Theses›Datacentre & AI compute›Beyond-CMOS devices

Beyond-CMOS devices

Beyond-CMOS device layer: memory beachhead, IP-layer capture

  • Bull
  • Reopening
  • $3.6B by 2035

The callThe drivers win: embedded flash stops below 28nm and the foundries are committing to MRAM and ferroelectric roadmaps now. The bet is the layer one off the foundry volume.

D1D2R1R2C1foragainst D1D2R1R2C1foragainst
Fig. 1The balance. Drivers and catalysts against restraints, each block sized by its weight; catalysts hang dashed because they have not happened yet, sized by their odds. It tips toward the drivers.
  1. D1Embedded flash hits its mask-cost wall below 28nm, so every foundry is committing to an embedded non-volatile memory roadmap now, with wafer volume growing about 80% a year to 2029.
  2. D2AI's memory and energy wall pulls the device layer beyond flash replacement into DRAM-class cache and compute-in-memory, which is the step FMC is taking with DRAM+.
  3. R1The graveyard of independents: Antaios closed in 2024, Spin Memory shut, Crocus left MRAM, and Samsung and TSMC built eMRAM without licensing anyone, which caps the royalty.
  4. R2No independent has ever captured a new logic device class, and imec's roadmap puts 2D channels and CFET ahead of spin and ferroelectric logic.
  5. C1FMC launches DRAM+ and 3D CACHE+ ferroelectric products from 300mm foundry volume production, opening the DRAM-class chiplet market to an independent.2027
Rests onthe beliefs underneath
Supported by · confidence
What would prove itdated tests, open and resolved
20272028202920302031today123456predictionswatch 20272028202920302031today123456predictionswatch
Fig. 2Each ring is a dated test: dashed while open, struck through when overdue, ticked or crossed once resolved.
  1. 1The IEDM 2026 technical programme (December 2026) includes at least one paper from TSMC, Samsung, Intel, GlobalFoundries or UMC on embedded MRAM or ferroelectric memory at a 16nm-class node or below.Dec 2026 · open
  2. 2FMC's DRAM+ and 3D CACHE+ reach 300mm foundry volume production.Dec 2027 · open
  3. 3Embedded MRAM/FeRAM (eNVM) is in volume production as the eFlash successor at ≤22nm at ≥4 merchant foundries/IDMs, and embedded-eNVM revenue exceeds ~$1.5B, by end-2028.Dec 2028 · open
  4. 4Two of Everspin, FMC, Numem, Avalanche or an imec spinout reach a major foundry licence or $50M of IP revenue; if not, the case fails.Dec 2028 · open
  5. 5Embedded NVM revenue over $1.5B with four or more foundries shipping at 22nm or below; foundries keeping stacks in-house closes the window.Dec 2028 · open
  6. 6No beyond-CMOS LOGIC device reaches a committed foundry product roadmap (named node) by end-2030.Dec 2030 · open
Expressionslisted, private, technologies
Private · 5 private companies, 3 of them in the US.

Numem · Avalanche Technology · Antaios · Crocus Technology · Kepler Computing

3 more private companies tracked, not shown

Technologies
Who has already bet on itprivate rounds, from public sources

$468M raised by 2 companies, $644M of it in the last three years. Wave under way: Large rounds in the last three years; the category is funded and getting crowded.

Backed by a strategic investor: Kepler Computing.

What changedthe record, newest first
  • 22 Jun 2026The end of embedded-flash scaling and the AI memory wall are moving magnetic and ferroelectric memory (MRAM, FeFET) into production now. The device layer beyond the transistor is arriving via memory, not the decades-away "new logic switch". The investable question is not the physics but the capture: the durable independent money is in the IP, materials and chiplet layer that licenses into the foundries (Everspin's royalty model, FMC's licensing model), not in out-manufacturing them and not in the pure-logic moonshot, whose track record for independents is a graveyard.