Exhaust Gas Analysis Helps to Reduce Costs

Gas-based processing forms the backbone of the modern semiconductor fab. As the industry has moved to embrace advanced technologies such as deep UV lithography, copper/low-k interconnects, and strained silicon transistors, a host of new challenges have emerged from narrower process windows, exotic specialty gases, and innovative gas generation and delivery systems. All these changes tend to increase the cost of wafer manufacturing.

Future manufacturing productivity improvements and cost reduction efforts will arise from improved utilization of consumables and tool uptime, in addition to traditional improvements from dimensional shrinks and wafer size increases. Increased sensor integration and advanced process control (APC) are considered by many to be the principal enablers for these productivity improvements. In particular, inline and in situ gas analysis holds great promise for improvements in semiconductor manufacturing, with benefits from reduced scrap, better consumable utilization, reduced greenhouse emissions, and faster yield ramping.

Robust and quantitative sensors can be inserted at a variety of points in the process, including gas supply lines, the process chamber, the tool exhaust and, finally, the abatement system and stack. Knowledge of gas concentrations at various measurement points provides leverage for improved cost savings as well as environmental compliance.

Several techniques can be used to measure the process and exhaust gases used in mainstream semiconductor processes such as etch, CVD, ALD, PVD and ion implant. Two leading gas analysis technologies, quadrupole mass spectroscopy (QMS) and infrared spectroscopy, can monitor and control gas streams in various locations, including downstream from the chamber, between pump and scrubber, and after scrubbing.

In recent years, Fourier transform infrared (FTIR) spectrometry has emerged as a robust, production-worthy technique capable of measuring most gases used in semiconductor processing. The technique is quantitative, with permanent calibrations, and is well suited to typical pressures found in all parts of the exhaust line.

Mass spectral analysis (i.e., residual gas analysis) is a well understood technique with the capability to measure a similarly wide breadth of molecular and atomic species. Because residual gas analysis (RGA) requires a fairly low vacuum for the measurement, only a small sample is pulled from the exhaust line into a low-pressure area created by the RGA pump.

In the past, RGA and FTIR analyzers were considered too large and costly, and the data too difficult to interpret to use for anything but R&D and troubleshooting. In the past 10 years, however, technology advances have reduced the size, weight and cost of RGA and FTIR sensors, enabling them to be easily mounted on vacuum lines. Both of these techniques are now also available with sophisticated analysis packages that provide easy-to-interpret results in the form of gas traces, rather than more complex spectral time series.

The latest FTIR sensors can be moved or swapped without breaking vacuum, minimizing maintenance costs and enabling one sensor to provide cost savings for multiple chambers. When the sensor is placed on the foreline, the analyzer can monitor the entire process gas stream, since it must pass by the analyzer. Both FTIR and RGA measurements are rapid — on the order of one second at low ppm concentrations or mTorr pressures — for immediate feedback of fault conditions. The Table documents the features of each sensor type. We will show that both techniques work well at all points in the exhaust stream of a target process.

While it may seem that, for process control, transport delays and mixing times might limit the usefulness of exhaust gas measurements between the chamber and pump, it is not hard to demonstrate that exhaust gas flow velocities in most processes are high, and changes to the process or wafer state are reflected in changes in the exhaust gas concentrations in less than a second. As a consequence, real-time measurements of gas composition provide a very sensitive and fast indication of the process state, and a number of crucial process parameters such as etch rate, endpoint and reaction efficiency can be well characterized using exhaust gas data.

Real-time foreline measurements are particularly useful data sources for process development and optimization. Using design of experiment (DOE) methods, the process engineer can systematically map a process space, and perform an intelligent trade-off analysis to identify the optimal process operating point, balancing multiple objectives, including materials cost, device yield and tool throughput. In addition to the data coming from the sensor, the data used in this analysis might include consumable usage, waste gas generation, process times, endpoint detection, or wafer quality data.

For example, a recent study by Mitsubishi Silicon America used RGA to analyze and optimize an epitaxial silicon process. Although the process was operated at atmospheric pressure, the RGA was able to identify unique effects such as byproducts recombining with the deposited material to create additional precursor gas. They also demonstrated that, by adjusting the process temperature, they could minimize this "reverse deposition" or etching of the material.¹

In another recent study, Samsung used FTIR as the principal tool for the multivariate optimization of a C₃F₈ chamber clean process. This optimization resulted in reductions of C₃F₈ usage by 30%, and of clean time by 15%. Overall, the project enabled significant reductions in PFC emissions and gas costs, while increasing tool availability.²

Environmental health and safety (EHS) groups also require analytical characterization of a new process, such as those performed at Motorola during a study of new gate materials.³ This group investigated the point-of-use abatement requirements for various process chemistries and their associated precursor byproducts such as nitric oxide (NO), nitrous oxide (NO₂) and ammonia (NH₃), which require special handling. The chemistry data proved essential to selecting appropriate scrubber technology, and for modeling abatement costs in the precursor selection process.

Once a process is characterized, the gas concentration variation over time as measured in the foreline can be used as a template or "fingerprint" for the process. Statistical analysis of multiple wafer studies can then provide either univariate or multivariate control limits for the process. Concentration variations that are outside normal process behavior can trigger process excursion alarms, reducing scrap, and improving device quality. Additional species observed in the exhaust may indicate the presence of contamination. Fluctuating traces may indicate hardware problems, such as backpressure on a mass flow controller (MFC, Fig. 1 ). Real-time gas analysis can also identify gas ratio faults (due to MFC faults or assignment of the wrong recipe), RF plasma problems, chamber-matching problems, first wafer effects, or poorly tuned chamber clean processes.

