HN Explorer

Ask HN: Whatever Happened to Memristors?

13 points by LarsDu88 ↗ HN
It seems like a huge part of latency in machine learning training is simply shuttling data between SRAM and DRAM... A decade ago there was talk of memristors blowing up that whole dichotomy. What happened? I'm aware of Intel's failures to commercialize optane... why can't memristors be used in place of SRAM or DRAM?

5 comments

[ 0.21 ms ] story [ 21.9 ms ] thread
mikewarot
I was hoping that Memristors would be the perfect storage for the 64 bits of mostly read memory in every cell of a BitGrid. Then they thought that analog mixers were the best use of it. 8(

I'm sad that we're in one of the worst timelines.

LarsDu88
My takeaway is that GPUs are not necessarily the best hardware for AI, but they rode 30 years of economic demand for gaming hardware
brudgers
I’ve been regularly watching Asianometry. New semiconductor ideas that only get implemented decades later is a pretty consistent theme. Memsistors will happen if and only if they solve a critical problem for the chip industry as a whole.

That might happen in the future. Or not. Nobody knows what will be the roadmap twenty years hence. Maybe memsistors will be the right technology. Maybe something better will come along. Good luck.

tmaly
I think it was all fluff. Articles were coming out saying they were close, but nothing ever materialized.