Continuous particle monitoring inside minienvironments improves wafer yield but requires new monitoring tactics
BY BILL BELEW, RAYMOND B. LUCERO, STEVEN D. KOCHEVAR & SCOTT JORGENSEN
Continuous monitoring of particles inside critical minienvironments improves wafer yield, potentially reducing the need for witness wafers. This monitoring application, however, requires a number of departures from traditional particle-counting tactics.
The jury is in. The data are clear and convincing. It is important to monitor particles inside critical 200-mm and 300-mm minienvironments.
The minienvironment is a key component in the manufacturing of 300-mm wafers. By isolating the tool, manufacturers can cost-effectively reach ISO Class 2-3 cleanliness levels in critical wafer-handling areas. At these levels—even with greatly reduced line and die sizes from increasing the number of die per wafer—high yields can be achieved, allowing a 30+ percent reduction in the manufacturing cost per chip.
With such significant improvements in cleanliness, some have asked whether particle monitoring can be eliminated within minienvironments. The following article summarizes a set of studies we recently conducted in 200-mm and 300-mm minienvironments. Based upon these results, we have found that particle monitoring in minienvironments remains an essential component of maximizing chip yields.
Particle incidence studies
In our tests, when the tool, load lock, front opening unified pod (FOUP), etc., all performed correctly, minienvironments kept the wafers extremely clean. This did not mean that there were no particles inside the enclosures; in normal operation, many tools generated high levels of particles. But if the components all performed correctly, then the air-handling system, positive pressure, etc., kept these particles away from the wafer. Die kills occurred when, for example, the fan filter units malfunctioned, allowing the particles to gather near the wafer instead of automatically being swept down and out of the minienvironment.
As a part of these studies, for both copper and non-copper processes, we monitored minienvironments protecting a wide range of tools, including, wafer/lot sorters, wafer probers, surface scanners, ion implanters, photolithography trackers, film thickness trackers, vertical diffusion furnaces and chemical vapor deposition tools.
Results showed that:
- Approximately 90 percent of the particles were generated by the tools, often as a normal part of operation.
- The remaining 10 percent generally were related to the load lock or FOUP. When all worked properly, the minienvironment system kept the particles away from the wafer.
- The number of aerosol particles sampled near the wafer correlated with the number of particles observed with surface scanners, which in turn correlated with the number of die kills. In other words, these particles killed die.
- Some types of tools were more problematic than others.
- Numerous examples were observed of properly installed and operating minienvironments that later malfunctioned. Causes included fan failures, gasket leaks, filter failures, misaligned robotic chucks, cross-contamination from other tools, vortices due to wafer movement, and bearing failures.
Particle distribution and re-circulation
Particles generated at a single point tended to travel like a smoke plume—the stronger the laminar flow, the more compressed the particle plume (see Figure 1). The typical laminar flow, combined with the minienvironment's lowered ceiling, often evacuated the particle plume before it could spread wide enough to be detected by a single-point particle counter.
The laminar flow sweeps particles out before they can spread widely.
Minienvironments depended on laminar flows designed to sweep particles away from the wafers; however, sometimes laminar flow alone proved inadequate. For example, as the robotics moved wafers across the laminar flow, the flow was disturbed, forming vortices below wafer level (see Figure 2).
In many cases (based on before versus after changes in particle counts and surface scan data), these vortices caused problems— collecting particles from below the wafer, carrying them upwards (despite the prevailing downward laminar flow), and re-depositing them on the wafer surfaces where they were held by electrostatic forces. Thus, particles residing below the wafer level were re-circulated, causing die kills.
The movement of the wafer through the laminar flow causes vortices, which can recirculate particles from below to above the wafer.
As a part of this study, microcontamination experts in operating 300-mm fabs discovered that a new set of tactics was required to monitor minienvironments—tactics different from those used in monitoring ballrooms or bay-and-chase layouts.
As seen in Figure 1, a single-point particle counter often missed detecting particle events inside a minienvironment. Particle Measuring Systems' Ensemble Manifold was incorporated into the process monitor. This device allowed a single particle sensor to sample simultaneously from multiple locations within the minienvironment (see Figure 3).
The Ensemble Manifold proved to be a cost-effective solution that provided a significantly increased coverage for detecting particle events. Multi-point sampling also allowed sampling in different planes or areas, which is quite useful in complex tools.
Detecting events instead of particles
With the movement to smaller line widths, quite tiny particles can kill die. In qualifying a minienvironment, you need to be able to detect 0.1 µm particles. For continuous monitoring, however, it proved more cost-effective in our studies to monitor at the 0.3 µm level.
Particle events did not generate uniform particles of one size, but a distribution of particles with varying sizes. Because minienvironments were so clean, the goal was no longer to eliminate all particles over a certain size, but to eliminate all particle events.
The Ensemble Manifold allows the MiniNet to simultaneously monitor up to seven points with a single particle sensor.
In these studies, particle events in minienvironments generated enough large particles to be detected with a 0.3-µm particle counter. For example, in wafer crashes (caused by misaligned robotic chucks), more than 90 percent of the particles that adhered to the wafer's top surface were ≥ 0.25 µm.
Since particles were swept out of the minienvironment so quickly, it sometimes was difficult to detect intermittent particle events. Continuous monitoring maximized the probability of detecting intermittent particle events.
Traditionally, many users have placed their particle-sampling probes at wafer level, but there are two primary reasons for placing the probes lower in minienvironments. First, since the plume spreads out, lower probes do not need to be directly under the particle source to detect the plume (see Figure 4). In our studies, as long as the particles were still numerous enough, lowering the probes increased the overall coverage area (against particle plumes from all possible locations).
Sample where particles are at medium density to maximize plume detection.
Based upon our experience, we recommend starting with the probes about six inches below wafer level.
Second, as seen in Figure 2, re-circulation due to vortices was capable of lifting particles from below, up and onto the wafer. A major finding from our studies is that you cannot assume that the product is safe just because the particles reside at lower levels than the wafer.
Once a particle event is observed, our studies showed that pressure data greatly accelerated locating the particle source. Thus, differential pressure (minienvironment vs. cleanroom) should be monitored for all minienvironments. If a particle event was observed, and the differential pressure had changed simultaneously, the problem generally was connected to the opening of ports, failure of fans, major leaks, etc. Conversely, if the pressure had not changed, the problem was more likely to be connected to the robotics or tool.
Moving semiconductor manufacturing to the 300-mm wafer has necessitated the use of minienvironments. While minienvironments make the critical areas extremely clean, continuous monitoring is required to keep many tools functioning properly. The significantly increased die density, greater susceptibility to small particles, and increased cost of slow response have all combined to make particle monitoring a necessary and cost-effective component of yield maximization.
In fact, particle monitoring in minienvironments has been so effective that studies are underway to evaluate whether this monitoring will allow the reduced use of test wafers, potentially offering major reductions in costs.
BILL BELEW is aerosols product manager for Particle Measuring Systems. He can be reached at: [email protected] RAYMOND P. LUCERO is responsible for all certification and sustaining of minienvironments at Intel's 300-mm facility in Rio Rancho, NM. He can be reached at: [email protected] STEVEN D. KOCHEVAR is Particle Measuring Systems' lead applications engineer for minienvironment monitoring. He can be reached at: [email protected] SCOTT L. JORGENSEN is a microcontamination engineer at Intel/Rio Rancho. He can be reached at: [email protected].