Study On Model-Based Fragmentation In OPC

As the IC process enters the UDSM era, the complexity of IC has increased dramatically and the critical dimension has become smaller. Now the leading manufacturers have been intensively working at 22nm node. Currently light of wavelength (193nm broadly used at present) employed by the steppers and scanners is far larger than the features on layout, therefore the patterns fabricated on the wafer exhibit apparent distortions. Semiconductor manufacturing community adopts many kinds of resolution enhancement technology to tackle this challenge. Optical proximity correction is an indispensable and broadly used technology in current IC manufacturing. Choosing appropriate fragmentation schemes for polygons in layout is the key step in OPC practice, which will directly and definitely shape the outcomes of OPC. However, prevalent fragmentation schemes in OPC recipe are all based on empirically rules, which lacks global universality in a complicated 2-D context, causing various defects originated from insufficient coverage of recipe. As a result, engineers have to spend a great amount of time on tuning recipe, then the whole product turn-around time suffers.This dissertation studies and proposes a new model-based fragmentation algorithm, which lays its foundation on ripple-suppressing on aerial image contour, achieving better OPC results by eliminating ripples on simulated contour. The algorithm firstly acquires aerial intensity curve at specified path with a calibrated OPC model, then it extracts extremes on the curve and assigns fragments with certain length to construct a new fragmentation scheme with the aid of a grid-searching or gradient method to optimizie the core parameters. Also, this dissertation explores how to incorporate this model-based fragmentation scheme into current mainstream OPC flow by proposing a hybrid OPC fragmentation framework. We conclude this paper with sophisticated experiments data on the validity and practicality of our proposed algorithm and framework. Experiment results show our algorithm can improve 1%-18% OPC fidelity with stable run time, comparing to a reference recipe. And the example using hybrid OPC fragmentation framework improves 2% OPC fidelity and reduces mask complexity slightly, but receives penalities on OPC run time.