Semiconductor processes continue to achieve smaller line widths to create products with greater capacity. As a result, processes that directly affect wafer thickness, like Chemical Mechanical Planarization (CMP), require greater efficiency and accuracy. Also, CMP processes continue to become increasingly expensive to sustain in terms of consumable materials, maintenance, uptime, and equipment. Therefore, any upgrades that canimprove process efficiency need to be fully investigated and utilized where feasible.
Hynix Semiconductor (www.hynix.com), a global supplier of memory chips for the semiconductor industry, sought to improve CMP process efficiency and uptime by reducing consumable and maintenance costs. Four objectives were proposed:
1. Improve point-of-use (POU) slurry flow accuracy.
2. Maintain equal or better process characteristics for particle reduction and wafer surface scratches.
3. Reduce overall slurry consumption.
4. Reduce preventive and unscheduled maintenance.
The existing CMP platform utilized peristaltic pumps to regulate the slurry flow from the pressurized global loop to the POU. The pumps deliver fluid by repeatedly squeezing a flexible tube in the same location with circular rollers. Each turn of the roller moves a small volume of fluid forward. However, a number of problems exist in using peristaltic pumps to regulate to desired CMP slurry flowrates.
The flow from the pumps is generally worse than 15 percent accuracy, affecting removal rate.
• The flow from the pumps is generally worse than 15 percent accuracy, affecting removal rate.
• Reducing flowrate lower than 180 ml/min is difficult to maintain within necessary tolerances.
• Particles generated from tubing wear are introduced into the downstream polishing process.
• Tube burst and leak events are not detected within the tool.
• Preventive and scheduled pump maintenance requires service expense and equipment downtime.
|Figure 1. Cross-section of NT differential-pressure flow controller.|
After evaluating several technologies, a differential pressure (DP)-based flow controller from Entegris (www.entegris.com) was selected for in-depth investigation on the process (Figure 1). The unit provides 1 percent accuracy, measures flow steadily (even in the presence of bubbles), has no moving parts in the flow measurement mechanism, maintains accuracy at different feed pressures, requires little maintenance or calibration, and can repeatedly achieve low flowrates.
DP-based flow measurement has been widely used for a number of years in semiconductor CMP applications. To determine liquid flowrate, the DP unit measures pressure before and after a fixed orifice in the flow stream. As the liquid passes through the orifice, the pressure loss is measured. The application flowrate and fluid type determine the orifice diameter built into the flowmeter. Per Bernoulli’s equation, as flow passes through an orifice it increases in velocity, causing a subsequent increase in kinetic energy. The increase in kinetic energy causes a corresponding loss of static energy, reflected by a pressure drop across the orifice. Flow increase causes a predictable DP increase. Flowrate is proportional to the square root of the pressure differential across the orifice, as shown in the equation:
Flow = k √(P1 – P2)
k = a constant based on the application properties
P1 = pressure before the orifice
P2 = pressure after the orifice
The DP-based flow controller integrates well into the CMP tool software and fits well within the spot previously occupied by the peristaltic pump. The installation within the CMP tool was not overly complicated and was performed in house. The tool supplies the same zero to 10 VDC setpoint signal to Entegris’s NT flow controller that previously went to the peristaltic pump motor driver. The unit compares the setpoint signal to the actual flow signal coming from the flow module. Based on the comparison, the control valve will either close or open as needed to maintain desired flow. The flow measurement signal is fed to Entegris’s MIRRA AIO board
that can generate alarms as necessary for nonstandard flow conditions (Figure 2).
|Figure 2. Theory of operation|
Objective #1: Improve point-of-use (POU) slurry flow accuracy. After installation, the NT flow controllers were calibrated and the information fed into the MIRRA slurry lookup tables. Blank wafers were processed through the tool to perform operational checks. The slurry setpoint was operated at 200 ml/min and during wafer processing flow maintained between 198–202 ml/min. The flowrate was then changed to 130 ml/min and performed a second series of tests as noted above. The slurry flowrate held stable at 128–132 ml/min.
Objective #2: Maintain equal or better process characteristics for particle reduction and wafer surface scratches. Having installed and calibrated one tool with NT model 6500 flow controllers, the next step in the investigation was to process production wafers to ensure that equal or better process characteristics are maintained.
Production wafers using NT flow controllers were processed through tool #105 at a setpoint of 130 ml/min. All the other tools in the investigation used the old peristaltic pump flowmetering method and were run at 170–220 ml/min (best obtainable flow control using pump method). The wafers processed with NT flow control units at 130 ml/min slurry flowrate processed as well as (or better) than the wafers processed with peristaltic pumps at 170–220 ml/min slurry for THK (post layer thickness), range, and CMP defects.
Figure 3 shows the THK within the desired bandwidth and range below the desired upper limit. Tool 105 performed as well as, or better, than the peristaltic pump-controlled tools.
|Figure 3. CMP post thickness and range|
Figure 4 shows the number of defects from tool 105 compared to the peristaltic pump-controlled tools. Tool 105 outperformed the other tools on average. The increase in performance is assumed to be due to both slurry flowrate and fewer contaminating particles from the peristaltic pump tubing.
|Figure 4. CMP defects level|
Objective #3: Reduce overall slurry consumption. The in-depth evaluation compared tools run by peristaltic pumps to tools run by NT flow controllers. This included tests with blank wafers and tools run with actual long-term production wafers. The results demonstrate that with the NT flow controllers running at 35 percent less slurry, the wafers’ quality is the same, or better, than wafers processed with peristaltic pumps at higher slurry flow ranges.
Based on slurry savings alone, the return on investment for the NT flow controllers is less than four months. This includes costs for installation and the NT controllers.
Objective #4: Reduce preventive and unscheduled maintenance. Peristaltic pumps require calibration 2–3 times per month and tube change every 2–3 months. This process can take as long as three hours. The new system allows Hynix to check the controllers only once ever 12 months in just 15 minutes time, saving considerable expense in both service hours and loss of production uptime.
Figure 5 is an in-fab photo of the actual upgrade installation on the CMP tool. The upper pump drawer contains three original peristaltic pumps, and the lower pump drawer holds two flow controllers. The NT flow controllers are a little smaller than the peristaltic pumps and fit easily into the pump drawer, replacing the original pumps.
|Figure 5. In-fab photo of the actual upgrade installation on the CMP tool.|
As a result of this investigation, Hynix chose to upgrade all of its CMP tools with NT flow controllers, resulting in short-turn ROI with considerable savings.
Darren Richards is the CMP and clean equipment engineering manager for Hynix Semiconductor. He has worked in the semiconductor industry for the past 12 years. He holds an NCEA in Electronics from Waterford Institute of Technology (Waterford, Rep. of Ireland) and a bachelor”s degree in Information Technology from the University of Phoenix.
The author would like to thank David Albrecht for his technical knowledge of flow instrumentation and assistance in developing this article. For more information, contact Entegris at 952 556-3131 or firstname.lastname@example.org.
1. Michael McCoy, “Many Roads for CMP,” Chemical and Engineering News, Volume 84, Number 26, (June 26, 2006): 17-19, pubs.acs.org/cen/coverstory/84/8426cover2.html.
2. Hu Taft, “Improving Yield Rates on Existing CMP Tools,” Cleanrooms, (April 2004).