My first job out of grad school was with a biotech company in Cambridge, Massachusetts. It was a small outfit, and everyone had a “lab job” in addition to whatever science they were hired to do — a task to maintain the common areas of the lab. My job was to maintain the water purification systems that made sure everyone had an ample supply of pure, deionized water to work with. The job consisted of mainly changing the filter and ion-exchange cartridges of the final polishing units, which cleaned up the tap water enough for science.
When I changed the filter packs, I was always amazed and revolted by the layers of slime and sediment in them. A glimpse out the window at the banks of the river Charles — love that dirty water — was enough to explain what I was seeing, and it was a lesson in just how much other stuff is mixed in with the water you drink and cook with and bathe in.
While we humans can generally do pretty well with water that rates as only reasonably pure, our industrial processes are quite another thing. Everything from power plants to pharmaceutical manufacturing facilities needs water of much, much higher purity, but nothing requires purer water than the specialized, nanometer-scale operations of a semiconductor fab. But how does ordinary tap water get transformed into a chemical of such purity that contaminants are measured in parts per trillion? And how do fabs produce enough of this ultrapure water to meet their needs? With some big chemistry.
Just the Water, Please
Although standards vary by industry, in general the level of purity reached by ultrapure water (UPW) is almost beyond belief, and suffers by comparison to something like drinking water. Even the purest drinking water is really a complex mixture of minerals and gasses dissolved in water, with a fair number of particulates suspended in it as well. As an example of how different UPW is from drinking water, the US Environmental Protection Agency sets the limit of chromium in drinking water to a mere 0.1 parts per million. But for semiconductor-grade UPW, the limit is 2 parts per trillion — 50,000 times less!
When you think about the scales involved in semiconductor manufacturing, the stringent UPW standards make perfect sense. The size of the features being etched onto silicon wafers varies by process node, but current processes can easily be killed by a particle only a few nanometers in diameter. For scale, a coronavirus particle is on the order of 100 nm. Control of particulates in UPW can be vexing, mainly because particles can come from just about anywhere in the piping, tanks, pumps, and vats of chemicals used in the purification process.
Particles aren’t the only contaminants that have to be dealt with. While the smooth, clean surfaces of a fab plant’s UPW equipment might seem like a poor place for life to flourish, bacteria have a proven ability to colonize even the most unlikely ecological niches. Biofilms can present a huge problem to UPW systems, and they can form anywhere that water is allowed to pool. Biofilms can contribute to both particulate contamination as well as total oxidizable carbon (TOC, aka total organic carbon), which is essentially the remains of dead bacteria.
Aside from particulates and TOC, the other main contributors to UPW contamination tend to be water-soluble substances, like minerals and gasses. Sodium is a big concern, mainly because it tends to be a leading indicator of trouble in ion exchange resins used to process UPW — more on that below. Silicates are a concern, too, as are dissolved gasses — oxygen is highly reactive and can easily oxidize the metal layers needed to build a chip, and carbon dioxide easily dissociates in water to form carbonic acid, which increases the conductivity of water and is detrimental to wafer processes.
Putting on the Polish
In a nutshell, water that’s going to be used to build chips needs to be as close to “just water” as possible. Getting it that way, though, requires a remarkable amount of effort. And it’s not just the purity — it’s the volume, too. A semiconductor fab uses a mind-boggling amount of UPW — two to three million gallons (7 – 12 million liters) per day. Building processes that can purify that much water to such stringent requirements, keep it at that purity until it’s needed, and recycle it where possible is a huge challenge.
UPW production begins with bulk treatment of the raw feed water. Steps here include processes that appear in most municipal wastewater treatment plants — the addition of flocculant and coagulant compounds to clump any suspended solids together, sedimentation to let the clumps settle out, and bulk filtration to remove the rest. These steps serve to remove the biggest, nastiest chunks — relatively speaking; the feed water for most fabs is municipal water that would be fine for human consumption — and prepare the water for the processes that will remove sequentially finer contaminants.
The next step is usually one or two stages of reverse osmosis, or RO. As the name suggests, reverse osmosis is the opposite of the natural process of osmosis, which occurs when an imbalance exists between the concentration of two solutes across a semipermeable membrane. The solvent tends to migrate from the side with low solute concentration to the more concentrated side, to even out the imbalance. In RO, the osmotic pressure is overcome by putting energy into the system with a pump, which forces the solvent (water) to migrate across a membrane to the side with lower solute concentration, leaving the solutes behind. The semipermeable membrane is engineered from a non-woven fabric support layer topped with layers of polymers such as polysulfone and polyamide, which form a barrier through which water can pass, but larger solutes cannot.
Ultraviolet light is used in several stages of UPW production. UV of the correct wavelength not only kills any bacteria left after the pretreatment steps, but also tends to degrade the biopolymers, like proteins, DNA, and RNA, in the bacterial remains. The more these macromolecules are chopped up at this stage, the easier they’ll be to remove during later stages of processing.
In order to remove electrically charged contaminants from the process water, ion exchange treatment is used. Ion exchange uses special polymer resin beads that have binding sites on their surfaces. The binding sites are either positively charged (cation exchanger) or negatively charged (anion exchanger). When process water flows over a vat of ion exchange resin, the charged ions that are in solution tend to bind to the sites in the resin with the opposite charge, effectively cleaning them out of the process water.
A variation on ion exchange, called electro deionization (EDI), is sometimes used as well. EDI basically combines ion exchange with reverse osmosis and electrolysis, using an electric current passing through multiple resin beds separated by semipermeable membranes to remove ions from the process water.
Measuring the Unmeasurable
After some final degassing step, the UPW is finally pure enough to enter fab processes. Or is it? That’s hard know, since UPW that’s clean enough to meet the fabrication process requirements is too clean to measure with any current technology. That puts engineers in a difficult position, as often the only reliable way to know if the UPW process is defective is by seeing a decrease in chip yields, and expensive and wasteful assay to say the least.
Despite that fact, there are some metrology methods that are employed to monitor the UPW process. The primary measurement is the conductivity of the water, which can be used to judge the presence of a number of contaminants. For practical reasons, the reciprocal of conductivity, resistivity, is usually measured, with pure water reading 18.18 MΩ⋅cm at 25°C. Resistivity can be exquisitely sensitive — the addition of sodium to just 0.1 parts per billion will drop the resistivity to 18.11 MΩ⋅cm, and such a drop may indicate that an upstream ion exchange bed is in need of attention. And as mentioned before, air leaks in the system can be detected by the resistivity change caused by CO2 dissolving into the system.
But when it comes to measuring particulates, there aren’t many methods available that can detect such dilute and minute particles as are required for fab-grade UPW. One technology that’s coming close is dynamic light scattering (DLS), which bounces polarized laser light off a water sample. The scattered light passes through another polarizer on its way to a detector, where a snapshot of the scattered light is taken. The process is repeated a short time later, on the order of microseconds to nanoseconds, and the images are compared. The difference between the two images can be attributed to the Brownian motion of whatever particles are in the sample, and the constructive or destructive interference caused by the particles’ movements. This results in a measurement of the particle count and size distribution, potentially down into the sub-nanometer range.
There’s a lot more to UPW systems for fabs, including incredibly complex recycling systems that take the waste-laden water after it’s used in fab processes and reclaims it for further use. And UPW standards are very much a moving target, too. Contaminants that would have gone unnoticed in the larger process nodes of the past are now considered killer particles for 5 nm nodes, so it stands to reason that UPW standards will have to become even more stringent as process nodes advance. And engineers will have to keep up, somehow building systems that can turn out oceans of water that’s purer than can be measured.