Issue



Taming thick-film nitride


03/01/1999







Taming thick-film nitride

Daniel Lefevre, IBM Essonnes, France

Mark Hope, Mike Czerniak, BOC Edwards, Crawley, UK

LPCVD nitride is an extremely challenging application for vacuum pumps due to the presence of solid particles and condensable vapors. Most pump failures occur because of either the sudden ingestion of solids or a steady accumulation of condensate in the pump mechanism, leading to frictional seizure. This is especially true when the thickness of the deposit (and hence the duration of the process run) increases by a factor of four, as was the case at IBM Essonnes.

The LPCVD stoichiometric reaction chemistry is described by:

3 SiH2Cl2 + 10 NH3 -> Si3N4 + 6 NH4Cl + 6 H2

where SiH2Cl2 = dichlorosilane (DCS); NH3 = ammonia; Si3N4 = silicon nitride; NH4Cl = ammonium chloride; and H2 = hydrogen.

The reaction chemistry can produce a range of by-products, and their concentrations depend on the detailed gas flows and reactor conditions. The effluent stream contains some silicon nitride particles and ammonium chloride vapor. Unreacted DCS and ammonia may also be present, as well as hydrogen chloride gas due to the nonstoichiometric conditions usually employed in the reactor. The pump therefore needs to withstand both solids ingestion and condensables handling.

The LPCVD nitride furnaces operate at 115 mtorr and 770?C, with a NH3:DCS ratio of 5:1. The original process (i.e., prior to 1996) produced a layer approximately 400 ? thick and the dry mechanical pumps had a mean time between service (MTBS) of >12 months.

During 1996, a newly introduced process increased the layer thickness to around 1800 ? and the pump MTBS fell to 2-3 months. Future deposited film thickness would be in excess of 2500 ?, so a reliable pumping solution was definitely necessary.

The first step was to implement best-known practices regarding vacuum installation:

 Installing a modified trap in the vacuum foreline just after the tool to condense some of the nitrides present in the gas;

 Fitting a deadleg above the pump to catch solid particles and to prevent them from entering the pump;

 Installing a gate valve between the deadleg and pump to allow the pump to run continuously, minimizing the risk of restart issues;

 Interfacing the tool and pump so that there was additional purging of the pump whenever the tool was not processing, increasing the removal rate of solids from the pump by evaporation (sublimation);

 Increasing the running temperatures of the pumps, limited to an extra 15-20?C due to the construction of the original pumps; and

 Removing residual NH4Cl from the pump exhaust stream by using a water-cooled trap, significantly reducing exhaust maintenance and the risk of blockages.

Click here to enlarge image

Figure 1. Variation of the pump motor current with time during a friction "event" in the original pump design. The shaded area is a measure of the energy associated with the spike.

Under these conditions, the pump MTBS remained around three months on all three thick-film tools tested, but random failures (short MTBS) were eliminated.

Current monitoring

To monitor the pumps` state of health, we started to monitor the pump motor current. An "event" has a typical signature. For a pump ingesting a solid particle from the vacuum foreline, there is a rapid rise in current as the particle is crushed, followed by smaller spikes as the debris works its way through the pumping mechanism. In practice, the majority of the spikes corresponded to friction caused by the accumulation of solid, condensed NH4Cl (Fig. 1). Indeed, we have been able to induce similar behavior in the laboratory by purposely simulating the process conditions.

This data could be used for predictive maintenance by analyzing the power in the frictional current spike (the shaded area in Fig. 1) as a function of time and comparing this to the characteristics of the electrical contactor on the pump. Previous experience showed that at <30% of the latter, it was safe to leave the pump in service; above this level, the pump required maintenance services. The decay in performance beyond this point was usually very rapid.

Since implementing this monitoring, there have been no pump seizures during process or wafers lost to equipment failure (inspection of pumps that had been removed confirmed they were heavily contaminated), but the MTBS remained around three months.

Pump improvements

To improve the dry pump performance and extend the MTBS, we investigated a new pump development that features a roots-claw mechanism with specific improvements to combat process by-products and with a lower cost of ownership. The new pump has a five-stage mechanism incorporating three different rotor profiles on the same shafts to provide optimum pumping in the 10-2 to 10-4 torr range. A cantilever shaft design eliminates high-vacuum bearings that are vulnerable to process by-products and require regular maintenance. The high-efficiency claw rotor stage design can handle the higher particle loads in many CVD tools. An additional, low-pressure, multilobe roots stage provides good vacuum performance and eliminates the need for an exhaust silencer that can act as a particle trap, thus removing any service needs for the exhaust system.

Click here to enlarge image

Figure 2. Pump and booster motor current characteristics over a 150-day period for the a) old and b) new pumps. The lack of stress in the new pump is evidenced by the absence of large current spikes (except restarts).

The pump is designed to operate up to twice as long as conventional dry pumps and demonstrate a 50% improvement in particle handling. In addition to the enhanced powder-handling capabilities, the pump runs at gas temperatures over 200?C to minimize the deposition of condensable solids such as NH4Cl from LPCVD nitride and AlCl3 from metal etch processes.

An integral purge system takes nitrogen from a single feed and distributes it throughout the pump. Diluting the process gas with nitrogen gas reduces the partial pressure to prevent condensation of products into solids and collecting on the pump mechanism, as well as maintaining toxic, corrosive, and pyrophoric gases at safe concentrations. In addition to using less nitrogen, the pump mechanism uses less power, and since it is not cooled, water consumption is also reduced.

Finally, by integrating the motor into the gearbox of the pump, the pump has a reduced footprint, a big added advantage when sub-floor is a premium in new fab designs.

Trial results

Monitoring the test systems showed that the two tools using the original pumps displayed pump motor current spikes after only one week, indicative of the presence of mechanical stress. These spikes increased in amplitude and frequency with time. After approximately three months, we serviced the pumps preventatively, based on the changes in the motor current.

The tool with the enhanced pump, however, exhibited completely different behavior (Fig. 2). There were no current spikes except for restarts when we stopped the pumps to change the water-cooled-trap disposable liners, three off-ingestion spikes, and a fab-side mains electricity spike. Detailed analysis of the ingestion spikes showed that the solids arrived in the booster approximately 1 sec before they reached the pump. At this stage, the solids broke up and the resultant powder caused only the slightest ripple in the dry pump motor current. A second burst of solids entered the booster about 1 min later and were similarly dispatched, without causing the booster or pump any significant stress.

During the 9.5 months of operation, the tool with the enhanced pump processed approximately 174,000 wafers, with 40 kg of ammonium chloride by-product passing through, yet the pump showed no signs of mechanical stress at all. We would have expected at least two maintenance services for the original dry pumps. As a result of the improvements, the pump lifetime increased by a factor of at least 3-9 x over the original pumps.

For further information, contact: Mark Hope, Section Leader, Semicon Product Support Group, ph 44/1273-444-338, or Mike Czerniak, International Applications Department, ph 44/1293-603-466, fax 44/1293-533-453, BOC Edwards, Crawley, UK.