Doing electronics stuff without using a processor is getting weirdly difficult. I mean it was already quite a lot of work to do things that would take no time flat given any microcontroller, but I think it's now also harder than it was back in the day because all those specialty ICs in the niche between plain logic gates and processors are going away.
@danderson I think the flip-side is, apart from the joy of doing things using non-MCU past-tech (which certainly exists!), modern MCUs are so crazy cheap, flexible, and available, that in most cases there are no shortage of decent options in the $1-5 range (and much cheaper in quantity) for throwing together little MCU-based solutions to problems.
The software tooling has been slower to catch up, but it is nonetheless so much better than it was a decade or two ago.
@swetland Oh absolutely. My entire thread comes with the caveat of "if processors are off limits". It's an extremely weird constraint and if you drop it, it's an absolute buffet of wildly powerful chips you can just buy.
I've spent a bunch of time on stm32f0 MCUs, and was amazed at the bang for buck you get out of them. Then I saw the stm32h7 that just low-key gives you 2 cores at 600MHz, and at least two copies of every peripheral imaginable including some I'd never heard of. Wild.
@swetland I immediately assumed that would be one of those "call us to discuss your requirements" thing, but no, digikey will just sell me one of these ridiculous things, for $10-20 depending on how much stuff I want.
Okay admittedly the dual-core version that comes with a Cortex-M4 sitting on the side of the primary Cortex-M7 to help with realtime stuff, that's only available as full reels. But if you can find a reason to own 500 of them... Also $10-20 depending on what trim you want. Wild.
@danderson Honestly one of my personal favorites is the RP2040 with its pair of 125MHz M0+s, generous 245KB of on-board SRAM, and nifty PIO blocks that give you little state machines that can be a reasonable alternative to an FPGA solution for simple peripheral interfaces. These critters are $1/ea in singles of $4/ea in singles of a PCB (rPi Pico) with MCU + Flash + Clock, etc. It's got some quirks but it's crazy flexible, inexpensive, and pretty fun to work with.
@swetland Yeah I have the RP2040's datasheet open in one of these tabs. In its price category it's a remarkable design! The future is definitely very cool, when you're not deliberately tying your legs together and taping one arm behind your back. Alas, here I am 😂
Like, display a hex digit on a 7-segment display for example.
Back in the day: grab a 7-segment decoder IC with 4 bits in, 7+1 drive lines out, done. Cheap, common, easy.
Today, if you're being normal: grab any old Cortex M0 micro and just bit-bang the display module directly. Maybe through a port expander or, if you're really fancy, a 7-segment driver IC that speaks I2C and runs like a dozen digits at once, with dot support and like 4 bonus GPIOs on the side because the package had free pins.
Today, if you can't have a processor: you can still buy some parallel input 7-segment decoders, but they only do BCD (0-10).
You can get decoders with hex support, but they speak serial or I2C in ways that are difficult to drive without software.
You can get exactly one 7-segment IC that does hex and parallel input. It's actually quite fancy and can drive multiple digits and other bells and whistles. It's not trivial to interface with, but just easy enough to do with hardwired circuitry.
It's also $20, and literally the last of its kind. Okay technically there's 3 companies selling it, but they're all "second sources" for an original source that went out of business decades ago. I can only assume there's some military control panel somewhere that uses them, which justifies keeping them in production and charging the price of two entire 400MHz SoCs for it.
Or... you can build your own decoder, out of a couple dozen raw logic gates and a nest of wires.
Oh, and for 7-segment decoding specifically, you can also use an EEPROM as a cheeky decoder: input bits as the address, output value is the drive signals for the display.
But it's also starting to get tricky to find EEPROMs with parallel I/O, everthing's SPI these days to feed the newfangled microcontrollers and SoCs. And you also need to program the EEPROM, which is certainly doable without a processor, but well off the beaten path these days.
@danderson I think the answer is we need to hold out for another decade or two until hobby ASIC fab gets to a suitable level of affordability to start re-manufacturing the classics of the 70s (and modern variants on their themes).
I can dream! Maybe that'll end up being a nice hobby for me in my 60s or 70s.
@danderson@swetland Are you aware of #TinyTapeout? It’s maybe not exactly what you want here, but an indication things are going in the direction of small hobby ASICs being possible.
@mike@swetland Yes, it's a very cool program! It drastically reduces the amount of ceremony (and money, and NDAs) needed to even start touching ASICs.
Last time I looked at it, I didn't really have anything I wanted that would make sense given the cost per gate and the 6-12 month lead time (my attention span often struggles to last a week :( ). But also, it looks like the recent runs have a different set of features/constraints, so may be worth another look!
@danderson@mike The most recent iterations move away from the serial interface for IO, giving you a clock pin, reset pin, 8 inputs, 8 outputs and 8 bidirectional pins available when your design is enabled. Sounds like 66MHz potential max clock rate: https://tinytapeout.com/specs/
The main limitation is that there's not really a simple path to getting your design back in quantity. Very cool to actually be able to get a design implemented in silicon, but it's pretty much just a single prototype.
@swetland@mike yeah I remember this being firmly for one person wanting one or two of their design, not "I could sell 1000 of these but that's several orders of magnitude too low for anyone to pick up the phone." Maybe that's just a sign that we need to keep generating a market for fpgas like the ice40 and ecp5...
@swetland@mike I guess the next step up would be going straight to efabless and doing a whole run for yourself. That lets you have 300 chips for $10k, so 35/ea give or take. And you can scale up to 1000 chips for 20/ea... Or run a z80 emulator on an rp2040 for 1/ea :/
Add comment