HP's Voyager calculators originally ran on custom silicon-on-sapphire chips (which gave them extremely long battery life). After these stopped production, HP changed the calculator design to use an off-the-shelf microcontroller that runs the original firmware in an emulator. This reduces the battery life, but the emulator runs the firmware 60-90 times faster than the original calculators. This is baseless speculation, but I imagine what happened was that they either tried switching microcontrollers or otherwise updated the emulator and introduced an emulation bug that gets exposed by the calculation in the blog post.
Do such long-battery-life silicon-on-sapphire chips (or an alternative) still get produced anywhere today, or is this a lost technology and we have to make do with less battery life now?
Not to my knowledge. But new processors solve the problem differently, by running active code very fast on very small feature size lithography, and then entering deep sleep to achieve really low duty cycles. They do really well, achieving impressive numbers of microwatts per MHz:
I expect that if you wrote code using the power management features of modern platforms, you'd blow that sapphire process on an obsolete node size out of the water.
No idea whether a modern node size on that low-leakage substrate would be any better if it existed...
