Materials loss per cleaning cycle must be limited in advanced semiconductor process flows. Using dilute chemistries is necessary to achieve this goal. Fab engineers need to be able to analyze the concentrations of each component in processing solutions in real time to ensure that the chemistries are always in a narrowly defined process window. This article describes how multivariate methods can be applied to analyze chemistries of concentrations ranging from conventional to ultra-dilute levels.
Material loss during cleaning is particularly problematic for advanced technology generations where there are very shallow junctions at the drain and source areas, and any excess removal of silicon can be detrimental to the resulting transistors. The International Technology Roadmap for Semiconductors (ITRS) therefore limits the materials loss of each cleaning cycle at 0.5Å for both oxide and silicon for the 65nm technology node, and 0.3Å for the 45nm technology node .
To achieve minimal materials loss, leading R&D groups have resorted to using dilute chemistries for cleaning processes. However, if the solution becomes overly dilute, then cleaning becomes ineffective. If the solution is too concentrated, then materials loss will exceed that permitted by the process specification. Therefore, concentrations of all components in a cleaning solution must be precisely controlled at all times. Effective in situ methods are required to monitor the concentrations of cleaning solutions in order to make necessary timely adjustments by spiking or diluting key components.
Currently adapted monitoring methods such as resistivity/conductivity techniques, though widely used for single component solutions (HF, for example), cannot differentiate contributions from different components in multicomponent solutions, such as SC1, SC2, DSP, etc. Titration and other laboratory techniques, though highly accurate and fully capable of differentiating co-existing components in the same solution, are relatively slow and not ideal for real-time control of solution concentrations. Conventional optical methods, typically measuring absorbance at discrete wavelengths corresponding to the absorption peaks of each component, worked well for legacy concentration ranges such as 1:1:5 or 1:1:20 SC1. But as the solutions become more dilute, the absorption signals are very weak even at the peaks, and the corresponding signal-to-noise ratio becomes very small. In some cases, absorption peaks of key components overlap each other. Such conventional optical methods lack the sensitivity mandatory for analyzing dilute chemistries. An improved method is necessary to analyze highly dilute solutions, such as 1:1:100 SC1 and 1:200 HF for advanced technology nodes.
To reliably analyze ultra-dilute multicomponent wet processing solutions in real time, the method must be noninvasive to both chemicals and wafers, meeting the automatic operation requirements for wafer fabs. The corresponding analytical results can then be fed back to the wet processing equipment through closed-loop control, so that the concentrations are maintained within process windows.
A broadband absorption monitoring hardware system can analyze optical absorption signals of semiconductor wet processing solutions in the near-infrared (NIR, 700-2500nm) wavelength range. However, unlike simpler optical systems that measure at only a very limited number of wavelengths corresponding to absorption peaks, this hardware measures the absorption spectra of the chemical solutions across a wide range of wavelengths at very small intervals. The corresponding spectrum, which includes several hundreds of data points, is then analyzed by a proprietary software routine using a multivariate method  to resolve the limited number of unknowns.
In SC1 for example, the only unknowns are the concentrations of NH4OH and H2O2 balanced by DI water. The abundance of data points ensures high accuracy of the analysis; the redundancy of information compensates for noises associated with signal collection. Figure 1 shows examples of as-collected spectra by the hardware before being fed into the microprocessor for further processing.
Figure 01. As-collected absorption spectrum by the hardware. The two bands indicate the limited portion of these data captured by conventional optical methods
The solutions were buffered HF, and each widely separated and differentiable spectrum corresponds to a distinctly different NH4F concentration. The two wavelength bands show examples of the discrete ranges typically used by conventional optical methods.
In many real fab cases, particularly for dilute solutions, minute changes in concentrations cannot be easily differentiated from raw data, rendering the conventional optical methods ineffective. In such cases, further processing of the spectrum by a multivariate method is mandatory to achieve the desired accuracy to determine the component concentrations. A schematic of such a setup is shown in Fig. 2a.
Figure 02a. The set-up of a monitoring system whereby light emitted from a source is diverted by a multiplexer into multiple channels through optical fibers to individual measuring modules
Light emitted from a high-stability broad-wavelength-range light source is diverted by a multiplexer into multiple channels through optical fibers to individual measuring modules. The setup is capable of measuring eight or more samples/points of different compositions.
Within each measuring module, piping (in a re-circulation loop or at point-of-use) of the wet processing equipment is encased by a flow-through cell (Fig. 2b).
