Buried in the silt of the Forth and Clyde Canal in Scotland lies a box of custom silicon that was right about almost everything — but simply arrived four decades too early.
In 1988 a Glasgow hi-fi company shipped a processor that checked the type and bounds of every memory access in hardware, garbage-collected its own heap in silicon, and treated memory and disk as a single persistent object store. Today those features are being shipped in production chips from Arm. The processor's story is no longer just a curiosity about a failed hardware venture — it is a case study in a computing era whose economics have finally reversed.
The processor — the Rekursiv, built by Linn Products, the Scottish firm famous for its Sondek LP12 record player — moved security and persistence out of the programmer's responsibility and into the silicon. Every memory access was checked for type and bounds at the hardware level, eliminating entire classes of memory corruption bugs. A built-in garbage collector freed memory automatically, and a persistent object store meant programs could stop and restart without reloading from disk. In 1988 the chip cost far more than any market could justify. Today, as memory-safety vulnerabilities are a leading cause of exploits and software complexity keeps growing, the cost of doing the same work in software is far higher than the cost of doing it in silicon.
The Rekursiv was not a toy and it was not a misunderstanding of what computing needed. Its creators were designing against the dominant von Neumann architecture — where programs and data share one memory and where safety is bolted on top — and proposing a different, safer model. They lost not because they were wrong, but because a single custom processor could not compete with cheap commodity chips on price, even if it did far more of the right work in hardware.
That economics gap has inverted. The demand for safe, reliable compute is now enormous, and the tools to build such silicon are far cheaper than in the 1980s. The ideas that ended up at the bottom of a Scottish canal are finally the ones ending up in the chips that run the world. It is a reminder that in computing, the difference between a breakthrough and a footnote is often not whether an idea is correct, but whether the market can afford it.
Knowledge takeaway: memory safety in hardware eliminates whole classes of software bugs before they become exploits; a persistent object store unifies RAM and disk so programs survive restarts without reloading; the economics of putting safety into silicon that once made the idea impossible now make it profitable.