Manufacturing-aware physical design techniques
- Author(s): Sharma, Puneet
- et al.
CMOS scaling has outpaced manufacturing technology advancements, and consequently process variability continues to increase. Manufacturing non-idealities induce variations in lateral dimensions and topography, stress variations, and material variations. These are manifested as circuit delay and power variations, and consequently low parametric yield, which is the percentage of chips that, though functional, fail to meet delay and power specifications. Design for manufacturing (DFM) refers to measures taken during design to enhance yield. Traditional DFM techniques are essentially geometric operations with limited electrical interactions or awareness. These include resolution enhancement techniques to improve fidelity of optical lithography, design rule checks to restrict the use of layout patterns not amenable to manufacturing, and guardbanding to keep margins for process variability in design. As the extent and complexity of process variations increases, and suboptimality due to conservative design threatens to offset the benefits of scaling, these traditional DFM techniques, while still crucial, are no longer adequate. DFM techniques to improve parametric yield can be classified according to their approach. A considerable fraction of variability is systematic in nature and can be predicted using layout and process knowledge. Examples of such variations are pitch-dependent lithography variations and layout-dependent stress effects. These variations can be predicted and compensated for in physical design to improve yield. A second class of DFM techniques enhances design robustness to process variations. Examples include gate length biasing and redundant link insertion in clock trees, which respectively reduce leakage and clock skew variations even when the gate length variability remains the same. A third class of parametric yield-directed DFM techniques reduces process variations themselves, and includes dummy fill insertion and the increased use of layout pattern regularity. In this thesis we propose novel DFM techniques that explicitly target parametric yield. We present three techniques for analysis and optimization of circuit leakage and delay that are knowledgeable of systematic lithography variations due to pitch, defocus, and lens aberration. Stress variations, due to width of shallow trench isolation (STI) wells, can lead to considerable delay variations. We propose timing analysis and optimization methodologies to account for STI width- dependent stress, which is highly systematic in nature. Variations in gate length arising from a variety of process variations are a major cause of leakage variability, an important problem being faced by the designers today. We propose gate length biasing, which leverages the threshold voltage roll-off to significantly reduce leakage and its variability. The technique is non- obtrusive to existing flows, easy to adopt, and inexpensive to manufacture. We also present our contributions to front-end of the line (FEOL) and back-end of the line (BEOL) fill. Our FEOL insertion methodology considerably improves topography after chemical mechanical polishing for STI and may avoid the need for reverse- etchback process steps. In BEOL fill insertion, a primary concern is the capacitive impact of inserted fill and the corresponding increase of delay and crosstalk. We describe a systematic study of the capacitive impact of inserted fill, and develop guidelines that reduce capacitive impact without sacrificing metal density