UC Berkeley

After decades of research, extreme ultraviolet lithography (EUVL) is being used in high volume manufacturing of semiconductor chips. However, extension of EUVL beyond the current manufacturing nodes requires advancement in nearly every step of the lithographic process, from the source used to generate EUV photons, to the etch used to create a physical structure on the silicon wafer. This thesis focuses on understanding the major bottleneck to further scaling of EUVL, the photoresist material responsible for capturing the light. These incredible materials capture incoming ionizing radiation, use the resulting photoelectrons to excite target chemical processes, drive that chemistry to induce a differential solubility between exposed and unexposed material, and ultimately are developed to leave behind the desired pattern on the wafer for further processing. The interplay between all of these processes needs to be nearly perfect for the process to achieve acceptable yield for high volume manufacturing, let alone for achieving the desired performance of the integrated circuit product. However, many intrinsic, random processes, including photon absorption, photoelectron generation and propagation, and the chemical distribution of the photoresist material itself, conspire to degrade the fidelity between the design and realization of the pattern on the wafer. In this thesis, these effects are studied through a combination of stochastic modeling and experimental techniques, with the goal of providing insight into the workings of photoresist materials and to shine light on how they may be improved for future lithographic nodes.

After an introductory first chapter, the second chapter is focused on the results of a stochastic resist model used to study in the influence of various resist parameters on material performance. A key metric used in the semiconductor industry to quantify resist quality is the roughness at the line edge of line/space patterns. Known as line edge roughness (LER) for a single edge or line width roughness (LWR) for roughness in width, this parameter must be kept to a minimum, both to enable accurate overlay of subsequent lithographic layers, and for device performance purposes. To study the impact of chemical additives to the host photoresist polymer on the resulting roughness of resist materials, the stochastic Multivariate Poisson Propagation Model (MPPM) was used, in which the initial distribution of photons and additives are treated as a random, Poisson variables and propagated through the exposure and bake steps to produce the final lithographic structure. The modeling study showed the critical tradeoff between chemical noise and chemical image slope that results from the exposure and subsequent post-exposure bake (PEB) in the workhorse material of the industry, chemically amplified resist (CAR). CAR materials are comprised of a host polymer and photoacitve additives. The modeling study showed that by changing the photoactivity of the additives, one can improve LWR by improving the chemical gradient that separates exposed and unexposed regions of resist, at the cost of additional noise coming form the photon and chemical shot noise in the material. This fundamental tradeoff underpins current resist development efforts, with some resist suppliers moving to single-component photoresist materials to minimize chemical shot noise contributions, while other seeking to increase the nonlinearity in CAR-like materials to improve the exposure-induced chemical gradient.

A second modeling study expanded the 2D resist model to a full 3D model, with a focus on another stochastic failure modality, closed contact holes. This study showed that small ``roadblocks'' of material with too little dose or the wrong combination of photoactive compounds can lead to a failure of the contact to develop. This process is sensitive to the precise nature of the development process; in particular, developer that is able to remove small road blocks has the ability to lower missing contact rates, at the cost of shifting the contact to a larger size at equivalent dose. Furthermore, the study was used to examine the possible benefits of specialized materials or processing that alter the diffusion of photoactive compounds during the post-exposure bake. In particular, the model verified what had been predicted previously by simplified algebraic models; increasing diffusion perpendicular to the wafer leads to a lower rate of closed contacts without the negative impact on the chemical slope.

The next two chapters focus on experimental means to measure input parameters to the MPPM. The first technique explored was atomic force microscopy (AFM), a technique in which a sharp probe is scanned across a sample, producing, among a few different possibilities, a height map of the sample surface. This technique was applied to measure the latent chemistry present in photoresist exposed to radiation but not yet developed. The exposure and bake process in resist materials often leads to outgassing of the chemically altered material, leaving behind a topgraphic structure. The experiment showed that roughness measured in the post-develop structure via scanning electron microscopy is present also at the latent stage, and that the line width and LWR prior to dissolution can be characterized via AFM. Furthermore, the results suggested that the topographic transition from the exposed to the unexposed regions of the latent image seemed to agree with an increase in chemical slope predicted by modeling.

AFM was also used to perform high-speed, in-situ measurements of the dissolution process. This was achieved using a specially designed flow cell to control the injection of the developer, a critical component of the experiment, as the development occurs on a sub-second time scale. These experiments were able to show that the exposed regions of resist swell prior to removal, a result seen previously with diluted developer, but not thought to occur when in the full-strength process. Additionally the results show that the development process itself serves as a sort of low-pass filter on the roughness; the material removed most quickly leaves behind a number of bumps on the line edge that are removed by further exposure to developer. Finally, using spiral scan trajectories instead of traditional raster scans, the technique was able to image capture a \qty{250}{\um} by \qty{250}{\um} image at a \qty{10}{Hz} frame rate, enabling measurement of the dissolution dynamics of the resist material. Together, these results provide new insight into the development process that will inform future dissolution modeling efforts as well as guide dissolution process improvements.

The final chapter is devoted to resist characterization using resonant soft X-ray scattering (RSoXS), a technique that combines near edge fine structure spectroscopy with X-ray scattering. This technique shows promise for measuring both the chemical noise within the photoresist and the measurement of latent images in the material. This thesis discusses the work performed to date using RSoXS. In measuring polymer-based samples that are either exposed, or exposed to a uniform radiation pattern, RSoXS showed subtle differences in scattering signature near the carbon edge as a function of exposure and incident X-ray energy, suggesting sensitivity to the acid-driven deprotection reaction in the film. On the same set of samples, RSoXS was unable to measure aggregation of photoacid generator molecules, suggesting that any aggregation that may occur does so on a length scale smaller than the \qty{10}{nm} resolution of the experiments that were conducted. In a different, metal-oxide based material, scattering experiments near the tin L edge revealed cluster size changes as a function of exposure and metrology wavelength, which suggest clusters approximately \qty{1}{nm} in size on average in the unexposed film, with a broadening in the distribution upon exposure. These early results highlight the potential utility for RSoXS to measure chemical homogeneity in resist films, a critical characteristic for materials that must pattern reliably in the sub-\qty{10}{nm} regime.

RSoXS was also explored for profilometry of the latent exposure pattern in the material. A combination of experimental results and rigorous electromagnetic simulation highlight the challenges that must be overcome in interpreting the collected data. Notably, proper fitting of the underlying latent structure must separate the scattering signal into contributions from exposure-induced shrinkage, chemical modification of the underlying film, and thin-film interference effects, particularly for conducting experiments in a grazing-incidence configuration. This picture is further complicated by modeling and experimental results that suggest that the Born approximation, a commonly used model used to interpret X-ray scattering results, may not sufficiently capture all the physics of the scattering in the X-ray regime, as highlighted by a discrepancy between the Born approximation and rigorous modeling for transmission based samples. Nevertheless, the modeled and experimental results do show that the technique is sensitive to both physical and chemical structures in latent photoresist; further development my yield a useful technique for latent photoresist metrology.

Together, these studies contribute to the foundational understanding and measurement techniques required to design materials that can pattern trillions of features without error. It is my hope the findings and techniques presented in this thesis will serve as a guide for future researchers as the community works towards extending EUVL, ultimately pushing the limits of technological achievement.