Industries / Technology / Electronics and semiconductors
Circuit boards and assembly
This work designs printed circuit boards (flat boards with copper wiring), makes them, solders parts onto them and builds the finished boards into devices.
7
Milestones achieved, from 1942 to 2016.
3
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1942
Wiring printed on a board
Before printed boards, the parts of an electronic product were joined by hand-soldered wires. Paul Eisler, an Austrian refugee in London, built a radio whose connections were a copper pattern on a board, which the Science Museum Group calls the first working device to use a printed circuit board.
Credit
Paul Eisler
1949
Solder every joint at once
Two US Army Signal Corps scientists pushed the parts' wire leads through holes in a printed board and dipped it in molten solder, so every joint was soldered in one step. The Army received the patent in 1956, and the method grew into the standard way of making boards.
Credit
Moe Abramson and Stanislaus F. Danko, US Army Signal Corps
1985
Parts move onto the surface
Surface-mount parts are soldered onto pads on the board's surface instead of having leads pushed through holes. The Semiconductor History Museum of Japan says the methods were developed mainly in Japan in the 1980s, driven by demand for cheap portable electronics, and that from 1985 a committee of the Electronic Industries Association of Japan ran campaigns to spread the new soldering methods, materials and equipment.
Credit
Japanese chip and equipment makers; Electronic Industries Association of Japan
1990
Testing joints without probes
The IEEE approved Standard 1149.1, widely called JTAG, on 15 February 1990. Chips that follow it carry a chain of test cells at their pins, reached through a few test pins, so a tester can check solder joints between chips on a finished board, including joints a probe cannot reach.
Credit
IEEE, from an industry draft
1992
Free board design software
Jean-Pierre Charras, a teacher, first released KiCad, free and open-source software for drawing circuits and laying out boards, as a tool for his students. CERN later paid developers to improve it, and in May 2026 released the library it uses for its own boards, more than 17,000 KiCad parts, under an open-source licence.
Credit
Jean-Pierre Charras; KiCad developers; CERN
2006
Lead leaves the solder
An EU directive (known as RoHS) required that new electrical and electronic equipment sold in the EU from 1 July 2006 contain no lead, so makers selling there moved to lead-free solders.
Lessonlead had held back tin whiskers, thin metal hairs that grow from pure tin and cause short circuits; NASA's parts programme lists satellites lost to them, such as Solidaridad 1 in August 2000.
Credit
European Parliament and Council (Directive 2002/95/EC)
2016
Galaxy Note 7 recall
In September 2016 the US Consumer Product Safety Commission recalled about 1 million Samsung Galaxy Note 7 phones after reports that they overheated and caught fire. Investigators traced the fires to two different battery cells, one with too little room inside so its electrode bent, the other with poor welding and missing insulation; the phone's own electronics were not the cause.
Lessona device is only as safe as the parts it is built from, and Samsung added an eight-point battery check.
Credit
Samsung (recall); UL, Exponent and TÜV Rheinland (investigation)
Watch for
AI designs working devices
Anthropic says its Claude Fable 5.2 model chose the parts, drew the schematics and four-layer board layouts, and wrote the cases and software for three devices (a USB speakerphone, a voice-assistant hub and a handheld virtual pet), and that all three worked with no electrical changes. Diode Computers, a board maker with a commercial tie to Anthropic, built and tested them, and Anthropic says the devices have not been through electrical validation, thermal testing or certification. Note: "Three working devices".
Achieved whenan outside lab builds the boards from published design files and confirms they work, or a device passes an accredited lab's electrical and radio tests.
Claimed by
Anthropic, November 2026
AI lays out a computer
Quilter says its AI laid out a two-board Linux computer with 843 parts that booted on the first try, with 38.5 hours of human work against 428 hours quoted for a manual layout, and it has released the design files. All About Circuits' report cites no independent check.
Achieved whensomeone outside Quilter builds the boards from the released files and reproduces the boot, or publishes signal measurements from them.
Claimed by
Quilter, December 2025
Humanoid robots on server lines
Foxconn said its Houston plant, which assembles NVIDIA AI servers, would put humanoid robots running NVIDIA's Isaac GR00T N model on production lines from early 2026, and in April 2026 one of its engineers said "fewer than 10" would start on a server line. Note: "Five demonstrations per task" covers a later GR00T model.
