[hakfoo]

The other draw of an established "old architecture" is that it's fairly fixed and sourcable.

There are a bazillion Z80s and 8051s, and many of them are in convenient packages like DIP. You can probably scavenge some from your nearest landfill using a butane torch to desolder them from some defunct electronics.

In contrast, there are a trillion flavours of modern MCUs, not all drop-in interchangeable. If your code and tooling is designed for an ATSAMD20, great, but I only have a bag of CH32V305s. Moreover, you're moving towards finer pitches and more complex mounting-- going from DIP to TSSOP to BGA mounting, I'd expect every level represents a significant dropoff of how many devices can be successfully removed and remounted by low-skill scavengers.

I suppose the calculus is different if you're designing for "scavenge parts from old games consoles" versus proactively preparing a hermetically sealed "care package" of parts pre-selected for maximum usability.

[kragen]

It's a good point that older hardware is less diverse. The dizzying number of SKUs with different pinouts, different voltage requirements, etc., is potentially a real barrier to salvage. I have a 68000 and a bunch of PALs I pried out of sockets in some old lab equipment; not even desoldering was needed. And it's pretty common for old microprocessors to have clearly distinguishable address and data buses, with external memory. And I think I've mentioned the lovely "external access" pin on the 8048, 8051, and 80C196 family, though on the 80c196 it's active low.

On the other hand, old microcontrollers are a lot more likely to be mask-programmed or OTP PROM programmed, and most of them don't have an EA pin. And they have a dizzying array of NIH instruction sets and weird debugging protocols, or, often, no debugging protocol ("buy an ICE, you cheapskate"). And they're likely to have really low speeds and tiny memory.

Most current microcontrollers use Flash, and most of them are ARMs supporting OCD. A lot of others support JTAG or UPDI. And SMD parts can usually be salvaged by either hot air or heating the board up on a hotplate and then banging it on a bucket of water. Some people use butane torches to heat the PCB but when I tried that my lungs were unhappy for the rest of the day.

I was excited to learn recently that current Lattice iCE40 FPGAs have the equivalent of the 8051's EA pin. If you hold the SPI_SS pin low at startup (or reset) it quietly waits for an SPI master to load a configuration into it over SPI, ignoring its nonvolatile configuration memory. And most other FPGAs always load their configuration from a serial Flash chip.

The biggest thing favoring recent chips for salvage, though, is just that they outnumber the obsolete ones by maybe 100 to 1. People are putting 48-megahertz reflashable 32-bit ARMs in disposable vapes and USB chargers. It's just unbelievable.

In terms of hoarding "care packages", there is probably a sweet spot of diversity. I don't think you gain much from architectural diversity, so you should probably standardize on either Thumb1 ARM or RISC-V. But there are some tradeoffs around things like power consumption, compute power, RAM size, available peripherals, floating point, GPIO count, physical size, and cost, that suggest that you probably want to stock at least a few different part numbers. But more part numbers means more pinouts, more errata, more board designs, etc.