Speaker 0:  we've been told for years that computers are hitting a wall, but researchers just found a way to smash through it using the laws of physics instead of raw code. if you ever wondered how your phone handles five g or how self driving cars see the world so fast, a massive breakthrough from peking university just changed the game. scientists have created a multi physics computer architecture that stops forcing every calculation through traditional digital circuits. instead, it uses a mix of light electricity in memory processing to handle complex math nearly four times faster than current tech. it's a specialized engine designed specifically to speed up the front fourier transform, the mathematical backbone of almost all modern communication and ai. what do you think about this, john?

Speaker 1:  so first off, i hated fast for your transformations in school. but, let's let's dive into this. so, yeah, this architecture, it can do about 500,000,000,000 operations per second, which is compared to about a 130,000,000,000 operations per second in prior systems for the same fundamental task. so we're talking about a fourfold performance improvement. i'm excited about this, because it's not, increasing energy use proportionally. and so let's dive into kind of the integration. so the architecture synthesizes multiple nodes of computing into one system. you've got different physical domains such as electrical current, charge dynamics, and potentially photonic effects, and they're matched to the parts of computation that they handle best. so by aligning which physical mechanism does what yeah. the system is able to avoid typical inefficiencies that are associated with traditional digital pipelines that use a single domain, like electrical voltage switching, which is used in chips from everything. right? so the heterogeneous physical design, it reduces the need to shuttle data back and forth between memory and processors, which

Speaker 0:  if

Speaker 1:  we're talking about it, it's a key bottleneck in conventional ai hardware. so this could significantly improve speed and energy efficiency. and so one of the ways that this architecture achieves this is what's called processing in memory or pim, which embeds computational capability directly where the data is stored instead of moving data to a separate processor. and that dramatically reduces data movement overhead that typically consume up to 60% of energy in conventional designs, and it also introduces latency. so by reducing this overhead, it allows the system to accelerate key kernels like for your transforms while cutting both time and energy per operation. i think this is a game changer with ai computing, with ai chips. and i don't know where nvidia is in all this if they're involved in this, but, i think this is legit. well, i think this could be a potential game changer, for ai. what do you think?

Speaker 0:  well, to me, it's just like adding, you know, math, coprocessors into the chip. you know, this gonna be the new, video cards, for example, the process ai will actually have, these processors built on or into size of the chip or as a third party add on that you can, you know, plug into your processor to process kind of stuff. so anytime you go in silicon, it's gonna be faster. and, you know, i interviewed and talked to a lot of companies that were doing in silicon stuff for bitcoin mining and whatnot. and it was so much faster, when they put the, code in the chip rather than running it on software, and it made it things so much faster. so when you start looking at this and that it's just the way it's architected, and then they hard they make the hardware optimized for this type of process, it's gonna be so much faster. so i'm excited. you know, i think of the traditional computer like a multitool that tries to use a screwdriver for everything. this tech is like a specialized workshop where the computer picks the perfect physical tool like light waves or electrical current to solve specific math problems naturally and instantly. you know? and it solves the data traffic jam problem by doing math right where the data is stored, like you said, in memory in real time and using natural physics. it increases the speed, like, from a 130,000,000,000 to 500,000,000,000 operations per second and lose while using way less power. i love that everything we're talking about these days you know, we're gonna get a four times or 10 times increase in speed, and the power drops. it's like, okay. great. you know, we're definitely heading in the right direction. so some of the impacts we might see on our life is your phone could download massive files even faster with zero lag because it processes wireless signals way more efficiently. smarter robots. home robots or self driving cars could react to obstacles in real time without needing massive battery draining or, in real time without needing massive battery draining or, computer in the trunk. and we talked a while ago about how they are putting quantum mechanics in the hand or in the elbow of robots to make them smoother and to understand. this obviously will allow for that. longer battery life. wearables, smartwatches could perform advanced health scans like heart or brain monitoring locally without killing, the battery. currently looking at, with successful research breakthrough published in natural ex electronics, we expect to see this hitting specialized industrial hardware within three to five years.

Speaker 1:  yeah. it's funny. yeah. when you say you could download a file much quicker, i immediately go to the show silicon valley with with, pied piper. but, anyways, we don't need to get into that. so what i like about this is so, like, what current architecture is like is they're using essentially, graphics cards, right, as these chips for ai. and nvidia is developing their own, but there's i said they're gpus. right? and there are tpus, but we're talking gpus. so what i like about this is now they're taking kind of a more fundamental approach to developing architecture that is ideal for ai, which i think makes the most sense if that makes sense. instead of trying to adapt an old architecture, let's create new architecture. let's look at what is needed. right?

Speaker 0:  yep.

Speaker 1:  k. so, yeah, i'm i'm super excited for this. like, about, like, we're talking edge ai systems, neuromorphic computing, brain inspired ai, communication signal processing. that's just some of the examples that can be done. but, yeah, if we can reduce that memory bottleneck using tim, i think, this this is gonna be phenomenal. did you read anywhere? because i know you always come up with the timelines. did it give us a timeline as to when this might go into, like, production kind of testing?

Speaker 0:  well, as i mentioned, we're hitting in specialized industrial hardware in the next three to five years because it's proven a study of the research papers launched. but consumer tech will follow that. so we're looking at a five to seven or five to ten year for consumer tech.

Speaker 1:  cool. yeah. i'm excited for that.

Speaker 0:  so when you combine that with all the other things we got going on, like, can we do this architecture with molecular computing and with all the other things, i'm excited. you know, the fourier transform math that this chip accelerates is the same math your brain uses to recognize a specific voice in a crowded, noisy room. so it's complicated, and it's been known for a long time. so we're moving out of the one size fits all chips and towards physics based hardware that works with nature rather than against it. and do you see that we're going to get, specialized chips, that say, this one is for doing stuff with video, and this one is doing stuff for this, and this one's doing stuff for that because they're wired specifically, for example, with the hardware codec with the acceleration and encryption for all that stuff so your cpu and your bandwidth can be even smaller. so let us know. is this the end of traditional processes as we know it, or are we just sticking fast? are we just really fast sidekick? you're gonna plug this on at the end. let me know if you trust an analog chip to drive your car in the comments. we'll be tracking this one closely.