Advances in Compound Semiconductor Processing
2
Joseph Sweeney, James Dietz, Karl Olander, Paul Marganski, Josep Arnó
ATMI, Inc.; Danbury, Connecticut, USA
2
2
Abstract—Compound semiconductor manufacturers employ a variety of techniques for managing toxic gases and solids in the effluent of MOCVD reactors. For gas abatement, chemical wet scrubbers, chemical dry scrubbers, and burners have been employed with varying degrees of success. Wet scrubbers and burners have become popular solutions due to their lower operating costs. Unfortunately, these systems can require frequent maintenance and generate large volumes of secondary waste, e.g., arsenic laden waste water in the case of chemical wet scrubbers. Dry scrubbing is the simplest and perhaps least maintenance intensive solution for compound semiconductor abatement of arsine and phosphine; however, efficiency challenges at high flow rates, thermal issues due to hydrogen reduction reactions, and material costs have limited their use.
This paper discusses new options for simplifying and improving the abatement of gases from MOCVD reactors depositing GaAs and InP films. The performance of several new dry scrubber materials is presented and proven to reduce materials costs and total operating costs when compared to existing treatment solutions.
Keywords-Compound semiconductor; abatement; effluent treatment; chemisorption; gas adsorption, cost of ownership, arsine, phosphine
I. Introduction
Various effluent treatment challenges arise in compound semiconductor applications. First, the process gases used typically have multiple hazards associated with them. For example, two of the most widely used deposition gases, the hydrides arsine and phosphine, are extremely toxic as indicated by their respective threshold limit values (TLV’s) and immediately dangerous to life or health values (IDLH’s); see Table 1. In addition, phosphine is pyrophoric, meaning that it will ignite spontaneously in air, while arsine is flammable. Occasionally, it is necessary to use hydrogen chloride gas (toxic and corrosive) to clean the tool chamber of unwanted deposits. Such hazards require effective and efficient abatement methods.
Traditionally, there have been three common abatement methods used in the treatment of effluent gases from compound semiconductor processes. These are gas absorption (water scrubbing), thermal oxidation, and gas adsorption (dry scrubbing).
Gas absorption entails contacting the gas stream with an aqueous stream typically over a packed bed. Although such a system is ideal for HCl abatement, in order to effect the removal of arsine and phosphine gases, a strong oxidizing agent (e.g. sodium hypochlorite) must be added to the aqueous phase since neither gas is soluble in or reactive with water alone. The benefit of gas absorption is its low operating cost. Although water usage is being scrutinized more and more, its cost is still relatively low. In addition, the cost of an oxidizing agent like sodium hypochlorite is not high. A disadvantage of gas absorption is its dependence on moving parts, e.g. pumps, in order to maintain an effective gas removal rate. The most notable disadvantage, however, is the large volume of arsenic laden aqueous waste that is produced. Not only must this waste be specially treated, it represents an elevated risk within the fab. For example, if a gas absorption system were to suffer a large leak, the cleanup costs could be significant due to arsenic contamination.
TABLE I. Exposure Guidelines for Typical Compound Semiconductor Gases
Gas / Concentration Limits /TLV-TWA (ppm) [1] / IDLH (ppm) [2] /
Arsine (AsH3) / 0.05 / 3
Phosphine (PH3) / 0.3 / 50
Hydrogen Chloride
(HCl) / 5 / 50
Thermal oxidation coverts toxic and hazardous effluent gases to their respective oxides through a reaction with oxygen at elevated temperatures. For example, arsine is converted to arsenic(III) oxide (As2O3), while phosphine is converted to phosphorus pentoxide (P2O5). These oxides are still hazardous and must be collected prior to venting to the atmosphere. A typical collection technique is to filter the solids from the gas stream (e.g. with a bag house filter system). As with gas absorption, the main advantage of such a system is its relatively low operating costs, which include fuel or electricity to operate the oxidizer as well as the need to potentially replace filter elements. The major disadvantage is the need to perform frequent maintenance. The significant quantity of solids produced in the oxidation process leads to arsenic contamination within the exhaust ducting and may also require frequent replacement of the filter elements. In addition, the oxidizer unit itself may require frequent maintenance due to solids accumulation. Another effect of using a thermal oxidizer (which may be viewed as a positive or negative depending on the user) is that in addition to treating the hydrides, it will also react the hydrogen present in the gas stream. Because the hydrogen flow rate is commonly very high (50 – 120 slpm is typical), the oxidizer may require active cooling in the form of large amounts of dilution air or cooling water, both of which add to the cost of ownership. A final disadvantage of a thermal oxidizer is that it will not abate HCl. If left untreated, the HCl has the potential to cause corrosion within the house exhaust system.
