to buffer command and address (CA) signals between the memory controller and the DRAM devices.
[0002] Although standard DIMMs offer scalable memory capacity and performance, limitations in bandwidth, signal integrity, and power efficiency at higher data rates have motivated the development of alternative memory device and memory module architectures.BRIEF DESCRIPTION OF THE DRAWINGS
[0003] The present disclosure is illustrated by way of example, and not by way of limitation, in the figures of the accompanying drawings.
[0004] FIG. 1 is a block diagram of a memory device, referred to as a multi-channel DRAM (MCDRAM) that supports a two-channel (2CH) mode and a one-channel (1CH) mode according to at least one embodiment.
[0005] FIG. 2 are timing diagrams of the 2CH mode and the 1CH mode according to at least one embodiment.
[0006] FIG. 3 A illustrates a hybrid bonded MCDRAM with a face-up memory layer according to at least one embodiment.
[0007] FIG. 3B illustrates a hybrid bonded MCDRAM with a face-down memory layer according to at least one embodiment.
[0008] FIG. 4A is a block diagram of a registered dual in-line memory module (RDIMM) according to at least one embodiment.Attorney Docket No.: 27170.1116 (L1074PCT)
[0009] FIG. 4B is a block diagram of a multi-channel dual in-line memory module (MCDIMM) according to at least one embodiment.
[0010] FIG. 5 are timing diagrams of the 2CH mode and the 1CH mode according to at least one embodiment.
[0011] FIG. 6 is a block diagram of modal data paths of an MCDRAM according to at least one embodiment.
[0012] FIG. 7 is a block diagram of modal column paths of an MCDRAM according to at least one embodiment.
[0013] FIG. 8 is a block diagram of modal column paths of an MCDRAM according to at least one embodiment.
[0014] FIG. 9 is a block diagram of a multi-channel registered clock driver (MCRCD) of an MCDIMM with MCDRAMs according to at least one embodiment.
[0015] FIG. 10 is a flow diagram of a method of operating an MCDRAM in either a one-channel mode or a two-channel mode according to at least one embodiment.DETAILED DESCRIPTION
[0016] Technologies directed to multi-channel DRAM devices are described. The following description sets forth numerous specific details, such as examples of specific systems, components, methods, and so forth, in order to provide a good understanding of several embodiments of the present disclosure. It will be apparent to one skilled in the art, however, that at least some embodiments of the present disclosure may be practiced without these specific details. In other instances, well-known components or methods are not described in detail or presented in simple block diagram format to avoid obscuring the present disclosure unnecessarily. Thus, the specific details set forth are merely exemplary. Particular implementations may vary from these exemplary details and still be contemplated to be within the scope of the present disclosure.
[0017] Aspects and embodiments of the present disclosure present a DRAM that is configurable into two distinct modes: the standard two-channel (2CH) mode and the one-channel (1CH) mode (also referred to herein as the MCDRAM mode). In the standard 2CH mode, the DRAM can be used to construct registered DIMMs (RDIMMs), featuring two independent interfaces and memory sections. In the 1CH mode, the DRAM interleaves the two similar accesses, which are controlled through a common command and address (CA) interface and transmitted using an interleaved protocol, akin to a multiplexed-channel DIMM. In the 1CH mode, the memory space may be sharedAttorney Docket No.: 27170.1116 (L1074PCT)between the two accesses, resulting in improved bank availability and optimized performance.
[0018] Aspects and embodiments of the present disclosure can provide for increased data bandwidth and capacity compared to a traditional RDIMM. By enabling the simultaneous combination of column accesses — either two reads or two writes — the data rate is effectively doubled without altering the DRAM. Aspects and embodiments of the present disclosure can maintain consistent access granularity, core frequency, and data I / O width, ensuring seamless integration with existing systems. Aspects and embodiments of the present disclosure can provide an MCDIMM without a buffer to interleave data from two column accesses, while preserving the latency for both operations. Aspects and embodiments of the present disclosure can double the capacity of traditional RDIMM with the same DRAM, at the same DIMM pin count and loading, without the need for an additional data buffer. This results in substantial cost and power savings in the DIMM, contributing to the overall efficiency of the system. Aspects and embodiments of the present disclosure can provide socket compatibility between MCDIMM and RDIMM sockets (i.e., same as DDR5 MRDIMM). Aspects and embodiments of the present disclosure can function similarly to RDIMM, with the ability to support DRAM configurations that cater to both RDIMM and MCDIMM setups. This flexibility provides a significant advantage across a range of application scenarios. Aspects and embodiments of the present disclosure can use hybrid bonding technology, enabling further scaling of DRAM bandwidth without increasing the DRAM core frequency. Two-channel DRAM die architectures may share timing signals, overlapping the internal data bus, which aligns with non-MCDIMM implementations. This approach ensures scalability and adaptability in evolving technological landscapes. Only a small die adder is needed for overlapping CA busses. Aspects and embodiments of the present disclosure can be configured as a package option (external pin) or mode register setting (MR setting), ensuring enhanced reliability and compatibility with existing and future systems. Aspects and embodiments of the present disclosure can enhance overall system performance and efficiency, and can ensure optimal utilization of each DRAM’s address space.
