Before he could graduate, Grigori’s Russian university program required him to complete a large-scale, real-world project. Like most of his peers, he planned to use this as an opportunity for job experience, which meant partnering with an outside company. Since Grigori did low-level development and microelectronic engineering, he found a paid internship position with the Russian Automation Institute. RAI has one major client: the company responsible for managing Russia’s nuclear reactors and supply parts for nuclear weapons.
Before Grigori could start working, his soon-to-be mentor assigned him a “test”. “Before you begin, you must implement this conversion.” The conversion in question was to turn IEEE754
floating points into a “secret” format. The spec document was a three column spreadsheet- a floating point number, a binary32 floating point number, and the “secret” format.
“Can I have more details?” Grigori asked Aleksandr. “Is there any documentation about the format?”
“Not that you can have, no. We use custom CPUs, and they are very secret. You must work from this document.”
It wasn’t a huge challenge. By comparing the columns in the spreadsheet, Grigori was able to discover that this “secret” format used a 6-bit exponent instead of binary32’s 8-bits. He handed his program over to Aleksandr, who showered him with praise. “Amazing. No one else has done this yet. This will go into immediate use.”
Grigori, still young and naive and lacking job-experience, just accepted the praise with a smile. He didn’t think about what it implied. Instead, still beaming from the ego boost, he showed up to work the following Monday.
The building he worked in had been designed in a particularly paranoid phase of the USSR. It was an anonymous, windowless cube, surrounded by a twisting intestine of barbed wire and security guards. After a series of searches, ID checks, logbook sign ins, and lots of dark look from men with automatic weapons, the bowels of the security apparatus released Grigori into the building. Aleksandr floated by the door, waiting for him.
Aleksandr didn’t give him a tour, except to make clear that Grigori should only ever open the door to his office and the restroom, and under no uncertain terms should ever touch anything else in the building. Grigori’s office looked more like an interrogation chamber- there was a single folding chair, a desk that had last seen use in 1967, and a naked fluorescent bulb. A pair of binders were chained to the desk. They were the documentation.
“Your computer comes soon. We got you a new one,” Aleksandr said. “In the meantime, read the documentation.”
The documentation was surprisingly helpful, in part because it hadn’t been made by RIA. RIA had purchased the chip designs from a third party in the mid–90s, and had spent the next 15 years getting their own assembly line off the ground. The chips were a fairly standard 80186, aside from their unusual floating point implementation.
When Grigori’s computer finally arrived, with its 13" CRT and 256MB of RAM, his first task was to move some of the documented circuit diagrams to SCADE. It was dull and brainless, and after a week, Grigori was ready to quit. That’s when Aleksandr called Grigori and several other interns in for a special project.
“Our circuit designs haven’t changed since we bought them,” Aleksandr explained, “but they are very expensive to manufacture. We must analyze the designs and find ways to remove components.”
“Is that really going to make them cheaper?” Grigori asked. “Won’t that change entail a lot of testing? Won’t the whole assembly line process need to be retooled? And what about building new test masks for the boards?”
“We are engineers, not accountants. We must find ways to make the boards cheaper.”
Over the next few weeks, they started to find ways to remove components. Grigori identified one mildly expensive capacitor that would be replaced. His peers found some resistors that could be safely removed. Aleksandr took that as inspiration, and decided that instead of having a pull-down resistor for each sensitive component (to protect against variations in current), he could have a single resistor on the entire board which would protect all the components.
Pleased with the success of his dream-team, Aleksandr promised them all a bonus, then sent the designs down to manufacturing. Manufacturing immediately sent them back, and along with their objections:
- The new designs would require extensive testing before they could be used.
- Entirely new test-masks needed to be made, which would cost millions OR they’d need to do a slow and expensive manual test of key components.
- The line would need to be retooled OR many assembly steps would have to be manually performed.
Aleksandr looked at these objections. “Testing? These are little modifications. Our designs are good. I vouch for them as safe. And manual assembly? Manual testing? I have a solution for that.”
His solution was to shove Grigori and the other interns behind desk with an oscilloscope, multimeter, and soldering iron. “You need this to graduate. Do the work.”
The promised bonus never arrived, but Grigori did eventually escape and graduate. It was at his graduation that Grigori saw Aleksandr again. The mentor had cornered the dean of the engineering school. “You send us good students, but could you maybe send us more next semester? A lot more?”