The editorial argues the Z80's survival isn't sentimental but structural: it won by being a strict superset of the dominant CPU, simplifying power requirements, and shipping exhaustive documentation. Those traits — backward compatibility, engineering simplicity, and debuggability — are precisely what modern chip design deprioritizes in favor of performance and feature velocity.
The editorial points to concrete active deployments — TI-84 Plus calculators, Roland MT-32 revivals, boutique European synth builders, industrial controllers, and arcade restoration kits — as evidence the Z80 keeps winning designs on merit. The parts are cheap, the datasheets are exhaustive, and any engineer over 40 can debug them with an oscilloscope, which is a real economic and operational advantage.
By submitting the IEEE Micro piece framed as 'still alive,' the poster signals that the Z80's continued production is the noteworthy hook — not its history. The submission's traction (129 points) implies the developer community agrees the ongoing shipment of 1970s silicon in 2026 is itself the story worth surfacing.
The IEEE Micro piece emphasizes Faggin's specific design choices: one power supply versus the 8080's three, built-in DRAM refresh that eliminated support chips, an alternate register set, and source-level 8080 compatibility that inherited the entire CP/M software catalog for free. These technical decisions are framed as the reason the Z80 displaced its predecessor and seeded a fifty-year dynasty.
The IEEE Micro article making the rounds on Hacker News is not really news — it's a reminder. Federico Faggin's Z80, designed at Zilog starting in 1974 and shipped in July 1976, is still in production. Not in a museum, not as a retrocomputing curiosity, not as an FPGA soft core someone drops into a hobby kit. Actual silicon Z80s and Z80-family derivatives are still being fabricated, sold, and designed into new products fifty years after tape-out.
Zilog formally discontinued the standalone Z80 die in April 2024, but the eZ80 — a pipelined, backward-compatible descendant that runs at up to 50 MHz — remains in production, and third-party clones and licensed cores continue to ship in industrial controllers, graphing calculators, MSX-style hobbyist boards, arcade restoration kits, and a surprising amount of embedded audio gear. The TI-84 Plus, which every US high schooler has held at some point, still runs a Z80-family core. The Roland MT-32 sound module, revived by hobbyists as MUNT, is Z80-driven. Newer synths from small European boutique builders still spec Z80s because the parts are cheap, the datasheets are exhaustive, and any engineer over 40 can debug them with an oscilloscope and a printout.
The IEEE Micro piece traces the technical lineage — the Z80's clever use of a single 5V rail versus the 8080's three supplies, the built-in DRAM refresh, the alternate register set, the source-level compatibility with Intel's 8080 instruction set that let CP/M and its enormous software catalog run unchanged. Those were the 1976 selling points. What's interesting in 2026 is which of them still matter.
The Z80's survival is not sentimental. It's a case study in what actually gives a technology decades of legs, and almost none of it is what the industry currently optimizes for.
The Z80 won its original market because it was a strict superset of the dominant CPU (the 8080), needed one power rail instead of three, and had a manual that a competent engineer could read in a weekend. Compatibility, simplicity, and legibility. Not performance — the 6502 was cheaper and, clock-for-clock, often faster on real workloads. Not elegance — the instruction set is a baroque mess of undocumented opcodes and index-register hacks that assembly programmers still argue about. It was, however, unreasonably easy to design a working board around, and once CP/M and later the Sinclair, Amstrad, MSX, and TRS-80 ecosystems piled on top, the software gravity became self-sustaining.
What kept it alive past its commercial prime is a different property: the Z80 is small, deterministic, and has no surprises left. Every timing edge case has been characterized in print. Every errata is known. There is no microcode update, no speculative execution side channel, no firmware blob, no vendor telemetry, no BGA package that requires a $40,000 rework station. If you are designing a medical infusion pump, an industrial valve controller, or a nuclear-plant panel that has to be certifiable for the next 25 years, a chip whose behavior is fully documented and whose supply chain includes multiple second-source foundries is worth more than another 100 MHz.
Compare this to any modern SoC: the datasheet is under NDA, the reference manual is 4,000 pages, half the peripherals have silicon bugs the vendor patches with kernel workarounds, and the part goes end-of-life in seven years. That is fine for a phone. It is a disaster for a factory floor with a 30-year depreciation schedule. This is why the embedded world is still full of 8051s, PIC16s, MSP430s, and yes, Z80s — parts that are boring enough to be trustworthy.
The Hacker News thread on the article is worth reading for the anecdotes: a commenter designing new Z80 boards in 2025 for arcade PCB replacements; another running a small business selling Z80-based MIDI hardware; a firmware engineer at a European industrial-automation company describing how their PLC line still ships new Z80-derivative products because the recertification cost of switching architectures exceeds the entire product margin for a decade. This is the hidden long tail of computing that never shows up in AWS revenue charts.
Most readers of this site are not going to design a Z80 board next quarter. The applicable lesson is about architectural choice, not silicon.
When you pick a database, a runtime, a framework, or a cloud primitive, you are implicitly betting on how long its behavior will remain legible to the humans maintaining it. The Z80's fifty-year run happened because the mental model fit in one engineer's head, the documentation was complete, and there was no hidden state. Systems with those properties compound in value; systems without them accumulate technical debt at a rate that eventually exceeds their utility.
Concretely: prefer boring Postgres over a 'purpose-built' database that will be acquired and sunset in four years. Prefer a small, well-documented job queue you can read end-to-end over a distributed workflow engine whose failure modes require a support contract to diagnose. Prefer a monolith you can attach a debugger to over a mesh of microservices whose observability story depends on a vendor's proprietary trace format. The Z80 principle is not 'use old things.' It's 'the surface area of what you are trusting should be small enough that a smart person can hold all of it at once.'
The second-order lesson is about second-sourcing. The Z80 outlasted Zilog itself as an independent company (it's now part of Littelfuse, a passive-components conglomerate) partly because the design was licensed, cloned, and second-sourced from day one — NEC, SGS-Thomson, Sharp, and later Chinese foundries all shipped compatible parts. In software terms: prefer standards with multiple independent implementations over any single-vendor runtime, no matter how convenient the developer experience.
The eZ80 will probably still be in production in 2035, and someone will still be posting to Hacker News about it. The interesting question isn't whether the Z80 survives — it's which of today's dominant technologies will look like the Z80 in 2075, and which will look like Itanium. The honest answer is that most of what dominates the current stack — the specific LLM APIs, the current cloud primitives, the framework of the month — will be gone. The parts that survive will be the ones with small, legible interfaces, complete documentation, and multiple independent implementations. Postgres. SQLite. POSIX. HTTP. Maybe Linux. Bet accordingly.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.