[0019] FIG. 1 is a block diagram of a memory device, referred to as MCDRAM 100 that supports a two-channel (2CH) mode and a one-channel (1CH) mode according to at least one embodiment. The MCDRAM 100 can be implemented as one or more DRAM dies. The DRAM die can be configurable into the 2CH mode having dedicated addressAttorney Docket No.: 27170.1116 (L1074PCT)space and dedicated command and address (CA) interface and data interface per channel (labeled as “DQ CHA” and “DQ CHB”) or the 1CH mode having a combined address space with one CA interface (e.g., first CA interface 106 labeled "CA CHA") and data interface (e.g., first data interface 102 labeled “DQ CHA”). During operation of the MCDRAM 100 in the 1CH mode, two access commands with independent memory addresses are received at the CA interface, in a combined format, and data interface and the data encoding for both accesses is interleaved at the data interface, as described in more detail below.
[0020] The MCDRAM 100 includes a first data interface 102, a second data interface 104, a first CA interface 106, a second CA interface 108, a first set of bank groups 110, a second set of bank groups 112, a first CA interface circuitry 114, a second CA interface circuitry 116, a first data interface circuitry 118, and a second data interface circuitry 120. Bank groups are subdivisions within a memory rank that help improve memory access efficiency and bandwidth. A bank is a sub-array of memory cells that can be independently activated, read from, or written to. Each bank contains a row decoder to select rows of memory cells, a sense amplifier to read and write data, and a column decoder for selecting bits within a row. Each bank group can contain multiple banks, and only one bank within a group can be accessed at a time with minimal delay. However, different bank groups can be accessed simultaneously or in rapid succession with reduced timing penalties, allowing for greater parallelism and higher data throughput.
[0021] The first CA interface circuitry 114 is coupled between the first CA interface 106, the first set of bank groups 110, and the second set of bank groups 112. The second CA interface circuitry 116 is coupled between the second CA interface 108 and the second set of bank groups 112. The first data interface circuitry 118 is coupled between the first data interface 102, the second data interface 104, and the first set of bank groups 110. The second data interface circuitry 120 is coupled between the first data interface 102, the second data interface 104, and the second set of bank groups 112.
[0022] During operation, in the 1CH mode, the first CA interface circuitry 114 can receive a first access command to access first data in either the first set of bank groups 110, or second set of bank groups 112 and a second access command to access second data in the set of bank groups not accessed by the first command. In at least one embodiment, the first CA interface circuitry 114 includes a first decoder and driver 122 and a second decoder and driver 124. The first decoder and driver 122 can receive the first access command to access the first data in the first set of bank groups 110. TheAttorney Docket No.: 27170.1116 (L1074PCT)second decoder and driver 124 can receive the second access command to access the second data in the second set of bank groups 112. In the 1CH mode, the first data interface circuitry 118 can interleave the first data and the second data on the first data interface 102. In at least one embodiment, the first and second access commands have shared memory addresses in the 1CH mode. It should be noted that the second CA interface circuitry 116 and the second data interface circuitry 120 are disabled in the 1CH mode.
[0023] During operation, in the 2CH mode, the first CA interface circuitry 114 can receive a third access command to access third data in the first set of bank groups 110. The second CA interface circuitry 116 can receive a fourth access command to access fourth data in the second set of bank groups 112. The first data interface circuitry 118 can send or receive the third data on the first data interface 102. The second data interface circuitry 120 can send or receive the fourth data on the second data interface 104. In at least one embodiment, the second CA interface circuitry 116 includes a third decoder and driver 126. In at least one embodiment, in the 2CH mode, the first decoder and driver 122 can receive the third access command to access the third data in the first set of bank groups 110, and the third decoder and driver 126 can receive the fourth access command to access the fourth data in the second set of bank groups 112.
[0024] In at least one embodiment, in the 2CH mode, the first set of bank groups 110 and the second set of bank groups 112 each has a dedicated address space. In at least one embodiment, in the 1CH mode, the first set of bank groups 110 and the second set of bank groups 112 have a combined address space. In at least one embodiment, in the 1CH mode, the encoding for the two access commands is interleaved at the CA interface, and the two access commands are not to the same bank or bank group. In at least one embodiment, the data transfer for the 1CH mode is the data interface from one channel of the 2CH mode. In at least one embodiment, the CA interface for the 1CH mode is the CA interface from one channel of the 2CH mode. In at least one embodiment, the CA interface for the 1CH mode is a subset of pins from both CA interfaces of the 2CH mode. In at least one embodiment, the DRAM die is hybrid bonded from two wafers. In at least one embodiment, one wafer is processed in a DRAM process and the other wafer is processed in a logic process. In at least one embodiment, the mode is set by a voltage level on a package pin. That is, the pin can receive a pin signal that indicates the mode. In at least one embodiment, the mode is set by one or more bits in a mode register.Attorney Docket No.: 27170.1116 (L1074PCT)
[0025] In at least one embodiment, the first data interface circuitry 118 includes a first receiver 128, a first deserializer 130 coupled to the first receiver 128, a second deserializer 132 coupled to the first receiver 128, a first output driver 134, a first serializer 136 coupled to the first output driver 134, and a second serializer 138 coupled to the first output driver 134. As illustrated in FIG. 1, the first data interface circuitry 118 includes a first interleaver 148 and a second interleaver 150. The first interleaver 148 and the second interleaver 150 can interleave the data for the first data interface 102. The first interleaver 148 is coupled between the first deserializer 130, the second deserializer 132, and the first receiver 128. The second interleaver 150 is coupled between the first serializer 136, the second serializer 138, and the first output driver 134. In at least one embodiment, the first interleaver 148 is a first-stage deserializer and the first deserializer 130 and the second deserializer 132 are second-stage deserializers. In at least one embodiment, the first serializer 136 and the second serializer 138 are first-stage serializers and the second interleaver 150 is a second-stage serializer.
