New AI Transistor Works Just Like the Human Brain (studyfinds.org) 44
Longtime Slashdot reader FudRucker quotes a report from Study Finds: Researchers from Northwestern University, Boston College, and the Massachusetts Institute of Technology (MIT) have developed a new synaptic transistor that works just like the human brain. This advanced device, capable of both processing and storing information simultaneously, marks a notable shift from traditional machine-learning tasks to performing associative learning -- similar to higher-level human cognition. This study introduces a device that operates effectively at room temperatures, a notable improvement over previous brain-like computing devices that required extremely cold conditions to keep their circuits from overheating. With its fast operation, low energy consumption, and ability to retain information without power, the new transistor is well-suited for real-world applications.
"The brain has a fundamentally different architecture than a digital computer," says study co-author Mark Hersam, the Walter P. Murphy Professor of Materials Science and Engineering at Northwestern's McCormick School of Engineering, in a university release. "In a digital computer, data move back and forth between a microprocessor and memory, which consumes a lot of energy and creates a bottleneck when attempting to perform multiple tasks at the same time. On the other hand, in the brain, memory and information processing are co-located and fully integrated, resulting in orders of magnitude higher energy efficiency. Our synaptic transistor similarly achieves concurrent memory and information processing functionality to more faithfully mimic the brain."
Hersam and his team employed a novel strategy involving moire patterns, a type of geometric design formed when two patterns are overlaid. By stacking two-dimensional materials like bilayer graphene and hexagonal boron nitride and twisting them to form a moire pattern, they could manipulate the electronic properties of the graphene layers. This manipulation allowed for the creation of a synaptic transistor with enhanced neuromorphic functionality at room temperature. The device's testing involved training it to recognize patterns and similarities, a form of associative learning. For instance, if trained to identify a pattern like "000," the transistor could distinguish that "111" is more similar to "000" than "101," demonstrating a higher level of cognitive function. This ability to process complex and imperfect inputs has significant implications for real-world AI applications, such as improving the reliability of self-driving vehicles in challenging conditions. The study has been published in the journal Nature.
"The brain has a fundamentally different architecture than a digital computer," says study co-author Mark Hersam, the Walter P. Murphy Professor of Materials Science and Engineering at Northwestern's McCormick School of Engineering, in a university release. "In a digital computer, data move back and forth between a microprocessor and memory, which consumes a lot of energy and creates a bottleneck when attempting to perform multiple tasks at the same time. On the other hand, in the brain, memory and information processing are co-located and fully integrated, resulting in orders of magnitude higher energy efficiency. Our synaptic transistor similarly achieves concurrent memory and information processing functionality to more faithfully mimic the brain."
Hersam and his team employed a novel strategy involving moire patterns, a type of geometric design formed when two patterns are overlaid. By stacking two-dimensional materials like bilayer graphene and hexagonal boron nitride and twisting them to form a moire pattern, they could manipulate the electronic properties of the graphene layers. This manipulation allowed for the creation of a synaptic transistor with enhanced neuromorphic functionality at room temperature. The device's testing involved training it to recognize patterns and similarities, a form of associative learning. For instance, if trained to identify a pattern like "000," the transistor could distinguish that "111" is more similar to "000" than "101," demonstrating a higher level of cognitive function. This ability to process complex and imperfect inputs has significant implications for real-world AI applications, such as improving the reliability of self-driving vehicles in challenging conditions. The study has been published in the journal Nature.
And the BS gets more stupid... (Score:4, Insightful)
I hope that this is a sign that in a little while we can actually look rationaly at the still pathetic state of "AI" again.
Re: (Score:3)
sadly, with a lot of free money, no end is in sight. some taxes are way too low, apparently.
Cool! (Score:3, Funny)
Apparently now we know exactly how the human brains works, such that we can determine something else works "just like it".
Surely the article links to a paper explaining exactly how the brain works, what conciousness is, etc. I look forward to reading it.
Re: (Score:2)
No (Score:5, Insightful)
I hate headlines like this.
No, it doesn't work "just like the human brain" which consists of wild flesh which is not fully understood yet. Not only all neurons and connections between them are slightly different, they are not set in stone and change outside of their primary functions of controlling network weights (or so we think).
And then we have functionally different parts of the brain and even different neurons.
Re:No (Score:5, Insightful)
Indeed, maybe "a little more like a human brain works" could have been acceptable if they insisted on the dubious analogy.
Re: (Score:3)
wanted to write : "a little more kind of like a human brain works"
Re: (Score:2)
well, "a little more kind of like a human brain works" would still have been acceptable I guess, "kind of like" doesn't imply we understand everything on how the human brain works IMHO
Perhaps (Score:3)
Maybe your brain doesn't function that way, but I'm certain I've run across people amongst the Slashdots whose brain does. Perhaps, it might have been something they assimilated.
