The work on photoresist fundamentals aims to tackle generic material related issues in advanced photolithography. Many of these are applicable to multiple imaging wavelengths. One goal of the work is to increase the understanding on the root causes of these issues and demonstrate proof of concept of innovative solutions or workarounds. The other goal is to explore material based solutions for enhancing lithographic performance. The work is done in close collaboration with the technology specific (immersion extensions and extreme ultraviolet (EUV)) lithography sub-programs (see section EUV lithography), (see section Immersion lithography extensions program).
Line width roughness
Line width roughness (LWR) continues to be a major concern for variability at small feature dimensions. Efforts have continued for understanding the root causes to LWR, for reducing LWR by post-processing techniques and for assessing the impact of LWR on the final device performance. For understanding the root causes contributing to LWR, a detailed analysis has been made of the impact of mask absorber roughness. Using EUV masks with programmed absorber roughness of varying frequency and amplitude, the transfer function from mask into photoresist on wafer has been determined P22914. Based on measurements of natural absorber roughness on state-of-the-art EUV masks and the resulting LWR on wafer C21278, the mask absorber is concluded to contribute ~15% to the final roughness in some of the earlier EUV masks P22891 and <5% in more recent ones. Also the impact of photon shot noise on LWR due to the discrete nature of light has been explored C22911. Stochastic simulations based on calibrated EUV resist models allow studying the impact of each process step individually. The problem is nicely illustrated by the results for an individual 22nm contact hole, see fig. 1. Counting the number of photons absorbed for patterning such a structure (fig. 1, top left) appears to indicate quite some noise. Indeed repeating this same process 25 times results in significant variability (fig. 1, top right). Owing to the high quantum efficiency, the acid distribution is a lot smoother and also variability is reduced (fig. 1, middle). However, during the polymer deprotection at the post-exposure bake step, significant variability is added again (fig. 1, bottom). This result suggests that focusing material optimization on that step will be the key for reducing LWR through resist optimization. For layers with high resist coverage such as via and metal, use of a bright field mask combined with a negative tone process is expected to reduce the impact of photon shot noise. As a consequence the interest on negative tone developed EUV resists has increased, and simulation methods developed for 193nm immersion based negative tone development C22479 are now also implemented in stochastic EUV simulators.
Figure 1: Stochastic simulation and analysis of a EUV exposed 22nm contact hole through the lithography process. The photon distribution of a single contact is noisy (top left), resulting in apparent high variability when this process is repeated 25 times. The high acid quantum yield causes a less noisy acid distribution (middle left) and less variability of the same 25 trials in the acid profile (middle right). Final polymer deprotection of these contacts after post-exposure bake (bottom right) again shows more variability.
For reduction of remaining roughness after development, the use of resist smoothing techniques is gaining increased interest. Approaches based on post-development rinse, solvent vapor treatment, ion implantation and plasma treatment have been studied and compared C21321. Total roughness reduction of up to 40% may be achieved. All techniques are primarily effective for reduction of high and middle frequencies. Low frequencies have so far not been subject to smoothing. On the contrary, for all cases care needs to be taken that roughness at low frequencies is not deteriorated as mid and high frequencies are improved. This turns out to be achievable with an appropriately optimized process P23108.
The link between LWR and device performance continues to be a source of debate. In order to tackle this, simulations have been done in collaboration with imec's memory INSITE program C22531. Experimental LWR input (including frequency information) has been used as input to estimate the bit-error rate in a 16nm NAND flash device C22317. This allows to estimate the complexity of the required bit error correction coding for various levels of LWR (see section INSITE memory: technology and circuit assessments).
In addition to LWR, there is also increasing need to determine variability and roughness of contact hole structures. A methodology for the characterization of contact edge roughness and extraction of the frequency components has been set up C22433. A first assessment of the impact of resist composition has been made through studying the impact of the variation of photo-acid generator concentration on contact edge roughness C24135.
EUV resist fundamentals
A lot of effort has been invested in setting up procedures for building accurate resist simulation models for EUV resists. This has allowed describing and understanding the impact of EUV sensitizers on EUV resist performance. The use of these sensitizers in a polymer-bound resist platform was demonstrated to increase the yield of acid formation by ~50% P22626. These models allow for accurate extraction of the acid diffusion length in these systems and therefore estimation of the ultimate resolution of the materials. Also specific attention has been paid to accurate description of the development process. In order to accomplish this, input from a dissolution rate monitor setup has been used as input for the resist models C22435.
Due to the excitation mechanism in EUV lithography, acids are not generated at the exact location where the EUV photons are absorbed. This phenomenon is termed secondary electron blur and there has been concern that it may significantly reduce processing latitude at the 16nm half pitch and prevent resolution below 11nm. However, the effect has proven to be very difficult to access experimentally. Through a combination of experiments with systematic variation of post-exposure bake conditions at 22nm half pitch and modeling as described above, it has been possible to estimate the secondary electron blur length C21317. The maximum probability for acid generation is estimated to be 2.4nm removed from the photon absorption site, which is sufficiently low not to have any impact down to 16nm half pitch P23909 and allow 11nm half pitch resolution. This value should actually be regarded as an upper limit and higher resolution EUV lithography is required for a more accurate (and possibly even lower) determination.
