Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/intel-prepares-hbm-killer-hb3dm-memory-stacks-with-z-angle-technology.25025/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2031070
            [XFI] => 1060170
        )

    [wordpress] => /var/www/html
)

Intel Prepares HBM Killer: HB3DM Memory Stacks with Z-Angle Technology

XYang2023

Well-known member
Intel and SoftBank, through their subsidiary Saimemory, have been developing an alternative technology to the popular high-bandwidth memory (HBM) to provide more bandwidth and capacity for memory modules used with powerful AI accelerators. At VLSI 2026 in June, Saimemory is scheduled to present a paper on the newly developed HB3DM memory, which is based on Z-Angle Memory (ZAM) technology. This name refers to the vertical (Z-axis) stacking of dies, similar to traditional HBM. However, Intel aims to achieve impressive results using state-of-the-art manufacturing technology. The first generation of HB3DM will feature a total of nine layers, stacked using a hybrid bonding technique for 3D chip placement. At the base will be a logic layer that manages data movement within the chip, with eight DRAM layers on top for data storage. Each layer will include about 13,700 TSVs for hybrid bonding.

 

This is basically HBM, next generation. It's still vertically stacked DRAM dies with TSV connections on top of logic base die.
 
Last edited:
Yay, Intel is getting back into memory! Or who is going to manufacture this?
I think Intel is just responsible for the stacking part. The DRAM is from PSMC apparently. The logic base die, if it had been made on Intel 18A or Intel 3, I'm sure that would have been called out.
 
Last edited:
Fred, can you please cite your source? I don't think ZAM is "still vertically stacked DRAM". There is a "vertical" component, but why is it called "Z-Angle" if it's not that different from HBM?
 
I thought ZAM had the HBM stack rotated 90 degrees and somehow bonded along the edge (saw some patents on this). But the figures in the article don't show this. Confusing.
 
Yay, Intel is getting back into memory! Or who is going to manufacture this?

I’ve lost track of how many times Intel has entered and exited the memory business, whether through manufacturing, joint ventures, partnerships, development, specification/standard setting, or licensing. So I asked my unpaid assistant Mr. Claude to find out.

Here's the full list:

Intel Memory History - Entries, Exits & Partnerships

Summary: 6 entries · 7 exits · 6 JVs/partnerships · 2 licensing/spin-outs across 57 years

Era 1 — DRAM & SRAM (Founded on Memory)

  • 1968 — ENTRY: Intel founded on SRAM/DRAM chips; 1103 DRAM displaced magnetic core memory as industry standard
  • 1971 — ENTRY: EPROM/ROM business launched; Intel invented the EPROM
  • 1985 — EXIT: Exited DRAM & SRAM manufacturing; Japanese competitors destroyed margins; pivoted to microprocessors

Era 2 — NOR Flash (1988–2010)
  • 1988 — ENTRY: Launched world's first NOR flash chip, replacing EPROM for firmware/boot code storage
  • 2008 — SPIN-OUT: NOR flash assets spun into Numonyx JV with STMicroelectronics (Intel 45.1%, ST 48.6%, Francisco Partners 6.3%); ~2,500 employees transferred; ~$300M impairment charge
  • 2010 — EXIT: Numonyx sold to Micron for $1.27B; Intel exited NOR flash entirely

Era 3 — Rambus RDRAM (1996–2003)
  • 1996–97 — PARTNERSHIP: Intel signed exclusive development & license agreement with Rambus; committed all chipsets (i820, i840, i850, i860) to RDRAM; agreement secretly barred Intel from shipping DDR chipsets until 2003
  • 2001–03 — EXIT: Abandoned RDRAM after it proved too expensive, hot, and high-latency; DDR matched RDRAM performance at far lower cost; Intel quietly launched i865/i875 DDR chipsets in 2003, ceding years of market leadership to AMD

Era 4 — NAND Flash (2005–2022)

2005–06 — JV: Formed IM Flash Technologies (IMFT) with Micron (Intel 49%, Micron 51%); each contributed ~$1.2B; fabs in Lehi UT, Manassas VA, Boise ID; Apple pre-paid $250M to each partner
  • 2007 — JV EXPANSION: Formed IM Flash Singapore (IMFS); opened April 2011; pioneered 34nm then 20nm NAND
  • 2018–19 — EXIT FROM JV: Micron exercised call option in Oct 2018; deal closed Jan 2019; Intel retained standalone Dalian NAND fab and SSD business

  • 2020–25 — FULL EXIT: Sold NAND + SSD business to SK Hynix for ~$8.85B in two parts (Phase 1: $7B Dec 2021 — Dalian fab + IP; Phase 2: $1.9B Mar 2025 — remaining IP + employees); rebranded as Solidigm

Era 5 — 3D XPoint / Optane (2012–2022)
  • 2012–15 — JV: Jointly developed 3D XPoint with Micron using IMFT Lehi fab; announced July 2015 as 1000× faster than NAND
  • 2017 — ENTRY: Launched Optane SSDs and Persistent Memory DIMMs; ~$500M/quarter operating losses
  • 2018–19 — JV DISSOLUTION: Ended joint 3D XPoint development with Micron after 2nd-gen node; Micron sold Lehi fab to Texas Instruments for $900M
  • 2021–22 — FULL EXIT: Consumer Optane killed Jan 2021; full division wind-down announced July 2022; $559M inventory write-off
 
I’ve lost track of how many times Intel has entered and exited the memory business, whether through manufacturing, joint ventures, partnerships, development, specification/standard setting, or licensing. So I asked my unpaid assistant Mr. Claude to find out.

Here's the full list: ...

Good summary. I think this history is relevant. This is particularly true if Intel is not building the DRAM silicon.
 
Fred, can you please cite your source? I don't think ZAM is "still vertically stacked DRAM". There is a "vertical" component, but why is it called "Z-Angle" if it's not that different from HBM?
This is from the VLSI Technology 2026 symposium highlights recently released: https://www.vlsisymposium.org/wp-co...hnical-Tipsheet-REVISED-FINAL-4.25.26-1-1.pdf

1777612253147.png

PSMC is a DRAM foundry: https://www.powerchip.com/en-global/services/foundry-services
 
I thought ZAM had the HBM stack rotated 90 degrees and somehow bonded along the edge (saw some patents on this). But the figures in the article don't show this. Confusing.
Yes, I agree, I probably saw the same figures. So I think this paper is not about Z-Angle Memory. HB3DM is just something else they are trying. ZAM might still be under development. It is not mentioned in the abstract at all. TechPowerUp jumped the conclusion.
 
Back
Top