Spectral imaging for high-throughput metrology of large-area nanostructure arrays

Modern high-throughput nanopatterning techniques such as nanoimprint lithography make it possible to fabricate arrays of nanostructures (features with dimensions on the 10’s to 100’s of nm scale) over large area substrates (in² to m² scale) such as Si wafers, glass sheets, and flexible roll-to-roll webs. The ability to make such large area nanostructure arrays, or “LNAs” as we will call them, gives birth to an extensive design space enabling a wide array of applications. For instance, LNAs exhibit nanophotonic properties enabling optical devices like wire-grid polarizers (WGPs), transparent conducting metal mesh grids (MMGs), color filters, perfect mirrors, and anti-reflection surfaces. LNAs can also be utilized for increasing surface area as well as generally creating large arrays of discrete features to be utilized as building blocks for electronic components in memory storage devices, sensors, and microprocessors. These unique properties make LNAs immediately attractive to certain industries such as the display and photovoltaic industries. As fabrication methods for LNAs are becoming viable, various industries are becoming interested in pursuing high-volume manufacturing of LNAs for these applications. Unfortunately, metrology methods are currently rudimentary outside of the silicon integrated circuits industry, impeding manufacturing scalability in applications such as displays and photovoltaics. Metrology is essential in the manufacturing context, because it provides invaluable feedback on the success of the fabrication process, both during new process development and large-scale production by tracking of device quality metrics, including performance and reliability metrics, and enables classification of defects that cause devices to not achieve desired quality metrics. Traditional nanometrology methods have fundamental issues which make their applicability to LNA manufacturing difficult. In particular, their low throughput is a major deal-breaker. Fortunately, the nanophotonic properties of LNAs offer a convenient basis for metrology which offers the potential to bridge the gap between the macro and nano scales. This is because the nanophotonic properties of LNAs are inherently geometry dependent, meaning that the optical effects observed from LNAs on the macroscale give direct insight into what is happening on the nanoscale. These optical properties can be characterized using spectral imaging methods such as RGB color imaging, multispectral imaging, and hyperspectral imaging. The throughput of these systems can be extremely high relative to traditional metrology approaches. For instance, a hyperspectral imaging system, when optimized, can achieve throughput of 2.6 m²/hr with 61 spectral bands (wavelength centers of 400 to 700 nm in steps of 5 nm) and a resolution of 10 x 10 µm. An RGB imaging system can achieve an even higher throughput of 15.3 m²/hr. The 10 x 10 µm lateral resolution is often adequate for display and photovoltaic applications. The high throughput makes this approach is incredibly attractive. In this dissertation, we show how spectral imaging techniques can be applied to metrology characterization tasks including defect detection and classification as well as providing a geometric measurement capability via a technique called optical critical dimension (OCD) scatterometry. In this work, we utilize exemplar manufacturing methods, namely JFIL nanoimprint lithography, to create a variety of exemplar LNAs on which we demonstrate the various metrology capabilities of spectral imaging. These LNAs include plasma etched vertical Si nanopillar arrays, metal assisted chemical etching (MACE) vertical Si nanowire arrays, WGPs, and MMGs. Each of these devices has unique manufacturing processes, and we show how the various manufacturing process steps can create a variety of different defects. Naturally, many of the defects originate in the nanoimprint process which lithographically defines the features. We show how defects like particle contamination, non-filling, residual layer thickness (RLT) variations, and adhesion failure uniquely manifest as changes in the optical signatures of the LNAs and use this principle to provide a basis for defect detection. Then, we show how image processing methods can be used to classify what types of defects have occurred over large areas such as wafer scale. Furthermore, we demonstrate that spectral imaging can be used as a geometric metrology using the OCD method, and show how hyperspectral imaging, in particular, can provide geometric measurement on wafer scale areas. The large field of view (FOV), high spatial resolution, and high speed offered by the spectral imaging approach allows for identification of a variety of interesting defect signatures that would be difficult, or nearly impossible, to observe using other metrology approaches. Finally, we discuss ongoing development of a spectral imaging system for roll-to-roll (R2R) LNA manufacturing. Construction of this system will begin in the months following this dissertation and will primarily be applied to manufacturing of WGPs and MMGs on R2R. In summary, these demonstrations are intended to serve as a demonstration of the use of spectral imaging wherever possible in LNA manufacturing. Naturally, this requires that the LNAs being manufacturing exhibit significant enough optical effects for the approach to work, but when this is the case, the advantages of the approach appear outstanding and thus have the potential to be utilized in volume manufacturing of LNAs.