Theses›Consumer devices›Ferroelectric memory

Ferroelectric memory

Hafnium-oxide-based ferroelectric memory (FeFET, FeRAM) captures ≥5% of embedded non-volatile memory units by 2030, up from <0.5% in 2024.

  • Bear
  • Shut
  • $9.4B by 2035

The callFlash cannot scale below 28nm and hafnium-oxide memory drops into standard logic, but the best results sit captive inside the foundries; nothing for a startup to own.

D1D2R1R2R3C1C2foragainst D1D2R1R2R3C1C2foragainst
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 restraints.
  1. D1Embedded flash has no cost-effective path below 28nm, and every microcontroller, sensor and edge AI chip needs memory that holds code and weights with the power off.
  2. D2Hafnium oxide is already in every transistor gate, so ferroelectric memory drops into a standard logic process with few added steps; TSMC and Samsung have shown endurance past 10^12 cycles.
  3. R1The good numbers stay captive inside TSMC and Samsung, where FeFET is one in-house option among MRAM, RRAM and flash, with no kit for outsiders.
  4. R2MRAM scales and takes the slot while flash keeps 40nm and above, leaving FeFET at about 1% of embedded memory by 2030.
  5. R3A memory cell needs a foundry process module and kit, FMC raised €100M to reach volume, and the best exit comparable is Adesto to Dialog for about $500M.
  6. C1GlobalFoundries offers ferroelectric memory on 22FDX to outside customers with a design kit, turning the June 2026 Fraunhofer integration into a product any chip designer can order.end-2027
  7. C2FMC, the one merchant, breaks ground on its Saxony-Anhalt fab with a dated first-wafer plan, after raising another EUR 100M in January 2026.end-2027
Rests onthe beliefs underneath
Supported by · confidence
Threatened by · confidence
What would prove itdated tests, open and resolved
20272028202920302031today12345predictionswatch 20272028202920302031today12345predictionswatch
Fig. 2Each ring is a dated test: dashed while open, struck through when overdue, ticked or crossed once resolved.
  1. 1GlobalFoundries 22FDX FeFET reaches volume customer adoption by 2027.Dec 2027 · open
  2. 2GlobalFoundries disclosing three or more volume design wins on 22FDX FeFET or a successor, on an offering open to outside customers.Dec 2027 · open
  3. 3HfO2-based ferroelectric memory captures ≥5% of embedded NV memory units by 2030.Dec 2030 · open
  4. 4FeFET does NOT replace HBM at AI accelerator memory layer through 2030.Dec 2030 · open
  5. 5FeFET's share of embedded non-volatile memory units heading toward 5%, and the separate view that it does not replace HBM in accelerators.Dec 2030 · open
  • ·NaMLab spinouts, Sony FeRAM volume and imec's roadmap at 14nm and 7nm.
Expressionslisted, private, technologies
Private · 4 private companies.

Antaios · IMEC · NaMLab · Kepler Computing

2 more private companies tracked, not shown

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

$468M raised by 1 company, $578M 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.

  • $468MKepler Computing
    • Founded 2018
     Late-stage, Sep 2026 $468M (GlobalFoundries, Intel Capital, AMD Ventures)
What changedthe record, newest first
  • 23 Jun 2026The endurance problem that held back hafnium-oxide ferroelectric memory has quietly been solved at the lab bench (over a trillion write cycles now), but the gating question has moved: the strong results sit captive inside the big foundries with no third-party offering and no volume design wins, so the value pools with the incumbents, not a startup.
Reading
Show the reading