Green Manufacturing

Achieving closed-loop process control of
DI-water replenishment

Mark Simpson, Texas Instruments

Semiconductor devices have sensitive metal layers that can be damaged by exposure to traditional aqueous cleaning methods. Hence, IC manufacturers rely on specialized solvent-chemical mixtures to prepare the wafer surface. However, many commercially available solvents still rely on water as a critical component to remove etch residue, although water can cause corrosion or erosion in metal interconnect structures. Controlling the water concentration in such specialized solvents is crucial for achieving both effective cleaning and damage prevention. While a heated solvent bath can help remove water from the cleaning process, the bath’s usable life can be drastically limited by water evaporation, increasing solvent consumption and costs.

This article, based on work performed at Texas Instruments’ DMOS5 wafer fab in Dallas, presents a method for controlling water concentration in a heated solvent bath using a closed-loop system that replenishes water at the same rate that it evaporates. Providing real-time feedback of water-concentration data, the system is fine-tuned and controlled using a chemical analyzer, a programmable logic controller (PLC), an electronic flowmeter, and an electronic metering valve. The article also details the benefits of controlling solvent water concentration by eliminating nonstandard process conditions and extending the bath life in solvent hoods.

Initial Strategy: Replenishing DI Water at a Set Rate

Solvent chemicals used in postmetal semiconductor processes are designed
to operate in heated tanks to remove polymers from wafers effectively. When solvent is delivered into the hood via a bulk chemical delivery system, its initial water concentration is too high for safe metal processing. By testing water concentration versus cleaning efficiency, particle contamination, and corrosion over a period of time, it was determined that the bath life is typically 12 hours. A fresh bath takes about 30 minutes to reach operating temperature. Since the concentration of water in incoming solvent is too high and corrosion is likely to occur, a mixing step is performed while the bath is idle to recirculate the fluid at operating temperature and allow some water to evaporate. This procedure, however, shortens the effective bath life, as illustrated in Figure 1.

Figure 1: Percentage of DI water in heated solvent during a 12-hour bath life without water replenishment. (The results are from a lab analysis of hand-pulled samples.)

Dumping and replacing the chemicals in a bath with a 12-hour lifetime, especially when semiconductor-grade chemicals with an environmental overhead are used, is costly. The deployment of multiple hoods and the need to perform nonproduction mixing steps multiply costs. Since water is a key component in wafer-surface preparation processes, it is reasonable to replace it after it evaporates. The utility of metering water back into the solvent is driven by cost and throughput.

To replace spent water, DMOS5 asked the hood manufacturer to design a manual replenishing system that can drip water into the bath at selected rates. This redesign resulted in a supply line that was tied to the facility’s DI-water system and incorporated a pressure regulator, valve, and manual rotometer. The valve was controlled by software that permitted it to open at predetermined times (set by the engineer) and inject DI water through the rotometer. As shown in Figure 2, this continuous spiking system, when optimized, extended bath life from 12 to 48 hours. For all four baths at DMOS5, the new design resulted in a 75% cost savings.

Figure 2: Percentage of DI water in heated solvent during a 48-hour bath life with manual water replenishment. (The results are from a solvent concentration analyzer.)

The manual-replenishment system was closely monitored to ensure that control remained in the effective operating range. To accomplish that control, a solvent analyzer was installed to take samples of the solvent, analyze the solution, and provide a reading of water concentration once every 30 minutes, both on a display screen and a 4–20-mA analog output. The signal was integrated into the fab’s tool interdiction and monitoring system (TIMS), a fault detection and data collection system developed by Texas Instruments. Used to monitor equipment in real time and compare signals using a logic model programmed by the engineer, TIMS can take a series of predefined actions, such as initiating an alarm at the hood, shutting down a hood, paging a technician, or even sending e-mail alerts.

Current Strategy: Replenishing DI Water at a Variable Rate

Set-rate-replenishment systems work reliably only under optimal operating conditions. Because such systems do not provide direct feedback from the analyzer to the hood, they do not account for concentration variability caused by process loadings, solvent drag-out, or tank replenishment. In addition, solvent mixture is not stable enough over time to increase bath life longer than 48 hours.

After DMOS5 engineers had concluded that manual spiking control involved significant risks, an alternative method was pursued. Several different automatic direct-feedback methodologies to control water flow into the bath were considered, but because chemical concentration control can be challenging, the ideal instrument was difficult to find. An automatic control system that remains accurate under a variety of conditions was needed. Several criteria were evaluated:

• Low Flow Control: DI water had to be metered into a heated solvent tank at a low flow rate to compensate for evaporation. A metering valve capable of controlling well at low flows was required. Automatic control of such a valve necessitates an electronic flow measurement device
to ensure that the exact amount of water required is being metered.
• Type of Control: There are many methods to accomplish flow measurement and control. Device accuracy and particle contamination were major concerns.
• Electronic Input and Output: External feedback from the concentration analyzer was essential. The 4-20-mA output signal had to interface with DI-water replenishment control.
• Process Compatibility: A flow controller that is compatible with DI water had to have wetted parts that do not generate particles and had to be able to contend with variable incoming water pressure.
• Ease of Installation: A new automatic system had to be a drop-in replacement for the manual replenishment system. It had to be up and running quickly to prevent excessive downtime at the solvent hood.

