Devices die socially before they die technically.
A lot of electronics become 'obsolete' because the people, notes, and build paths scatter. The silicon often still has plenty left to give.
Measured Studios · Public program
Silicon gets a second act.
Second Silicon is Measured Studios' public home for the Freecycled worldview: older electronics as recoverable creative and computational material. We use LLMs, coding agents, and web archaeology to reconnect the manuals, forum threads, and toolchains that make those devices buildable again.
5case files in progress
12+candidates researched and scored
Greenpublic discipline that keeps the work credible
The thesis
“Most devices called obsolete are not technically dead — they died socially, when the knowledge to work on them scattered. AI changes that.”
Modern LLMs and coding agents are very good at the work that kept older hardware out of reach: reading decades of forum archaeology, translating vendor SDKs across eras, summarizing datasheets, and generating build configs that actually compile today.
Second Silicon is where we practice that work in public. Each case file pairs what the research recovered with what the bench still demands, so the next builder gets a clearer on-ramp instead of another myth that AI solved embedded engineering overnight.
The Freecycled worldview
Older hardware becomes usable again when the knowledge becomes reachable again.
Second Silicon carries a bigger belief than one repo or one studio page: millions of devices are still capable, but the path back into them is fragmented across manuals, forgotten repos, archive snapshots, teardown notes, and forum archaeology. AI helps recover that path.
Manuals, forums, repos, and notes are latent infrastructure.
The value is not just in the board. It is in the hidden map around it: support pages, issue threads, community guides, and archived vendor knowledge that can be reconnected into a working path.
AI widens participation instead of replacing expertise.
LLMs can surface and sequence the path, but they do not replace probing, soldering, firmware judgment, or knowing when a device is not worth the risk. The win is a bigger on-ramp for makers, artists, students, and curious engineers.
That is the Freecycled idea inside Second Silicon: reclaim access to older hardware honestly, publish the path clearly, and leave enough evidence for the next person to continue.
What AI actually changes
Research cost collapses. Debug cost doesn't.
We publish the distinction so builders, collaborators, and future engineers know exactly what modern tooling earns and what it does not.
AI collapses the research cost.
- Finding and sequencing obscure vendor manuals.
- Translating forum archaeology into a linear build path.
- Drafting register-level code from datasheets.
- Generating toolchain configs for modern Rust or C targets.
- Converting vendor SDK examples across toolchain eras.
- Explaining boot flows once the logs exist.
AI does not collapse the debug cost.
- Reading oscilloscope traces when nothing boots.
- Bring-up debugging on silent hardware.
- Bricked-device recovery.
- Undocumented timing that only a probe reveals.
- Physical skill — solder, JTAG, reflow, thermal care.
- Knowing when to stop before you damage the board.
The pre-bench loop
From prompt to probe.
Simulation creates an evidence checkpoint between an agent's proposal and physical hardware. It can retire obvious mistakes and clarify what the bench still needs to prove, without turning a virtual result into a hardware claim.
01
Recover primary documentation
Find the manuals, datasheets, toolchain references, and community evidence.
02
Propose firmware and circuitry
Let the agent produce a reviewable implementation hypothesis, not a fact.
03
Compile and simulate
Exercise the proposal in an emulator or circuit simulator when coverage exists.
04
Record evidence and limits
Publish what passed, what failed, and what the simulator could not represent.
05
Verify on physical hardware
Move the named test to the bench and keep physical uncertainty visible.
06
Publish the delta
Document every meaningful difference between simulation and real silicon.
Scope discipline
Green only. We link, we do not circumvent.
The scope stance is public because movements lose credibility when their boundaries are fuzzy. Every case file declares its class. We only publish work that does not require DMCA §1201 circumvention.
Status is honest, not aspirational. A case file reads documented until a milestone is verified on the bench.
Community-sanctioned paths.
Nintendo DS (BlocksDS), Game Boy Advance (agb), PSP (rust-psp), Kindle (KOReader), end-of-life OpenWrt routers, rooted Androids with vendor unlock, ESP-era boards with official esp-rs.
