DDR5 SDRAM

Advantages of Migrating to DDR5

DDR5 is the next evolution in DRAM, bringing a robust list of new features geared to increase reliability, availability, and serviceability (RAS); reduce power; and dramatically improve performance. Some of the key feature differences between DDR4 and DDR5 are as follows:

Feature/Option DDR4 DDR5 DDR5 Advantage
 Data rates  1600-3200 MT/s  3200-6400 MT/s  Increases performance and bandwidth
 VDD/VDDQ/VPP   1.2/1.2/2.5   1.1/1.1/1.8   Lowers power
 Internal VREF  VREFDQ  VREFDQ, VREFCA, VREFCS   Improves voltage margins, reduces BOM costs
 Device densities    2Gb-16Gb    8Gb-64Gb    Enables larger monolithic devices 
 Prefetch    8n   16n    Keeps the internal core clock low
 DQ receiver equalization   CTLE  DFE  Improves opening of the received DQ data
 eyes inside the DRAM
 Duty cycle adjustment (DCA)   None   DQS and DQ  Improves signaling on the transmitted DQ/DQS pins
 Internal DQS delay
 monitoring 
 None   DQS interval oscillator   Increases robustness against environmental changes 
 On-die ECC  None  128b+8b SEC, error check and scrub   Strengthens on-chip RAS
 CRC   Write   Read/Write    Strengthens system RAS by protecting read data 
 Bank groups (BG)/banks   4 BG x 4 banks (x4/x8)
 2 BG x 4 banks (x16)
 8 BG x 2 banks (8Gb x4/x8)
 4 BG x 2 banks (8Gb x16)
 8 BG x 4 banks (16-64Gb x4/x8)
 4 BG x 4 banks (16-64Gb x16) 
 Improves bandwidth/performance
 Command/address interface   ODT, CKE, ACT, RAS,
 CAS, WE, A<X:0>
 CA<13:0> 

 Dramatically reduces the CA pin count

 ODT  DQ, DQS, DM/DBI   DQ, DQS, DM, CA bus    Improves signal integrity, reduces  BOM costs 
 Burst length  BL8 (and BL4)   BL16, BL32
 (and BC8 OTF, BL32 OTF) 
 Allows 64B cache line fetch with only 1 DIMM subchannel. 
 MIR (“mirror” pin)   None  Yes  Improves DIMM signaling
 Bus inversion   Data bus inversion (DBI)  Command/address inversion (CAI)   Reduces VDDQ noise on modules
 CA training, CS training   None   CA training, CS training   Improves timing margin on CA and CS pins  
 Write leveling training modes   Yes  Improved  Compensates for unmatched DQ-DQS path
 Read training patterns   Possible with the MPR  Dedicated MRs for serial
 (userdefined), clock and LFSR
 -generated training patterns
 Makes read timing margin more robust
 Mode registers  7 x 17 bits  Up to 256 x 8 bits
 (LPDDR type read/write) 
 Provides room to expand
 PRECHARGE commands   All bank and per bank  All bank, per bank, and same bank   PREsb enables precharging-specific bank in each BG
 REFRESH commands   All bank   All bank and same bank  REFsb enables refreshing of specific bank in each BG
 Loopback mode  None   Yes  Enables testing of the DQ and DQS signaling 

Featured Blog Posts

DDR5: The next step in system level performance

As CPU core counts continue to increase, bandwidth per core cannot continue to scale with DDR4. New memory architectures are required to meet next-generation bandwidth per core requirements in x86 CPUs. Enter DDR5.

DDR5 early collaboration with validation silicon enables the data currency of tomorrow

As systems feel the pressure of improving, the drive for increasing bandwidth for next generation memory is priority #1.

Micron leads the charge with DDR5 to improve the performance baseline for tomorrows applications

Providing products and solutions with confidence is really about getting applications to higher bandwidths faster.
+