The Hook. In a retro-technical teardown of the first BlackBerry device, a detail flashed by that an engineer's eye simply cannot miss: inside a pager from 1996–1998 sits "a real, full-fledged Intel 386EX" — an embedded version of that very legendary 386 on which the entire x86 kitchen was built. In a pocket device from the '90s — a processor architecture that was already considered museum-worthy by then! But the tastiest part came later, in the wiki article about the chip: the 386EX "was used aboard several orbiting satellites and microsatellites" and was part of the NASA FlightLinux project. Meaning the same conservative, long-"obsolete" silicon worked both in a businessman's pocket and in the vacuum of orbit. I wanted to understand how this happened — and why it's actually a brilliant engineering pattern, not a curiosity.
In 1994, Intel released something strange at first glance: a variation of the 80386SX intended for the embedded market — the i386EX. By high standards, the processor was already outdated: 16-bit external bus (like the SX), core up to 33 MHz, addressing up to 64 MB of memory. The Pentium was already selling like hotcakes, and it seemed that the 386 was a story about museums and recycling.
But the essence of the 386EX wasn't in performance, but in architectural density. Intel packed onto the chip essentially half of an IBM PC motherboard:
The peripherals were compatible with standard PC/AT components. The result was a system-on-chip for $39 (in batches of 5000, end of 1993 data): instead of routing a scattering of discrete chips around the processor — one package, minimal support circuitry, predictable behavior.
And here are two properties that meant nothing for desktop but meant everything for embedded systems.
First — static core. The core can be clocked arbitrarily slowly — down to completely stopping the clock — without losing state. For desktop this doesn't matter, the fan's going to whir anyway. But for a system on an AA battery, this is the main trick: the processor sleeps at zero frequency, wakes on an incoming packet, processes, and goes back to sleep. This is exactly how the pager lived on a single battery for two weeks.
Second — reputation. The 386 — Intel's most polished, most widely deployed design: millions of machines, thousands of applications, decades of field testing. In the engineering culture of embedded systems, this turned into the argument "we know everything about this silicon."
The wiki phrase "used aboard several orbiting satellites" turned out to be verifiable — and this is where truly beautiful things begin.
SUNSAT (Stellenbosch University, South Africa). South Africa's first microsatellite, launched in 1999, was built on a budget under $2 million by students. Its flight documentation explicitly states: "General flight management tasks — scheduling, CCD imager and communications management — are performed by an Intel 386EX processor, backed up by an Intel 80C188EC." And the T800 transputer dedicated to the precision orientation system, in case of failure, also handed tasks back to the 386. So in the fault-tolerance architecture of the South African satellite, the "Soviet" in spirit principle of "main + cold backup" was implemented through a pair of gray Intel chips.
UoSAT-12 (SSTL, Surrey). On this 350-kilogram research vehicle, SSTL tested autonomous orbit control (Orbit Control Kit from Microcosm with AFRL participation): autonomous maneuvering software lived "on a customized 386 onboard computer, developed by SSTL." For the first time in history — a demonstration of autonomous satellite station-keeping without a human in the loop — and doing it with hardware from the desktop segment from a decade earlier.
NASA FlightLinux. Pat Stakem's project (Goddard) was going to make Linux a flight OS for science satellites, and the 386EX was listed as one of the target platforms. The project never actually saw flight: funding ended June 30, 2002 with the language "not enough time for a flight demonstration," and the source code remained under ITAR restrictions. But the gesture itself matters: NASA seriously considered mass-market x86 for spaceflight.
Why did this work? Not because the 386EX was a "space" processor — it wasn't, it had no radiation hardening whatsoever. The answer lies in microsatellite economics: classic space CPUs (RAD6000, RAD750) cost hundreds of thousands of dollars per unit. A $2 million microsatellite cannot afford a CPU more expensive than the most expensive instrument on board. Instead, engineers took commercial off-the-shelf and wrapped systemic protection around the chip: watchdog, triple redundancy, EDAC memory, two-machine agreement. One processor failure didn't kill the mission — it was simply bypassed. The "cheap commercial hardware + smart redundant architecture" scheme became an SSTL classic and later spread throughout the world of cubesats. The 386EX here — the perfect illustration: a conservative, universally known, cheap chip trusted with a spacecraft.
The second and less known front — pocket devices. For the 386EX, this was also not a niche for fun, but a logical application of its main quality — self-sufficiency.
RIM Inter@ctive Pager 900 (1996) and BlackBerry 950/957 (1998–2000). Research In Motion from Waterloo, Canada started not with smartphones but with two-way pagers. The Inter@ctive Pager 900 (aka RIM-900) worked on the Mobitex network and was first in the industry to give the world things we think appeared much later: peer-to-peer message delivery, read receipts, and even text-to-speech. The unit verified at York University Computer Museum — 4 MB flash, full keyboard, one AA battery.
And then — the main thing. BlackBerry 950/957 were built around the same Intel 80386EX: 4 MB flash, 512 KB SRAM, 132×65 monochrome display, AA battery for two weeks, and a 900 MHz radio modem for Mobitex. RIM OS used the 386's MMU for memory protection — cooperative multitasking with message passing, C++ applications in PE format (yes, the same as Windows), which were sideloaded through Desktop Manager, linked with code on the device, and executed directly from flash. The system was written by people who grew up on PC architecture — and they got PC-compatible logic in their pocket, without IBM PC compatibility overall: the device called itself x86, but BIOS, PCI, and the entire desktop stack were absent as a class.
