SMIC 90nm arrived this year, which puts the Shanghai foundry about two generations behind the leaders and closing, building 90-nanometer chips that TSMC was shipping a couple of years ago. SMIC 90nm is not a breakthrough and nobody pretends it is. It is a rung, one more step up a ladder the company has been climbing since Richard Chang started pouring fabs outside Shanghai at the turn of the decade, and the only remarkable thing about it is the speed of the climb. Six years from farmland to 90 nanometers is fast. It is also still years from the front.
Ninety nanometers is where the industry leaned hard into copper interconnect and low-k dielectrics, and where Intel and TSMC pulled away by adding strained silicon to squeeze more speed out of the transistor itself. SMIC’s version of the node is plainer: a basic copper flow, dry 193nm lithography, none of the exotic strain engineering the leaders use to widen the gap. It works and it yields, two years late and a notch simpler. That is exactly what catching up looks like at this stage, not matching the leading recipe but running a clean, cheaper imitation of where it was a while ago.
The business underneath is ugly, and that is normal for what SMIC is. The company is the world’s fourth-largest foundry now, behind TSMC, UMC and Singapore’s Chartered, and it loses money doing it, because building and running fabs costs billions and SMIC sells mature-node capacity into a market where the leaders already compete on price and scale, with DRAM prices sliding underneath the whole thing. A normal company bleeding like this would be in trouble. SMIC is not a normal company; it has the patience of a state behind it, and a state building strategic industry does not panic over losses the way shareholders do. The red ink is the cost of admission, and Beijing is willing to pay it for as long as the climb continues.
One and two nodes back from the leading edge is where the money SMIC can realistically win actually lives, not in a race with TSMC it cannot afford. SMIC sits fourth in foundry revenue, a hair behind Chartered and a rounding error against TSMC’s roughly half the market and UMC’s mid-teens, all of them chasing the same 90 and 130nm business. Underneath sit the other mainland fabs, Hua Hong NEC and Grace Semiconductor, climbing from further down and undercutting on price. Above, the leaders keep older fabs running long after they are paid off and dump cheap mature-node capacity whenever the front end goes quiet. SMIC is caught in between, building expensive new capacity to make commodity chips in a market where everyone older already wrote theirs off.
The first fight with TSMC is settled, not forgotten. The 2003 suit closed early last year, SMIC agreeing to pay TSMC 175 million dollars over six years and the two cross-licensing their 180nm patent portfolios through 2010. That bought peace, not trust. TSMC’s underlying complaint was always that SMIC’s newer processes carry Taiwanese fingerprints, and SMIC’s 130nm and now this 90nm are precisely the processes that argument points at. Whether that grievance stays in a filing drawer or comes back to court is an open question in 2006. The pattern it points at is the real story, which is that SMIC climbs by absorbing what the leaders already know, through hired talent, through licensed technology, through foreign equipment, through every channel a determined latecomer can open. There is no shame in it; every foundry learned from someone. SMIC just has to learn faster and from further behind than anyone before it.
Every advanced thing in SMIC’s fabs comes from outside China. The lithography machines are Dutch and Japanese, ASML and Nikon, the deposition and etch tools American, Applied Materials and Lam, the materials and the methods imported, because the equipment that makes leading chips is built by a handful of companies and none of them are Chinese. This is simply how the industry works in 2006: chipmaking is global, the tools flow to whoever pays, and SMIC pays. The arrangement suits everyone. The toolmakers get a fast-growing customer, SMIC gets the machines it cannot build itself, and the chips get made. Nobody treats the supply of those tools as a question. It is just the plumbing of a global business.
At 90nm SMIC can take real business, the chips that do not need the leading edge, the controllers and the connectivity parts and the cheaper application processors that fill the electronics China assembles by the container-load. It is widening that base the way a state-backed champion does, taking over the running of new fabs in Wuhan and Chengdu to spread its process across more of the country. What it does not have is what its rivals just built: IBM has pulled Chartered, Samsung and Infineon into a joint development club to share the brutal cost of 65 and 45nm, and SMIC climbs essentially alone, on licensed tech and hired engineers, which is slower and lonelier and entirely the point. SMIC does not need to beat TSMC to be useful to China; it needs to exist, to grow, to keep climbing, and to be there when the country needs a foundry it controls. At 90nm, it is becoming exactly that.
The climb from here looks like more of the same, steady and state-funded and gradually closing the gap. Perso, I expect SMIC keeps gaining a node every few years and is a genuine second-tier global foundry within the decade, not at the front but solidly in the race, the way the will and money behind it suggest. The lawsuits are noise; they settle, the engineers stay, the technology keeps coming. The one thing the whole plan rests on is the thing nobody questions, the steady supply of foreign machines and methods that every rung of this ladder is built from. As long as the tools keep arriving in Shanghai, SMIC keeps climbing. In 2006 there is no version of the future where they stop arriving. So the climb continues, quarter after quarter, node after node, exactly as planned.