Figure 02b. The flow-through cell hardware used to take the measurements
Incoming light is partially absorbed by the solution inside the piping before being collected by the optical fiber at the other side of the tubing. Light signals from multiple locations are then detected in sequence by a diode array detector before being forwarded to the computer for signal processing. The flow-through cell blocks stray light from the cleanroom and holds optical fibers onto precision locations and orientations, with respect to piping, to minimize measurement noises.
The flow-through cell is an integral part of the fluid path and does not alter the diameter from upstream or downstream piping. This eliminates concerns about optical distortion due to compression of fluid or release of dissolved air. No optics are in physical contact with the solutions in the piping and the reliability of the setup is expected to be high. Each analysis takes 1.5 sec, including software processing and collecting multiple full spectra at 0.4 millisec/spectrum.
One significant improvement of the current work over the authors’ earlier results [3, 4] is temperature correction. Small temperature variations at measurement points are unavoidable because of temperature changes in the solution and in the ambient, among other factors. Optical absorption signals are known to be extremely sensitive to minute temperature changes. For example, measurements of a dilute HF solution prepared at 620ppm varied perceptibly due to slight changes in temperature despite real-time correction. To resolve this problem, new spectral temperature correction software was developed that fundamentally minimizes measurement variability. The new spectral approach ensures that the measurement is insensitive to temperature variations over several degrees Celsius.
Multivariate measurements of NH4OH in ultra-dilute 1:1:500 SC1 solutions correlated better than 0.995 with the known concentrations of as-prepared solutions. Statistics of measuring H2O2 in SC1 showed similar reliability.
Figure 3 shows concentration changes of the two key components of an originally 1:1:50 SC1 mixture over a 16-hr period in a recirculation loop without active concentration control.
Figure 03. SC1 concentration changes over a 16-hr period in a recirculation loop without spiking/dilution, showing the ability to clearly resolve the components
NH4OH decreases much more slowly than H2O2, thus the latter should be replenished more frequently into the tank. Such different trends cannot be resolved by resistivity/conductivity analytical techniques since contributions from ammonium hydroxide and hydrogen peroxide cannot be differentiated.
The multivariate technique measured several different concentrations of dilute HF between 1000 and 3000ppm (corresponding to approximately 1:570 and 1:190 dilutions, respectively) over an extended period of time. The temperature of the measuring environment was not actively controlled, varying between 20° and 25°C. Results show that all datasets were within 2.5% of the known as-prepared concentrations, and the repeatability was better than 0.7% in all cases. In general, when measuring HF of a 50,000ppm concentration, a standard deviation of 200ppm (0.4% σ) is achievable.
We monitored the concentration of a dilute HF solution as it changed from 1500ppm to 8350ppm by spiking and then back to 1500ppm by dilution. Because of the short 1.5 sec analytical time of each measurement, several data points during mixing were captured in the graph (Fig. 4).
Figure 04. Measurement of dilute HF solution mixing, showing that a 1.5 sec analysis time allows for control of dynamic changes to recipes
The effectiveness of the current method to measure solutions in transition can help fabs precisely control tank or point-of-use concentrations associated with process recipe changes.
Other cleaning solutions, including BEOL organic solvents, aqueous solutions, HF/H2SO4 blends, as well as many proprietary mixes were measured using the same setup, and similar accuracies as the previously mentioned examples were achieved.
A proven multivariate method measures ultra-dilute single- and multi-component wet processing solutions in real time. Simpler optical methods to measure concentrations often have problems associated with unavoidable small temperature variations. The multivariate method and system can be used to help control the concentrations of critical processing solutions in fabs, vital for necessary wafer cleaning without excessive material removal.
Chenting Lin, Peter Bratin, Guang Liang, Michael Schneider, Eugene Shalyt
ECI Technology, Totowa, NJ
1. “Surface Preparation Technology Requirements: Near-term Years,” Table 68a, Front End Process, ITRS, 2005 Edition.
2. I.T. Jolliffe, Principal Components Analysis, 2nd ed., Springer, 2002.
3. E. Shalyt, et al., “Real-time Monitoring of Dilute Multicomponent Wet Processing Chemistries,” pp. 163-172, 25th SPWCC Proceedings, 2005.
4. Y. Shekel, et al., “Real-time Chemical Monitoring by NIR Spectroscopy,” pp. 245-250, Proceedings of the 208th ECS Meeting, Los Angles, CA, 2005.