ULSI Research Laboratory

Hewlett Packard Company

rparkerREMOVE_FOR_SPAMERS8@pacbell.net

Electrodeionization applied to a recycle system has shown the remarkable gains projected at the outset of this project. Cost of Ownership (CoO) has decreased significantly at our site because of it. There is still some characterization left to do, however, because recycle systems are subjected to organic species from process rinse waters. These species are not normally found in industrial feed water, which has fairly well known properties in RO and TOC-destruction equipment. Further, these species are process dependent, and were generally not part of the initial EDI study for recycle application. Here, then, are some semi-quantitative results from experiments to assess how our recycle system, with EDI equipment as the primary means of water treatment for rinses, is able to cope with intentional TOC upsets.

The water system used for these tests was the same as in the previous testing. The figure is reproduced below. Samples were injected at point A, a filter housing (not shown in the diagram), when the fab was still in operation. After a partial shut down of the fab, samples were introduced at point B, a QDR tank in a fab wetbench.

For testing of a small set of organics, the carbon bed was bypassed to prevent 'treatment' by the carbon (activated carbon has interesting properties on weak organic solutions). TOC sample points were inserted just upstream of the EDI, and just upstream of injection point A. TOC was measured with a Sievers Instruments model 800. The input of the TOCA (analyzer) alternated between sample streams to allow observing signals from before the EDI and after it. There may be some noise from varying baselines due to process rinse water changes, but the majority of time the baseline is very quiet (<.5 ppb noise). Testing in this way, however, was problematic. Switching the sample lines at the input of the TOCA produced transients in both the organic and inorganic carbon components of the stream. These confused the analyzer (and me) to the point that it was at times unclear whether the test specimen had made it through the EDI and produced a signal.>

So, testing of the remainder of organics involved leaving the carbon in place, and adding various organics at injection point B. The TOCA at the 450-gallon tank was disabled: no matter how large the TOC signal was at that point, the water was allowed to go into the 3300 gallon tank and ultimately into the main water train at the RO storage tank. Organics would later show up on the polish loop TOC analyzer if they were of sufficient concentration, and escaped total destruction by 185 nm UV lights. By the way, TOC destruction in the primary and polish loops downstream of the RO storage tank is typical of that used in the semiconductor industry. It's not total; a fraction of incoming TOC will always be present in the output.

1. The plot below shows typical responses to organic injections. Five mls. of isopropyl alcohol (IPA) were introduced into the system at point B, a rinse tank. Here's how to read the chart.

A. Ignore the straight green line, the wavy or straight brown/tan line, and the gray line.

B. The cycling green line is the TOC analysis from the 450-gallon tank; the scale is 0-500 ppb. The wave action in it reflects the fill/transfer cycle of the tank (it has a strong impact on the TOC--never mind why, or email me). The offset from zero is typical of the drift of this instrument. Though the offset is at about 150 ppb, the water is really about 5 ppb, to which is added the cycling amount. Get it? Oh well...

The injection occurred at about 10:30. The plot shows a space where the value shot off scale, then a steep vertical where it returned.

C. The red line is the TOC in the polish loop; full scale for this parameter is 50 ppb. Each step is one ppb. and typically rattles between 5 and 6. The event on the chart, the mound (time goes: left is yesterday, right is today), is the result of the IPA addition. The peak in the red line (effect) occurs about six hours after the injection (cause).

2. Five mls. of glacial acetic acid were added to the rinse bath. The result is below.

Actually, this is not a good graph. The red line peak would be to the right of the chart's right end! I've run this one several times, and the peak never shows up. The EDI removes acetic acid. I'll run this again though.

A few days later... Here's an experiment designed to make sure I had enough signal to watch. Twenty mls. of glacial acetic acid were dumped into a rinse tank. Ten minutes later, 200 mls of conc. ammonium hydroxide were added. The idea was to try to make the rinse water basic and look for a change in EDI product resistivity. That part didn’t work well, but the acetic acid (or acetate) was fully rejected by the EDI. The chart below shows TOC at the first (450-gallon) collection tank, so it's quite high--off scale. The green trace is the polish loop TOC (bouncing between 4 and 5 ppb), which showed no effect at all.

3. 200 mls. of 2% tetramethyl ammonium hydroxide were injected into a rinse (equivalent to 4 mls. of 'neat' TMAH). TMAH is used in resist developing, which is not likely to be a process included in water recycling. But, it's being used in some cleaners and other process chemicals, and may be found in several workcells.