Achieved whenan outside reporter, customer or auditor confirms the robots doing daily production work and publishes what they do and how often they fail.
Claimed by
Foxconn, October 2025 and April 2026
Open
01
Robot plugs in board cables
A low-cost robot arm that plugs a wire-to-board connector, such as a 2.54 mm JST-style plug, into its header on a fixed circuit board, followed by an automatic check that every pin connects.
The briefCredit · unclaimed02
Build and measure AI layouts
Take one open-source board design whose human-made layout is public, have an AI tool lay out the same schematic, then make and assemble five boards of each version and measure all ten the same way.
The briefCredit · unclaimed03
Check AI footprints against CERN
A footprint is the pattern of copper pads a part is soldered to. Have an AI model read the package drawing in each part's datasheet and draw a KiCad footprint, score it with a script against the footprint for the same part in CERN's library, and test a sample on a real board.
The briefCredit · unclaimed
Open 01 · the brief
Robot plugs in board cables
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A low-cost robot arm that plugs a wire-to-board connector, such as a 2.54 mm JST-style plug, into its header on a fixed circuit board, followed by an automatic check that every pin connects.
Why it is still open
Fraunhofer researchers write that "Cabling tasks (pulling, clipping, and plug insertion) are today mostly manual work". A February 2026 study passed 90% insertion success, but on an industrial arm, and Foxconn's 2025 humanoid demonstrations of plugging cables have not been reported in production.
Built with what exists
an SO-101 arm with LeRobot (Note: "No download") and one of its imitation-learning policies, such as ACT; a USB camera; a 3D-printed fixture; an Arduino wired as a continuity checker; 5 V USB power only.
First step
Print the fixture, wire the continuity checker, and record 50 remote-controlled demonstrations.
Done looks like
twenty consecutive filmed attempts, with the board shifted by up to 5 mm between attempts, and a published tally; an attempt counts only if the plug latches and every pin passes the continuity check; code and training data public.
Open 02 · the brief
Build and measure AI layouts
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Take one open-source board design whose human-made layout is public, have an AI tool lay out the same schematic, then make and assemble five boards of each version and measure all ten the same way.
Why it is still open
AI board layouts so far are judged by the people who made them. Anthropic says its three devices have not had electrical validation, All About Circuits found no independent check of Quilter's computer, and PCB Gym, a NeurIPS 2026 benchmark, scores layouts in software, not on built boards.
Built with what exists
KiCad; an AI layout tool such as Claude Fable 5.2 (Note: "Three working devices") or Quilter; a low-cost board making and assembly service; a USB oscilloscope, multimeter, thermocouple, and a near-field probe with a low-cost spectrum analyser; USB power only.
First step
Choose the reference design, run the AI layout, and post both layouts and the measurement plan before ordering boards.
Done looks like
a public report with both sets of design files and, for all ten boards, power-supply ripple, start-up behaviour, USB connection, temperature rise and a scan of radio emissions near the board; a second person repeats at least one measurement on the same boards.
Open 03 · the brief
Check AI footprints against CERN
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A footprint is the pattern of copper pads a part is soldered to. Have an AI model read the package drawing in each part's datasheet and draw a KiCad footprint, score it with a script against the footprint for the same part in CERN's library, and test a sample on a real board.
Why it is still open
Researchers found that current image-reading AI models have "inaccurate geometric perception" of footprint drawings, and vendor libraries still rely on staff checks; SnapMagic says its footprints are "verified with a three-step verification process". CERN's May 2026 release of more than 17,000 KiCad parts gives a ready answer key.
Built with what exists
CERN's KiCad libraries; KiCad's Python scripting; an AI model that reads images; public datasheets; a board making and assembly service.
First step
List 50 CERN parts with public datasheets across several package types and run the first scoring script.
Done looks like
200 parts scored on pad count, pad position, pad size and pin-1 location against CERN's footprints, with per-part results published; plus 20 AI-made footprints on a test board, assembled with the real parts, every pin continuity-tested and the results published.