[0026] During operation, in the 1CH mode, the first deserializer 130 can receive the first data from the first receiver 128, deserialize the first data, and provide the first data to the first set of bank groups 110, and the second deserializer 132 can receive the second data from the first receiver 128, deserialize the second data, and provide the second data to the second set of bank groups 112. Or, in the 1CH mode, the first serializer 136 can receive the first data from the first set of bank groups 110, serialize the first data, and provide the first data to the first output driver 134, and the second serializer 138 can receive the second data from the second set of bank groups 112, serialize the second data, and provide the second data to the first output driver 134.
[0027] In at least one embodiment, the second data interface circuitry 120 includes a second receiver 140, a third deserializer 142 coupled to the second receiver 140, a second output driver 144, and a third serializer 146 coupled to the second output driver 144. As illustrated in FIG. 1, the second data interface circuitry 120 also includes a fourth deserializer 152 and a fourth serializer 154. In at least one embodiment, the third deserializer 142 is a first-stage deserializer and the fourth deserializer 152 is a second-stage deserializer. In at least one embodiment, the third serializer 146 is a first-stage serializer and the fourth serializer 154 is a second-stage serializer.
[0028] During operation, in the 2CH mode, the first deserializer 130 can receive the third data from the first receiver 128, deserialize the third data, and provide the third data to the first set of bank groups 110, and the third deserializer 142 can receive the fourthAttorney Docket No.: 27170.1116 (L1074PCT)data from the second receiver 140, deserialize the fourth data, and provide the fourth data to the second set of bank groups 112. Or, in the 2CH mode, the first serializer 136 can receive the third data from the first set of bank groups 110, serialize the third data, and provide the third data to the second output driver 144, and the third serializer 146 can receive the fourth data from the second set of bank groups 112, serialize the fourth data, and provide the fourth data to the second output driver 144.
[0029] In at least one embodiment, the MCDRAM 100, which is configurable into a 2CH mode or a 1CH mode, can be used to build a regular RDIMM in the 2CH mode (also referred to as a “standard mode”) or an MCDIMM in the 1CH mode (also referred to as an "MCDRAM mode"). In the standard mode, the DRAM can have two independent interfaces and memory sections. In the MCDRAM mode, the DRAM interleaves two accesses, where each access is controlled through a common CA interface, such as the first CA interface 106 illustrated in FIG. 1, and transmitted with an interleaved protocol. The memory space for both accesses is shared for both accesses. This can improve bank availability.
[0030] As illustrated in FIG. 1, the 2CH Mode involves two separate channels. On the other hand, the MCDRAM Mode, also referred to as the 1CH mode, has several distinct features. In this mode, Channel B's command / address (CA) and data interface are disabled. Channel A's CA interface decodes both commands, such that one command is processed on even beats and another on odd beats. In the context of DRAM or memory interfaces, even and odd beats refer to individual data transfers within a burst on a DDR (Double Data Rate) bus. A "beat" is a single data transfer operation, and in DDR memory, data is transferred on both the rising and falling edges of the clock. Even beats correspond to data transferred on the rising edge of the clock, while odd beats correspond to data transferred on the falling edge. This terminology is used to distinguish between the two edges of the clock during burst operations. The data interface (DQ interface) of Channel A interleaves data from both commands, like during a read operation. Additionally, the address space overlaps in this mode, and the same bank group (BG) access timing restriction applies. In 2CH mode, internal buses dedicated to Channel A / B are repurposed for access 1 / 2 in MCDRAM mode.
[0031] Example operations in the 2CH mode and the 1CH mode are illustrated and described below with respect to FIG. 2.
