Instrument commonality aids industrial spectroscopy
A flexible spectrometer engine allows cost-effective use in both laboratory and production settings.
The pairing of compact array-based spectrometers with on-board microprocessors has resulted in a new generation of streamlined, portable instruments. These spectrometers are well suited to on-line and at-line industrial measurements and can even be used for closed-loop process control. The latter is based on calibrations and standards determined in a laboratory setting, and the key to their successful, robust transfer into production is to use the same spectroscopy “engine” in all the instruments involved: benchtop, on-line, portable, and integrated OEM application. In one example, a novel semiconductor application of portable spectrometers has been developed by Blue29 (Sunnyvale, CA) that uses spectroscopy in its technology for depositing metal-alloy barrier layers on wafers without the use of vacuum processing.
The modern spectroscopy engine
In analytical-chemistry applications, the data from a spectrometer can be used to determine the composition of a sample or, more commonly, the concentration of specific constituents in that sample. Most modern spectrometers operating in the near-UV through near-IR (NIR) comprise collection optics, a slit, a diffraction grating and a CCD. Light enters the instrument through collection optics, passes through the input slit, and is then dispersed by the grating and detected by the CCD. This is typically a linear array consisting of 1024 or 2048 photodiodes or CCD pixels, although high-performance research instruments sometimes use a 2-D array to enable simultaneous analysis of light from multiple fiber inputs stacked in the nondispersive direction. The performance of these components and their precise configuration in this spectrometer engine determines the system spectral resolution, spectral range, wavelength linearity, signal to noise, signal gain as a function of wavelength, and so on.
The raw data from a spectrometer is converted into the parameter(s) of interest, which is often the concentration of one or more chemical species in the sample under test. This conversion is performed by a computer algorithm and can be based on a simple ratiometric comparison with reference data or can involve complex data fitting, such as regression and partial least squares, or, in the case of near-IR, can be a full spectral fit using chemometrics. In all cases, the ultimate goal is to make robust, accurate, and repeatable measurements. The fitting of the data is thus specific to the particular spectrometer engine.
Use of selective features
Because repeatable data fitting requires consistent instrumentation, it would seem simple to conclude that identical instruments should be used at all stages of calibration and implementation for a given application. However, for most manufacturers, this doesn’t make sense from a cost standpoint. This is because the applications developer typically needs an instrument with greater flexibility and functionality, and hence higher cost, than the production-line user.
To meet these diverse needs, today’s instrument providers offer a wide range of optional functions, beyond just dispersion and detection. These functions can be summarized in four categories: data storage, data display, data analysis, and data transfer.
Optional data storage can take several forms. Traditionally, disk drives were the method of choice for data storage, but today’s technology offers solutions that draw much less power and occupy less space. Data storage now is typically offered in the form of random-access-memory chips supplemented by removable flash-memory cards or sticks, not unlike the arrangement in a mass-market digital camera.
For data display, a smart touchscreen enables the user to interact with the system software through a graphical user interface, for example to select out spectral features for fitting or storage, or to operate data-acquisition, storage, display, and transfer functions.
The incorporation of data-analysis hardware and software enables an operator to perform different types of measurement, including absolute measurements, optical density, background subtraction, and so on. It also permits other mathematical manipulations of the data, such as addition or subtraction of spectra or individual data points or values. The software typically provides a choice of data formats, including SPC (GRAMS Standard) and CSV.
Data transfer can take several forms. Serial (RS-232) interfaces are still common, but most high-performance instruments now offer a choice of a GPIB parallel interface as well as an Ethernet input/output and a USB link. These interfaces not only allow data transfer but also enable remote operation of the spectrometer from a host computer.
FIGURE 1. Instruments based on the same spectrometer engine can offer quite different levels of functionality.
The ready availability of these higher-level functions means that instrument manufacturers often offer two or more versions of the same spectrometer. The most basic version has virtually none of these higher-level functions and is intended for operation with a dedicated personal computer in the case of an end user, or by the system computer in the case of an OEM application. An example of this type of instrument is the OSM-100 from Newport. Conversely, self-contained instruments, such as the OSM-400 from Newport, can be based on the same spectrometer engine, but offer some or all of the above functions, and may even include battery operation for portability (see Fig. 1).
