Francis Stokes

Can't be upset about losing what you never had

Brexit, but we exercise our isolated Britishness by returning to Druidic Paganism, taken from us by the Roman Empire - the original EU.

Actually I'll probably write it in Typescript, just because I think that'll be a good learning oppertunity for me

I'm seriously thinking about writing a HDL library for JavaScript (something like nMigen). I think it'll be hilarious to be able to tell the elitist on hn and reddit that people are writing hardware in JS. I can already feel the blood vessels popping in their eyes

Didn't know about this feature. I have a feeling this might make regexp just a tiny bit more approachable for peoplehttps://twitter.com/Carlillo/status/1217880779079999488 …

I'm not trying to scare monger here. Regardless of whether it makes it into the mainstream, I'm fully behind this being an avenue of research. But I think it deserves just as discussion and thought as we're seeing in the wider AI space right now.

We've seemingly already giving over control to ML for everything in our daily lives, including control of cars. But that's software, right? The hardware is the immutable, predictable layer. Is it different to derive circuits we don't understand?

And at this level, there's not a lot of abstraction left to break! We're talking about logical gates here. But those abstractions are what allow for the formal verification. So should this creep into rockets, into space exploration, into nuclear power plants?

Personally I find this amazing on a bunch of different levels. But there is an interesting question that arises: a genetic algorithm, much like other kinds of ML, produces something unexplainable and non-intuitive. On top of that, it completely breaks out of the FPGA abstraction.

Obviously this isn't magic! The unconnected circuit was interacting via a subtle electromagnetic effect. This is exactly the kind of weird illogical thing that evolution would find and exploit. It even threw away the system clock! Basically went from ordered Bach to free jazz.

The crazy thing that happened in Sussex is that the hardware that evolved had a circuit inside that was unconnected to the main circuit. But if anything changed on that unconnected circuit, the whole thing didn't work anymore. Somehow the unconnected part affected the main part!

Excluding researchers and compiler devs, formal verification is normally not a big thing for software developers (no, tests are not the same!). But if you're building critical hardware for cars, rockets, or nuclear power plants, it's important to know your circuit's correct.

And all of that is formally verifiable. That means you can know - logically and mathematically - that what you've designed is correct (for whatever correct means in your case).

Want to build a 128 bit CPU from scratch? You can do that with an FPGA. Want to process or modify a complex signal like HDMI in real time? FPGA. What about a rocket controller that can handle more peripherals than a microcontroller could even dream about? You guessed it.

For those who haven't come across them before, FPGAs allow you to create digital circuits with tons of IO, speed, and flexibility. When you "program" an FPGA, you're essentially creating your own hardware.