In order to avoid using a filter unit to remove the solid byproducts emitted from a thermal oxidizer and in order to treat HCl, an integrated scrubber system containing a thermal oxidizer followed by a wet gas absorption column can be employed. Such a solution can significantly reduce the footprint compared to a thermal oxidizer used in conjunction with a filter system. Although an oxidizing agent is not required, the problem and risk associated with aqueous arsenic containing waste is the same as described for the stand-alone gas absorption system. In addition, care must be taken in the design such that even very small particles of arsenic oxide are captured in the absorption column. For example, a typical gas absorption column will only remove particles down to 2-10 microns, depending on specific process conditions [3]. Additional separations techniques may be necessary to remove smaller particulate. Such technologies may increase the capital cost and complexity of the equipment. Finally, as with the thermal oxidizer solution, the integrated scrubber will require significant cooling to manage the heat liberated from the abatement of hydrogen gas.
The final major technology used is gas adsorption – specifically chemisorption. In chemisorption (often referred to as dry scrubbing), the effluent gas stream is contacted with a fixed bed containing solid particles having very high surface areas. The chemistry of the solid material is such that a reaction occurs when the effluent gas molecules adsorb onto the surface. Byproducts of the reaction are non-volatile solids and cannot be desorbed from the surface. A major benefit of such a system is its simplicity. The system does not require moving parts, fuel, or water, and consumes minimal electricity. The system is passive in operation, meaning that the only requirement for abatement is the presence of the chemisorbent material itself. This is in stark contrast to other methods of treatment where loss of water, air, fuel, or electricity can mean a partial or total loss of abatement efficiency. In addition, a dry scrubbing system contains the arsenic, phosphorus, and hydrogen chloride waste in an isolated, well contained, and compact vessel. The biggest disadvantage of gas adsorption systems is their cost of operation. This arises from the need to replace the vessels containing the chemisorption media (from hereafter to be termed resin) on a basis prescribed by the capacity of the given resin material in conjunction with the total gas loading onto the resin. In addition, the relatively large effluent flow rate common to most compound semiconductor processes tends to reduce a resin’s capacity due to the shorter residence time the effluent stays within the media bed. Added to this challenge is another in which a secondary reaction can occur between hydrogen and common metal oxide resins if the bed temperature rises above 110-120°C. This hydrogen reduction reaction of the resin is undesirable simply due to the large flow of hydrogen. Not only can the resin capacity be used up, the temperature of the bed can increase significantly.
It is in this light that the authors have embarked on a major effort to improve the cost of ownership of ATMI’s dry scrubbing solution, as well as to further enhance our resins’ resistance to the secondary hydrogen reduction reaction.
II. Resin Development Program
In order to reduce the cost of ownership of dry scrubbing, the focus was placed on significantly improving the capacity (lifetime) of the chemisorbent resin media. This challenge was tackled in three steps. First a program was developed to screen resins of various chemistries. Second, the top performing materials from the screening experiments were characterized more thoroughly to determine the effect of process parameters on the resins’ capacity and operating temperature. Third, resin optimization techniques were evaluated.