Figure 3: The Model 6500 integrated flow controller.

In addition to these requirements, the device had to be able to integrate a flow controller with an automatic metering valve. The automatic replenishment system and the flow controller had to work together to measure and adjust flow automatically according to a variable setpoint.

After researching and bench-testing several devices, DMOS5 chose the NT Model 6500 integrated flow controller manufactured by Entegris (Chaska, MN). The unit, pictured in Figure 3, is a drop-in replacement for the set-rate control system. It can receive and output control signals, has all-PTFE wetted surfaces, and is both an integrated flowmeter and a metering valve together in one compact package. In addition, the unit offers 1% full-scale accuracy with fast response time, is available with a very low flow range, and controls DI-water replenishment properly, despite changes in feed pressure.

The integrated flow controller consists of three subcomponents: a differential-pressure flowmeter, a pro- portional control valve driven by a stepper motor, and proportional integral derivative control electronics. The unit receives a 0–10-V dc or 4–20-mA setpoint signal of the desired flow range and controls the proportional valve’s position to achieve the desired flow rate.

Figure 4 shows a schematic diagram of the integrated system together with the concentration analyzer. A PLC was added to the system to determine the setpoint sent to the flow controller based on the concentration of water in the bath.

Figure 4: Schematic diagram of the integrated flow controller together with a concentration analyzer and a PLC.

Using differential-pressure technology, the flowmeter in the integrated flow controller determines flow by measuring fluid pressure before and after an orifice in the flowmeter. The application flow rate and fluid type determine the size of the orifice built into the flowmeter. As described in Bernoulli’s principle, fluid flow passing through an orifice increases in velocity, causing an increase in kinetic energy. The increase in kinetic energy causes a corresponding loss of static energy, reflected by a pressure drop across the orifice. An increase in flow causes a predictable increase in differential pressure. Flow rate is proportional to the square root of the pressure differential across the orifice, as shown in the equation

where k = a constant based on the application properties, P1 = pressure before the orifice, and P2 = pressure after the orifice.

Figure 5: Cross section of the NT electronic flowmeter.

Figure 5 presents a cross section of the flowmeter with differential-pressure technology. Additionally, the flowmeter provides analog signals for both flow and pressure measurement, triggering alarms for nonstandard processing conditions.

The necessary flow range of DI-water replenishment was determined using the manual system. The integrated system was then mounted into a panel that included a display with a manual dial-in setpoint. A toggle switch was added to switch from the manual setpoint to the PLC setpoint. The PLC was programmed to give a varying output voltage to the integrated flow controller according to its input from the concentration analyzer. With that initial setting and historical readings from the analyzer, the right amount of DI-water flow was determined.

Before the new system was installed, investigators confirmed that automatic control would be error-free. To do so, they filled a test bath with solvent and heated it to temperature. The concentration and set- point from the PLC were both interfaced to TIMS to collect data. After the system was set up, test wafers were processed to check drag-out and incoming DI-water pressure variability. Other tests were performed to check the bath for particles, trace metals, and contamination. The results of these tests did not deviate from the baseline. Initially, bath life was 48 hours; later, it increased to 72 hours, leading to even greater cost savings.

Conclusion

The integrated flow controller with PLC feedback greatly improved the performance of the replenishment system. Figure 6 presents data from the solvent analyzer over a 72-hour period, and Figure 7 illustrates a graph of the TIMS historical viewer. The peaks in the graph show the concentrations of water after the bath has been freshly exchanged and automatic control has started following a short delay. The automatic replenishment system guarantees tighter control limits than the manual system and triggers an alarm through the TIMS system if the water concentration falls outside those limits. The integrated system allows for true loop feedback control in a critical application.

Figure 6: Percentage of DI water in heated solvent during a 72-hour bath life with automatic feedback control of water replenishment.

Texas Instruments improves manufacturing processes continually to provide customers with the highest-quality electronic devices. Since low yields and scrap are costly and affect customers negatively, the fab must be able to exert strict control over all processes. The method detailed in this article transformed solvent processing at DMOS5 from a process in which there was no statistical control over water concentration to one with consistent 6s capability. Greater stability has led to a sixfold increase in solvent bath life, resulting in a sixfold increase in utilization times and a >75% reduction in solvent costs. Additional benefits include a reduction in engineering time and maintenance by approximately 5 hours per week.

Acknowledgments

The author would like to thank Evelyn Lafferty, a plasma process engineer at Texas Instruments, for her assistance and for helping to clarify some of the work discussed in this article. He would also like to acknowledge David Albrecht of Entegris Sensing and Control for his technical knowledge of flow instrumentation.

Mark Simpson is an equipment engineer in the diffusion/wet group at Texas Instruments’ DMOS5 wafer fab in Dallas, where he has been focusing on optimizing wet-process equipment and prevention of failures. He has 17 years of experience working in the semiconductor industry. Before joining the company, he worked for several other semiconductor industry leaders, including Sony Semiconductor, Motorola, and STMicroelectronics. He received a bachelor’s degree in electrical engineering technology from Texas A&M University in College Station. (Simpson can be reached at 972/995-1590 or m-simpson3@ti.com.)