How Does a Mass Flow Controller Cause Semiconductor Process Failure?

The mass flow controller (MFC) is a critical component in semiconductor manufacturing. Control variability or manufacturing defects in an MFC can cause reduced wafer yields, or even a loss of wafer lot.  http://bit.ly/1Rr9NKZ

To provide some context, I need to state that most MFCs in semiconductor applications are based on thermally-derived mass flow measurement. The flow sensor is usually a capillary tube with a resistance wire heater wound around the outside of the tube with temperature sensor wires would around the tube, upstream and downstream of the heater. The amount of heat transferred to the downstream sensor is proportional the mass flow and inversely proportional to the heat capacity of the flowing fluid (gas or liquid).

Pressure Variations

A sudden change in upstream or downstream pressure can cause a flow rate transient in an MFC. Most MFC manufacturers try to tune the MFC valve response to be as fast as possible in order to react to the pressure surge without over-reacting. Another approach by gas panel manufacturers is to put a pressure regulator upstream. This can slow down and mitigate the pressure spike from upstream, but does not help for downstream pressure changes. A secondary effect is that with a change in pressure, the density of the gas will change affecting the heat capacity. This will cause a slight shift in flow calibration. Also, if the MFC has a bypass, which is required for flow rates above about 50sccm, the flow split between the sensor and bypass may change causing a measurement error. Several manufacturers try to match the pressure-drop between the sensor and bypass for a range of pressures, but it is difficult to get them to match due to differences in geometries.

Calibration Shift

As electronic components age, their properties will change causing a shift in the flow measurement and valve response. Zero-shift is one of the more common effects. Some manufactures have implemented an auto-zero function that resets the zero when the MFC valve is closed. An unintended consequence occurs if the valve leaks and the sensor “zero” is set on a finite flow rate. Another unintended consequence occurs when the MFC is mounted with the flow output facing either up or down. The valve is closed, but a flow rate is registered because the gas that is heated in the capillary becomes buoyant, flowing upward. This pushes the cooler gas from the bypass to into the capillary tube causing circulatory flow. Auto-zero again zeros on a finite flow causing calibration shift. MFC manufacturers have battled this for years and have given rise to a variety of sensor tube designs.

Another cause of shift can also come from the capillary sensor tube becoming coated by reacting vapors flowing through the MFC. The extreme condition is the tube becoming fully clogged. I consulted with Applied Materials and Intel some time back where an MFC on a process tool kept clogging. It was downstream of a TEOS bubbler. I determined that a small air leak in the bubbler was causing the TEOS to polymerize in the MFC, clogging the sensor tube.

Mitigation approaches include installing a flow reference standard (another MFC with better calibration) to spot-check the MFCs periodically, installing in-line flow verifiers to check MFC calibration, or integrating a reference volume in the MFC to check the flow by pressure decay. Obviously, this adds space, complexity and cost to a system.

Particulate Contamination

The premier MFC manufacturers test in manufacturing for particles 100% of the time per SEMI standard. The MFCs are double-bagged in a nitrogen environment within a cleanroom. Under most circumstances this is sufficient. However, particles that may be entrained during manufacturing can become dislodged during shipment or installation. Particles may also be introduced during installation if strict cleanroom protocols are not observed. Particles may form by entrained air reacting to air-reactive process gases like silane. Generally, particle emission from all sources is monitored during process tool commissioning.

Molecular Contamination

This is a more difficult to measure in manufacturing. Very expensive equipment, such as atmospheric pressure ionization mass spectrometers (API-MS), are required to measure contamination to the sensitivity needed by chip manufacturers. The approach is to clean the MFC parts often as they move through the manufacturing process. After assembly, the MFC is cycle-purged with very pure nitrogen, ends capped and immediately bagged. The cause of molecular contamination usually comes from the design of the MFC itself. If the flow path is not fully swept, that is, if there are “dead-spaces”, then air can be entrained. Cycle-purging is more effective than using a straight purge, but it is difficult to tell when you are done without the previously mentioned “very expensive equipment”.

The second source of molecular contamination is cross-contamination. If there are dead-spaces in the MFC, process gas can be entrained. If a second process gas is used as part of the process cycle, the first gas will contaminate the second. Simple nitrogen purging between will not eliminate this.

A third source of molecular contamination comes from under-specifying the MFC. A temptation is to get the least expensive product for a system. For example, someone specifies an MFC that has elastomeric seals for a process gas that attacks the elastomer. The reaction products will contaminate the process. In some cases, the process gas will react exothermally resulting in a thermal run-away, causing the MFC to respond erratically. This is in addition to delivering unwanted reactants to the wafer.

Because of these and other issues with current MFCs, I have been working on a successor that will reduce process variability and be easy to install and use. Contact me if you have questions.