Concentration Sensors Curb Rising CoO

Developments in IC manufacturing are increasing the demand for in situ concentration measurement as liquid chemicals are consumed in significantly greater quantities for chemical mechanical planarization (CMP), photolithography, electrochemical deposition (ECD) and surface preparation. At the 65 nm node and beyond, process windows are narrowing to the point that increasingly smaller variations in chemical concentration adversely affect wafer defect levels and, most importantly, yield. In addition, there is increasing pressure to minimize the environmental impact by reducing liquid chemical waste streams.¹ The impact: overall cost of ownership (CoO) is on the rise.

A solution to rising CoO is process control of liquid chemicals with highly accurate concentration and analysis systems that take measurements continuously, in real time and at point-of-use, where chemicals are blended or delivered. In addition to increased yields, real-time sensors contain costs by reducing chemical consumption, waste streams and wafer scrap while increasing efficiencies and productivity.¹

Existing technologies for measuring concentration include conductivity, near-infrared (NIR) absorption spectroscopy, ultrasound, titration and index of refraction (IoR). However, proven effective in other industries, the unique demands of liquid concentration sensors for IC manufacturing may make a particular technology less appropriate, whether caused by in situ location, real-time response, resolution or size.

This article discusses existing technologies, as related to semiconductor applications, and presents case studies comparing IoR technology to auto-titration and other sensor technologies.

Based on a fundamental laboratory technique, titration remains the gold standard for sensor accuracy and resolution. This method collects a sample from the process line, combines it with reagents, and analyzes the concentration. A titration unit may be linked to the process line (auto-titration) or operate as an unconnected unit (manual titration). If connected to the process line, it may be called "inline," but fluid only travels in one direction, toward the unit, where it is processed, analyzed and then disposed of, creating a waste stream. It is not "inline" in the sense that process fluid flows through it on the way to its destination in real time.

In spite of its large size and slow speed, where one reading may take 45 minutes, titration is widely employed in semiconductor manufacturing today. Even if readings are only intermittently taken, titration remains the point of reference. In processes such as CMP and wafer cleaning, batches of chemicals are circulated, spiked and disposed of in accord with the experience of the technical staff, which is informed by a history of titration measurements.

Titration, however, is impractical for measuring the concentration of more than one liquid chemical at a time. The ongoing expense of chemical reagents and replacement parts, such as O-rings and probes, leads to high CoO.

Conductivity works by measuring voltage across cylindrical probes placed in a liquid chemical. This method provides reliable measurement at a low cost across a narrow dynamic range of concentration measurements, beyond which resolution deteriorates. A single unit is typically calibrated for a particular dynamic range. If measurements are required in an additional range, a second sensor is calibrated. Common applications in semiconductor manufacturing, in fact, often exceed a single unit's dynamic range.

A second limitation is that only chemicals containing ions can be measured, which excludes the measurement of solvents, such as benzene, methanol, acetone and hydrogen peroxide. Finally, the probe must make contact with the fluid stream, which may affect flow or raise the prospect of contamination.

For NIR absorption spectroscopy, a fiber optic probe ("light pipe") is placed in the fluid stream and carries a light signal back to a box for analysis. Unique to this technology is its ability to read and identify the individual "fingerprints" of several chemicals inside a single blend. NIR spectroscopy also measures concentration, but with limited resolution or accuracy. Therefore, it is most often used in cases where identifying the chemicals in the mix is more important than tracking concentration.

Measurements are slow, produced at a rate of about one every 30 seconds and, in a fab environment, the long fiberoptic cable may present challenges in terms of its stability.

Furthermore, NIR spectroscopy units are comparatively expensive, have a large footprint, and the sensors may present challenges in terms of calibration.

A similar form of this technology, ultraviolet (UV) visible spectroscopy, measures UV wavelengths instead of infrared. Its benefits and limitations are otherwise similar to those of NIR spectroscopy.

Ultrasound is based on measuring the speed of sound through a fluid. Because the speed of sound is dependent on a liquid's pressure, temperature and flow rate, these variables must be controlled or compensated for. With the addition of a pressure regulator and flow controller, variables may be controlled and ultrasound can provide a high degree of resolution and accuracy in concentration measurements. Bubbles in the liquid will affect the accuracy of the reading, but this problem may be addressed through digital signal processing techniques.

