Are memristors real? I feel like I've been hearing about how memristors are going to revolutionize computing for as long as I've been paying attention (~10 years). And yet, they never arrive. I also recall some detractors who say it's all marketing and "true" memristors aren't going to be viable for a super long time. Can anyone from HN summarize what the deal with memristors is?
I work on memristor based crossbars for neural network computation. Memristors are real, but we’re pretty far away from mass scale commercial products.
This looks like a snake oil salesman rebranded CBRAM as a memristor and hoped nobody qualified would notice that they're claiming a revolutionary breakthrough.
They explicitly say that isn't the case: " SDC memristors are distinct from (and superior to) Conductive Bridge RAM (CBRAM) technology and represent their own class of memristor."
No, they make a claim and their only sources are their own work that doesn’t appear to have been peer reviewed or even vouched for by anyone that doesn’t work for them.
A true memristor as envisioned by Chua would be a revolutionary device of extreme utility. If knownm has what they say they have, the company would be worth billions. But they’ve been around for years and their website looks like every other scammy snake oil site out there.
Hah, I just noticed, the founder and the CTO have no industry experience and I can’t find anyone who does connected to the company. They started this scam fresh out of college.
And more: Their "angel investor" and "chief administrative officer" is the founder's mother, but this is very carefully sidestepped in the bios. Every time I look closer, it's one more classic sign of a scam.
Who are their process engineers? Where is their fab? Are they contracting with a third party? Who? Will they vouch for the technology?
Ahahahaha, as it were:
"In January of 2018 AHaH Computing was proven feasible with Knowm SDC Memristors on the open-source Memristor Discovery platform. As of January 2019, Knowm SDC Memristors passed four years of shelf-life testing and have shipped to researcher in over forty countries."
The "Memristor Discovery Platform" was made by... Knowm. (in other words, "it works because I say so").
Are you sure this isn't a practical joke?
We've been here before. Companies make all sorts of microchips that do a lot, a little, or nothing at all, and many are counterfeit and outright frauds. When a small company in New Mexico with no experience tells you they're going to revolutionize computing, just walk away.
To be fair, MS did not revolutionize computing when it was founded. They didn't make anything that everyone else wasn't already making. I'm trying to think of a single thing that they did before the company was worth a billion dollars that was in any way revolutionary, technically. I can't think of a single thing (lots of stuff since then, BTW, but I think that's a bit unfair ;-) ). They executed well, both in development and in business and they had a good combination of good business foresight and lucky breaks. They were also pretty ruthless.
I know. :) It's fun in two ways, in fact. In one way, I could say "See? New Mexican companies are all criminals!", but in the other, I can point out that Microsoft, while scrappy, was basically trying to sell something they knew they could build - and everyone else more or less did, too. They made something important, and evolutionary, but not revolutionary.
Any company that says "The revolution is here!" is suspect, but if they can't actually give you a device that brings major, practical, commercially important changes, it's probably bullshit.
Through-hole scale components are valuable for people to play with. They're accessible for researchers and hobbyists, but don't necessarily present potential for massive, commercial scaling.
Why don't you just talk to them yourself instead of ranting on YCNews. It amazes me how many people don't follow the guidelines on this site and the mods just let it fly. Every single one of the members of Knowm you can directly email. I know because I have actually talked to them. One lady is not on Linked in so you will have to talk to her through her her academic email.
Also your "ha no industry experience" is pretty laughable considering the entire Quantum Computing Industry is blossoming out of universities and PHDs. Musk had no Rocket ship industry experience, Zuckerberg had no social media industry experience, Paulmer Luckey had no hardware or entertainment industry experience, etc. I can go on forever.
Maybe instead of mocking someone elses work and slandering their name with barely a whiff of an idea of who they are and what they are doing you should engage in some sort of meaningful criticism or debate.
I'll not be responding or reading your reply. So feel free to do whatever with that knowledge. Just thought I should mention that this site is apparently supposed to be about hackers conversing in useful conversations and enlightening themselves. So maybe try to stick to that.
