Canyon sound

See the cross-section before you cut the wafer.

Physics-based, AI-assisted process modeling. Predict profiles, calibrate to your fab, and screen for systematic defect hotspots.

Join the waitlist See how it works

One model of the process, three jobs.

Predict a step, match it to your fab, then screen a whole design for the systematic defect hotspots your process will cause.

Run: via-middle Cu TSV, 5 × 53 µmExample output

Flow

Bosch DRIE
130 cyc
SACVD liner
300 nm
ALD TaN
10 nm
iPVD Cu seed
793 nm
Cu ECP
bottom-up

Snapshot

ECP time
14.3 min
Fill front
21.0 µm
Depth52.5 µm
Aspect ratio10 : 1
Seed, 90% depth20 nm

Predict

Predict what a step produces. Physics-based models return the profile, thickness and coverage from the recipe, fast enough to explore the process window.

  • Etch, deposition, plating, CMP and bonding
  • Parameters you already tune at the tool
  • Feature-scale transport, not lookup tables
Calibration: Bosch via profile vs. SEMExample output

Fitted parameters

Sticking coeff.
0.09
Passivation yield
0.61
Ion spread
4.8°

Residual, RMS

Before fit
0.58 µm
After fit
0.08 µm

Floor depth error

Before fit
+3.6 µm
After fit
+0.1 µm

Data

SEM edge points
128
Residual, RMS0.08 µm
Cross-sections4
FitConverged

Calibrate

Make it specific to your fab. Fit the models to your tools, chemistries and recipes with measured cross-sections, so rates, sidewall coverage and undercut match what you see.

  • Fit from a few cross-sections
  • Parameters stay physical and inspectable
  • Residuals reported against your data
Hotspot scan: layout tile 4 mm × 4 mmExample output

Ranked by risk

  • H-01 pad jog, 0.320.91
  • H-02 line tips, 0.450.64
  • H-03 dense, 0.500.47

Check

Fill
PECVD oxide
Metal height
0.8 µm
Safe space
≥ 0.55 µm

Top cause

H-01
Keyhole void
H-02
Void at tips
H-03
Seam
Tiles screened3,600
Flagged3
Re-checked3 of 3

Screen

Find systematic defect hotspots before silicon exists. Scan a layout for failures that come from process, pattern, their interaction, and process and overlay variation.

  • Fast screening across the whole design
  • High-accuracy checks on what it flags
  • Ranked by risk, traced to cause
Run: via-middle Cu TSV, 5 × 53 µmExample output

Flow

Bosch DRIE
130 cyc
SACVD liner
300 nm
ALD TaN
10 nm
iPVD Cu seed
793 nm
Cu ECP
bottom-up

Snapshot

ECP time
14.3 min
Fill front
21.0 µm
Depth52.5 µm
Aspect ratio10 : 1
Seed, 90% depth20 nm
Calibration: Bosch via profile vs. SEMExample output

Fitted parameters

Sticking coeff.
0.09
Passivation yield
0.61
Ion spread
4.8°

Residual, RMS

Before fit
0.58 µm
After fit
0.08 µm

Floor depth error

Before fit
+3.6 µm
After fit
+0.1 µm

Data

SEM edge points
128
Residual, RMS0.08 µm
Cross-sections4
FitConverged
Hotspot scan: layout tile 4 mm × 4 mmExample output

Ranked by risk

  • H-01 pad jog, 0.320.91
  • H-02 line tips, 0.450.64
  • H-03 dense, 0.500.47

Check

Fill
PECVD oxide
Metal height
0.8 µm
Safe space
≥ 0.55 µm

Top cause

H-01
Keyhole void
H-02
Void at tips
H-03
Seam
Tiles screened3,600
Flagged3
Re-checked3 of 3

Five process modules.

Each is built on the physics of the step. The knobs you turn here are the knobs you turn at the tool.

ModulePhysics modeledOutputs

Etch

Plasma and Bosch etch in feature-scale transport. Ion and neutral flux, passivation, bowing, aspect-ratio-dependent etch rate, scalloping and SOI notching.

Profile, depth,
sidewall angle, scallop pitch

Deposition

Line-of-sight and surface transport for PVD and related processes. Sidewall and bottom coverage, mouth overhang and pinch-off.

Thickness, step coverage,
pinch-off

Plating

Seeded, seedless and electroless fill. Additive-driven superfill, seams and trapped voids as a function of bath behavior and geometry.

Fill state, seams, voids,
overburden

CMP

Metal, dielectric, silicon and barrier polish across slurry and pad conditions, with dependence on line width and pattern density.

Dishing, erosion,
residual thickness

Bonding

Fusion, hybrid, thermocompression, eutectic, anodic and polymer bonding, with the interface physics that decides void formation.

Void risk,
interface conditions

How the physics gets checked.

Every module is compared against measured cross-sections. Results are published here as they complete.

Etch profile

Bosch and continuous etch compared against published and partner SEM cross-sections.

Reported as error and spread
  • Sidewall angledeg
  • Depth and aspect-ratio effectµm, %
  • Scallop pitch and amplitudenm
Results published as they complete

Deposition and plating

Step coverage and fill behavior compared against measured fills across feature aspect ratios.

Reported as error and spread
  • Sidewall and bottom coverage%
  • Superfill and void onsetµm
  • Overburdennm
Results published as they complete

CMP and bonding

Dishing, erosion and interface behavior compared against measured profiles and inspection data.

Reported as error and spread
  • Dishing by line widthnm
  • Erosion by pattern densitynm
  • Interface void fraction%
Results published as they complete
Offer a cross-section to test

Join the waitlist.

ab Studio is in alpha. A small number of design partners get early access and a direct line to how it develops.

  • How it worksYou send a process flow or a question. We run it and walk you through the results in a screen-share session.
  • Who it suitsProcess and integration engineers in MEMS, RF, power and photonics fabs, university cleanrooms and teams without TCAD budgets.
  • What we askHonest feedback, and a conversation about where the predictions miss.
  • TermsAlpha, as-is, no service-level agreement. We sign a mutual NDA before you share any process data.
Processes you care about

We use your answers only to decide who to contact and what to build. We do not sell or share them.

Questions engineers ask first.

Can I use it today?

Not as a download. ab Studio is in alpha, and capability varies by module. Design partners send us a process flow and we run it with them, so we can tell you up front what is ready and what is not.

Do I need to share fab data?

No. Uncalibrated predictions run from the recipe alone. Calibration to your fab needs measured cross-sections or similar data, and we sign a mutual NDA before any of it changes hands.

Where does the physics come from?

Published literature and first-principles models of plasma, surface chemistry and transport. We do not use proprietary recipes or data from any company.

How is this different from full TCAD or 3D virtual fabrication?

Those tools are broad and heavy. ab Studio is aimed at fast, physically grounded answers for specific steps, calibration to your own tools, and screening layouts for systematic defects across a design space.

What does it cost?

Pricing is not set. Alpha design partners will help decide what a fair price looks like.

Process knowledge should not live in one engineer's head.