CONTAMINATION CONT

Characterization of filters used in recirculated buffered oxide etch baths Joseph Zahka, Ven Anantharaman, Mark Carroll, Karim Vakhshoori

Millipore Cop., Bedford, Massachusetts /

i

The physical, performance,and cost characteristicsof f i b for recirculatingetch baths are discussed. However, to choose the proper filter for a productionoperation, knowledgeof the etching application and practical experience ate also needed. n ideal chemical filter should provide total retention and complete cleanliness; no particles should appear downstream, and there should be no extractables. The filter should behave like an open pipe (i.e., no pressure drop). Its initial cost should be low; it should not degrade or "plug in" the chemical and so should not need to be changed.In real world applications, some of these characteristics are mutually exclusive and tradeoffs must be made. Particle retention and flow rate, for instance, are in contlic~the more retentive the fiiter, the lower the flow rate. The desire for low cost conflicts with the high expense of filter materials required to achieve low extractables in aggressive chemical environments. To achieve downstream cleanliness, filters must be manufactured in clean environments and cleaned in postmanufacture pmcesses, which also add to cost. This article examines these issues and describes the testing of filtration devices for recirculated buffered oxide etch (BOE) applications. Here, flow rate and freedom from particles are very important and, due to the high surface tension and low gas solubility of many etch bath formulations,fdter wettability is also signifcant.

A

Figure 1.Recirculation etch bath system.

Treated polysulfone

Recirculatingetch baths BOE solutions are used extensively in silicon processing to remove oxide layers while leaving Si unaffected. A traditional BOE bath consists of 7:l 40% ammonium fluoride, 49% HF [l].More recent fondations use lower concentrationsof ammonium fluoride, which results in an unsaturated solution with lower surface tension [21. Bufferedoxide etching is often accomplished in recirculating etch baths (REB) (Fig. 1). Fluid in the tank flows over a weir into the suction of a pump (either a singlestagecentrifugal pump or a positive displacement diaphragm pump), through a particle-removing filter, back into the tank through eductors in the bottom of the tank, and then up past the wafers to complete the loop.

Filter characterization Depending on the application, some properties of filters are more critical than others. For BOE REBs, flow rate, extractables, and particle retention are very important, but filter wettability is also significant, as mentioned above. In the sections that follow, filter properties are described in the order of importance for the REB application. The laboratory tests used to quantify each property were performed on three filters designed specifically for etch bath applications (Table 1). June 1993 Solid State Technology 63

Y

% %

ed. Filter C has about one-half the pressure drop of Filter A.

w

%@

Flow rate Flow rate through a filter affects its particle removal rate. At high flows, there is faster bath tumover and increased particle removal. High flows also increase agitation and eliminate dead zones in the bath. Flow is controlled by the resistance of the piping, eductors, and filter [3J.Reducing filter resistance increases flow, espeto pipe resistance. cially if filter resistance is high In the current work, deionized (DI) water was circulated through the filter to measure flow rate. DI water was used rather than BOE solution. However, membrane filter samples were tested with BOE to ensure that the flow rate was proportional to viscosity with no additional flow loss due to other effects.The results of pressure drop testing are shown in Fig. 2. Pressure drop versus flow is shown for all the filters tested (housing resistance is included). Pressure drops at 10 gpm water flow are given in Table 2. Note that Filter C has about one-half the pressure drop of Filter A. Because all these filters have such low resistance, the effect on bath performance is not significant. When Filter A replaced Filter C in a 7:l BOE with a single-stage centrifugal pump, flow decreased only from 6.0 to 5.5 gpm.

