JEDEC and the JC-42.6 Subcommittee for Low Power Memories has announced the publication of the new JESD209-5B standard which now includes improvements to LPDDR5, as well as an extension for the standard in the form of new LPDDR5X.

The new LPDDR5X standard is an evolutionary step over LPDDR5, further increasing the data rates possible by 33% from 6400Mbps to 8533Mbps.

The industry had first shifted over to the LPDDR5 memory standard back in 2020, with the first generation SoCs and memory modules running at a speed of 5500Mbps which had gotten an upgrade to 6400Mbps speeds in recent 2021 flagship devices.

As IP vendors and DRAM manufacturers have matured their LPDDR5 products, there’s an increasing need to look for further updates to the LPDDR5 standard as we’ve essentially reached the maximum data rates of the standard in current generation implementations.  

LPDDR Generations
  LPDDR3 LPDDR4 LPDDR4X LPDDR5 LPDDR5X
Max Density 32 Gbit 64 Gbit 32 Gbit
Max Data Rate 2133Mbps 4266Mbps 6400Mbps 8533Mbps
Channels 1 2 1
Width x32 x32 (2x x16) x16
Banks
(Per Channel)
8 8 8-16 16
Bank Grouping No No Yes
Prefetch 8n 16n 16n
Voltage 1.2v 1.1v Variable
(Max 1.1v)
Vddq 1.2v 1.1v 0.6v 0.6v

While we currently don’t have access to the official documentation to detail the exact changes, back in February Cadence had written more extensively about the new improvements in LP5X over LP5:

  • To Improve READ SI performance in the dual rank system at high speeds that Lpddr5X devices support a Unified NT-ODT Behavior has been defined. Unified NT-ODT is a requirement for all LPDDR5X devices
  • To support high data rates for Lpddr5X, we need a way to compensate for transmission loss. This has been achieved by defining the pre-emphasis function. Lpddr5X devices have pull up or down pre-emphasis for each of the lower/upper byte lane programming.
  • Rx Offset Calibration Training - LPDDR5X SDRAM provides Offset Calibration Training for adjusting DQ Rx offset and Offset Calibration Training is recommended for every power-up and initialization training sequence to cope with the SDRAM operating condition change
  • Extended Latencies - LPDDR5X SDRAM devices support extended Read, Write, nWR, ODTLon and ODTLoff Latency Values to account for longer number of cycle it takes to do the data access to memory array. WCK2CK Sync AC Parameters are also extended.
  • LPDDR5X SDRAM Devices support Per-pin controlled Decision Feedback Equalization: DFE. This includes new Mode Registers 70/71/72/73/74.
  • New LPDDR5X SDRAM Device specific Clock AC Timings for 937.5/1066.5MHz and Write Clock AC Timings for 3750/4266.5MHz.
  • New Mode register fields or additional conditions on the use of existing fields have been added to several Mode registers for LPDDR5X devices. Some of the examples of changed MR are MR0, MR1, MR2, MR13, MR15, MR41, MR58, MR69, etc.
  • LPDDR5X SDRAM devices do not support 8 Bank Mode of operations. 8 Bank Mode doesn’t offer the architectural benefit of more bank interleaving resources and core operation timings at high speed that 16B and BG Mode have. It is specially limiting for high speed LPDDR5X devices support leading JEDEC to drop 8 Bank Mode support for LPDDR5X.

In short, the new standard covers deeper tweaks to the architecture and operating mode of the memory standard to achieve higher data-rates.

For eventual mobile SoCs using 8533Mbps memory, the peak theoretical available bandwidth would grow from 51.2GB/s to 68.26GB/s, allowing future designs to further increase CPU and GPU performances. It’s to be noted, that we haven’t heard much about power efficiency improvements of the new LP5X standard, so I assume that we’ll be relying on DRAM vendors to improve power efficiency via more advanced manufacturing nodes to keep total power usage inside of devices in check.

I estimate we’ll be seeing LPDDR5X in future late 2022 or 2023 SoCs.

Related Reading:

POST A COMMENT

17 Comments

View All Comments

  • lemurbutton - Thursday, July 29, 2021 - link

    Nice, it'll be great for Apple's M3/M3X SoCs. Reply
  • darkswordsman17 - Thursday, July 29, 2021 - link

    Won't be enough outside of iPhones (which likely will stick to A series SoCs). I'm curious why Apple would bother with LPDDR if they add HBM, but seems like they plan on a hybrid of the two, but I'd think a unified memory pool would work better. Reply
  • okoroezenwa - Thursday, July 29, 2021 - link

    It probably would if the M1 is fine with 68GB/s. In the same config this will be ~135GB/s in a hypothetical M3, however no idea of a hypothetical M3X considering we don't know what type of RAM we'd get even in the supposed M1X. Reply
  • ikjadoon - Thursday, July 29, 2021 - link

    ? The M1 uses LPDDR4X without issue; still near-chart-topping in SPEC2017 & SPEC2006. Its GPU is also closest to a GTX 1650. What's your thinking that its LPDDR4X RAM has been a significant bottleneck to the M1's passive & light systems?

    As Apple goes wider + more cores & especially bigger GPUs (!), LPDDR5X sounds like a straightforward upgrade.

    //

    Apple shares its SoC (and thus its memory controller) with a lot of systems, so ideally it'd only differentiate memory types once it's worthwhile; maybe the "M1X/M2X/M3X" desktop-class variants might splurge for HBM?
    Reply
  • RSAUser - Thursday, July 29, 2021 - link

    Because HBM uses more power and is more expensive? Reply
  • BillBear - Thursday, July 29, 2021 - link

    Apple just announced that their usual level of spending to prepay for components was about to go up by ~25%, so it's entirely possible that something new and expensive is in the offing.

    https://twitter.com/tim/status/1420340541762453509
    Reply
  • at_clucks - Monday, August 2, 2021 - link

    Titanium body for the iPhone might account for some of that. Reply
  • mode_13h - Monday, August 2, 2021 - link

    It's interesting that you say it uses more power, since it actually uses less power than GDDR. Not sure about LPDDR5X, but then there's also the question of what they're both clocked at and how many stacks of HBM you're using. Reply
  • Wardrive86 - Thursday, July 29, 2021 - link

    ARM v.9 based CPUs, LPDDR5x, Potentially Adreno 700 series, Qualcomm Nuvia, next couple of years could get quiet interesting for Android users Reply
  • Wereweeb - Thursday, July 29, 2021 - link

    v9 sounds absolutely boring. Nuvia CPU's and GAA transistors are much more exciting.

    And my personal wish is to see better desktop modes and a RISC-V Android device. Two steps closer to a usable open-source and all-purpose pocket computer.
    Reply

Log in

Don't have an account? Sign up now