Future Wei Technologies, Inc.

Huawei Technologies Co., Limited

Xiaobing Lee

Feng Yang

Yunxiang Wu

Yu Meng

Ting Li

Yansong Wang

Yong Tai Chen

Michael Young

Xiangyu Tang

Chunlei Dong

Ken Hu

Shaojie Chen

A method performed by a processor to improve wear-leveling in a cross-point (XD) memory, comprises detecting, by a processor coupled to the XD memory, a trigger event, wherein the XD memory comprises a first section of memory units and a second section of memory units, and in response to detecting the trigger event, relocating, by the processor, data stored in a first memory unit of the first section of memory units to a memory unit adjacent to a last memory unit of the first section of memory units, and relocating, by the processor, data stored in a first memory unit of the second section of memory units to a memory unit adjacent to a last memory unit of the second section of memory units.

Wear-leveling method for cross-point memory for multiple data temperature zones

A transaction-based hybrid memory device includes a host memory controller to control operation of the device. A hybrid memory controller is coupled to the host memory controller over a memory bus. The hybrid memory controller includes non-volatile memory control logic to control operation of non-volatile memory devices and cache control logic to accelerate cache operations, a direct memory access (DMA) engine to control volatile cache memory and to transfer data between non-volatile memory, and cache memory to off load host cache managements and transactions. A host interface couples the host memory controller to the memory bus.

Transaction-based hybrid memory

A hybrid DDR5 DIMM device includes a PCB board with a host interface through one of two DDR5 sub-channels, and a plurality of DDR4 or slow DDR5 SDRAM chips on the PCB coupled to this single channel DDR5 host interface. An embodiment processing system includes a host CPU to access one or more pairs of hybrid DDR5 DIMM devices for 4&#xd7;DDR5 memory capacities (4DPC), a first or second hybrid DDR5 DIMM including a plurality of half-speed SDRAM chips, and a first or second DDR5 sub-channel coupled the host with slow SRAM chips on DIMM. Mounting same data-buffer and RCD chips on hybrid DIMM to a server motherboard can double available DDR4 DIMMs' speed to DDR5 speed rate. Pairs of hybrid DDR5 DIMM devices cascaded one-by-one can aggregate more DDR5 DIMM devices to expand memory capacities at double speed of DDR4 or DDR5 SDRAM chips, beyond current DDR5 speed limit 6400 MT/s.

Systems and methods for utilizing DDR-DRAM chips in hybrid DDR-DIMMs and for cascading DDR-DIMMs

An apparatus comprises a hybrid-memory multi-chip package (MCP) including a non-volatile memory (NVM) in an NVM die; a dynamic random access memory (DRAM) in two or more DRAM die, wherein a portion of the DRAM is allocated as a cache memory for the NVM; and a hybrid controller for the NVM and DRAM. The hybrid controller includes an NVM interface to the NVM; a DRAM interface to the cache memory; a host interface to communicate data with a host processor, wherein the host interface includes a parallel data bus for reading and writing data directly with both of the DRAM and the NVM; and logic circuitry configured to interleave access by the host processor and hybrid controller to the DRAM and NVM.

Hybrid LPDDR4-DRAM with cached NVM and flash-NAND in multi-chip packages for mobile devices

System and method for a unified memory and network controller for an all-flash array (AFA) storage blade in a distributed flash storage clusters over a fabric network. The unified memory and network controller has 3-way control functions including unified memory buses to cache memories and DDR4-AFA controllers, a dual-port PCIE interconnection to two host processors of gateway clusters, and four switch fabric ports for interconnections with peer controllers (e.g., AFA blades and/or chassis) in the distributed flash storage network. The AFA storage blade includes dynamic random-access memory (DRAM) and magnetoresistive random-access memory (MRAM) configured as data read/write cache buffers, and flash memory DIMM devices as primary storage. Remote data memory access (RDMA) for clients via the data caching buffers is enabled and controlled by the host processor interconnection(s), the switch fabric ports, and a unified memory bus from the unified controller to the data buffer and the flash SSDs.