You don’t ask a scammer “are you a scammer”. And there is no “Quantum Computing Industry” yet outside some scams and hype. We remain nowhere near real quantum viability. And Musk hired people with experience. I can’t find anyone with experience connected to this company. Nobody talks about it publicly except the founder and hyped up press releases, and their documented staff are mostly dedicated to “intellectual property”. This is a scam and/or patent troll. End of story.
Obligatory reminder that "memristor" is mostly a marketing hype word at this point. No clear example of the original memristor hypothesis has ever been crafted, and there is wide disagreement on what should or shouldn't be called a memristor, and it's not even agreed whether a memristor can physically exist at all. And hearing about the "first" memristor <whatever> is basically a biannual occurrence.
Also, reminder that even if someone manages to make a real memristor and everyone agrees it is a memristor, it may not end up being a big deal. They have some nice hypothetical properties, but they're not magic, and any widespread use would probably take decades.
In the 60s/70s, an EE/CS guy named Leon Chua did some clever math that suggested we could build transistor-like devices that had a non-linear relationship between charge and flux and called it a memristor. He didn't build one, and as far as I know it's not something he's spent all that much time on since.
In the 90s somebody at HP must have mentioned memristors in hearing range of their marketing department and suddenly HP was going to have a product to market in a few years.
Yeah. Never happened. Nobody's done it.
In theory, such a device could allow us to do things like build extremely low power non-volatile memory (a flash replacement, basically). A lot of marketing hype has talked about completely replacing computer RAM with non-volatile memristor-backed memory, but again, neither it nor anything like it has happened. But as it turns out, battery-backed DRAM covers a multitude of sins anyway.
The actual debate over physical possibility is because it's too low-power, to the point that an actual implementation might violate lower theoretical limits on the amount of energy required to change the state of a bit. In other words, it might violate the second law of thermodynamics, which is something of a problem.
Yes, memristors are real, and no, their properties are not magic[1]. As I stated elsewhere on this thread, you can even make them at home[2]. It's even been discussed here on Hacker News before[3].
They are real if you accept the branding of of companies trying to sell products that don’t really do much to people who are lured in by buzzwords. Those of us who actually do real work have serious doubts.
The original hypothesis was based on the existent resistor, capacitor, and inductor. The memristor would complete the symmetry between these components and the underlying physics.
And I'm going to assume GP is basically referring to the first paragraph of the next session regarding definition and criticism:
According to the original 1971 definition, the memristor was the fourth fundamental circuit element, forming a non-linear relationship between electric charge and magnetic flux linkage. In 2011, Chua argued for a broader definition that included all 2-terminal non-volatile memory devices based on resistance switching.[2] Williams argued that MRAM, phase-change memory and ReRAM were memristor technologies.[21] Some researchers argued that biological structures such as blood[22] and skin[23][24] fit the definition. Others argued that the memory device under development by HP Labs and other forms of ReRAM were not memristors, but rather part of a broader class of variable-resistance systems,[25] and that a broader definition of memristor is a scientifically unjustifiable land grab that favored HP's memristor patents.[26]
This paper rejects the existing definitions of memristors as in inappropriate expression (composition) of devices using magnetic flux and inductance. It really doesn't cover the systems that use electrically adjustable resistance.
Bit depth becomes continious, essentially. Instead of a 32 bit number, you just have any number you please, as it's continuous. Granted, there are restrictions in translating this to memristors, but thats essentially the gist.
The real key feature of memristors is not the math or any of that stuff, it's that all that math is stable under power cycling and it take nearly no power to compute things. The article mentions that they preformed 188 GigaFlops with about 0.3W of power. The human brain is about 20W at about 1 ExaFlop (a VERY coarse estimate) for comparison.
This power savings opens up the doors for very good computation (ML stuff) on very tiny power sources.
The 'standard' summing circuit [0] typically requires an op-amp, which is in itself a multi-element component, and a few resistors. With a memristor, you can theoretically skip the op-amp and just use the passive elements, though an op-amp is likely still needed to read out the very small signals [1].
Analog computers in the past were mostly used to solve differential equations. The input was voltage levels. You could add, multiply, integrate or combine multiple inputs. The is what is done in artificial neural networks if you imagine the voltage levels as input.