W b l e S In BOE applications, a bath volume of about 20 liters was run in a totally recirculated mode. Extrdctables from the filter cartridge remain in the bath until the chemical is hanged (typically once a 64

Solid State Technology June 1993

Figure 3.Resistivity and TOC flushup of Filter A. Both parameters

,iter,minute were measured with through an Anatel the filter, TOC Analyzer Model A-100 with one

week). If extrdctablesare inithfly high or there is continuous e m c tion at a low rate, recirculation exacerbates the problem. DI water was chosen as one extraction solution because BOE solutions are also aqueous-based and the low interferenceof water with analytical techniques allows for low-level measurement. A second, more aggressive extraction sohtion, 10% HCI, was selected to identlfy the potential for release of metallics from the filter. Extractables were measured by soaking the filter cartridgesin a liter of DI water at room temperature for four hours. Concentrations of ionic species were measured with ion chmmatography.The filters were then soaked in 10??HCI for about 16 hours, and Cu, Fe, and Al concentrations were measured with atomic absorption. Table 3 shows results for static extraction. The three filter types released approximately the same extmctables. As-shipped chemical cleanliness appears adequate.

t

Table 4. Downstream cleanliness summary time to reach background at steady flow Time Filter

(min)

A

90

6

90

C

>240

Effect of pulsing

-

slight increase initially no effect after 4th pulse slight increase initially no effect after 4th pulse significant increase

all pulses (1.5 to 2.5 log) Notes: Downstream cleanliness measured with PMS M50 [>0.05-pmparticles). Background = 5 particles (>0.05pm ) per milliliter. Steady flow at 12 liter/min for 120 min. Pulsed flow at 15-minintervals for 120 min.

Figure 4. Latex bead retention of etch bath filters. Monolayer coverage is 0.1%. Filters A and B showed equivalent retention. Filter C had poorer retention, with the curve shifted by 0.1 pm. Published filter retention ratings of 0.1 pn and 0.2 pn do not correlate with these hard particle retention results.

The dynamic flusliup of Filter A is recorded in Fig. 3. The resistivity and TOC of 111 water downstream of a filter as it is initially flushed with clean DI water is measured with an Anatel TOC analyzer. Resistivity peifonnance is good, and the TOC recovery is excellent.

Retention Particle retention was tested by challenging filter rneinbranes with a monolayer of latex b e d s of fixed size [41. These lxads were stabilized with 0.1% Triton X-100. After the challenge, the filters were flushed with surfactant, and the bead concentration in solution was determined by nephelometry (a method for comparing tlie brightness of light passing through a solution or suspension with light passing through a stanciard solution). Beads passing the initial challenge and the flush are sunmed and used to calculate total passage and retention. The results of separate tests on a c h fdter for each size of bead are presented in Fig. 4.Filters A and B have better retention than Filter C. Retention can also be characterized by alcohol porosinietry [51, a test in which air flow through a dry membrane is compared to air flow through an alcohol-wetted membrane over a range of pres sures. The alcohol stays in the pores of the wet filter at low pres sures due to capillary forces but is forced from the pores at higher pressures. The results of such tests indicate the distribution o f pore size in the filter. The bubble points of Table 2 correspond well to the hard particle retention data of Fig. 4. A lower bubble point corresponds to greater porosity and, hence, lower retention. Wettability Membranes are difficult to wet spontaneously with 4Wi NH/,Fsince it is a nearly saturated solution with a surface tension of 95 dynes/cm [61; some membrane filters actually dewet in concentrated baths of NH4F. Dewetting, which causes a loss in flow and an increase in particle counts on wafers, can be e l h a t e d by using a BOE sohition with lower surface tension and higlier g~ssolubility,i.e., a sohtion with a lower concentration of amnioniiiin fluoride. Wettability was measured by placing a drop of 40% NH4F on the surface of the fdter and recording the time to wetting. Wettzability was also measured for NaCl solutions of various concentrations 66

Solid State Technology June I993

(surfwe tensions). The Same inembranes used for downstream cleanliness and flow tests were used for wettability tests. They had been water wet and dried at 60°C. A sumniary of filter wettability is given in Table 2. Filters A and C were spontaneously wettable with 40??ammonium fluoride. Filter 13 would not even wet with water. The flushing and drying may have had a deleterious effea on the wettability of Filter 13.