Due to the large improvements that have been made in controlling acid diffusion in chemically amplified EUV resists, the failure mechanism for ultimate resolution for most of the current materials has shifted from loss of 'intrinsic resolution' to pattern collapse P22857. Controlling and quantifying collapse is thus becoming of increasing importance. A method, based on atomic force microscopy (AFM), has been developed for measuring the collapse force of free standing isolated resist lines. In principle this method will allow quantification of the material stiffness and/or the adhesion strength of the line to the substrate. At first the line is imaged in non-contact mode and after that the line is broken by the AFM tip. During this process the strength required to break the line is measured. After the process the line is again imaged in non-contact mode to verify the breaking mode . Experimentally the force to collapse is found to increase with critical dimension, as expected. A simple description has been derived to extract a collapse coefficient from this data. Comparing two state-of-the-art EUV photoresists (A and B) with a known collapse prone material (C) indeed results in a substantially lower collapse coefficient for the latter material. For quantification of collapse as seen on wafer an optical defect inspection based methodology has been developed. Here, large arrays of line doublets are imaged. The line doublets are especially sensitive for collapse due to the large capillary forces exerted on these structures during the drying process after rinse. Since collapse tends to propagate along the line, the defects are relatively large. This allows for maximizing defect classification based on optical patches and minimizing the need for scanning electron microscopy (SEM) based review. The methodology has been set up using 193nm immersion lithography, and optical inspection recipes down to 40nm half pitch line doublets have successfully been implemented C23902. Future work will focus on transferring this approach to EUV studies.
Figure 2: Breaking an isolated resist line by atomic force microscopy (left) allows to measure the breaking force as a function of line CD (middle). Analysis of these curves allows to extract the collapse coefficient (c) which is indicative for the polymer strength and correlates well with pattern collapse behavior (right).
For EUV-induced resist outgassing most of the efforts have been focused on setting up an ASML-qualified procedure (see section EUV lithography). In addition some work has been done to gain better understanding of how various resist components contribute to contamination growth on EUV optics. For this study the non-chemically amplified poly-olefin sulfones have proven to be an interesting set of materials P21202. Introduction of small oligomers of polymethyl methacrylate ('plexiglass') (PMMA) in the main chain allowed to significantly reduce the outgassing and improve the imaging performance of these materials, without measurable traces of the oligomer groups in the outgassing spectrum P22886. Also a more detailed analysis has been done to identify the important components (photo-acid generator, quencher, protecting group, etc.) and material parameters (glass transition temperature, polymer molecular weight, etc.) in chemically amplified resists that contribute to EUV optics contamination C21319. Based on this work grouping of resists into families has been proposed, along with a method to identify the worst case member of that family for outgassing. This approach would help to reduce the amount of outgassing testing that is required prior to exposure on the ASML NXE:3100.
Several approaches for low-cost frequency multiplication (compared to self-aligned double patterning) have been explored, including direct frequency doubling through use of a photo-base and a photo-acid generator in a single formulation C22429. However, it is becoming clear that such approaches will not be adopted by the industry. Directed self-assembly (DSA) offers a different approach for such frequency multiplication, but has the added advantage that much higher orders of frequency multiplication may be achieved. In addition it offers pathways for repairing critical dimension (CD) uniformity and LWR. As a consequence DSA has recently gained increased attention. Methods for optimizing DSA process flows and approaches to combine it with available 193nm lithography to generate the required prepatterns have rapidly matured. In 2011, imec has started efforts to evaluate DSA for line/space frequency multiplication. The initial goal is to compare various integration approaches. On the longer term, the goal is to identify whether DSA is able to offer the high pattern fidelity that is required for CMOS manufacturing. The integration approaches are largely divided into two groups: grapho- and chemo-epitaxy. In both cases the morphology (line/space or hole) and dimensions of the patterns is determined by the block-copolymer composition. The approaches differ in the method to position these structures. In grapho-epitaxy, topography on the wafer surface is used to guide the self-assembly process. In chemo-epitaxy this is accomplished through local variations in the surface energy. A method for grapho-epitaxy based DSA has been optimized at imec using 193nm dry lithography for the prepatterns C22912. However, this process shows a mismatch in the optimal process window for the prepattern and for the self-assembled block-copolymer. Also the process is very sensitive to imperfections in the prepattern, as shown in fig. 3. Finally, the guide structures in grapho-epitaxy consume surface area, which is uneconomical. Chemo-epitaxy does not suffer from this, since the guide structures are smaller and underneath the patterns. Moreover, chemo-epitaxy is considered to be able to provide more robust results and thus preferred for frequency multiplication of line/space structures. Efforts to make an existing chemo-epitaxy flow compatible with requirements for a manufacturing environment have resulted in successful pattern generation P22923. This process is now under further optimization in the imec 300mm facility.
Figure 3: Impact of programmed defect on patterning in grapho-epitaxy DSA. Protrusion defects of varying amplitude have been designed (top row) and are well replicated on reticle (2nd row). The defects cause local line widening in resist (3rd row) and after block-copolymer coat and anneal (with 4 line/space periods in each prepattern space) only the smallest defect does not print.