[0032] FIG. 2 are timing diagrams of the 2CH mode and the 1CH mode according to at least one embodiment. A first timing diagram 202 shows timing of the MCDRAM 100Attorney Docket No.: 27170.1116 (L1074PCT)for four read commands in the 2CH mode, where two read commands are issued at the same time on the separate command channels, channel A (COM CH A) and channel B (COM CH B). The first CA interface 106 receives a first read command (RD BGO) on command channel A to read first data from the first set of bank groups 110, and the second CA interface 108 receives a second read command (RD BG1) on command channel B to read second data from the second set of bank groups 112. The first CA interface circuitry 114 can decode the first read command (RD BGO) and provide CA signals (RD BGO CH A) to the first set of bank groups 110 to read the first data from the first set of bank groups 110. The second CA interface circuitry 116 can decode the second read command (RD BG1) and provide CA signals (RD BG1 CH B) to the second set of bank groups 112 to read the second data from the second set of bank groups 112. A column address strobe (CAS) is the signal used to initiate a column read or write. Before another set of commands can be issued to the MCDRAM 100, there is a CAS-to-CAS delay, referred to as tCCD. The tCCD is a timing parameter that defines the minimum number of clock cycles required between issuing two successive column access commands (typically read or write commands) to different banks within the same rank. The tCCD determines how soon after one column command the next can be issued, ensuring data integrity and proper timing on the memory bus. As illustrated in the timing diagram 202, the tCCD is 8 clock cycles (8tCK). The value of tCCD affects memory throughput and efficiency, especially in high-performance systems that rely on tight command scheduling. After the tCCD, the first CA interface 106 receives a third read command (RD BG7) on command channel A to read third data from the first set of bank groups 110, and the second CA interface 108 receives a fourth read command (RD BG6) on command channel B to read fourth data from the second set of bank groups 112. The first CA interface circuitry 114 can decode the third read command (RD BG7) and provide CA signals (RD BG7 CH A) to the first set of bank groups 110 to read the third data from the first set of bank groups 110. The second CA interface circuitry 116 can decode the fourth read command (RD BG6) and provide CA signals (RD BG6 CH B) to the second set of bank groups 112 to read the fourth data from the second set of bank groups 112. A second timing diagram 204 shows timing of the MCDRAM 100 for the four read accesses in the 2CH mode in response to the read commands in the timing diagram 202.
[0033] A third timing diagram 206 shows timing of the MCDRAM 100 for four read commands in the 1CH mode where the two read commands are interleaved on a singleAttorney Docket No.: 27170.1116 (L1074PCT)CA interface, i.e., the same command channel, channel A (COM CH A). At odd beats (rising edges), the first CA interface 106 receives a first read command (RD BGO) on command channel A to read first data from the first set of bank groups 110. At even beats (falling edges), the first CA interface 106 receives a second read command (RD BG1) on command channel B to read second data from the second set of bank groups 112. As such, the first read command and the second read command are interleaved on one CA interface. The first CA interface circuitry 114 (e.g., first decoder and driver 122) can decode the first read command (RD BGO) and provide CA signals (RD BGO CH A) to the first set of bank groups 110 to read the first data from the first set of bank groups 110. The first CA interface circuitry 114 (e.g., second decoder and driver 124) can decode the second read command (RD BG1) and provide CA signals (RD BG1 CH B) to the second set of bank groups 112 to read the second data from the second set of bank groups 112. As illustrated in the third timing diagram 206, the MCDRAM 100 in the 1CH mode has the same tCCD as the MCDRAM 100 in the 2CH mode (e.g., 8 clock cycles (8tCK)). The interleaved protocol maintains constant access latency using one interface. In at least one embodiment, in 1CH mode, a controller can schedule the second read command (RD BG1) to CH A address through channel B interface versus the 2CH mode, where the address space is channel specific. This provides more flexible scheduling in the controller. A fourth timing diagram 208 shows timing of the MCDRAM 100 for the four read accesses in the 1CH mode where the two read accesses are interleaved on a single data interface in response to the read commands in the third timing diagram 206. In this example, the data bit rate (toggle frequency) is 2x the bit rate of the second timing diagram 204.
[0034] It should be noted that the second CA interface circuitry 116 and the second data interface circuitry 120 are disabled in the 1CH mode.
[0035] As illustrated, in the 1CH mode, Channel B's command / address (CA) and data interface are disabled, consolidating operations into Channel A. Channel A decodes commands on alternating beats and interleaves data from each command via its DQ interface during read operations. This results in a single command / address interface serving two similar access operations, effectively maintaining consistent access latency.
[0036] The 1CH mode utilizes an interleaved protocol that, in an embodiment, provides for constant access latency by relying on a single interface. This mode also provides more flexibility in scheduling, allowing the controller to manage read operations to Channel A's address through the previously designated "Channel B" interface. ComparedAttorney Docket No.: 27170.1116 (L1074PCT)to the 2CH mode, where address spaces are specific to each channel, the 1CH mode allows for versatile scheduling within the controller.
[0037] Overall, the 1CH mode repurposes internal buses and maintains the same bank group access timing restrictions as in 2CH mode in an embodiment.
[0038] FIG. 3 A illustrates a hybrid bonded MCDRAM 300 with a face-up memory layer according to at least one embodiment. Hybrid bonding (also called direct bond interconnect or DBI) is a die-to-die or wafer-to-waf er bonding technique where copper pads and dielectric surfaces from two chips (or wafers) are directly bonded together at the atomic level. Hybrid bonding does not require TSVs or microbumps for interconnects, enabling extremely fine pitch (e.g., <10 pm), allowing millions of vertical connections between stacked dies.
