Memristors Were Going to Replace RAM and SSDs. What Happened?:
Memristors promised to revolutionize how we store and access data. However, despite hype surrounding memristors, they haven't quite replaced traditional storage technologies like RAM and SSDs. In this article, we'll explore the journey of memristors and where they stand today.
Memristors, or memory resistors, are a type of electrical component that can change their resistance based on the flow of electrical current. This unique property allows them to store information in a fast and energy-efficient way. Memristors were first theorized by Professor Leon Chua in 1971, but it wasn't until 2008 that researchers at HP Labs created a working prototype.
The primary advantage of memristors over traditional storage like RAM and SSDs is their non-volatile nature. This means memristors can retain data even when the power is turned off. In contrast, RAM (Random Access Memory) is volatile, which means it loses all data when the system is powered down. SSDs (Solid State Drives) are non-volatile but have limited read/write cycles, which can lead to wear and eventual failure. On the other hand, memristors have been touted to offer virtually unlimited read/write cycles, leading to a longer lifespan.
In other words, memristors promise to be the perfect combination of RAM and SSD storage if we can get them to work commercially.
When working memristor technology was first announced, it generated a lot of excitement within the tech industry. This was mainly because memristors promised faster, more energy-efficient, and longer-lasting storage solutions than traditional RAM and SSDs. This led to high expectations and a flurry of investments in research and development. However, the hype surrounding memristors has significantly waned in recent years, with the technology yet to impact the market.
Despite their potential advantages, several challenges have hindered the widespread adoption of memristor technology. Manufacturing memristors at a large scale has proven difficult and expensive. This has limited the number of companies willing to invest in developing and producing memristor-based devices. Also, while memristors have shown promise in laboratory settings, their real-world application performance has not always met expectations. Factors such as temperature fluctuations and material inconsistencies have led to variable performance and stability issues. To quote Wikipedia "Experimentally, the ideal memristor has yet to be demonstrated." So those diligently working on the problem are looking for a tantalizingly close breakthrough.
(Score: 5, Funny) by Rosco P. Coltrane on Thursday May 25 2023, @10:23AM
And I call BS: the guy who invented the first memristor is Thomas Edison, and that's called a fuse: it's normally one, and you program a zero in it by passing a very strong current through it. You set it back to one by changing the fuse - which, admittedly, is a bit inconvenient.
(Score: 4, Informative) by driverless on Thursday May 25 2023, @11:23AM
They were going to generate lots of clicks because they were a fancy solution in search of a problem, so stories imagining a use for them were perfect clickbait for the sort of people who swallow everything organisations like Gartner say.
(Score: 5, Informative) by VLM on Thursday May 25 2023, @12:16PM
The short version of the long article is they're too big and slow compared to competing 'inferior' technologies. Hundreds of thousands of capital invested in memristors, trillions in dram and flash ... Maybe in the long run memristors will catch up.
"Back in the old days" there's a thesis from a guy in texas with the name "a high frequency memristor emulator circuit" high being about 280 Hz (a sharp musical C4 aka "middle C", or close enough, IIRC, although I'm no musician...). This is an emulator in the sense that a pile of opamps and inductors can emulate a capacitor and a pile of capacitors and opamps can emulate an inductor. The thesis is pretty awesome and I verified this morning its still online. In my infinite spare time I intended to F around with memristors, and as usual I have WAY too much other stuff to do so I never did. Page 19 of the thesis has the schematic and reading the schematic probably explains the concept of a memristor to non-EE people better than any pile of verbiage. So using stereotypical opamp circuits "everyone" has seen, there's a high impedance buffer, an integrator, and a kind of wild feedback optocoupler as "memory". Why an opto instead of a big resistor? Well, in a sense optos are a perfect current diode, the reverse impedance so to speak is infinite as nothing short of lightning will send even a tiny signal in the reverse path.
Now because the performance of an analog solution is so shit, even the thesis guy promoted the idea that a "better" memristor can be emulated using a cheap microcontroller of the era with a ADC and DAC running at 10 KHz or whatever.
This is obviously for training and proof of concept as a SSD or computer DRAM that has access times around 'three hundred hertz' will be unhappy unless he's using a relay-era computer. That's too slow even for vacuum tube computers. But its good enough for training and experiments.
The memristor idea is kind of like hysteresis loops in a metal transformer where the next pass thru the magnetization curve will depend on the history of past loops thru the curve; literally like core memory. What makes it a memristor is generalizing the concept, abstracting it away from an effect that makes inductors non-linear, and a general belief you can make them from silicon instead of literal power transformers.
If, by some miracle, you could make a memristor that's fast and efficient and small and easy to use (which, apparently, we cannot) then the topology to apply them to make physical neural networks is pretty easy. You could make 2-d neural network chip arrays just like you make 2-d memory arrays now. Cool.