The Chemistry Behind the AI Memory Boom: The Industrial Chemicals That Build Every DRAM, NAND & HBM Chip

The AI boom has made memory the hottest corner of the chip industry — Micron, SanDisk, and Marvell are household tickers again, and high-bandwidth memory is sold out years in advance. Behind every one of those chips is a surprisingly old-school list of industrial acids and hydrogen peroxide doing the unglamorous work of cleaning and etching silicon. Here is the actual chemistry behind the memory supercycle — and an honest look at what grade of each chemical the work really needs.

What is driving the 2026 AI memory boom?

Memory is the bottleneck of the AI build-out, and that is what is driving the boom. The graphics processors that train and run large AI models are starved for bandwidth, and the way you feed them is with high-bandwidth memory (HBM) — stacks of DRAM dies bonded vertically and wired together with through-silicon vias. Each AI accelerator needs many gigabytes of it, demand has outrun supply, and that shortage has rippled into ordinary DRAM and NAND flash prices as well.

That is why the memory names are back in the headlines. Micron (MU) makes DRAM and HBM and has reported record memory revenue on AI demand. SanDisk — spun back out as a standalone NAND-flash company — sits at the center of the storage side. Marvell (MRVL) designs the custom silicon and custom-HBM architectures that hyperscalers wrap around their AI chips. Their stocks move on the same story: the world cannot make memory fast enough.

The scale is what makes it a supercycle rather than a normal upturn. A single high-end AI accelerator can carry well over a hundred gigabytes of HBM, and each HBM stack is itself a tower of eight, twelve, or more DRAM dies thinned and bonded together — so every accelerator effectively consumes many full memory chips at once. Multiply that by the millions of accelerators going into data centers and you get a demand curve that DRAM and NAND fabs, which take years to build, simply cannot match on short notice. Prices rise, lead times stretch, and every wafer a fab can clean and finish becomes more valuable.

What rarely makes the headlines is the supply chain underneath the supply chain. Building a single memory wafer takes a thousand-plus process steps, and a large share of them are wet-chemical steps: baths of acid and hydrogen peroxide that clean, oxidize, and etch the silicon between every patterning operation. The chips are exotic; the chemistry is not. It is mostly the same acids that have run industrial plants for a century — just held to extraordinary purity. That is the part we can actually explain, because we sell those chemicals.

How is a memory chip built — and why does it need so much chemistry?

A memory chip is built by repeating a small set of operations — deposit a film, pattern it with light, etch it, clean it — hundreds of times to stack up billions of identical cells. The cleaning and etching between those steps is where the wet chemicals live, and there is a simple reason the bar is so high: at these dimensions, a single stray particle or a few atoms of metal contamination can kill a memory cell.

Contamination control is the whole game. Modern DRAM features are measured in the low tens of nanometers, and a 3D NAND stack is now well over 200 layers tall. A speck of dust, an organic smear left by photoresist, or trace iron, copper, and sodium ions on the surface will each ruin the device in their own way — particles cause physical defects, organics block subsequent reactions, and mobile metal ions poison the transistor electrically. So after almost every major step the wafer goes through a cleaning sequence designed to remove exactly one class of contaminant at a time.

Two cleaning recipes do most of that work, and they have barely changed in principle since the 1960s and 70s: the Piranha clean (also called SPM) and the RCA clean (its two halves, SC1 and SC2). Both are built almost entirely from chemicals on this page. A third operation — selectively stripping silicon nitride with hot phosphoric acid — turns out to be the linchpin of how 3D NAND is built at all. Let us take them one at a time.

What is a Piranha clean, and which chemicals make it?

A Piranha clean is a mixture of concentrated sulfuric acid and hydrogen peroxide that strips organic material off a silicon wafer — most importantly the leftover photoresist after a patterning step. In the industry it is called SPM (sulfuric–peroxide mixture). It is one of the first and most frequent cleans a wafer sees.

The chemistry is aggressive and self-heating. Sulfuric acid (H₂SO₄) and hydrogen peroxide (H₂O₂) are typically blended around three or four parts acid to one part peroxide; they react to generate Caro’s acid (peroxymonosulfuric acid) and a flood of reactive oxygen, and the mixture heats itself to well over 100 °C without any external heater. That hot, strongly oxidizing bath turns organic films — photoresist, fingerprints, carbon residue — into carbon dioxide and water, leaving a chemically clean silicon surface behind.

What is the RCA clean (SC1 and SC2)?

The RCA clean is the two-step sequence that removes particles and metal contamination from a wafer, and it is arguably the most important cleaning recipe in all of microelectronics. It was developed by Werner Kern at RCA Laboratories in 1965, and the two baths — SC1 and SC2 — are still run, in updated dilute forms, in every memory fab on earth.

SC1 (APM): ammonium hydroxide + hydrogen peroxide

SC1, the “Standard Clean 1” or ammonia–peroxide mixture (APM), removes particles and organic residue. It blends ammonium hydroxide, hydrogen peroxide, and water — classically about 1:1:5 — at roughly 70–80 °C. The peroxide gently oxidizes the silicon surface while the ammonia etches that thin oxide away, and this slow lift-and-renew action undercuts particles so they float off. It is the particle-removal workhorse.

SC2 (HPM): hydrochloric acid + hydrogen peroxide

SC2, the “Standard Clean 2” or hydrochloric–peroxide mixture (HPM), removes metallic contamination. It blends hydrochloric acid, hydrogen peroxide, and water — classically about 1:1:6 — again around 70 °C. The hydrochloric acid forms soluble chloride complexes with metal ions like iron, copper, zinc, and the alkali metals, pulling them off the surface and keeping them dissolved so they rinse away. After SC2 the wafer is electrically clean.

