AI News
All stories
Official

Anthropic says Claude designed and prototyped three devices with Diode Computers

Anthropic reports that Claude Fable 5.2 designed circuit boards, enclosures and firmware for three devices, and that Diode built working prototypes without circuit fixes.

AI in Practice · · 4 min read

Anthropic says its Claude Fable 5.2 model designed circuit boards, enclosures and firmware for three electronic devices. Diode Computers, a circuit board maker, built working prototypes from those designs without any fixes to the circuitry. The devices are a USB speakerphone, a home voice-assistant hub and a handheld virtual pet. Anthropic says Claude started from a short prompt. For the speakerphone, that prompt asked for "a puck that plugs into a computer over USB-C and includes microphones and a speaker, supporting Bluetooth with a PCB antenna."

From there, according to Anthropic, Claude wrote the device specifications, drew schematics, chose where each component sits on the printed circuit board, routed the copper connections, designed a 3D-printable enclosure, wrote the firmware and prepared a parts list. Diode assembled the boards and ran the firmware under Claude's direction. Anthropic says the speakerphone still works after weeks of use, and that the other two prototypes are also in good working order. It does not yet know how the devices will hold up over time or whether they would pass commercial certification.

Routing is where Anthropic says the newer model differs most from its predecessors. Earlier models, it says, placed supporting parts too far from the chips they served and oriented connectors wrongly. Fable 5.2 routed the most sensitive connections first: high-speed, high-current or noise-sensitive signals. For the voice-assistant hub, Claude routed the USB and Ethernet pairs first, keeping the two tracks in each pair equal in length and correctly spaced. It also added tuning patterns, small zigzags that lengthen a track so that signals arrive together.

Anthropic cites two benchmark results. On PCB Gym, an internal benchmark that scores how well a model places components and reroutes connections on boards whose layouts have been removed, Fable 5.2 scored 69%, against 25% for Fable 5. On BenchCAD 2.1, a public test of rebuilding 3D parts and boards from drawings, it scored 83.5%, nearly double its predecessor, Claude Fable 5.1. A footnote in the post also discloses that Anthropic has a commercial relationship with Diode.

Anthropic says earlier models rarely changed course. Over the past year, Diode had built many boards with them that were 95% complete but unusable. On the speakerphone's four-layer board, Claude reassigned signals to different pins on two chips so that the tracks under the touch wheel could stay on one layer. The reason is physical: moving a track between layers needs a drilled hole, and under the wheel that hole would go through one of the touch electrodes. When one area of the back of the board became crowded, Claude swapped signals on interchangeable pins and laid out the whole corner again, rather than adding jumps between layers.

Claude also consulted documentation when a choice depended on a specific part. Asked whether the handheld pet would run without its battery, it read the charger's datasheet. It said the device would run on USB power, but the charging light would flicker, because the charger keeps recharging a small capacitor that stands in for the battery. For the hub's long connector tracks, Claude judged that KiCad's built-in router, which draws one track at a time between two points the user chooses, would need too much step-by-step guidance. It wrote its own grid-based router instead. Anthropic says that took about half an hour and could lay each long track in roughly a second.

Anthropic also describes Claude checking how decisions interact. For the speakerphone, Claude found the board could not be lowered straight into its enclosure because the mute switch would hit the back wall. It worked out that tilting the board about 25 degrees during installation would clear it. For the hub's microphone port, Claude calculated that a 25 mm port would resonate near 3.4 kHz, which it called "right in the voice band" and "not ideal though still usable." A wedge-shaped enclosure shortened the port to about 10 mm and moved the resonance to about 8.9 kHz, at the edge of the range used for speech.

On manufacturability, the post says Claude ran design-for-manufacturing and sourcing checks itself, then sorted the results into problems to fix and flags to ignore. The hub's checks flagged six slivers of solder mask as too thin, which Claude fixed by widening the openings. A ground-to-ground connection was flagged as an error because the checker didn't recognise it as deliberate. Claude judged that it would not affect performance and ignored the flag. Diode then manufactured all three boards directly from Claude's design files, with no adjustment rounds. Diode's team said they were surprised by that, since factories and engineers often work through several rounds of changes.

Anthropic is clear about the limits. All three devices are early prototypes that have not been through electrical validation, antenna tuning, thermal testing or certification, and they were not designed for mass production. Claude was not connected to test equipment or to the machines that make and assemble boards, so Diode engineers did the physical build and testing. The post notes that OpenAI recently wrote about using its models to design a handheld device. Anthropic says its next step is to check whether Claude can design denser, faster and more complex boards, verify them in simulation and learn from measurements of the finished devices.

What changed

According to Anthropic, Claude Fable 5.2 designed circuit boards, enclosures and firmware for three prototype devices, which Diode Computers built as working prototypes without circuit fixes.

Type: Other · Applies to: Claude Fable 5.2, Claude Fable 5.1

Why it matters

According to Anthropic, an AI model can now take a hardware design from concept to a working prototype with little guidance, which could shorten hardware development cycles. The results still come from a vendor-run test with early prototypes, so certification and production readiness remain open.

What you can do

Hardware teams could trial AI-assisted board layout on a low-risk prototype, keeping human review, DFM checks and physical testing in the loop before relying on it for production designs.

Original source

Where we saw it

1 source

Anthropic Research

AnthropicDiode Computers

AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works