Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Ask HN: Let's generate some startup ideas in 2024

news.ycombinator.com

Post reply on HN
Print view View original

1–10 of 23 posts

Re: Ask HN: Let's generate some startup ideas in 2024

#6

Post by pantantrant »

mikewarot wrote: post #3

FPGAs, but you don't care about propagation delay, so you throw out all the switching fabric and put in latches instead to smooth out the timing. Upside: everything is pipelined, and memory does compute directly. Should be good for Petaflops

I like this, want to brainstorm with me further?

I raised $372.03 for my FPGA chip on my profile and I've only sketched out a rough mock prototype of what I want the architecture to be.

So far, it works with any amount of silicon defects, can be up to the size of a silicon wafer, and the larger it is, the more it can reconfigure itself in parallel, which is directly proportional to its size growth.

Re: Ask HN: Let's generate some startup ideas in 2024

#8

Post by smadam9 »

paulcole wrote: post #5

Soylent for cats Uber using skateboards FTX in biotech Craigslist for hookups Like Blue apron but they also ship you a Casper mattress Put your kids on the blockchain

"Put your kids on the blockchain", as in setting them up with a wallet from early on so that it can be accessed later?

Re: Ask HN: Let's generate some startup ideas in 2024

#9

Post by mikewarot »

pantantrant wrote: post #6
mikewarot wrote: post #3

FPGAs, but you don't care about propagation delay, so you throw out all the switching fabric and put in latches instead to smooth out the timing. Upside: everything is pipelined, and memory does compute directly. Should be good for Petaflops

I like this, want to brainstorm with me further? I raised $372.03 for my FPGA chip on my profile and I've only sketched out a rough mock prototype of what I want the architecture to be. So far, it works with any amount of silicon defects, can be up to the size of a silicon wafer, and the larger it is, the more it can reconfigure itself in parallel, which is directly proportional to its size growth.

I found your previous post - https://news.ycombinator.com/item?id=38170336

I find your ideas interesting, especially your strong interest in routing around any faults, but still ending up with a relatively usable compute fabric.

My approach is much, much simpler.... just a grid of cells, with the programming lines unspecified. Each cell has a 4 bits in from their neighbors, a 4x4 LUT, a latch on each of the 4 bits of output. Clocking on the latch is in 2 phases, like the colors on a chess board.... this makes everything deterministic, but fully Turing complete.

The fact that the logic does the routing means that you can route around anything at runtime. The homogeneity and simplicity of the fabric means the run-time libraries on the host computer could do that routing, provided it had a map of the defects.

Here's a write up as an Esoteric Language: https://esolangs.org/wiki/Bitgrid

Here's an emulator written in Pascal: https://github.com/mikewarot/Bitgrid

I think the best approach is for both of us to try to get our ideas in the next Google open source chip shuttle.

Re: Ask HN: Let's generate some startup ideas in 2024

#10

Post by paulcole »

smadam9 wrote: post #8
paulcole wrote: post #5

Soylent for cats Uber using skateboards FTX in biotech Craigslist for hookups Like Blue apron but they also ship you a Casper mattress Put your kids on the blockchain

"Put your kids on the blockchain", as in setting them up with a wallet from early on so that it can be accessed later?

No like proving they are your kids and not somebody else’s. It would also allow for more trusted adoption procedures. And let parents buy/sell children more securely without government intervention.
Post reply on HN