Why does 3D NAND need hot phosphoric acid?

Three-dimensional NAND flash — the memory inside every SSD and phone — is built by stacking alternating layers of silicon oxide and silicon nitride, and at a key moment the nitride has to be removed selectively while the oxide stays put. The chemical that does that, and essentially only that, is hot phosphoric acid. No hot phosphoric strip, no 3D NAND.

Here is the sequence in plain terms. To build a tall NAND stack, fabs deposit dozens to hundreds of oxide–nitride pairs (an “ONO” or O/N stack), etch deep holes and slits through the whole tower, and then flow hot phosphoric acid in to dissolve away every nitride layer at once, leaving horizontal gaps. Those gaps are then filled with metal to become the word lines — the control gates of the memory cells. The nitride was a sacrificial placeholder; phosphoric acid is what removes the placeholder. This is the “gate-replacement” or “nitride pull” step, and it is unique to how 3D NAND and many advanced DRAM and logic structures are made.

What makes phosphoric acid special is selectivity. Hot, concentrated phosphoric acid — run at roughly 85% strength near 160 °C — etches silicon nitride (Si₃N₄) far faster than it touches silicon oxide (SiO₂). That difference is everything: it lets the process remove the nitride layers cleanly while leaving the oxide scaffolding intact. Controlling the acid’s temperature, water content, and dissolved-silicon level to hold that selectivity steady is one of the quiet arts of NAND manufacturing.

The difficulty is that the acid does not stay constant on its own. As phosphoric acid strips nitride it accumulates dissolved silicon, and both the temperature and the water content drift as the bath runs hot — and all three of those variables change the etch rate and the precious nitride-to-oxide selectivity. Hold them loosely and you either leave nitride behind or start eating the oxide scaffolding the structure depends on. Managing that window across a bath that has to behave identically for every wafer, all day, is a large part of why NAND process engineering is hard, and why the “boring” chemical step gets so much attention inside a fab.

DRAM and HBM lean on the same toolkit. HBM is not an exotic new chemistry — it is ordinary DRAM dies, made with the same cleans and etches, then thinned, stacked, and wired together with through-silicon vias. The wafer thinning and via etching in that packaging flow add their own cleaning steps, which is one more reason the memory boom pulls through extra volume of exactly these wet chemicals. More memory, in every form factor, means more acid and more peroxide moving through the fabs.

Where do nitric acid, hydrofluoric acid, and IPA fit in?

Beyond the four core chemicals, a handful of others round out the wet bench — and being honest about which ones we sell matters here.

Nitric acid (HNO₃) is a strong oxidizer used in metal etching and cleaning, in some surface-preparation and silicon-etch blends, and in tool and parts cleaning across the fab. Because trace metal and particle contamination is the enemy, the semiconductor world prizes low-particle nitric acid — which is exactly the spec name on one of our grades. It is a good example of how an ordinary industrial acid is differentiated for electronics work by how clean it is, not by what it is.

Hydrofluoric acid (HF) is the other giant of wafer processing — it etches silicon oxide and, as buffered oxide etch, strips native oxide before critical steps. We mention it because no honest account of fab chemistry can leave it out, but Alliance Chemical does not sell hydrofluoric acid. It is exceptionally hazardous to handle, and we have chosen not to stock it. If your process needs HF, it should come from a specialist supplier with the right handling and medical-response framework around it.

Isopropyl alcohol (IPA) closes out the cycle: after the final rinse, ultra-dry IPA vapor is used in Marangoni drying to pull water off the wafer without leaving spots or watermarks. We have written about that in depth — see our guide on isopropyl alcohol in semiconductor wafer cleaning and drying.

Do these chemicals make memory chips, or just clean them?

They clean and etch — they do not, by themselves, make the memory cell, and it is worth being precise about that because the distinction is the whole honest framing of this article. The transistor and capacitor that store your data are created by deposition, lithography, and ion implantation. What the acids and peroxide do is prepare and reset the surface between those steps and carve away sacrificial material so the real structures can form.

But “just cleaning” undersells it. In a process with a thousand steps and zero tolerance for contamination, the cleaning is the yield. A fab that cannot reliably strip organics, lift particles, and remove metal ions does not make bad memory — it makes no working memory at all. And in the specific case of the hot-phosphoric nitride strip, the chemistry is not cleaning up after the process; it is a load-bearing structural step that defines the device. So the accurate statement is this: these chemicals do not invent the memory cell, but every memory cell that works passed through them many times over.

What grade do these chemicals need — and what does Alliance supply?

This is the part we will be completely direct about. A production memory fab buys SEMI-grade chemicals — ultrapure material certified to parts-per-billion or parts-per-trillion metal limits under SEMI standards, from specialist electronic-chemical suppliers. Alliance Chemical does not sell that tier, and we will not pretend otherwise. What we supply is the same chemistries in ACS reagent, technical, USP, and low-particle grades.

That distinction decides who we are the right supplier for. We are a strong fit for the work that surrounds and feeds the industry rather than the high-volume production line itself:

Why grade matters even off the production line: trace impurities are exactly what corrupt early results. In a bench RCA clean, stray metal ions in the acid can deposit the very contamination you are trying to remove; in materials research, inconsistent assay scatters your data. A high, well-documented purity with a certificate of analysis on every shipment lets you change one variable at a time — the entire point of lab work. Tell us the application and the step, and we will help you match the grade honestly: ACS where it earns its keep, technical where it is plenty, and a straight answer when what you really need is SEMI-grade we do not carry.

Key numbers & authoritative sources