Case-by-case judgement.
Older car head units (non-traction), printers with undocumented firmware, e-readers outside vendor jailbreak tolerance, consumer IoT with closed but tractable firmware. Each entry names its specific concern.
Out of scope.
Current-gen consoles, iPhone secure enclave, medical devices, ICS/OT, Tesla MCUs on bench power, anything with active vendor takedowns in the last 24 months. If meaningful work requires §1201 circumvention, we do not publish it.
Current entry points
Five devices, five different ways into the program.
The case files span different toolchains, different eras of forum archaeology, and different kinds of LLM leverage — from a bounded ATtiny85 simulation-to-bench pilot to 16-bit co-processor archaeology, ARM handhelds, and E Ink. They are documented-first on purpose: honest starting states lower the barrier for the next builder.
2005 · Atmel / Microchip
ATtiny85 simulation-to-bench pilot
greendocumented
One small test, one explicit evidence chain.
The first paired pilot starts with a named PB0 blink-and-timing test: plan it in Velxio, repeat the same source on a physical ATtiny85, and publish the measured delta. The case file is real; its initial state is intentionally honest—no compile, simulator run, or bench result is claimed yet.
Toolchain
Velxio · Arduino CLI · ATTinyCore · AVRDUDE · ISP
2004 · Nintendo
Nintendo DS (Lite / Phat)
greendocumented
Dual-screen ARM handheld, refreshed toolchain.
BlocksDS replaced the older devkitPro-only path, and a Rust workflow via blocksds-sys lets you target the DS from a modern toolchain. LLM leverage is in translating a decade of devkitPro forum answers into working BlocksDS builds today.
Toolchain
BlocksDS · devkitARM · Rust nightly (armv5te-none-eabi)
Dual-screen Rust demoGBATEK translationEmulator-first via melonDS
2001 · Nintendo
Game Boy Advance
greendocumented
The cheapest path into real ARM silicon.
A clean-room Rust toolchain (agb) abstracts the memory-map math while leaving registers visible. Ideal for demonstrating AI-assisted DMA timing work, sprite-sheet pipelines, and GBATEK register translation into modern abstractions.
Toolchain
agb v0.23 (Rust) · rust-console/gba · devkitARM (C fallback)
Sprite demoDMA + IRQ explainermGBA developer tooling
2015 · Amazon
Kindle Paperwhite 3 / Voyage
greendocumented
E Ink silicon with a decade of forum archaeology.
KOReader keeps this era of Kindle alive; MobileRead holds the sequencing knowledge. The LLM-sweet-spot is summarizing scattered jailbreak threads into a current working path for a specific firmware version. Explicitly no DRM-removal topics.
Toolchain
KOReader · MobileRead community guides · cross-compiled C / Lua
Forum thread sequencingKOReader installRepurposed reader build
1990 · Nintendo
Super Nintendo Entertainment System
greendocumented
16-bit icon with a co-processor story built into the cart.
The SNES offloaded compute to the cartridge itself — SA-1, Super FX, DSP-1, and CX4 co-processors lived on the game board, not the motherboard. Fabien Sanglard's clock and cartridge teardowns plus the FullSNES register reference give the clearest modern entry. Homebrew routes through PVSnesLib (C) and WLA-DX with bsnes for cycle-accurate debug.
Toolchain
PVSnesLib (C) · WLA-DX · bsnes · Mesen2
Each entry point carries a declared status, legal class, toolchain, and next milestones. The repo is curated and documented first; bench notes accumulate over time as work is actually verified.
Join the program
Two ways into Second Silicon.
Second Silicon is meant to be useful whether you are here as a builder following the field notes or as a team looking for the same research discipline on your own hardware work.
Community lane
Start with the documented entry points, follow the research surface as it matures, and send the next device, note, or weird old platform that deserves a second act.
Studio lane
If your team is evaluating agentic AI against embedded, instrumentation, salvage, or legacy-system research, Measured Studios can bring the same discipline to your own program.