News
IndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value ChainIndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value Chain
EveryChip
Hybrid Memory Cube

Hybrid Memory Cube

Stacked DRAM with a serial link interface, built by Samsung and Micron

Hybrid Memory Cube is a RAM interface and package design for DRAM dies stacked vertically using through-silicon vias. Samsung and Micron created it together in 2011, and Micron revealed it that September. It aimed for far higher bandwidth than DDR3, but the product was dropped in 2018 after buyers stayed away.

Share on WhatsApp

  • Announced by Micron in September 2011 after joint work with Samsung
  • Promised about fifteen times the speed of DDR3
  • Stacks 4 to 8 DRAM dies using through-silicon vias and microbumps
  • HMC 2.0 (Nov 2014) supports SerDes from 12.5 to 30 Gbit/s
  • Micron discontinued the product in 2018 after weak adoption

What it is

Hybrid Memory Cube, usually shortened to HMC, is a way of building and talking to random-access memory. Several DRAM dies are piled on top of one another and wired together through tiny vertical connections called through-silicon vias, with a separate logic die holding the memory controller. Micron and Samsung developed the idea jointly in 2011, and Micron presented it publicly that September. The pitch was roughly fifteen times the speed of DDR3. It never became mainstream: Micron stopped selling HMC parts in 2018 after the market did not take them up.

How it came about

The design came out of work by Samsung and Micron, announced in September 2011. A supporting group, the Hybrid Memory Cube Consortium, drew in firms such as ARM, Open-Silicon, Altera, Xilinx, HP and Microsoft, though HP and Microsoft later left and Altera was bought by Intel. The Linley Group gave HMC its Best New Technology prize in 2011. A first public specification arrived in April 2013, and a second followed on 18 November 2014. Micron shipped engineering samples in September 2013.

Inside the chip

A cube holds four to eight memory-cell dies stacked vertically, joined by through-silicon vias and microbumps, with the controller on its own die underneath. The cells are ordinary DRAM, but the part carries more banks than a flat DRAM chip of equal capacity. Outside, the cube talks over differential serial links, either 16 lanes wide or a half-width 8 lanes, with each lane running at 10, 12.5 or 15 Gbit/s. Up to eight cubes can be chained together, some acting as pass-throughs. This interface shares no compatibility with DDR2, DDR3 or High Bandwidth Memory.

How it was made

Early demonstration parts stacked four 50 nm DRAM dies over a 90 nm logic die, giving 512 MB in a 27×27 mm outline and drawing 11 W at 1.2 V. Later samples packed 2 GB, built from four 4 Gbit dies, into a 31×31 mm body with four links; a smaller two-link version measured 16×19.5 mm. A four-link cube in the standard package uses 896 ball grid array pins and stands 3.8 mm tall.

Where it went

The first processor designed to work with HMC was Fujitsu's SPARC64 XIfx, which went into the PRIMEHPC FX100 supercomputer launched in 2015. Beyond that machine, adoption stayed thin. JEDEC's Wide I/O standards cover a similar stacked-die idea aimed at phones and other mobile gear, while HMC targeted desktops and servers. Micron walked away from the technology in August 2018, choosing instead to back GDDR6 and High Bandwidth Memory.

Versions and family

HMC 1.0, published in April 2013, allowed 4-link or 8-link packages and capped signalling at 10 Gbit/s when all eight links were used. HMC 2, released in November 2014, trimmed the choice to 2 or 4 links, added a quarter-width 4-lane option, and widened SerDes rates from 12.5 up to 30 Gbit/s. The rival High Bandwidth Memory, built by AMD and Hynix and used in Fiji and Pascal parts, solved the same bandwidth problem in an incompatible way.

Why it matters

HMC drew a lot of attention early on, winning a technology award in 2011 and gathering a consortium of major vendors around its specifications. Bandwidth figures looked strong: a 16-lane link at 10 Gbit/s signalling gives 40 GB/s in total, and a four-link cube at 15 Gbit/s per lane reaches 240 GB/s. Yet the standard never won broad commercial use, and its main backer gave up on it in 2018.

Problems and reception

Reviewers and engineers treated the concept warmly at first, and the Linley Group named it the best new technology of 2011. Practical uptake was another matter. Efficiency depended heavily on packet size, ranging from 33 to 50 per cent with 32-byte transfers and 45 to 85 per cent with 128-byte ones. Micron ended the product line in 2018 because it had not reached the market in volume.

Questions people ask about Hybrid Memory Cube

What is a Hybrid Memory Cube?

It is a RAM design in which several DRAM dies are stacked and linked by through-silicon vias, with a controller die included in the same package. A serial interface connects the cube to the processor. Samsung and Micron created it in 2011 as a way to raise memory bandwidth well above ordinary DDR3.

How does it differ from DDR3 or High Bandwidth Memory?

HMC uses its own serial link and cannot plug into DDR2 or DDR3 slots. High Bandwidth Memory, from AMD and Hynix, is a separate stacked-memory approach and is likewise incompatible. JEDEC's Wide I/O standards target mobile devices, while HMC was aimed at desktops and servers.

How fast is it?

A single 16-lane link at 10 Gbit/s signalling delivers 40 GB/s combined, split evenly between sending and receiving. A four-link cube using 15 Gbit/s lanes can reach 240 GB/s, and an eight-link cube 320 GB/s. Actual efficiency depends on transfer size, from about a third to 85 per cent.

Who used it?

Fujitsu's SPARC64 XIfx was the first processor built for HMC, and it powered the PRIMEHPC FX100 supercomputer introduced in 2015. Micron shipped engineering samples in 2013, but the technology never spread widely and was dropped in 2018.

Good reading and sources

Where this page comes from

The facts on this page come from the Wikipedia article "Hybrid Memory Cube", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryChip rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.

It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

Photo credits on this page

The small pictures above come from Wikimedia Commons under free licences. Some are cropped to fit.

The weekly digest

Every Monday: the chip of the day, new pages on EveryChip, and the conferences and anniversaries of the week ahead.