[0039] Hybrid bonding enables the direct integration of a logic base die — such as a memory controller or physical interface (PHY) — with one or more DRAM array dies. As illustrated in FIG. 3A, the hybrid bonded MCDRAM 300 includes a logic base die 302 (also referred to as logic layer) that is hybrid bonded to a DRAM array die 304 (also referred to as face-up memory layer). In at least one embodiment, the logic base die 302 is manufactured using a logic process, and the DRAM array die 304 is manufactured using a memory process. In this embodiment, the DRAM array die 304 is face up. In other words, the circuitry on the logic base die 302 and the circuitry on the DRAM array die 304 is facing the same direction. This close coupling of logic and memory supports significantly increased data throughput and localized processing, similar to high-bandwidth memory (HBM) architectures, but with improved vertical connectivity and interconnect density.
[0040] FIG. 3B illustrates a hybrid bonded MCDRAM 350 with a face-down memory layer according to at least one embodiment. As illustrated in FIG. 3B, the hybrid bonded MCDRAM 350 includes the logic base die 302 (also referred to as logic layer) that is hybrid bonded to the DRAM array die 352 (also referred to as face-down memory layer). In at least one embodiment, the logic base die 302 is manufactured using a logic process, and the DRAM array die 352 is manufactured using a memory process. In this embodiment, the DRAM array die 352 is face down. In other words, the circuitry on the logic base die 302 is facing a first direction and the circuitry on the DRAM array die 352 are facing a second direction towards the logic base die 302. In at least one embodiment, the DRAM array die 352 and the DRAM array die 304 are the same, but are hybrid bonded in the face-up direction or face-down direction relative to the logic base die 302.Attorney Docket No.: 27170.1116 (L1074PCT)
[0041] In some embodiments, the MCDRAM 100 can be used to build a registered dual in-line memory module (RDIMM) in the 2CH mode, as illustrated in FIG. 4A, and a multi-channel dual in-line memory module (MCDIMM) in the 1CH mode, as illustrated in FIG. 4B. The MCDIMM with MCDRAM 100 presents a significant advancement in memory technology. By utilizing twice the capacity at the same pin count and load as the RDIMM, it effectively eliminates the need for at least data buffers, leading to notable power savings for DIMMs. This MCDIMM with MCDRAM 100 can offer superior bandwidth, achieving 33% higher bandwidth at two-thirds the clock and DRAM core frequency, and doubling the bandwidth at full clock and core frequency.
[0042] FIG. 4A is a block diagram of an RDIMM 400 according to at least one embodiment. As illustrated in FIG. 4A, the RDIMM 400 includes multiple MCDRAMs 100, and a registered clock driver (RCD) 402 (also referred to as command buffer or CA buffer).
[0043] FIG. 4B is a block diagram of an MCDIMM 450 according to at least one embodiment. As illustrated in FIG. 4B, the MCDIMM 450 includes multiple MCDRAMs 100, and a multi-channel RCD (MCRCD) 452 (also referred to as command buffer or CA buffer). As described above, by the MCDIMM 450 utilizing twice the capacity at the same pin count and load as the RDIMM 400, the MCDIMM 450 effectively eliminates the need for at least data buffers. Example timing diagrams of the RDIMM 400 in the 2CH mode and MCDIMM 450 in the 1CH mode are illustrated and described below with respect to FIG. 5.
[0044] FIG. 5 are timing diagrams of the 2CH mode and the 1CH mode according to at least one embodiment. A first timing diagram 502 shows timing of the RDIMM 400 for six read commands in the 2CH mode, where two read commands are issued at the same time on the separate command channels, channel A (CA PCHA) and channel B (CA PCHB). The system runs at a lx CA clock frequency (tCK=lx). The RCD 402 receives a first read command (RD BG0A / 1 A) on command channel A to read first data, and a second read command (RD BG0B / 1B) on command channel B to read second data. After the tCCD, the RCD 402 receives a third read command (RD BG7A / 3 A) on command channel A to read third data, and a fourth read command (RD BG7B / 3B) on command channel B to read second data. After the tCCD, the RCD 402 receives a fifth read command (RD BG7A / 3 A) on command channel A to read fifth data, and a sixth read command (RD BG7B / 3B) on command channel B to read sixth data. A second timingAttorney Docket No.: 27170.1116 (L1074PCT)diagram 504 shows timing of the RDIMM 400 for the six read accesses in the 2CH mode in response to the read commands in the timing diagram 502.
[0045] A third timing diagram 506 shows timing of the MCDIMM 450, with the system running at 2 / 3 the frequency of the RDIMM, or tCK=1.5x for six read commands in the 1CH mode where the two read commands are interleaved on a single CA interface, i.e., the same command channel, channel A. At odd beats (rising edges), the MCRCD 452 receives a first read command on command channel A to read first data. At even beats (falling edges), the MCRCD 452 receives a second read command on command channel B to read second data. As such, the first read command and the second read command are interleaved on one CA interface. At the same time, at odd beats (rising edges), the MCRCD 452 receives a third read command on command channel B to read third data. At even beats (falling edges), the MCRCD 452 receives a fourth read command on command channel B to read fourth data. As such, the third read command and the fourth read command are interleaved on the other CA interface. A fourth timing diagram 508 shows timing of the MCDIMM 450 for the six read accesses in the 1CH mode where the two read accesses are interleaved on a single data interface in response to the read commands in the third timing diagram 506.