The chart shows three peaks. The far-left 'peak' is from another chemical. The two TMAH injections 'run together' (in the green wavy line, above the 'v' in 'prev') and appear as two descending lines. This is an anomaly of the data taking/plotting system. The first peak (the left of the pair) was from TMAH right out of the bottle. The second peak is from a 200 ml. sample that was used to strip resist from fifteen 150-mm wafers. The liquid was quite colored when it was poured into the rinse, indicative of substantial resist content. That extra organic might make recycling a bit tougher, in theory. It's likely, though, that the ionizable developer interacts with a similar ionizable, and thus removable-by-ion-exchange, resist moiety. Typically, wafers are developed and rinsed one-at-a-time, but I shortened the experiment a little here by treating them all at once.

Note there's no hump on the red line. EDI removes it completely, along with the resist, it seems.

4. Ten mls. of NMP (N-Methyl Pyrollidone) were added to the rinse. As it's a base (even though weak), it was completely rejected by the EDI.

5. Ten mls of an unspecified anti-foaming agent were added to the rinse. There's evidence that it's not rejected, as a TOCA just downstream of the EDI indicated a slight response. However, I don't know what else might be in that mixture (no MSDS!); it was pretty tenaciously stuck to the walls of my graduated cylinder, and so may be stuck to piping in the recycle system. Typically, anti-foamers are used in very small quantities, so my experiment was quite overkill. I'm staying away from this glop from now on.

6. Ten mls of an ethylene glycol mixture with water (50%) were injected into a wet bench rinse tank. The organic was expected to survive through the recycle system to the polish loop. The chart below shows the collection tank TOC (wavy green line that abruptly drops). The large step in TOC was due to my adjusting the zero shortly after the EG injection. About two hours later the EG started appearing in the polish loop (red line). Several hours later the level reached about 10 ppb. For the rest of the day and into today, the level lowered back to normal (4-5).

 7. Twenty mls of a Waco (601?) surfactant, composed of 30 % surfactant component and 70 % of who-knows, were added to a rinse tank. The chart below shows the collection tank TOC in green. Again, I fiddled with the TOCA baseline just before the injection, so the zero shifts from over 200 ppb to the tens range. The peak is the Waco being detected. Not shown are the results from a TOCA just after the EDI: no detectable TOC was detected during the run. It appears that the EDI can remove the material. Why?? It's a non-ionic type, so I'm told. We'll watch closely for a few days & see if it got stuck on the carbon bed. It should elute, eventually....right?

Waco ^^

One other interesting highlight. The red line, polish loop TOC, had been resting at about 9-10 ppb. This is because we'd sent some of the water, which would have been recycled through the EDI, to AWN. This was a short-term move to lower the gpm through the EDI back to its rated capacity of 50 from 60. Doing this resulted, of course, in recycling less water. The RO sent more city water to the RO Storage tank, raising the system TOC. For the Waco experiment, the collected waters were not partially diverted to AWN, and 100 % again went to the EDI. That brought the TOC back down to the 4-5 range.

More Interesting Observations 

We've been closing the fab for a while, and the EDI has had very little work to doWater quality feeding it was over a meg at times, and the EDI output was over 18.Then came some excitement. As the bulk chem distribution system was removed, the lines were purged of excess chemical by forcing it out open connections in the wet benches. Rather than lead the chemistry to the process sink and to AWN, the chemistry went to the water collection plenums, and thence to the recycle system.

Sulfuric Acid Overload:

An estimated 10 gallons of concentrated sulfuric acid hit the recycle system over a span of about 1/2 hour. The feed conductivity to the EDI went to the low hundreds of ohms for at least an hour. This was a serious overload for the unit. The output conductivity, expressed in megohms on the EDI's front panel as 'XX.XX', showed '00.00' for a comparable amount of time. When the event was diluted enough, all the values came back up to normal. It took a day for the system to fully rinse out, but water was recyclable as soon as it's resistivity rose to our minimum spec, 1 meg.

Was that it? No. Downstream of the EDI is a diverter that sends EDI product water to Acid Waste if it has a resistance of less than a meg. Well, when was the last time we tested this? Who knows.. It failed. All the water went to the RO product tank, then to the primaries. There was evidence that the sulfuric exhausted the cation resin.

Hydrogen Peroxide Overload:

Yet another estimated 10 gallons of 30% Hydrogen Peroxide--same thing. This time, there was no conductivity indication that it had occurred. Instead, the carbon bed upstream of the EDI bubbled profusely. Oxygen collected at the top of the tank, and was pumped to the EDI. After some time, the EDI pumps cavitated, system pressure went low, and the EDI stopped. All we lost was a bunch of water; no resin problems. After several more engineered leaks (that is, poor engineering judgment led to repeated attacks on the recycle system), everything went back to normal. Remember that during these excursions the fab was in shut-down, and it was OK to be cavalier about system upsets.

There may not be more. The EDI is maybe shipping to another Agilent Technologies Site.