1. An MFC backpressure fault is indicated by a fluctuating value in CF4 partial pressure (green line) vs. normal conditions (red line), detected by an FTIR sensor on the foreline of an etch tool.

Gas traces can also be used in a comprehensive e-diagnostics system, giving the OEM or subcomponent supplier additional information to use in off-line troubleshooting. The sensor quantitative information can also be fed into a tool APC system for univariate or multivariate closed-loop control over the process.

As shown in Figure 1 , gas analysis can also be used for endpoint detection. Over-etching not only damages the wafer, but causes excess etch gases to be sent to the abatement equipment, where the scrubber must work harder to eliminate these often highly polluting and greenhouse-producing gases.

Gas analysis-based endpoint detection eliminates these costly effects. Because the measurement is in the exhaust, this type of endpoint detection can be retrofitted onto existing process equipment as requirements become more stringent. Etch gases, such as trifluoromethane (CHF₃), carbon tetrafluoride (CF₄) and nitrogen trifluoride (NF₃) are expensive; endpoint detection can also save on direct processing costs by turning these gases off as soon as the wafer has been completely etched.

While leverage after the process may afford the most significant cost benefits, measurement after the pump has advantages as well (Fig. 2 ).

2. Depending on the process and the possible gains, the gas analysis system may be placed directly after the process chamber, between the pump and the scrubber and/or after the scrubber.

Pump-to-scrubber monitoring

Measuring after the pump provides fault detection indicators for the pumping system and enables adaptively scheduled maintenance in place of more costly scheduled maintenance. Gas analysis can identify corrosion of the pumping line before it begins to cause problems for the scrubber or process.

Fine-tuning the scrubbing process using real-time loading data may achieve higher savings potential than pump fault detection. Knowing what the scrubber needs to handle at any point in time may allow the scrubber to be turned off or put into a standby mode. Energy for thermal activation,⁴ diluents and reacting chemicals can be supplied in the amounts needed, eliminating further excesses.

By understanding the relationship between operating conditions and scrubber efficiency, and by taking measurements before and after the scrubber, the user can select the best equipment and maximize the efficiency of the abatement device. In addition, energy costs for thermal abatement systems can be considerable. Throttling the fuel consumption using real-time data to verify effective scrubbing efficiency has the potential for large energy savings, resulting in a rapid return on investment (ROI) for the automated, permanent FTIR sensor.

A recent study at Winbond used FTIR before and after the scrubber to look at efficiencies from point-of-use abatement using thermal oxidation, dry adsorption and wet scrubbing. The team also evaluated central scrubbers using web scrubbing and volatile organic compound (VOC) rotating drums. Beyond optimizing scrubber performance, the researchers indicated they were able to reduce their costs by more than $1M per year as a result of increased dry scrubber cartridge lifetime.⁵ The authors of the study expect significant additional savings from reduced insurance premiums.

Most semiconductor companies are working to voluntarily reduce perfluorocarbon (PFC) emissions. Up to 90% of a 200 mm fab's PFC emissions come from etch and CVD chamber cleaning processes.⁶ Several companies have used FTIR to analyze the emissions from various chamber clean processes and have achieved significant PFC reductions. The hexafluoroethane (C₂F₆) clean process, for example, typically generates CF₄, which can be difficult to scrub.⁵

A team at Applied Materials used FTIR analysis of stack effluents to optimize a C₂F₆ clean process, and demonstrated significant reductions in PFC effluent. This group was even more successful using remote NF₃ cleaning, showing a virtual elimination of PFC emissions.⁷ These last results were confirmed by a similar study by Motorola, which showed >99% destruction of NF₃, again using FTIR.⁶ In this second study, gas analysis was also used to optimize downstream handling and abatement of the fluorine.

In the absence of inline metrology, environmental managers typically rely on independent test firms to perform expensive, time-consuming pre-compliance tests to verify scrubber operation. These tests consist of an iterative process that requires testing, the resolution of inefficiencies and then retesting, which can cost up to $10,000 per year per stack.

Instead, the on-site environmental manager can measure the gas levels at the stack and perform the iterative process without having to absorb the added expense of the independent test facility. Should abatement efficiency degrade, the manager has an immediate warning and the means to determine when the system can be returned to optimal working conditions, long before annual compliance tests are required. Exhaust gas analysis can also provide immediate and low-cost compliance confirmation after a change in process or chamber clean procedures.

In the case of CVD and ion implant, where highly toxic gases are used, gas analysis can detect increases in trace amounts of hydrochloric acid (HCl) or phosphine (PH₃), for example, before a release activates plant or external toxic gas alarms, thereby eliminating evacuation and emergency response costs.

Readily usable gas analysis is providing fabs with cost leverage points in CVD, etch, PVD and ion implant. While the greatest leverage comes from better processes and yield enhancements from APC and process characterization, gas analysis also helps save on direct processing costs by reducing source gas usage, minimizing downtime through improved process understanding and e-diagnostics, and lowering scrubbing costs. FTIR and RGA sensors enable inline monitoring of process gases in production fabs.

Using gas analysis for fault detection affects both yield and maintenance costs. Indirect costs such as insurance premiums and environmental compliance can be affected, while reduced emissions of toxic, ozone-depleting, and global-warming gases improve overall environmental performance — a key area of corporate accountability.