Blue29 and cobalt barrier layers
Blue29 is a developer and vendor of electroless metal-deposition technology for the semiconductor wafer industry. The company is using spectrometers in an interesting process-monitoring application that demonstrates the value of having two different instruments that are based on a common spectrometer engine.
“As chip manufacturers strive to keep pace with Moore’s Law, the resultant miniaturization presents problems in both fabrication and device reliability,” says Dan Marohl, vice president of engineering at Blue29. “A key factor causing diminished reliability stems from the reduced size of the copper (Cu) interconnects used in multilayer devices. Specifically, these small interconnects translate into high current densities.” This causes the Cu atoms to migrate in the opposite direction to electron flow by a process called electromigration (EM), thereby thinning the interconnect and further raising the current density, notes Marohl. Eventually, this leads to the formation of voids and, ultimately, the failure of the interconnect.
The Cu-dielectric interface provides the main route for Cu-atom migration. In high-density devices, migration of the Cu into the dielectric causes the additional problem of modifying the so-called k value of the dielectric layer (a constant important in semiconductor processing). Several barrier materials had been previously tested to address the EM problem, but they suffered from two limitations: they were not sufficiently effective and they involved complex processes (for example, blanket metal-film deposition by physical vapor deposition on recessed Cu lines followed by chemical-mechanical polishing) that added cost and reduced overall process yields.
FIGURE 2. High current densities in miniaturized copper interconnects can result in migration of the copper atoms, leading to voids and failed interconnects. The problem is greatly reduced by the use of a cobalt tungsten phosphide barrier cap rather than a standard dielectric silicon carbide nitride cap (data courtesy of Intel). In fabricating these barriers, spectroscopy is used to monitor the deposition of cobalt from a solution of numerous compounds.
In response, Blue29 developed the use of a cobalt (Co) alloy as an alternative and effective barrier layer. Sometimes referred to as a cobalt cap, this cobalt tungsten phosphide (CoWP) layer has two advantages. First, the Cu-Co metal-metal bond is much stronger than a Cu metal-dielectric bond, so the interface between the Cu and the barrier layer prevents Cu migration. In addition, the Co itself is very resistant to migration. Blue29 developed a wet-chemistry process that directly deposits this CoWP layer on top of exposed Cu without resorting to electrochemistry or additional masking (see Fig. 2).
Process monitoring
A key component of the process is the catalytic deposition of Co from a complex solution containing Co ions plus numerous other compounds. For each customer’s specific need, hardware (chemical and wafer handling) and software to monitor the chemistry in a seamless manner is supplied. The wafer manufacturer then simply exposes the wafers to the chemistry for a finite time, producing a specific target thickness of CoWP.
Without replenishment of key components, the chemical composition of the deposition solutions changes with repeated use and eventually reaches a point at which the solution must be discarded or recharged with key materials. It is thus very important to monitor all key components in the solution, which is done using several techniques. The Co concentration is monitored using visible-absorption spectroscopy with a deuterium-tungsten light source.
In the laboratory, a spectrometer is used with a cuvette holder to study captured samples from test platforms, as well as new systems before customer delivery. In the final product, a spectrometer that contains the same engine as the lab spectrometer, but is otherwise simpler, is integrated into a rack-mounted system controller. The spectrometer is integrated with a 1 × 4 fiber switch, according to Tim Franklin, a mechanical engineer at Blue29. This allows the system controller to connect the spectrometer with four different fiber bundles. Two of the bundles are connected to flow cells in two separate process modules to maximize system throughput; a third is connected to a flow cell in a reference sample of deionized water; and a fourth is used to monitor the condition of the deuterium lamp.
The use of two equivalent spectrometers-one battery operated and with display, data-analysis, and memory-card functions for lab use, and the other much simpler, but with an ethernet input/output for integration into the product-gives a data-collection ability that is consistent for process development, production simulation during test and assembly, and use in the final product.
HECTOR LARA is senior product manager and RON HARTMAYER is director of applications and products at Newport, 1791 Deere Ave., Irvine, CA 92606; e-mail: [email protected].