Index of refraction (IoR), a well-established technology only recently used by chipmakers, works by shining a light beam on a liquid chemical and then measuring the IoR. If the mixture or concentration changes, so will the IoR.

Refractive index values are determined in two ways: transmission geometry and optical reflective geometry. In transmission geometry, light must travel through the liquid before it can be measured, while in optical reflective geometry, a light source is directed toward a sapphire crystal on the liquid and the light reflecting off the surface is measured.

Transmission geometry is affected by the opaqueness and other properties of the fluid. The light beam can be absorbed or diffracted, both of which can produce incorrect measurements. Absorption increases as the liquid becomes more opaque, and diffraction occurs when the light is knocked off course by particles or bubbles. To lessen the effects of absorption and diffraction, transmission geometry uses a narrow fluid path, typically a maximum of 0.5 in., which can present challenges in terms of flow.

In contrast, optical reflective geometry is not affected by the opaqueness of the fluid because the measurement is taken at the optical window-liquid interface. In the Swagelok CR-288 concentration monitor, for example, light from an LED is directed at the surface of a sapphire window in contact with the fluid (Fig. 1). It strikes the sapphire at multiple points and angles (u_s) and is either absorbed by the liquid or reflected and scattered upward into a photodiode array. For reflectance angles (u_S) smaller than the critical angle (u_C), light is mostly reflected off of the window. For angles greater than u_C, light is both reflected off of the window and transmitted through it. By measuring the change in scattered light intensity as a function of u_s, u_C can be determined. From Snell's law, u_C is related to the liquid's IoR. Snell's law (n₁sinu₁ = n₂sinu₂) is the relationship between angles of incidence and refraction for a wave impinging on a surface between two media of different IoRs.

1. The CR-288 concentration monitor sensor and its internal configuration shows the layout of the light source, sapphire window and photodiode array in relation to the liquid.

Because the IoR will change as temperature changes, the output is passed through an additional algorithm.

Software enables fine adjustments in resolution where concentration is plotted against time for extended data logging. Average time per reading may be adjusted from one reading every second to one every 10 seconds or one every minute, and so on. For example, one data point per 10-second interval means that each data point represents an average taken over 10 readings (one per second). Similarly, the Y-scale may be altered so that the user may zoom in on a transition of interest.

The CR-288 concentration monitor can measure four different chemicals at the same time over a refractive index (RI) range of 1.25–1.45, with resolution of better than 5 × 10^-6. Response time is <1.2 sec. The following case studies test IoR technology against an auto-titrator and other technologies in third-party simulations of semiconductor industry applications.

In a joint study conducted by Chartered Semiconductor Manufacturing (Singapore) and Kinetics Process Systems (Singapore), the CR-288 concentration monitor was installed in the Kinetics slurry blending system to monitor hydrogen peroxide concentrations in the slurry batch. An auto-titrator was also employed, and the two sets of results were compared.²

The blending and distribution system is composed of two blending stations that work together to provide a continuous supply of slurry to multiple polishers. The slurry is blended and tested in one system before it is available for distribution in the global loop, which services the polishers.

For the study, the concentration monitor was retrofitted to one of the blending systems containing copper CMP Slurry E. The auto-titrator remained in place, measuring the same slurry mixture as the concentration monitor. An ANOVA using the Tukey-Kramer test with SAS-JMP software showed that the concentration monitor's performance was comparable with the auto-titrator's. In fact, the concentration monitor's sample-to-sample variation was better and the time required to prepare the slurry through circulation in the loop was reduced (Fig. 2).

2. An analysis of variation using the Tukey-Kramer test with the SAS-JMP software shows the sample-to-sample variation for the concentration monitor is comparable with that of the auto-titrator.

The concentration monitor's digital display unit (DDU) made concentration trends and data available in the control room, which eliminated the need for an offline computer. With elimination of the auto-titrator, the concentration monitor enables an appreciable reduction in the slurry system's CoO.

