India’s chip plan finally says the word ‘Chemicals’

The least photogenic part of semiconductor manufacturing has made it into national policy at last. Whether it survives contact with reality is another matter

There is a rule inside a chip factory that visitors always find strange. Do not drink the water.

It is the purest water on the premises, purer than anything sold in a bottle, and that is exactly the problem. A fab run on ultrapure water, stripped of salts, metals, dissolved gases, bacteria and organic residue until almost nothing is left. Engineers measure how empty it is by electrical resistivity, and the number they want is 18.2 megaohm-centimetres, which is a technical way of saying there is nothing in it capable of carrying a charge. Water in that state is chemically ravenous. It pulls ions out of steel piping, out of glass, out of the lining of your gut. A large plant swallows millions of litres of it a day and must manufacture every litre on site, because the stuff begins spoiling the moment it stops moving.

I offer this because the semicon industry is almost always photographed from one angle: the cleanroom, the white suits, the wafer catching the light. The water tells you what the building actually is. It is a chemical plant that happens to produce computers, and chemistry, not silicon, is where India’s chip ambition will be decided.

Which brings me to a phrase in the Cabinet approval for Semicon 2.0, issued on 15 July, that almost no newspaper quoted. Past the paragraphs about fabs and packaging units and design startups, under the second of the programme’s six pillars, the Government commits to supporting the domestic manufacture and research of semicon equipment, specialty chemicals, industrial gases and critical materials. Four years into a national semicon mission, that is the first time the unglamorous half of this industry has been written into Indian policy with any seriousness. It deserved more attention than it got. It is the part of the mission most likely to fail without anyone noticing until the fix has become expensive.

Three hundred steps, mostly wet

The process itself is a long chemical sequence. A light-sensitive polymer, the photoresist, is spun across the wafer in a film a few hundred nanometres thick. Light prints a pattern into it. A developer dissolves away the parts that were exposed. Reactive gases eat the pattern down into the layer beneath. Metals and insulators arrive as vapour and condense into place. Abrasive slurries grind the surface flat to within a few atoms. Acids and solvents clean up after every one of these steps, and ultrapure water rinses away the acids. Then it happens again, and again, several hundred times over. The point worth holding on to is that machines are bought once and chemicals are bought forever. A fab that loses its supply of a single etch gas does not slow down. It stops, in days.

What three bottles did to Seoul

Korea learned this in the summer of 2019, when Japan placed three material categories under export licensing: photoresists, high-purity hydrogen fluoride, and fluorinated polyimide. Korean chipmakers were, and are, among the most advanced manufacturers on earth. It made no difference. Executives spent that year flying to Tokyo hunting alternate supply, and Seoul has been pouring money into domestic materials capacity ever since. Nobody had restricted a fab or a machine. Three bottles were enough.

India is currently building the kind of exposure Korea spent the last seven years trying to unwind. We have the world’s sixth-largest chemical industry and a real depth of skill in complex organic synthesis, which is why Indian firms already supply intermediates that end up inside advanced photoresists made elsewhere. What we do not have is electronic grade. Domestic production of finished photoresists, of high-purity hydrogen peroxide, of ultrapure water systems, ranges from negligible to nonexistent. For the specialty gases used in etching and deposition, along with high-purity hydrogen and helium, India imports well over half of what it consumes by value. Every fab we approve deepens that, unless something moves upstream at the same time.

So the second pillar is the right instinct. The difficulty is that a chemicals ecosystem cannot be built the way a fab is built, and a subsidy line written on fab logic will not produce one.

A millionfold tighter

Start with purity, because the gap is not what most people assume. Indian pharmaceutical manufacturing is world-class, and it controls impurities at parts per million. Several semicon inputs must be controlled at parts per trillion. That is a millionfold tightening, and it is not the same job done harder. One part per trillion is roughly one second out of thirty-two thousand years.

At that level your contamination is no longer coming from your process; it is coming from the valve, the drum lining, the tanker, the person who opened it.

Which leads to a problem that costs almost nothing to solve and has gone unsolved anyway. Nobody can sell a purity they cannot prove. Certification at trace levels needs metrology, the measurement infrastructure that turns a claim into a document a customer will accept, and India has barely any of it for electronic materials. An Indian supplier today ships samples abroad for verification, waits months, pays for the privilege, and in the process hands a foreign laboratory a fairly detailed picture of what it can make. A national trace-analysis facility, anchored in our standards institutions and open to any Indian firm, would cost a rounding error against Rs 1,27,500 crore. It is the single highest-leverage thing this programme could fund in its first year.

The two years nobody pays for

Then there is the money’s timing, which is where I suspect this pillar will actually break. A fab does not buy your chemical because you built a plant. It buys after qualification, meaning twelve to twenty-four months of sample runs, contamination testing and yield comparison, because one bad batch can scrap wafers worth several million dollars.

Production-linked incentives pay on output. Qualification is precisely the stretch when there is no output, only cost, and a customer who is entitled to say no at the end of it. That is the window in which an Indian supplier dies, and it is the window our incentive design does not cover. Paying against milestones instead, when samples are accepted, when a fab audit is cleared, when a trial batch passes, would cost a fraction of what we hand a single fab and would decide whether any of this becomes an industry.

There is a related fix that requires no money at all. Any fab drawing public support could be required to run a minimum number of qualification trials with Indian suppliers each year and to report the results. Not an obligation to buy, which would be bad policy and worse engineering. An obligation to test. At the moment Indian firms cannot even get through the door to fail, and failing usefully is how a supplier eventually passes.

The API film, a third time

None of this is exotic. It is the same argument India has had before, in pharmaceuticals, where we mastered the difficult chemistry, exported the intermediate, and let someone else own the finished product and the margin that came with it. The chip industry offers a chance not to run that film a third time, and the window is open for about as long as the current fabs take to reach volume.

The image that will tell us whether Semicon 2.0 worked is not the cleanroom. It is a tanker of electronic-grade hydrogen peroxide pulling up to a fab gate at three in the morning with its certificate in order, having driven there from Ankleshwar rather than sailed from Yokohama. We are not close to that yet. We have, at least, finally written down that we want it.

The author is a physicist at the University of North Carolina at Chapel Hill and a columnist on AI, infrastructure and global systems; Views presented are personal.