[0046] A fifth timing diagram 510 shows timing of the MCDIMM 450 for six read commands in the 1CH mode where the two read commands are interleaved on a single CA interface, i.e., the same command channel, channel A. The fifth timing diagram 510 is similar to the third timing diagram 506, except the clock cycle is lx in the third timing diagram 506, whereas the clock cycle is lx in the fifth timing diagram 510. A sixth timing diagram 512 shows timing of the MCDIMM 450 for the six read accesses in the 1CH mode where the two read accesses are interleaved on a single data interface in response to the read commands in the fifth timing diagram 510.
[0047] As illustrated in FIG. 5, the MCDIMM 450 has 33% higher BW at 2 / 3 clock and DRAM core frequency, as compared to the RDIMM 400. The MCDIMM 450 also has twice the bandwidth at lx clock cycles and core frequency, as compared to the RDIMM 400.
[0048] As described herein, in a 2CH mode (also referred to as RDIMM mode), the system operates on two channels with separate memory spaces. In a 1CH mode (also referred to as the MCDIMM mode), the system uses a single channel with interleaved access and one address space. This can be accomplished with an interleaver in one of the serializer or deserializer stages, as illustrated and described below with respect to FIG. 6.Attorney Docket No.: 27170.1116 (L1074PCT)
[0049] FIG. 6 is a block diagram of modal data paths of an MCDRAM 600 according to at least one embodiment. In this embodiment, the MCDRAM 600 includes multiple modal data paths to accommodate the RDIMM mode and the MCDIMM mode. In the MCDIMM mode, interleaving is managed by SER Stage 1.5 602. The SER Stage 1.5 602 can alternate between even and odd read data, while SER stage 2604 interleaves data from the first command (Com 1) and the second command (Com 2). It should be noted in the 2CH mode, the SER stage 2604 is responsible for interleaving even and odd data. Similarly, the write process, which is managed by the DES Stage 1.5 606 and DES Stage 2 608, follows the same interleaving pattern. Notably, the Read Write Data (RWD) for Channel B will overlap with that for Channel A, ensuring no die adder is necessary.
[0050] FIG. 7 is a block diagram of modal column paths of an MCDRAM 700 according to at least one embodiment. The MCDRAM 700 includes a serializer 702 (SER) that separates CA signals based on their arrival during even and odd beats. In a 2CH mode, both even and odd CA signals are decoded as a single command and address. Conversely, in a 1CH mode, even CA signals are decoded as Command and Address 1 (Com+Addl), while odd CA signals are decoded as Command and Address 2 (Com+Add2). Bank group (BG) multiplexers connect either CA to any bank group (BG) based on the BG address, ensuring efficient data processing. This implementation utilizes a subset of both channel pins in backup, with a small die adder for overlapping CA buses in 2CH DRAM efficiency mode.
[0051] FIG. 8 is a block diagram of modal column paths of an MCDRAM 800 according to at least one embodiment. The MCDRAM 800 is similar to the MCDRAM 700 except some CA pins can be repurposed for the 1CH mode. For example, to maintain the same command packet length in both 2CH mode and 1CH mode, some CA pins from channel B can be repurposed for 1CH mode. If the mode is set to 0, no pins are needed, but if the mode is set to N, all pins are required.
[0052] FIG. 9 is a block diagram of an MCRCD 900 of a MCDIMM with MCDRAMs according to at least one embodiment. The MCRCD 900 can manage CA packet width and clock frequency translation to ensure compatibility with socket pins, providing an alternative conversion between DDR and SDR. In this embodiment the clock frequency to the DRAM is fixed and the clock rate on the controller facing clock is changed according to the width ratio between host and DRAM facing CA rate.
[0053] FIG. 10 is a flow diagram of a method 1000 of operating a MCDRAMs in either a one-channel mode or a two-channel mode according to at least one embodiment. TheAttorney Docket No.: 27170.1116 (L1074PCT)method 1000 may be performed by processing logic that may comprise hardware (e.g., circuitry, dedicated logic, programmable logic, microcode, etc.), software (e.g., instructions run on a processing device to perform hardware simulation), or a combination thereof. In one embodiment, the method 1000 is performed by the MCDRAM 100 of FIG. 1, hybrid bonded MCDRAM 300 of FIG. 3 A, hybrid bonded MCDRAM 350 of FIG. 3B, RDIMM 400 of FIG. 4A, MCDIMM 450 of FIG. 4B, MCDRAM 600 of FIG. 6, or MCDRAM 700 of FIG. 7. Alternatively, other devices can perform the method 1000.