Only in 2002 (models 5810/5820) did RIM move to ARM7 and a Java environment — six years before the first Android, which was also built on Java over ARM, by the way. And the Nokia 9000 Communicator — the first "smartphone" slider — carried an honest Intel 386 at 24 MHz with GEOS/ROM-DOS on the PDA side.
The paradox: millions of people in 1998–2002 carried an i386 in their pocket without knowing it. The last mass-market 386 device in the world turned out to be not a computer, but a mail pager.
This story has a dramatic tail, without which it would be incomplete. The device with a 386EX heart in the winter of 2005–2006 nearly disappeared from the US entirely — and the culprit wasn't the chip, but the concept.
In 1990, Chicago inventor Thomas Campana Jr. (son of a milkman, Air Force sergeant, about 50 patents over his career) patented a series of ideas for wireless push email. In 1992, he and patent attorney Don Stout — a man who kept all papers in an ordinary filing drawer — created the holding company NTP (New Technologies Products): no employees, no office, no product, only patents. When BlackBerry took off, Stout offered RIM a license; RIM, according to Lazaridis, conducted an internal review and replied that there were no infringements — a written record of the response was never presented in court. In November 2001, NTP filed suit in the "rocket docket" of the Eastern District of Virginia.
What followed — a verdict worth remembering for any engineer: the jury figured it out in three hours. RIM's demonstration of a "prior" SAM system failed — the software in the demo turned out to be younger than Campana's patents, Judge Spencer later called the defense's actions "questionable" and even "fraudulent." They found the infringement "willful": royalties jumped from 5.7% to 8.55% of revenue, plus damages and a ban on BlackBerry sales in the US. The only thing saving the market — escrow and appeals. The moment of truth came when the DOJ and US Department of Defense stated in briefs: shutting down BlackBerry in government is unacceptable — this was, it seems, the first case where a patent dispute literally threatened to shut down email for the White House, Congress, and the Pentagon. The Supreme Court refused to hear the case, and on March 3, 2006, RIM paid NTP $612.5 million in "full and final settlement" — about $978 million in today's money.
A bitterly ironic ending: Campana died of cancer on June 8, 2004, not living two years to see the victory. By that point, the USPTO was already stamping his patents as invalid on reexamination — but the legal fiction "valid until appeal is heard" did its job. None of this had anything to do with the 386EX silicon: the conflict was about network layer ideas. Hardware built on the most mass-market and "obvious" processor in the world turned out to be hostage to the most invisible things — patent claim formulations.
The most non-obvious thing in this story — the fate of the direction. The 386EX turned out to be "substantially more successful than its predecessor, the 80376" (wiki), scattered across Garmin GPS III+, GPS 48, GPS 12, Globalstar/Qualcomm GSP-1600 satellite phones, USRobotics Courier I-"everything" modems, Swarco ITC-2 traffic light controllers, Sperry Marine gyrocompasses, and Ziton ZP3 fire panels. This really was one of the first (if not the first) Intel processors with a memory controller and clock right on the chip — 15 years before "integrated northbridge" became mainstream.
And yet Intel itself didn't notice. After the 386EX, the company did not release a single integrated x86 processor for thirteen whole years — until Tolapai (EP80579) in 2007. Even Intel itself, investing in the Pentium/MMX race against AMD, wasn't looking where its own product was quietly flying in space and lying in pockets. Embedded x86 was ultimately revived not by the architecture's creators, but by AMD with Geode and later Intel with Atom — but that's a different page of history, and it essentially starts from scratch, not from the 386EX legacy.
So, what's the point of all this. The 386EX story — a textbook lesson that silicon "obsolescence" is not absolute, but a question of context.
First. The 386 had a unique combination of qualities: huge knowledge base, simplicity, cheapness, complete documentation. For a workstation this meant nothing anymore — but for a pocket device on one battery and for a $2 million microsatellite, this was decisive. The right architecture — not the fastest, but the one with exactly the risk and cost profile your system needs.
Second. Radiation resistance — a systemic concept, not just a silicon one. A chip with no radiation hardening whatsoever flew in orbit and controlled satellite orientation because they built redundancy, watchdog, and "two-machine agreement" around it. Space reliability — a property of system architecture, not just lithography. This is exactly why small satellites today increasingly fly on consumer hardware with software wrapping, while large interplanetary probes — still on $200K RAD750. Both approaches are correct — for different risk budgets.
Third — and this is my favorite layer. The fate of the 386EX — the fate of all "boring" technologies. While the market watched Pentium megahertz, the veteran chip quietly carried businessmen's mail, traffic lights, gyrocompasses, and South African science in orbit. Without any fanfare. It's precisely in this quiet utilitarian role that real architecture testing happens: not in benchmarks, but in whether it survives 30 years after being declared obsolete.
Paraphrasing one Habr teardown of the BlackBerry 957, we can say simply: Intel's most underestimated architecture was carried in pockets and placed in orbit. Respect for code — and for silicon — must be earned. The 386EX earned it.