Memristors Run AI Tasks at 1/800th Power | Hacker News Reader

Memristors Run AI Tasks at 1/800th Power

spectrum.ieee.org

40 pointsmfiguiere11 comments

Memristors seem the same ‘always close by’ thing next to graphene, nuclear fusion, universal cancer medicine and definitely much better batteries.

Is any real happening that we can buy soon with memristors? I liked their properties a lot since I first heard about them some 15 years ago and I believed they were closed because it was HP talking about it. 15 years…

This is unfortunately true. And IEEE has become more like science.con, spreading hype instead of facts.

AFAICT the current owner of the patent for the memristor, HP, are just not too interested in developing it. I don't know the reasons; they can be anything from an apathy towards a technology which is not bringing a product quickly, to a plan to shelve it because else it would somehow cannibalize some of their other important products / technologies.

How many years are left on the patent?

I can't find a proper patent, but HP's scientists declared to have a working memristor in May 2008 [1], so a patent application should not be much earlier that that. (I suppose they filed for a patent before unveiling the device to the public.)

[1]: https://www.memristor.org/economics/288/memristives-seeking-...

There are different varieties of memristor but it looks like the main ones were invented 2004-2009, so 2024-2029.

My understanding is HP evaporated huge amounts of funds into memristor development, then simply stopped talking about. I suspect they never had any working product or prototype in any capacity.

They were literally calling the product that was using it "The Machine", like a 1980s Sci-Fi B-film:

https://www.hpcwire.com/2015/06/11/hp-removes-memristors-fro...

It's a crying shame that a commercial lab found this (and presumably patented it). If it had been a government lab, we could theoretically have dozens of groups all making progress with memristor tech.

I remember breathlessly telling friends and family how memristors were gonna change everything. I didn't (and still don't) have any particular understanding of memristors other than having drunk the cool-aid in terms of their popular science value. It feels like a huge letdown that we don't have memristor tech in the wild yet.

OT, but apparently ieee.org doesn't respect do-not-track requests, and there doesn't appear to be a way to deny cookies

Right, it doesn't. Ieee.org is one of the worst of the tech sites for news these days when it ought to be the best. Its cookies policy and aggressive use of JavaScript make for a horrible user experience.

BTW, I was an IEEE member for decades but I gave up my membership after I perceived that it was no longer acting independently. Unfortubately, these days, it seems corporations have a stranglehold over the organization and that's reflected in its editorial policy etc.

Worth noting is that stochastic computing devices would make reproducible experiments more difficult. For R&D, that's not a trivial thing to throw away. Indeed, one of JAX's motivations is finer control of pseudo-randomness for reproducibility of experiments.

Reply on news.ycombinator.com