[0054] Referring to FIG. 10, the method 1000 begins with the processing logic determining whether a pin signal or a mode register indicates the DRAM die is configured in the 2CH mode or the 1CH mode. At block 1004, the processing logic, responsive to determining that the DRAM die is in the 1CH mode, receives a first access command to access first data in a first set of bank groups and a second access command to access second data in a second set of bank groups (block 1006). The first access command and the second access command are interleaved on a first CA interface. At block 1008, the processing logic interleaves the first data and the second data on a first data interface. Or, at block 1010, the processing logic, responsive to determining that the DRAM die is in the 2CH mode, receives a third access command to access third data in the first set of bank groups on the first CA interface (block 1012). At block 1014, the processing logic receives a fourth access command to access fourth data in the second set of bank groups. At block 1016, the processing logic sends or receives the third data on the first data interface. At block 1018, the processing logic sends or receives the fourth data on a second data interface. In other embodiments, the method 1000 can perform other operations described herein.
[0055] It is to be understood that the above description is intended to be illustrative and not restrictive. Many other implementations will be apparent to those of skill in the art upon reading and understanding the above description. Therefore, the scope of the disclosure should be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled.
[0056] In the above description, numerous details are set forth. It will be apparent, however, to one skilled in the art that the aspects of the present disclosure may be practiced without these specific details. In some instances, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring the present disclosure.Attorney Docket No.: 27170.1116 (L1074PCT)
[0057] Some portions of the detailed descriptions above are presented in terms of algorithms and symbolic representations of operations on data bits within a computer memory. These algorithmic descriptions and representations are the means used by those skilled in the data processing arts to most effectively convey the substance of their work to others skilled in the art. An algorithm is here, and generally, conceived to be a self-consistent sequence of steps leading to the desired result. The steps are those requiring physical manipulations of physical quantities. Usually, though not necessarily, these quantities take the form of electrical or magnetic signals capable of being stored, transferred, combined, compared, and otherwise manipulated. It has proven convenient at times, principally for reasons of common usage, to refer to these signals as bits, values, elements, symbols, characters, terms, numbers, or the like.
[0058] However, it should be borne in mind that all of these and similar terms are to be associated with the appropriate physical quantities and are merely convenient labels applied to these quantities. Unless specifically stated otherwise, as apparent from the following discussion, it is appreciated that throughout the description, discussions utilizing terms such as “receiving,” “determining,” “selecting,” “storing,” “setting,” or the like, refer to the action and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (electronic) quantities within the computer system’s registers and memories into other data similarly represented as physical quantities within the computer system memories or registers or other such information storage, transmission or display devices.
[0059] The present disclosure also relates to an apparatus for performing the operations herein. This apparatus may be specially constructed for the required purposes, or it may comprise a general-purpose computer selectively activated or reconfigured by a computer program stored in the computer. Such a computer program may be stored in a computer-readable storage medium, such as, but not limited to, any type of disk, including floppy disks, optical disks, CD-ROMs, and magnetic-optical disks, read-only memories (ROMs), random access memories (RAMs), erasable programmable ROMs (EPROMs), electrically erasable programmable ROMs (EEPROMs), magnetic or optical cards, or any type of media suitable for storing electronic instructions, each coupled to a computer system bus.
[0060] The algorithms and displays presented herein are not inherently related to any particular computer or other apparatus. Various general-purpose systems may be used with programs in accordance with the teachings herein, or it may prove convenient toAttorney Docket No.: 27170.1116 (L1074PCT)construct more specialized apparatuses to perform the required method steps. The required structure for a variety of these systems will appear as set forth in the description. In addition, aspects of the present disclosure are not described with reference to any particular programming language. It will be appreciated that a variety of programming languages may be used to implement the teachings of the present disclosure as described herein.
[0061] Aspects of the present disclosure may be provided as a computer program product, or software, that may include a machine-readable medium having stored thereon instructions, which may be used to program a computer system (or other electronic devices) to perform a process according to the present disclosure. A machine-readable medium includes any procedure for storing or transmitting information in a form readable by a machine (e.g., a computer). For example, a machine-readable (e.g., computer-readable) medium includes a machine (e.g., a computer) readable storage medium (e.g., read-only memory (“ROM”), random access memory (“RAM”), magnetic disk storage media, optical storage media, flash memory devices, etc.).
Claims
Attorney Docket No.: 27170.1116 (L1074PCT)CLAIMSWhat is claimed is:
1. A dynamic random access memory (DRAM) die, configurable into at least one of two modes including:a two-channel (2CH) mode having dedicated address space and dedicated command and address (CA) interface and data interface per channel; anda one-channel (1CH) mode having a combined address space with one CA interface and data interface, wherein two access commands with independent memory addresses are combined at the CA interface and data interface and the data encoding for both accesses is interleaved at the data interface.
2. The DRAM die of claim 1, wherein in the 1CH mode:the encoding for the two access commands is interleaved at the CA interface; and the two access commands are not to the same bank or bank group.
3. The DRAM die of claim 1, wherein the data interface for the 1CH mode is the data interface from one channel of the 2CH mode.
4. The DRAM die of claim 1, wherein the CA interface for the 1CH mode is the CA interface from one channel of the 2CH mode.
5. The DRAM die of claim 1, wherein the CA interface for the 1CH mode is a subset of pins from both CA interfaces of the 2CH mode.
