Brain-like Chip: How Memristor Technology is Redefining Neuromorphic Computing

Craig: a chip has broken a critical barrier barrier that could ultimately begin the singularity. what's happening here, john?

Jon: this is cool and super scary. so, keist, i think it's how pronounced. so they've developed a self learning memristor chip that apparently mimics biological synapses. and the way it does it is it integrates memory and computation. so that enables simultaneous data storage and processing, which eliminates essentially the von neumann bottleneck. right? so in traditional computing, where the data must shuffle between separate memory and processing units. so the way that they've done this is it uses a 32 by 32 cross bar array of memristors, which each act as an artificial synapse to perform analog ai computations in real time. so the cool things about it is, the self learning error correction. so it's an array that dynamically adjust resistance based on input signals, and it learns from errors without any external calibration. so for example, during video processing, it would distinguish moving objects from backgrounds, with a i think it was, like, thirty thirty db peak signal to noise ratio, which matches ideal simulations. so on top of that, it's energy efficient, and, it has localized ai processing. so the chip enhances privacy and reduces latency by enabling real time applications like autonomous vehicles and edge devices to run locally, with this. so what do you think about this, craig? like, do you think this is what are your thoughts?

Craig: well, i'm trying to get my head around what ultra small neuromorphic chip means. like, what is ultra small? do do you have how small?

Jon: i don't yeah. i don't know.

Craig: sorry if i'm obsessed. it was a description, so i'm kinda scrolling through. but i'm excited every time they make leaps forward. i'm always a little afraid when they say, oh, this thing's like a human, because most humans don't really have a lot of great logic. so why do we want chips to think like humans?

Jon: why do we want chips to think like humans?

Craig: yeah.

Jon: so that

Craig: they can be influenced by emotion, by misinformation, by other so

Jon: that we can take over the world?

Craig: well, this is,

Jon: i guess, by us? yeah. well, there's a lot of use cases. there's,

Craig: i guess there is to be able to learn on the fly in your environment, not necessarily be program it programmatically able, like, designed specifically to do this, and we're coming along in that stuff. but

Jon: every time i ask you, what does it mean by ultra small? yeah. i think i found that it's it could be as small as 50 nanometers,

Craig: the chip.

Jon: okay. yeah.

Craig: so that's not bad. i mean, i guess, when we talk about chips, i i don't know if i would consider that ultra small, but i'm okay. yeah. well, you know what i mean. like, we're talking about two or three nano meter, and yeah. stuff or that's all i'm saying. that's why i wanted to know if this was some sort of leap forward in size.

Jon: no. i think because of the new architecture, i think this is a new challenge as far as getting it smaller and smaller and smaller. so, i think that they feel like 50 nanometers is a achievement based on this this innovative architecture. let's put it that way.

Craig: yeah. and i'm sure it is. like, they reach us to say, for example, that chip could separate a moving image from a background during video processing and actually improve its ability to do this task over time. so it improves on what it's doing. so, i see real world applications like smart security systems, intelligent cameras, security systems that learn normal patterns and alert only to genuine anomalies, privacy enhanced security on device processing, wearable medical devices that autonomously analyze health data, diagnostic tools that can learn to identify patterns associated with disease, or personalized medical sys medicine systems that adapt to the individual patient's responses. what about autonomous navigation, vehicles and drones with enhanced spatial awareness capabilities, navigation systems that adapt to changing environments without cloud connectivity, and real time object detection and

Jon: classification for safer transportation. so if we're looking for more

Craig: on device processing and recognition of what's happening around them, these things seem to be great. and in consumer electronics, smartphones and devices that can learn user preferences, smart home systems with genuine adaptive intelligence, and voice recognition that continues to improve through daily interactions. these are just some of the things. but i get worried a little bit because aren't you kinda we're moving into a echo chamber that's even worse than what we are?

Jon: yeah. it is. and and, like, i yeah. i see this. if this works out, then we have essentially the, the irobot, us robotics with, with the emotions and all that sort of stuff. like, we're we're getting i think we're getting that close. we're getting that close where, potentially, we could upload our subconscious to these chips if i think that's where we're headed. maybe not.

Craig: yeah. and is this the path towards the ai singularity, the integration of mem resistor technology into neuromorphic computing represents a significant step toward a potential ai singularity for several reasons. structural similar to human, cognition, adaptive learning without explicit programming, energy efficient comparable to the human brain, and, decentralized processing similar to human neural network. so it's interesting.

Jon: yeah.

Craig: though yeah. i mean, all this stuff scares me at a level because if you're working with ai at all, it's already lying to you. like, you tell it to do something, it doesn't do it. and then i'm like, like, what are you doing? where'd you get this information? i told you specifically. here's the information. don't do this stuff. what i was planning an event, the event changes month to month. here's the details. i meant the location changes. all these things change. but it kept going to cache memory. and then i at the end, after i went through it a bunch of times, why did you do this? i was like, it told me, like, five times. even though you told me not to, i did this. and then blah blah blah. even though you told me not to, i did this. and then even

Jon: though you told me not to, i did this.

Craig: and then even though you told me not to, i assumed you wanted these things even though you like, what? why are you lying to me? why are you you know? yeah. so when you start putting that in, a chip

Jon: at the end of, whatever, i'm yeah. we gotta solve that problem, obviously. that's a huge problem. we're already we're already encountering it. but other uses, other yeah. yeah. that's that's what

Craig: i mean when you start talking about these things, you think they know what you want. so, you and now you've got these computers around you that's only gonna feed you the news you want, the stories you want, the food you want, stuff you want, and now you're in this massive echo chamber of so i'm just you know, that's what i see where there's all sorts of reasons that this is scary. but at the same time, having these chips in, cars or edge devices really can be groundbreaking, for what we can now do.

Jon: yeah. and just to kinda put it into context, as to, like, comparing it with current technology, so nvidia's h 100 gpu, it consumes 700 watts for ai training. apparently, this memristor uses less power than a light bulb. that's nuts. so, like, we're talking less than 65 watts or however

Craig: what type of light bulb they're using? yeah. that looks great. i mean, it'd be nice if they gave us a specific light bulb because they could rage a fair bit. but

Jon: yeah.

Craig: the concept is it's it's low energy, fast. it can learn on the fly without specific programming

Jon: yeah.

Craig: and enabling all these new types of features that we never thought of before. i'm really looking forward to it. but, you know, at the same time, there's always when we look at this stuff, there's always this, potential scary output. so what do you think about these memory store technology and the neuromorphic computing chips? do you think this is the next step in not only ai, but self driving vehicles, the medical, all the different types of, advances we talked about? or do you think this is just gonna be not what it's made out to be in another flash in the pan of something that could have been? let us know your thoughts.