Downstream cleanliness Clean water was circulated through a new filter, and downstream particles were measured using PMS optical particle counter models M50 and HSLIS, which are capable of detecting particles greater than 0.05 pn and 0.10 pm, respectively. Initially, a steady flow ot 12 l i t e r / n ~was run for 120 inin. Then the flow was pulsed niomentarily (no flow for 2 s) at 15-min intervals for the next 120 min. Filter A reached background counts within 90 inin after staring tlie test. Pulsing had only a slight effect on patticle levels. Counts returned to the background level of tlie optical particle counter within the fourth pulse. The performance of Filter B was siniihr to Fdter A. Filter C, on the other hand, was very dirty. I’article counts never reached tlie Ixseline, and particle spikes were evident at each pulse throughout the test. A second Filter C was tested with the sanie results, but a third Filter c was slightly cleaner. “Particles”generated by Filter C were not easily removed by the 0.1-pii PTFE prefilter in the cleanliness system. The water in the tank had to be changed before the particle counts would return to the baseline. Table 4 suinnmrizes the data. Strength The REI3 application does not stress cartridge filters to their limits. The typical single-stage centrifugal pump produces a pressure differential of less than 20 psig across the fdter. Furthennore, aninioniuni fluoride and hydrofluoric acid are not aggressive toward most l strength pbstic nwcrials, and fdters must have only n ~ mphysical to have acceptable performance. One of each type of filter was subjected t o an integrity test that finds defects that may exist in the filter cartridge structure. Since filters were only exposed t o DI water at pressure differentialsof less than 20 psig, failure would indicate serious strength issues. Filters A and I3 were integral. Filter C failed integrity because o f a defect Where the lneInbl-dne Was Sealed to the e I l d Cap. Two Filter A cartridges were tested for graded failure [71.This test exposes filters to increasing pressure differentials in the fonmrct and reverse directions and t o pulsing conditions. The cartridge is continidly monitored t o determine the point o f failure. The car-

:;,::

Figure 5. Filter A aidwater diffusion curves. The data at 15, 20, and 25 psig is the average of fourteen cartridges. The remaining data is

for two cartridges. tridges survived pressure pulsing at 100 psig in the forward direction and a reverse pressure of 75 psig before failing at a 110 psig pressure differential in the forward direction. Such strength performance exceeds the needs of REBs. Membrane strength was also evaluated with a Mullens burst test [SI. In this case a rubber bladder expands under a piece of membrane that stretches until it ruptures. The bladder pressure at rupture indicates the strength of the membrane. PTFE membranes generally reach a yield point before they rupture; then yield pressure is recorded. As indicated in Table 2, the membrane in Filter C is very weak. This is one of the reasons why the cartridge has poor strength. Filter B is also relatively weak but probably strong enough for the application.

Integrity testable It is worthwhile to conduct integrity tests before installing filters in a REB El. An integrity test may indicate filter failure and the potential for particle passage. For REB applications, in which particles that pass through a small filter defect will be removed in the next tumover of the bath, integrity tests are not critical; here integrity testing need only detect a major failure of the filter. While particle monitoring of the test wafer surface can serve the same purpose, integrity tests have the advantage of specifically identfying the Filter as the source of particle problems. Integrity testing involves wetting the Filter with water and subjecting it to upstream air pressure. The pressure is increased, and the flow of air through the filter is measured. Excessive air flow, beyond difFusive flow through the membrane, indicates a hole in the filter device. All filters were tested in a water-wet condition. After flushing and drying, the Filter B membrane had a water bubble point of only 15 psig. This is lower than predicted by the mean IPA bubble point, probably due to poor wetting of the filter. Filter C was not integral after flow testing, but the defect could be detected with an integrity test.Figure 5 plots average values of diffusive air flow versus air pressure for water-wetted Filter A cartridges.