Re: (Score:2)
In fact, we already have one technology (in the lab still) that's closer to how actual neurons work than this technology - clustered memristors.
Mixing memory and switching, using (far) less power, capable of analog processing. All the good stuff. And while nothing's come out of the labs just yet, I still see the occasional paper published about the progress being made.
Re: (Score:2)
Sure it does. JUST like a human brain. (Score:2)
Re: (Score:2)
Can you mfers even go beyond an headline for once? (Score:1)
Re: (Score:2)
There's no need to read any further here. With a headline that phenomenally stupid, the odds of there being anything significant behind it are effectively zero.
This doesn't even deserve derision. It deserves to be ignored. "AI transistor" "works just like the human brain" it's offensive.
Let this be a lesson to you. If you want anyone to care about your whatever, don't smear shit all over it.
Re: Can you mfers even go beyond an headline for o (Score:2)
Shh, rational takes are "bad" (Score:2)
Don't feed the luddites. :)
Re: (Score:3)
Yes, I know. I did look at the paper. To the surprise of no one, it's also full of sensationalist nonsense:
The asymmetric gating in dual-gated moiré heterostructures realizes diverse biorealistic neuromorphic functionalities, such as reconfigurable synaptic responses,
If you're not familiar with synaptic transistors, they're very similar to memristors in that their conductance varies depending on past input. This is where the "just like the human brain" nonsense comes from. Just like the word "neural" in "neural network", the use of the word "synapse" is intentionally misleading. The authors are very clearly encouraging that misunderstanding with lines like I qu
Re: (Score:2)
Re: (Score:2)
If you found the paper, then you also know that it's also loaded with the same silly sensationalism.
As for the title, the key words here are "neuromorphic" and "synaptic transistor". If you think the terms AI and NN are misleading, they've cranked things up to 11! If you don't think the authors share any blame for those terms, remember that they chose to use the word "biorealistic" in the abstract all on their own.
You might still be inclined to give them a pass because the StudyFinds article also complet
Re: (Score:2)
Again, if you want someone to pay attention to your thing, don't smear shit all over it.
Apparently this is false, since it was the AI hype that got them into Nature, not the room-temperature stuff.
Re: (Score:2)
Apparently this is false, since it was the AI hype that got them into Nature
That's a depressing thought.
Copy/Paste Headlines (Score:3)
I better lay off this list before it repeats forever.
Joke Incoming! (Score:2)
The paper is available on arXiv ... (Score:2)
... if you actually want to understand what this research is all about:
https://arxiv.org/pdf/2306.039... [arxiv.org]
No, it does not work "just like the human brain", in which neural activations across synapses consist of streams of ions activated by molecules known as 'neurotransmitters'.
Instead, this research deals with something entirely different called 'ferroelectricity', which is a property of some materials that have a spontaneous electric polarization. This polarization can be reversed by applying an external el
Associative memory? (Score:2)
Re: (Score:2)
And yet you could simulate this device (Score:2)
Revised headline, from a different source (Score:2)
New transistor works just like a dog's brain. To claim it could think like a human was a bit of a stretch, said the janitor.
Re: (Score:2)
Re: (Score:2)
So why don't we have a working human brain model in a virtual jar?
It's too big. Wikipedia [wikipedia.org] lists the last attempt along those lines, in 2013, managed to simulate 1% of a human brain using a supercomputer. And it was slow: 40 minutes to simulate 1 second of activity.
Assuming, almost certainly incorrectly, that scaling this linearly would get us to 100% real-time emulation, and that Moore's law (both the real as well as the popular ones) stays validly for ever, that would require about 7 Moore doublings for the percentage, times about 11 doublings for performance, times 1.5
Not the Turing test (Score:2)
One way to make sure is to mention certain buzzwords e.g. anniversary, mother-in-law, and itcanhappentoanyone come to mind
clickbait (Score:2)
I'm thinking they made one device that can; 'distinguish that "111" is more similar to "000" than "101,"
Can they easily reproduce it? Does it communicate with other such devices to form a network of some kind? The article doesn't say, but probably not.
Re: (Score:2)
Re: (Score:2)
Good point. You would have to tell the device somehow what rules would indicate similarity in a set of inputs. Otherwise you are just interpreting whatever it says yourself.
pure flack (Score:2)
Here is what it actually is, from the abstract: (https://www.nature.com/articles/s41586-023-06791-1)
"Here we report the experimental realization and room-temperature operation of a low-power (20pW) moiré synaptic transistor based on an asymmetric bilayer graphene/hexagonal boron nitride moiré heterostructure. The asymmetric moiré potential gives rise to robust electronic ratchet states, which enable hysteretic, non-volatile injection of charge carriers that control the conductance of the devi
Nope. (Score:2)
Wow, a cure for cancer! (Score:2)
Let's drink to that!