Airgap Process

Airgap interconnect is the logical endpoint of the same k-reduction race that produced OSG and FSG — instead of adding still more porosity to an organosilicate matrix the way Black Diamond III already pushed carbon-doped oxide past its mechanical limit, airgap BEOL removes the dielectric between metal lines entirely, leaving a literal void with a dielectric constant near 1.0, the theoretical floor no further material substitution can beat, since air itself is the lowest-k dielectric available.

Non-conformal pinch-off deposition
PECVD tuned so oxide seals the trench top before filling the bottom between metal lines
Sacrificial fill + selective removal
Dielectric deposited full, then wet/dry-etched back out from between the lines
Two routes, same void
Pinch-off leaves an open seam
Fewer process steps, less control over final void shape
Etch-back leaves an engineered cavity
More masking/etch steps, precise void geometry
Thin conformal cap (SiCN or SiN liner)
Seals the void, blocks moisture ingress and copper migration, carries mechanical load
Survive the rest of BEOL
CMP downforce
Capping layer must not collapse under polish pressure
Remaining thermal budget
Must hold through later anneals and additional metal levels

IBM pioneered airgap BEOL in volume production, not Intel, and the two companies solved it differently. IBM introduced self-aligned airgaps on the 32nm SOI node and carried the technique through 22nm and 14nm POWER and z-Series processors, etching the inter-level dielectric back out from between adjacent copper lines after patterning and sealing the result with a thin SiCN cap that also acts as the copper diffusion barrier. Intel's version, introduced at 14nm for its tightest-pitch metal layers, uses a self-aligned non-conformal deposition that pinches off the gap during cap deposition itself rather than a separate etch-back, trading a simpler process for less control over exactly where the void forms. Both routes land in the same place: an effective dielectric constant of roughly 1.0-1.3 once the air fraction and remaining liner are averaged together, well under Black Diamond III's 2.4-2.5 floor and far under FSG's 3.5-3.7.

OSG/Black Diamond IIIAirgap/ULK
Effective k2.4-2.5~1.0-1.3
Mechanical riskModerate — modulus down to 4-6 GPaHigh — the liner alone carries load across an open void
Process complexitySingle PECVD deposition + UV porogen cureExtra etch-back or precisely tuned pinch-off deposition, plus a dedicated capping step
Moisture/particle riskModerate, open pore network until plasma-sealedHigh before capping, low after — a breached cap is a direct path into the void
Adoption nodesBroad, 32-14nm and still in useNiche — IBM 32-14nm POWER/z-Series, Intel 14nm select layers only
The Low-k Roadmap Hits a Floor at k=1 each dopant swap buys less k than the last; airgap is the only step left k 4.2 k 2.5 k 1.0 Generation (chronological, not to scale) USG k 4.1 FSG k 3.6 OSG BD-I k 3.0 OSG BD-III k 2.45 Airgap/ULK k ~1.0-1.3 diminishing returns each extra % porosity in OSG buys less k and costs more modulus air itself is k=1.0 — no further dopant chemistry can beat it USG → FSG → OSG/Black Diamond → airgap/ULK

Read against the osg-process chain, airgap is not a new dopant generation — it is the chain running out of dopants to try. USG held near k 4.0-4.2 by refusing any risk at all, FSG bought a modest drop to 3.5-3.7 by accepting bounded fluorine corrosion risk, OSG's Black Diamond generations traded that corrosion risk for a mechanical one and reached 2.4-2.7 by engineering porosity directly, and at each step the k reduction purchased per process change got smaller while the reliability cost got larger.

Airgap breaks that pattern instead of extending it: it does not try to out-dope OSG, it removes the dielectric from the equation entirely. There is no lower-k solid film waiting to be discovered the way methylsilane followed fluorosilicate glass — the only way below Black Diamond III's porosity ceiling is to stop filling the space between metal lines with anything at all, which is why IBM and Intel both reach for a capping liner and an engineered void rather than a fourth generation of organosilicate chemistry.

Take airgap process further

Ask the copilot about this term, or have our engineers assess it against your process.