RDMA-SSD dual-port unified memory and network controller

Various embodiments include methods and apparatus structured to provide synchronization of a transaction identification between a host and a memory module using a parity check. A transaction identification can be generated at both the host and the memory module independently using incremental counters of these apparatus. Synchronization of the transaction identifications generated by the host and by a controller of the memory module can be implemented using a parity bit sequences pattern of a combination of the generated transaction identification plus the corresponding transaction command and data address. Use of transaction commands modified with respect to transaction identifications can be used in initialization of the synchronization, in message passing, and in error detection and response to errors. Additional apparatus, systems, and methods can be implemented in a variety of applications.

Transaction identification synchronization

A computer-implemented method for storing and caching data in an all-flash-array includes erasing a TLC-NAND flash cell and programming the cell with a binary value multiple times in sequence corresponding to multiple sequential stages between erasures. The method also includes processing the binary value in relation to a respective threshold voltage at each of the multiple sequential stages. The method further includes storing metadata corresponding to a current stage associated with the number of times the TLC-NAND flash cell has been programmed since being erased.

All-flash-array primary storage and caching appliances implementing triple-level cell (TLC)-NAND semiconductor microchips

Hybrid LPDDR4-DRAM with cached NVM and flash-nand in multi-chip packages for mobile devices

As a solution to the type of problems noted above, this disclosure provides novel methods and systems that include dual-port solid-state drive (SSD) DIMM devices to provide primary storage capabilities with very low latency and better availability of DDR4 devices. The dual-port DDR4-SSD flash memory devices guarantee primary storage devices still accessible with one CPU or network failure. The novel DDR4 memory bus devices may be used not only for memory media and storage device buffers, but also to allow two CPUs to share data stored in flash SSD chips and to greatly improve DDR4 bus efficiency and bus utilizations by block accesses and eliminate PCIE-DMA data transfers. Through the features of the claimed subject matter described herein, dual-port DDR4-DIMM memory devices can be achieved that provide an All-Flash-Array storage system with substantially higher reliability, availability, and performance over conventional SATA/SAS-SSD, PCIE-SSD, and NVME-SSD solutions.

DDR4-SSD dual-port DIMM device

A memory system is disclosed that includes a first FPGA controller coupled to a first SSD cluster, a first DDR4 DIMM and a second DDR4 DIMM. A second FPGA controller is coupled to a second SSD cluster, the first DDR4 DIMM and the second DDR4 DIMM, where the first and second FPGAs are operable to share access to the first and second DDR4 DIMMs and provide connectivity to a plurality of network resources. The dual-port design enables the use of existing SDRAM, MRAM and RRAM chips at low speed rates to reach DDR4 2.0 speed DIMM devices. The dual-port DDR4 DIMM comprises 1-to-2 data buffer splitters and a DDR3 or DDR2 to DDR4 bus adaptation/termination/relaying circuits to increase (e.g., double or quadruple) the chip speed of SDRAM, MRAM, and RRAM chips.

Dual-port DDR4-DIMMs of SDRAM and NVRAM for SSD-blades and multi-CPU servers

System and method embodiments are provided for managing storage systems. In an embodiment, a network component for managing data storage includes a storage interface configured to couple to a plurality of storage devices; and a vector-direct memory access (DMA) cache-exclusive OR (XOR) engine coupled to the storage interface and configured for a multiple parities convolution codes (MPCC) erasure coding to accelerate M parities parallel calculations and the erasures cross-iterations decoding, wherein a single XOR-engine with caches and a vector-DMA address generator is shared by the MPCC erasure coding engine for pipelining external dual data rate (DDR4) memory accesses, where M is a positive integer greater than two.

Growing community of inventors

Santa Clara, CA, United States of America

Xiaobing Lee