semiAIfoundry

Fabrication · Atlas v1.3.0

Wafer and fabrication hierarchy

Provides the shared substrate on which many integrated circuits are fabricated.

Explore in Atlas300 mm example wafer carrying microscopic structures

Learning objectiveSeparate wafer area, die geometry, starts and good output.

The mechanism

A wafer is a thin, polished slice of highly purified crystalline silicon. Many die patterns are processed across its surface together before testing and singulation separate usable chips. Device formation near the surface is commonly described as front end of line, or FEOL; local connections bridge devices to the back end of line, or BEOL, wiring stack. These labels describe manufacturing regions, not software front ends and back ends. Repeated patterning, deposition, removal, cleaning, and measurement build the structures layer by layer. The atlas compresses this long chain into a navigable cross section so viewers can connect a macroscopic wafer to the tiny electrical structures it carries.

  1. Begin with a polished crystalline wafer that supports many die locations.
  2. Build device regions and wiring through repeated patterning and material-processing stages.
  3. Test the fabricated structures, separate die, and assemble usable die into packages.

Science and assumptions

A first defect-yield model

Y≈exp⁡(−D0Adie)Y\approx\exp(-D_0 A_{\mathrm{die}})

Poisson die survival is approximately the exponential of minus fatal-defect density times die area.

Symbols and units
Y
Die survival probability in the Poisson model

fraction

D₀
Uniform fatal-defect density

cm⁻²

A_die
Effective die area exposed to those defects

cm²

If random fatal defects occur at density D₀ per unit area, larger die area increases the chance of encountering one. Yield affects how many functional chips emerge from the same wafer.

Assumptions Illustrative Poisson model with independent uniform fatal defects; real yield uses process-specific models, clustering, repair, and many nonrandom factors.

Engineering example

A 300 mm wafer with edge exclusion can contain thousands of nominal sites, but the good-die count falls with die area and defect exposure.

Distinctions and tradeoffs

A wafer, a die, and a package are different objects; one patterned wafer normally contains many die.

Larger useful die area can reduce unit cost. Defects, edge loss, and yield complicate that benefit.

AI accelerators emerge from this manufacturing platform. Wafer yield and die size influence the cost of available compute.

Specifications and interfaces

Example wafer diameter
300 mm

Widely used advanced-logic manufacturing format; not universal.

Thickness
Less than 1 mm

General introductory wafer description, not a procurement specification.

Hierarchy
Wafer → die → package

Test and assembly intervene between stages.

Interfaces

Purified crystal Input
Mechanically supported semiconductor substrate
Fabricated dies Output
Patterned and tested integrated circuits

Check your understanding

Which order follows physical integration?

  1. Package → atom → wafer
  2. Wafer → fabricated die → package
  3. Metal gate → data center → dopant
Answer and reasoning

Wafer → fabricated die → package

Die are fabricated across a wafer and later separated for assembly. Testing and yield determine which proceed.

Evidence and scope

Gross-site geometry and yield are teaching estimates; reticle fields, scribe lanes and parametric loss are not hidden.

Explore gross sites and random-defect survival; geometric approximations and defect models do not replace wafer test.

Sources

Published lesson data · Atlas releases and licensing

Continue exploring

Foundations

Related lessons