Price and cost The most obvious cost in the REB application is the purchase price of the filter, but factors such as filter lifetime affect cost. Filter replacement has a labor cost associated with it. In REB applications, filter lifetimes can vary from one week to more than one year. 68

Solid State Technology June 1993

To the IC manufacturer who is using filters, yield and throughput can be the real determinants of filter cost. A filter that has no extractables and quickly removes particles from the bath has cost benefits compared to a filter with poorer performance. A true measure of cost would account for all economic and functional aspects of the filter, but price is often substituted for cost because it is easy to measure. In noncritical applications, price may be near the top of the attribute list. In critical applications, where a slight improvement in the REB perfonnance has potential for improving yield, price may move to the bottom of the attribute list. Table 2 shows that Filters B and C have the same price, while Filter A is substantially more expensive. There is insufficient information to estimate total lifetime cost for the filters.

Conclusions Knowledge of the intended application is required to determine the important performance attributes for a filter (Table 5). For REB applications, Filter flow rate, extractables, and retention are among the most important characteristics. Of the three filter types developed for the REB application and discussed in the present work, no single type was best in all categories. Thus, the filter user’s experience is required to subjectively rank and weigh each performance characteristic. Acknowledgment This paper was f m presented in May 1992 at the IESAnnual Meeting in Nashville, TN. References 1. W. Kern, C. Deckert. ChemicalEtcbing, Thin Film Processes, Chap. 6, pp. 401-

496, Acad. Press Inc., 1978. 2. T. Ohmi et al., “Optimization of the Wet Process by Controlling Composition, Reaction Products, Crystal Deposition, and Wettability,” Nikkei Microdevice, Feb. 1990. 3. J. Zahka, D. Grant, C. Myhaver, ”Modeling of Particle Removal from a Circulating Etch Bath,” in Particles in Gases and Liquids 2, Detection, Characterization, andcontrol, K.L. Mittal, ed.,Plenum Press: New York, NY, p. 367 (1990). 4. MilliporeTest Method 0001143TM, “Membrane Characterizationby Polystyrene (PSL) Latex Bead Challenge.” 5. T.D. Brock, “Membrane Filtration: A User’s Guide and Reference Manual,” p. 46, Science Tech Inc. 6. R. Matthews, J. Zahka, “Optimization of Recirculated Etch Baths,” presented at SEMICON/Europa 91 Technical Conference, Zurich, Switzerland, March 5-7, 1991. 7. Millipore Test Method 0001720TM, “Fluorogard Plus Hydraulic Stress Test,” 6/91. 8. “Standard Test Method for Bursting of Paper,” ASTM Designation: D 774-67, 1986 Annual Book of ASTM Standards, Sect. 15, vol. 15.09, p. 151, ASTM. Philadelphia, PA. continued on page 71

.‘I

JOSEPH

ZAHKAreceived his B.S. degree from MIT and his M.S. degree from RPI in Chemical Engineering in 1970and 1971,respectively. He is a consulting engineer in the Microelectronics Applications Department of Millipore’s Process Group. He has worked for Millipore for thirteen years, applying products to the microelectronics,pharmaceutical, and medical industries.

MARKCARROLL received the B.S. degree in Chemical Engineering from the University of New Hampshire in 1982. He is now a senior development engineer in the Microelectronics Div. of Millipore Corp., responsible for new product development in the liquid fdtrationarea. Prior to joining Millipore in 1986, Carroll was producVprocess engineer for Sprague Electric Co., where he was responsible for the design and manufacture of solid state devices.

VENANANTHARAMAN received the M.S. degree in in Analytical Chemistryfrom Indian Institute of Technology, India, in 1972, and the Ph.D. in Chemistry (spectroscopy) from the University of Toledo, OH, in 1983. He joined Millipore’s Analytical ChemistryDept. in 1984,specializing in trace contaminant characterization and analysis of organics and inorganics in water and chemicals

KARIM VAKHSHOORI received the associates degree in Mechanical Engineering and the B.S. degree in Petrochemical Engineering from Louisiana Tech University in 1983. He is an applications engineer in the Microelectronics Applications Dept. at Millipore Corp., responsible for evaluating the performance of Millipore and competitive products used in the microelectronics industry.

Circle 42

June 1993 Wid State Technolcgy 71