Design automation algorithms for advanced lithography

Electronic Design Automation;Compute-aided Design;Lithography;Self-aligned Double Patterning;Directed Self-Assembly;Machine Learning;Design-Technology Co-optimization

Xiao, Zigang

关键词: Electronic Design Automation;    Compute-aided Design;    Lithography;    Self-aligned Double Patterning;    Directed Self-Assembly;    Machine Learning;    Design-Technology Co-optimization;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/88991/XIAO-DISSERTATION-2015.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

In circuit manufacturing, as the technology nodes keep shrinking, conventional 193 nm immersion lithography (193i) has reached its printability limit. To continue the scaling with Moore's law, different kinds of advanced lithography have been proposed, such as multiple patterning lithography (MPL), extreme ultraviolet (EUV), electron beam lithography (EBL) and directed self-assembly (DSA). While these new technologies create enormous opportunities, they also pose great design challenges due to their unique process characteristics andstringent constraints. In order to smoothly adopt these advanced lithography technologies in integrated circuit (IC) fabrication, effective electronic design automation (EDA) algorithms must be designed and integrated into computer-aided design (CAD) tools to address the underlying design constraints and help the circuit designer to better facilitate the lithography process.In this thesis, we focus on algorithmic design and efficient implementation of EDA algorithm for advanced lithography, including directed self-assembly (DSA) and self-aligned double patterning (SADP), to conquer the physical challenges and improve the manufacturing yield.The first advanced lithography technology we explore is self-aligned double patterning (SADP). SADP has the significant advantage over traditional litho-etch-litho-etch (LELE) double patterning in its ability to eliminate overlay, making it a preferable DPL choice for the 14 nm technology node. As in any DPL technology, layout decomposition is the key problem. While the layout decomposition problem for LELE DPL has been well studied in the literature, only a few attempts have been made for the SADP layout decomposition problem.This thesis studies the SADP decomposition problem in different scenarios.SADP has been successfully deployed in 1D patterns and has several applications; however, applying it to 2D patterns turns out to be much more difficult. All previous exact algorithms were based on computationally expensive methods such as SAT or ILP.Other previous algorithms were heuristics without a guarantee that an overlay-free solution can be found even if one exists. The SADP decomposition problem on general 2D layout is proven to be NP-complete. However, we show that if we restrict the overlay, the problem is polynomial-time solvable, and present an exact algorithm to determine if a given 2D layout has a no-overlay SADP decomposition.When designing the layout decomposition algorithms, it is usually useful to take the layout structure into consideration. As most of the current IC layouts adopt a row-based standard cell design style, we can take advantage of its characteristics and design more efficient algorithms compared to the algorithms for general 2D patterns. In particular, the fixed widths of standard cells and power tracks on top and bottom of cells suggest that improvements can be made over thealgorithms for general decomposition problem. We present a shortest-path based polynomial time SADP decompositionalgorithm forrow-based standard cell layout that efficiently finds decompositions with minimum overlay violations. Our proposed algorithm takes advantage of the fixed width of the cells and the alternating power tracks between the rows to limit the possible decompositions and thus achieve high efficiency.The next advanced lithography technology we discuss in the thesis is directed self-assembly (DSA).Block copolymer directed self-assembly (DSA) is a promising technique for patterning contact holes and vias in 7 nm technology nodes.To pattern contacts/vias with DSA, guiding templates are usually printed first with conventional lithography (193i) that has a coarser pitch resolution. Contact holes are then patterned with DSA process. The guiding templates play the role of defining the DSA patterns, which have a finer resolution than the templates. As a result, different patterns can be obtained through controlling the templates. It is shown that DSA lithography is very promising in patterning contacts/vias in 7 nm technology node. However, to utilize DSA for full-chip manufacturing, EDA for DSA must be fully explored because EDA is the key enabler formanufacturing, and the EDA research for DSA is still lagging behind.To pattern the contact layer with DSA, we must ensure that all the contacts in the layout require only feasible DSA templates. Nevertheless, the original layout may not be designed in a DSA-friendly way. However, even with an optimized library, infeasible templates may be introduced after the physical design phase. We propose a simulated-annealing (SA) based scheme to perform full-chip level contact layer optimization. According to the experimental results, the DSA conflicts in the contact layer are reduced by close to 90% on average after applying the proposed optimization algorithm.It is a current trend that industry is transiting from the random 2D designs to highly regular 1D gridded designs for sub-20 nm nodes and fabricating circuit designs with print-cut technology. In this process, the randomly distributed cuts may be too dense to be printed by single patterning lithography.DSA has proven its success in contact hole patterning,and can be easily expanded to cut printing for 1D gridded designs. Nevertheless, the irregular distribution of cuts still presents a great challenge for DSA, as the self-assembly process usuallyforms regular patterns. As a result, the cut layer must be optimized for the DSA process.To address the above problem, we propose an efficient algorithm to optimize cut layers without hurting the original circuit logic. Our work utilizes a technique called `line-end extension' to move the cuts and extend the functional wires without changing the original functionality of the circuit. Consequently, the cuts can be redistributed and grouped into valid DSA templates.Multiple patterning lithography has been widely adopted for today's circuit manufacturing. However, increasing the number of masks will make the manufacturing process more expensive. By incorporating DSA into the multiple patterning process, it is possible to reduce the number of masks and achieve a cost-effective solution. We study the decomposition problem for the contact layer in row-based standard cell layout with DSA-MP complementary lithography. We explore several heuristic-based approaches, and propose an algorithm that decomposes a standard cell row optimally in polynomial-time. Our experiments show that our algorithm is guaranteed to find a minimum cost solution if one exists, while the heuristic cannot or only finds a sub-optimal solution. Our results show that the DSA-MP complementary approach is very promising for the future advanced nodes.As in any lithography technique, the process variation control and proximity correction are the most important issues.As the DSA templates are patterned by conventional lithography, the patterned templates are prone to deviate from mask shapes due to process variations, which will ultimately affect the contacts after the DSA process even for the same type of template.Therefore, in order to enable the DSA technology in contact/via layer printing, it is extremely important to accurately model and detect hotspots, as well as estimate the contact pitch and locations during the verification phase. We propose a machine learning based design automation framework for DSA verification. A novel DSA model and a set of features are included. We implemented the proposed ML-based flow and performed extensive experiments on comparing the performances of learning algorithms and features. The experimental results show that our approach is much more efficient than the traditional approach, and can produce highly accurate results.