Most of the resin experiments were performed using a test cell (Figure 1) with the following characteristics:
· 1” or 2” outer diameter (0.065” wall thickness)
· 8” resin bed height
· 5-point thermocouple centered within cell
· ~ 1” thick insulation surrounding cell
· Stainless steel construction
Figure 1. Test Cell used to Determine the Effectiveness of Various Resins
Tests were performed flowing either arsine or phosphine gas in a ballast stream composed of 80% hydrogen and 20% nitrogen by volume (this ballast stream was meant to approximate the actual conditions in the field). All gases were delivered to the test cell using mass flow controllers. The effluent of the test cell flowed through a MIDAC G2000 FTIR (Fourier Transform Infrared Spectrometer) with a 10 cm pathlength in order to determine the shape of the arsine or phosphine breakthrough curve (i.e., the effluent concentration of hydride gas as a function of time). In addition, a portable gas detector sampled the exhaust stream to monitor for TLV breakthrough, while a Fluke Hydra Databucket logged cell temperatures during the experiments. Figure 2 is a simple block diagram of the setup.
Figure 2. Major Components of the Experiment
The performance of any given resin was evaluated by three criteria:
· TLV Capacity
· Dynamic Theoretical Capacity
· Maximum Temperature
TLV capacity was defined as the amount of gas taken up per unit volume of resin at the time when the effluent concentration reached TLV levels as determined by the portable gas monitor. (Note that many commercially available gas adsorption systems are outfitted with a sensor that detects TLV breakthrough from the fixed resin bed. At this point, the resin bed is replaced with a new vessel containing fresh resin. Therefore TLV capacity is directly related to cost of ownership). Dynamic theoretical capacity was determined by recording the full breakthrough curve using the MIDAC FTIR. The term “full breakthrough curve” means the test was run until the outlet concentration approached 50%+ of inlet concentration. The area above this curve was used to determine the total amount of hydride gas that had been chemisorbed by the resin (note that the curve slope was extrapolated for calculation puposes). Finally, we monitored and logged the temperature of the cell interior at various depths in the resin bed. For each test, we recorded the maximum value attained.
A. Resin Screening Study
Resin screening consisted of testing various materials to quickly determine those that showed promise. Generally, resins are composed of various metal oxides that react with arsine, phosphine, and HCl gases in the following fashion:
AsH3 + MOx à MyAsz + As + H2O unbalanced (1)
PH3 + MOx à MyPz + P + H2O unbalanced (2)
HCl + MOx à MyClz + H2O unbalanced (3)
where “M” stands for a generic metal. Although chemistry of the resin is very important in terms of dictating the resin’s capacity, so is surface area, pore structure, density, water content, etc. In fact, a resin screening study could be almost limitless due to the large number of resin property variations one could imagine. For practical purposes, we chose to focus on only a handful of carefully selected candidates.
Because HCl flow rates are usually much smaller than the hydride challenges in compound semiconductor processing, the focus of this project was on the hydrides, notably arsine. In addition, resins that abate HCl are typically lower in cost and tend to have higher capacities than their hydride resin counterparts. Therefore, hydride capacity improvements have a much larger impact on cost of ownership.
TABLE II. Resin Screening Experiments
Resin / Resin Capacity (moles gas per liter resin) /TLV / Theoretical /
A / 1.16 / 3.10
B / 0 / NA
C / 2.47 / 4.87
D / 1.00 / 3.89
E / 1.02 / 1.44
F / 0 / NA
G / 0.16 / 1.53
H / 0.82 / 1.13
I / 0.55 / 0.82
Table II lists the results of the screening tests. Note that resin “A” represents the material ATMI sells today for arsine and phosphine abatement. All tests were performed using arsine as the hydride gas; the inlet concentration was 4%, while the superficial linear velocity through the cell was 4 cm/sec. These test conditions are quite challenging and were chosen mainly to decrease the testing time during the screening program. Actual field conditions are typically 0.5% - 2% inlet gas concentration with a linear velocity of less than 1 cm/sec (although this is dependent on the specific design of the resin containing vessel).
Three resins were chosen for further study based on the results of the screening program. These materials are shaded in gray in Table II: resins A, C, and D. It is noteworthy that a significant portion of the resins’ chemistry remained available for further reaction at TLV breakthrough. This is evidenced by the theoretical capacities being 2-4 times higher than TLV capacities for the three top performers. As an example, Figure 3 plots the exhaust concentration (breakthrough curve) from the test cell as a function of time for resin “A”. Whereas TLV breakthrough occurred at t = 60 minutes, the outlet had not reached 50% of the inlet concentration by t = 140 minutes.