6. The DRAM die of claim 1, wherein the DRAM die is hybrid bonded from two wafers.
7. The DRAM die of claim 6, wherein one wafer is processed in a DRAM process and the other wafer is processed in a logic process.
8. The DRAM die of claim 1, wherein the mode is set by a voltage level on a package pin.Attorney Docket No.: 27170.1116 (L1074PCT)9. A dynamic random access memory (DRAM) die comprising:a first data interface and a second data interface;a first command and address (CA) interface and a second CA interface;a first set of bank groups and a second set of bank groups;first CA interface circuitry coupled between the first CA interface, the first set of bank groups, and the second set of bank groups;second CA interface circuitry coupled between the second CA interface and the second set of bank groups;first data interface circuitry coupled between the first data interface, the second data interface, and the first set of bank groups; andsecond data interface circuitry coupled between the first data interface, the second data interface, and the second set of bank groups, wherein, in a one-channel (1CH) mode:the first CA interface circuitry is to receive a first access command to access first data in the first set of bank groups and a second access command to access second data in the second set of bank groups;the first data interface circuitry is to interleave the first data and the second data on the first data interface; andthe second CA interface circuitry and the second data interface circuitry are disabled in the 1CH mode.
10. The DRAM die of claim 9, wherein, in a two-channel (2 CH) mode:the first CA interface circuitry is to receive a third access command to access third data in the first set of bank groups;the second CA interface circuitry is to receive a fourth access command to access fourth data in the second set of bank groups;the first data interface circuitry is to send or receive the third data on the first data interface; andthe second data interface circuitry is to send or receive the fourth data on the second data interface.
11. The DRAM die of claim 10, wherein:in the 2CH mode, the first set of bank groups and the second set of bank groups each has a dedicated address space; andAttorney Docket No.: 27170.1116 (L1074PCT)in the 1CH mode, the first set of bank groups and the second set of bank groups have a combined address space.
12. The DRAM die of claim 10, wherein the first access command and the second access command have independent memory addresses.
13. The DRAM die of claim 10, wherein:the first CA interface circuitry comprises:a first decoder and driver;a second decoder and driver; andthe second CA interface circuitry comprises a third decoder and driver.
14. The DRAM die of claim 13, wherein:in the 1CH mode, the first decoder and driver is to receive the first access command to access the first data in the first set of bank groups;in the 1CH mode, the second decoder and driver is to receive the second access command to access the second data in the second set of bank groups;in the 2CH mode, the first decoder and driver is to receive the third access command to access the third data in the first set of bank groups; andin the 2CH mode, the third decoder and driver is to receive the fourth access command to access the fourth data in the second set of bank groups.
15. The DRAM die of claim 13, wherein:the first data interface circuitry comprises:a first receiver;a first deserializer coupled to the first receiver;a second deserializer coupled to the first receiver;a first output driver;a first serializer coupled to the first output driver; anda second serializer coupled to the first output driver;the second data interface circuitry comprises:a second receiver;a third deserializer coupled to the second receiver;a second output driver; andAttorney Docket No.: 27170.1116 (L1074PCT)a third serializer coupled to the second output driver.
16. The DRAM die of claim 15, wherein:in the 1CH mode, the first deserializer is to receive the first data from the first receiver, deserialize the first data, and provide the first data to the first set of bank groups; andin the 1CH mode, the second deserializer is to receive the second data from the first receiver, deserialize the second data, and provide the second data to the second set of bank groups; orin the 1CH mode, the first serializer is to receive the first data from the first set of bank groups, serialize the first data, and provide the first data to the first output driver; andin the 1CH mode, the second serializer is to receive the second data from the second set of bank groups, serialize the second data, and provide the second data to the first output driver.
17. The DRAM die of claim 15, wherein:in the 2CH mode, the first deserializer is to receive the third data from the first receiver, deserialize the third data, and provide the third data to the first set of bank groups; andin the 2CH mode, the third deserializer is to receive the fourth data from the second receiver, deserialize the fourth data, and provide the fourth data to the second set of bank groups; orin the 2CH mode, the first serializer is to receive the third data from the first set of bank groups, serialize the third data, and provide the third data to the second output driver; andin the 2CH mode, the third serializer is to receive the fourth data from the second set of bank groups, serialize the fourth data, and provide the fourth data to the second output driver.
18. A multi-channel dual in-line memory module (MCDIMM) comprising:a plurality of dynamic random access memory (DRAM) devices, wherein each of the plurality of DRAM devices is configured into at least one of two modes including:Attorney Docket No.: 27170.1116 (L1074PCT)a two-channel (2CH) mode having dedicated address space and dedicated command and address (CA) interface and data interface per channel; anda one-channel (1CH) mode having a combined address space with one CA interface and data interface, wherein two access commands with independent memory addresses are combined at the CA interface and data interface and the data encoding for both accesses is interleaved at the data interface.
19. The MCDIMM of claim 18, further comprising a command buffer coupled between the dedicated CA interface and the plurality of DRAM devices.
20. The MCDIMM of claim 18, wherein in the 1CH mode:the encoding for the two access commands is interleaved at the CA interface; and the two access commands are not to the same bank or bank group.