In a back-end-of-line (BEOL, post-etch cleaning) application, a mixture of solvent (ST250) and water is employed. As a normal part of the process, the water evaporates or is otherwise consumed.

The objective of the study, which was conducted by a third party, was to monitor water loss and replenish the water at a given concentration. The concentration monitor and a NIR spectrometer were employed in the process tool to measure water concentration. Figure 3 demonstrates that the resolution of the concentration monitor is higher than the NIR spectrometer.

3. In an application aimed at measuring the change in water concentration, the graph indicates that index of refraction (IoR) technology produces results with higher resolution than a near-infrared (NIR) absorption spectrometer.

In a separate but similar study, also conducted by a third party, the application used a mixture of solvent (EKC-265) and water. The concentration monitor was compared to a conductivity measurement instrument. Measurements were taken in the process tool. While the concentration monitor accurately plots the decrease in water, the conductivity tool is unable to accurately measure the change in concentration over the better part of the range (Fig. 4).

4. IoR technology plots a decrease in water concentration, but a conductivity-based measurement tool is unable to accurately measure the same change over the better part of the range.

Case study #3: differentiating among photoresists

In a study simulating a photoresist feed line, as one used for delivery to a lithography tool, the objective was to determine whether RI technology could be used to differentiate among all possible photoresists. The study was conducted by Swagelok at a third-party laboratory setting, Micron Technology Inc. (Boise, Idaho).³

The larger question in forming the study was whether the concentration monitor could be used to mistake-proof the lithography process by indicating in real time if and when the wrong photoresist was fed into the system, a mistake that could cost hundreds of thousands of dollars in scrapped wafers.

Two experiments are summarized here. In the first experiment, the goal was to determine if the CR-288 concentration monitor could draw clear distinctions among 15 different photoresists, including 10 deep ultraviolet (DUV) photoresists. The 15 samples were randomly fed into the concentration monitor and, for comparison, into a benchtop refractomer, Leica model AR600. In between samples, the two devices were cleaned. Each sample was fed into the two devices three times.

Initial results showed that the concentration monitor was able to produce distinct peaks for nine out of the 15 photoresists (Fig. 5). Two additional photoresists (samples G and K) became identifiable once the range and calibration of the sensor was adjusted. The four photoresists not detected (samples L, M, N and O) have RIs >1.46, which is out of range for the concentration monitor.

5. IoR technology was able to clearly distinguish 11 out of 15 photoresists tested. The initial nine detected are shown. Temperature has been corrected for 20°C.

For eight of the 10 DUV photoresists, the margin of error between the concentration monitor and benchtop device was <0.5%. Variance for repeatability between the three measurements taken of the 11 photoresists that the concentration monitor could detect was very good at 2.75 × 10^-4 RI.

The second experiment was designed to simulate an application where a continuous stream of different photoresists would be delivered to a lithography tool without the possibility of cleaning or purging in between. Different samples were injected continuously, sequentially and randomly into the same concentration monitor. Figure 6 shows that for nine out of 10 DUV photoresists, 20 mL of fluid was enough to generate a clearly identifiable step change. When the Y-scale was adjusted, a step change between the two DUV photoresists with the closest RIs (samples D and E) was also identifiable.

6. As indicated, in 9 out of 10 deep ultraviolet (DUV) resists tested, 20 mL of fluid was enough for IoR technology to generate an identifiable step change. With Y-scale adjustment, the step change between the DUV photoresists with the closest refractive indices (samples D and E) could also be detected.

The study concludes that the concentration monitor is able to draw clear distinctions between 10 out of 10 DUV photoresists and, therefore, would be able to identify an incorrect photoresist flowing into a lithography tool, trigger an alarm, and enable a system to divert flow away from the tool to waste, avoiding scrap or rework.

Sensors for semiconductor manufacturing are effectively meeting the new demand for miniature, accurate, real-time, inline measurement of chemical concentrations. In particular, IoR technology is proving effective in making fine determinations in applications concerning wafer cleaning, etching, lithography, ECD and CMP. These developments set the stage for significant reductions in environmentally sensitive waste, lost product and CoO for concentration sensors.