CD Bias Naming Iso Dense Bias
CD bias is named by subtraction, and the sign of that subtraction is the entire naming convention — take the printed or etched critical dimension (CD) of a feature and subtract the CD of a reference feature at the same drawn size, and whatever number falls out is the bias. "Iso-dense bias" specifically subtracts a densely packed feature's CD from an isolated feature's CD (or vice versa, depending on house convention — always check which), and the result tells you, in one signed number, how much a feature's neighbors changed its outcome even though both were drawn identically on the mask.
A few things explain why this bias gets its own name instead of being folded into generic process variation:
The bias exists because local pattern density changes the physics at every step, not just one. In lithography, dense lines reflect and diffract light differently than an isolated line of the same width, shifting the effective exposure dose it receives. In etch, a dense array of features starves the local plasma of reactant faster than an isolated feature does (the same microloading effect that drives RIE lag), slowing the etch rate inside dense arrays relative to isolated ones. Both mechanisms push the same direction — dense features tend to print or etch narrower than isolated ones at the same drawn size — which is exactly what the iso-dense bias number quantifies.
Naming the bias by its two endpoints, iso and dense, rather than by a single average number, is what makes it actionable. A single average CD across a chip hides this effect completely; reporting iso-dense bias forces every CD-control conversation to specify which context a number came from. Optical proximity correction exists largely to counteract this named bias directly: a mask's isolated features are deliberately drawn at a slightly different size than its dense features of the same nominal design width, pre-compensating for the bias before exposure or etch ever happens.
The same naming logic extends beyond a simple two-point iso-versus-dense comparison into a full pitch-dependent bias curve. Real process-control decks often report CD as a function of pitch (the spacing between repeating features) rather than just two bins, because the bias doesn't jump suddenly between isolated and dense — it varies continuously as pitch shrinks, which is why an engineer will talk about "bias versus pitch" data as the generalized, continuous version of the same iso-dense naming idea.
Siblings worth distinguishing: this entry is about CD bias as a geometric and optical-proximity concept specifically; it is not about etch microloading or RIE lag as physical mechanisms (covered elsewhere), even though those mechanisms are one of the causes of the bias this entry names, nor is it about resist tone or mask polarity, which determine which half of a pattern survives rather than how its size shifts with context.
Read "iso-dense bias" as a signed measurement, not a defect label: a positive number says isolated features print or etch larger than dense ones at the same drawn size, a negative number says the opposite, and the entire point of naming it this way — instead of just reporting an average CD — is to give optical proximity correction and etch-bias compensation a specific, signed target to cancel out before the wafer ever leaves the fab.