'Memtransistor' Brings World Closer To Brain-Like Computing

the gmr writes: According to a recent article published in the journal Nature, researchers at Northwestern University's McCormick School of Engineering have developed a "memtransistor," a device that both stores information in memory and processes information. The combined transistor and memory resistor work more like a neuron and purports to make computing more brain-like. The new "memtransistor" would use less energy than digital computers and eliminate the need to run memory and processing as separate functions while also being more brain-like. Lead researcher Mark C. Hersam clarified the brain-like efficacy of the memtransistor: "...in the brain, we don't usually have one neuron connected to only one other neuron. Instead, one neuron is connected to multiple other neurons to form a network. Our device structure allows multiple contacts, which is similar to the multiple synapses in neurons... [but] making dozens of devices, as we have done in our paper, is different than making a billion, which is done with conventional transistor technology today." Hersam reported no barriers to scaling up to billions of devices. This new technology would make smart devices more capable and possibly more seemingly-human. The devices may also promote advances in neural networks and brain-computer interfaces, new technologies also recently reported at Futurism.

Comments

For those unfamiliar with memristors...

[jouassou]

I recommend checking out e.g. Wikipedia's summary of the theoretical motivation behind them. It's not just about making "computers more like brains", it's rather that memristors are the fourth passive electronic component (the first three being the resistor, capacitor, and inductor). Once we've got a full set of passive electronic components, perhaps a lot of circuits that today have to be built using active components (transistors, op-amps, etc.) could be replaced by smaller and more efficient passive equivalents.

Re:For those unfamiliar with memristors...

[postbigbang]

The problem with the passive components is the same as it's always been, and why Von Neumann idealized state machines as computing elements. Adding a state value helps, but causes multiple concurrent states.

Memtransistors don't have checksums, and their state isn't arbitrated in such a way as to give them the capacity to be shared without other active components. Because there is no checksum or CRC easily possible, coupled to the logic that sets (and checks) their value, means that they have limited architectural applications until several facets of their nature can be changed.

Look at 100 people, and 33 of them have faulty neurons. Analogizing states in this way, memtransistors, could also be capacitive arrays, inductive arrays, LC/LCR arrays, and so forth. Their present state of changeability comes nothing close to the high speed memory (transient and charged state--think nv-ram) present today.

There are great potential applications in ASICs, (fp) gate arrays, and other constructions, but just as GPUs don't replace CPUs, arrays made of memtransistors aren't going to replace either CPUs or GPUs, FPGAs, ASICs, etc. They're not more "brain-like", rather, they're a different architectural models whose limitations still haven't been surmounted.

It's not a fully passive device-- it's a resistor with a third leg in terms of boolean logic.

Re:For those unfamiliar with memristors...

[SumDog]

Yea I found this summary pretty sensational. I've heard the term you used, "memristors," and how HP was working on them a few years back. I might have understood the concept wrong though. Typically memory is where you store data and registers are units on the processor that act on that data (add, subtract, bitshift, or more complex instructions).

As I understood it, memristors would allow instructions to operate directly on memory without having to load or store. If you had enough memristors to load your program into them, you'd effectively run everything incredibly fast in place. However this would change the entire way we write programs and compilers.

You'd probably no longer have a stack, the purpose of a program counter would change entirely, and you'd start to get into the very gritty details of immutability and self modifying code. Even concepts like branch prediction would have to be entirely rethought. It'd be a larger diversion than even VLIW (e.g. EPIC/Itanium).

It'd be like quantum computing; incredibly powerful but requiring entirely different computing processes and mechanics.

Re:For those unfamiliar with memristors...

[Hal_Porter]

Yeah, but because we're all stupid now Futurism has a to write the clickbait article as 'new magic component will lead to ELECTRONIC BRAINS'. As if the problem of AI is going to be solved by some low level component as opposed to understanding the high level organisation.

Re:For those unfamiliar with memristors...

[jouassou]

Good point. According to TFA, they're closely related like a diode and transistor, but not exactly the same device:

The memtransistor is essentially a combination of a memristor and a transistor. Memristors, or memory resistors, remember the voltage that has been applied to them but can only control a single voltage channel. By transforming such a memristor from a two-terminal to a three-terminal device in the memtransistor, the Northwestern team made this tech much more capable for complex circuits and systems.

Re:For those unfamiliar with memristors...

[Sir Holo]

I read both of those. This is a logic gate, a resistor that looks like a standard logic element transistor. If you want to be less baffled, look into electronics engineering as it applies to both computing, and also the history of how microcomputers were originally designed against the state machine model(s).

CORRECT.

1T1C (or any two-terminal implementation of a non-volatile memory element) is a goal of extreme interest for the big CMOS foundries. Density!

I believe that it is still 2T2C available on the market, but RAMtron is likely to have fixed that a while ago. It was a hot topic in 2000.

Re:For those unfamiliar with memristors...

[Sir Holo]

Except that a memresistor is not really a passive component as it has state. And there is no "missing" passive component either. I don't know why this obvious BS is being repeated and repeated all over the place. A great success for marketing nonsense, a great loss for actual truth in engineering.

The actual article was sensationalistic. All they did was implement a hybrid memristor (assumed, TL;DR) incorporated with a transistor. So they combined two of the four basic circuit components into one with two leads. Big deal --- Ancillary circuitry will be needed to send "write" currents sometimes, and "read" voltages other times. Those must be put somewhere on the die.

Call the memristor a passive component if you want. The fact is, though, that a memristor can be "set" to a certain resistance, meaning that this device has many, many possible states. . . as long as signal-read is by decade or so for ON/OFF voltage (or 10%-ON/OFF, etc.).

The paper's use of "monolayer" of amorphous MoS2 is a nonsense term, especially when they call it polycrystalline. Their ON/OFF was a measly 10x. There are materials with 10^5 ON/OFF ratios, at least in the bulk.

I need to get to campus to read this article. The Abstract sounds like many Nature Abstracts to unexciting work. You make it sound big to get it into Nature, and word-smith like crazy. But from the looks and smell of this abstract, I will be disappointed when reading the article.

Did they rule out an interfacial charge-trapping scenario as the mechanism of their charge storage? In a single monolayer, it would have nowhere else to go.

Re:For those unfamiliar with memristors...

[thegarbz]

This is a logic gate

So you DIDN'T read it then. A memristor doesn't have a gate. It's a 2 terminal passive device.

Keep trying. You'll get there eventually.

Re:Is Slashdot broken or something?

[sheramil]

Perhaps they're using a new server that runs on memristors.

"Okay, what's two plus two?"

-thinks- "... two."

"What happened to the other two?"

"... I forgot about it.

Re:Welcome to 2008

[dohzer]

Completely different. This is a memTrANSistor

Not entirely true

[Viol8]

We have a pretty good idea of how neurons work and behave individually and also some brain components are understood up to a point - eg visual system which has allowed some pretty good advances in artificial neural networks. However how individual systems in the brain link up and produce a conciousness - the ghost in the machine - is still frankly anyones guess. There are lots of idea but nobody really has a clue yet.

Re:Is Slashdot broken or something?

[TheRaven64]

Yup. SourceForge has also been suffering, so I imagine that the 486 in the corner that's handling both sites is now completely full of dust.

Re:Is Slashdot broken or something?

[AmiMoJo]

My bet is that they installed Meltdown patches and took a massive performance hit.