Thread by @alexandrosM

Variants are evolving. It may get better or worse.
Supply chains in chaos. Waiting times for some components now touching 2 years. It may get better or worse. As we hope for the best, let's plan for the worst. What happens to supply chains if a variant mostly escapes vaccines?

If our electronics ecosystem of 2019 was like an animal evolved for times of plenty, what does an electronics ecosystem evolved for famine look like? There will certainly be disruption and short term chaos, but afterwards, I suspect things may not be as bad as we fear.

I've been digging into the hardware supply chain with team balena for a few years and what I saw was disheartening. Many components have serious lead times (dozens of weeks) and high "minimum order quantity". Just this one fact makes hardware incredibly hard.

To make a custom piece of electronics, you need to design, prototype and certify, establish a supply chain, a manufacturing partnership, deal with inventory and distribution. And that's before we get into shipping and customs. And each of these is a mess and will break at will.

But it seems the pain wasn't painful enough, because we build elaborate hardware, with thousands of components. When we made our first board, the Fin, I asked our engineers to explain why we need so many tiny components on a pcb. From a software background it looked wasteful.

It turns out, every little component often does something very specific like delay a voltage by a couple milliseconds at boot. Many of these things can be done in software, or standardized. Some of them we're not even sure are necessary, but we add them anyway, just to be sure.

Why would we do things that we're not sure are necessary? Because iterations are so slow, we work like each iteration will be the last. There's no time to optimize, get more out of the hardware. Each chip, lookong at the data sheet, has several cool features, we don't use them.

Why don't we use cool features we're already paying for? Because hardware design is so cautious, you only connect the absolute minimum. Remember, every iteration is hoped to be the last. You don't want to mess it up.

What happens if the cornucopia of components dries up? We'll have to get a lot smarter, fast. What we learned over 40 years in software will have to come crashing into the world of hardware. The incremental improvements so far have staved the paradigm shift off, but no longer.

Here's a few things hardware will have to do to survive the world of thin supply chains:

1. Move as many components as possible to software. Use a general purpose microcontroller to code up a lot of that logic you now use components to realize. Io expanders, RTCs, muxes,...

... and every component you remove also drops supporting components, power circuitry, pcb space, etc.

2. Use edge connectors and other smart pcb-only board-to-board connectors. Implement usb with PCB. fewer components, less space, less cost.

3. Modularity and extensibility with daisy chaining. Things like CAN bus and i2c take an arbitrary amount of peripherals. Thus, you don't need to know in advance how many extension slots you'll need. Each peripheral adds an extension slot. Less components, less cost, less space.

4. Use polymorphic components. FPGAs and software defined radio are coming down the cost curve. Instead of single purpose chips, they cover many needs, current and future

5. Standardization: supply chains are a problem as long as components are not commoditized. Projects like zephyr show promise in making microcontrollers easier to interchange if one dries up.

6. Make the most of your PCB. It's a creative medium, not just a flat sheet to jam components on. Some of these uses can save you components, complexity, cost.

https://twitter.com/alexandrosM/status/1404967112108482560

7. Similarly, rely on other multipurpose tech. 3d printing, lasercut ting, cnc. It can adapt with you, rather than holding you back.

8. Modularity for smaller increments. By making complex hardware modular, you can iterate on each part separately, reducing the risk of experimentation, and increasing the parallelism of experiments you can run at once.

9. With so much complexity reduction, you're ready for the big one: use 2-layer boards that can be protoyped with a Voltera machine. Instead of waiting for weeks, get an iteration in under a day, and keep going. 1https://www.voltera.io/

10. Build a real post-sale feedback loop. Collect data from the field, understand real world use, improve your hardware, and improve your (now significantly increased amount of) software. Deliver your software with balena or similar, and get a lot more from your hardware.

Ultimately, I think we're on the cusp of radically simpler, radically more software-centric hardware, that benefits from fast, local iteration. A world where supply chains no longer spoil us might just speed up the inevitable.

A caveat: I'm not exactly a hardware engineer, just a pain in the ass to some good ones. So I can't vouch for every word in the above thread, but I think it's generally in the right direction.

What do you think hardware will have to do to adapt to a supply chain winter?

@mattparlmer @divalbanerjee @notAbtin @AlexRBucknall @rahulthakoor @anujdeshpandey @shaunmulligan9 hardware friends what think you?

Another great concept -- common part lists. Well-known parts with multiple sources.

https://twitter.com/idlethread/status/1408676544365559808?s=20