BRENNEN

Industries / Technology / Electronics and semiconductors

Semiconductor equipment

Semiconductor equipment is the industry that designs, builds and services the machines chip factories use to print, add, remove, measure and move the layers of a chip on a silicon wafer.

1967

Applied Materials was founded to supply chemical vapour deposition machines that grow thin crystal films on wafers, and began moving the development of manufacturing methods from chipmakers to their equipment suppliers.

computerhistory.org

37

Milestones achieved across 6 domains.

17

To watch for: announced, not yet confirmed outside.

13

Open deliverables. Nobody has shipped them.

In one minute

Where it started

The first outside suppliers sold single machines: the Computer History Museum names Electroglas, a 1961 spinout of Fairchild that built probe equipment for testing wafers, and Kulicke & Soffa, which introduced commercial contact aligners in 1965. Applied Materials, founded in 1967, encouraged chipmakers to hand the development of manufacturing methods to their toolmakers. The museum says this let chip companies focus on product design, and later helped the rise of foundries and of chip firms that own no factory.

Where it stands

SEMI, the industry's trade body, says world sales of chipmaking equipment rose 15% to $135.1 billion in 2025, and that China, Taiwan and Korea bought 79% of it, China the most at $49.3 billion. In July 2026 SEMI forecast $165.9 billion for 2026 and $229.5 billion for 2028. The research firm TechInsights ranks ASML first, followed by Applied Materials, Lam Research, Tokyo Electron and KLA, the same top five as in 2024, and says Chinese toolmakers gained as foreign suppliers pulled back from China under US-led export rules. Those rules now decide who may buy, and who may service, the most advanced machines.

What is open

Much of this trade's knowledge sits inside a few companies. We found no open model of how many wafers an hour an EUV scanner can print, no free count of lithography machines shipped, no public benchmark for measuring line roughness in electron-microscope images, and no shared scoreboard for the ceiling vehicles that carry wafers. Public datasets now exist for some steps, such as Bosch etching runs and fab layouts, but nobody has fixed them into tests others can score against. Low-cost robot arms and current AI models are enough to try handling and maintenance jobs on mock-ups, and to score how well a model reads a tool's manual.

Domains

AchievedWatch forOpen

Advances to build with

1 research note

Open in semiconductor equipment

3 of 13 open deliverables

Open · Lithography

Open EUV throughput model

An open program that works out how many wafers an hour an EUV scanner can expose from its light budget: the source power where light enters the scanner, the losses at each mirror and at the mask, the resist dose (the light energy per area the coating needs, in mJ/cm²), the layout of exposure fields, scan speed and the fixed handling time per wafer.

Built with what exists

Python, NumPy, pandas and Jupyter in a public GitHub repository; ASML product pages and published papers for inputs. The Note "400 wafers an hour" explains what a wafers-an-hour rating means for a scanner.

Credit · unclaimed

Open · Deposition

ALD growth-per-cycle table

An open table of the 30 or so ALD processes most used in chipmaking, such as aluminium oxide from trimethylaluminium and water, hafnium oxide, titanium nitride, silicon oxide and nitride, molybdenum and ruthenium. Each row gives growth per cycle (nm per cycle), the temperature window in which growth stays steady (°C) and the dose needed to cover the surface, tied to a paper's DOI and figure or table number, with a short script to query it.

Built with what exists

Python and pandas; WebPlotDigitizer to read values off published graphs; Zotero; the Open Research Knowledge Graph for publishing; Zenodo for a DOI; an AI model to draft rows from papers for a person to check.

Credit · unclaimed

Open · Etch

Open etch-recipe game

An open version of the "virtual process game" that Lam Research used in a 2023 Nature paper to compare how people and algorithms develop an etch recipe. Wrap a simulation of a silicon hole etch as a black box with hidden settings, publish a target hole shape, charge each attempt at Lam's rates ($1,000 per recipe plus $1,000 per batch, standing for wafer and measurement costs), and let people, Bayesian optimisers (programs that pick each next trial from the results so far) and AI agents compete, with a public tally of money spent to reach the target.

Built with what exists

ViennaPS, an open-source process simulator with Python bindings and an SF6/O2 hole-etch example; a Bayesian optimisation library such as optimas; a coding agent such as Claude Code for the agent players; a GitHub repository for scored run logs.

Credit · unclaimed

Where it sits