Hi I was just watching the Tesla AI day here. the artificial intelligence day. This is their uh second one. of course they infamously did the one last year where they announced the Tesla bot and they had the uh Spandex dancing robot and it was meme worthy.

It was hilarious and um, it's absolutely amazing what they've done in 12 months anyway. I don't really want to talk about the Tesla bot so much. There was something interesting about, uh, 58 minutes into the presentation which we'll talk about here. but we will go have a look at the Tesla bot because well, let's just go have a look.

Okay, they pretty much came out straight away and and boom here it comes here. it comes and he said you know, uh, lower your expectations and stuff but I was I was actually pretty impressed. And here's the new. Here's the new Tesla bot.

This is their prototype uh one. so this is not. this is the uh one, that's you know, fully naked and they did most of their own. They're all.

there are development on this, but they developed this. One of the presenters at the end of um. his speech said they developed this in like six to eight months or something like eight months. Nine months tops or something like this.

So yeah, they've gone from nothing to this. Um, and it's it's now I'm I'm actually blown away. There's no way that you can be an engineer and anyway, you've got to watch the whole presentation because it's very technical. It's of course there's wires running everywhere okay, but you can't help but be thoroughly impressed.

And then they bought out their one, which is uh yeah. so that's the you know, this one has a tether and that was literally on stage. There was literally the first time it had walked autonomously without the tether. Um, so yeah, like you got to take your hats off to them.

Okay, anyway, then they brought out this one here I won't bother trying to get the audio right I don't care. Okay, you can go watch it yourself. They actually bought out the more polished one which has could be going to be closer to the production version if they ever make it to production. Um, but it's just you've gotta take your hats off to what they've developed in Under 12 months.

It's absolutely remarkable and you've got to watch the full thing. as I said. but this isn't about the Tesla bot. and here it comes.

here. it comes. Here's the more polished version of the Tesla bot. Unfortunately, this one had to be understand because it's they.

reckon it's a couple of weeks away from actually autonomously walking, but it's not quite there yet. And they go into lots of technical details about the density of the battery pack and all the uh, pneumatics, all the actuators that they've actually developed in-house and stuff like that. Um, so anyway, it's it's. very impressive.

You have to take yeah, I am thoroughly from an engineering standpoint I am thoroughly impressed and I have to take my hat off to Elon and and the team for what they've developed here in less than 12 months. Anyway, to solve the AI humanoid robot problem is exactly the same as solving the full FSD the full self-driving thing I Believe Elon himself has said publicly that you have to solve the whole AI solution in order to get full self-driving and also for the humanoid uh robot thing for it to be you know, like a practical day-to-day helper like around the house. And no, you won't see it this decade. So like, no, that is my prediction.

Um, you will not see a practical humanoid robot like this doing useful stuff around your house this decade. Don't even think about it. Okay, but it's once again, absolutely hats off to what they've developed I was I was pretty blown away from it from a technical point of view. Anyway, I think one of the most interesting things about the Um and what I saw here I mean there's tons of detail I mean they go into like here it is lifting it all the actuators anyway like they go into their their actuator configurations and they're going into how they've like um, the the actuators are all the orange ones here and how they've uh, you know, like had to model and get the reverse kinematics of everything and how it works in the envelope, the performance envelopes and how they all worked.

and apparently they've done all this in like, yeah, like six to nine months. It's absolutely remarkable how they've modeled all this and how. the knee. Um, how their knee system.

You can actually see it transition here. There it is. From that to that, there you go. so it's all it's It's thoroughly impressive.

Um, stuff. And they hats off to the amount of technical detail that they're willing to give to share in these AI days. It's phenomenal. Anyway, I am raving around in full self-drive and I didn't really watch that, but I did cut in at an intern.

Interesting point. Okay, unfortunately the thing just re-rendered so I've lost my timeline spot. But basically they're developing all of their own uh, compute structure, compute infrastructure to uh to render all of the Tesla AI you know, full self-driving stuff, right? They have to do this. and they're developing their own compute tiles, their own uh processing chips, and ultimately their own supercomputer racks to actually do this.

So please forgive me I won't I haven't watched uh, the full thing I just jumped in and here is what I wanted to, uh, talk about I found this absolutely fascinating and if you watch my videos on ceramic capacitor crack in and the solutions for that and I've even done a video on singing capacitors and stuff like this before or I've mentioned it in, you know I've done a lot on how uh, ceramic capacitors? Uh, not only are not only can pick up vibrations, you can actually use them as microphones. You can very poor ones if you actually use them as oh, it's still playing. use them as microphones. Okay, not only that, but they can also, uh, vibrate.

They can sing. It's called. you know, in the industry it's called singing capacitors because you can physically hear them if you actually switch the ceramic capacitors at an audible frequency. Sometimes you can actually hear them sing.

and of course you can get this with Magnetic components as well. Transformers inductors and stuff like that. You can hear that. You know that high pitched? You know that 15 kilohertz wine from the 15 kilohertz switching frequency in the dodgy Power brick and stuff like that.

you can get the magnetics, but you can also get the capacitors. And they actually, um, talked about a Folia mode they had in this very dense power module. Here Here it is um to a United States quarter dollar. Um, it's a thousand amp uh with 12 independent phases so you can see the 12 switching mosfets in there.

Okay, two, four, six, eight, ten, twelve. yep and count them right. But they've also got other stuff on here due to the requirements. How how they this is used to power their uh, their compute module.

They're um you know little brick module that they've got inside all of these racks and they have to pack them in so densely that they've put the mems oscillators also on here as well. So they claim like 0.86 amps per square millimeter and unmatched, uh, density and stuff like that. But they're talking about. So yeah, here's the here are these training tiles which they, uh, develop their own chips for and these power develop these power modules for.

And they're incredibly dense. They're like Leading Edge stuff. so you know Tesla's in the super Computing business as well. but they're basically the reason that they include this slide is because they have to talk.

They're talking about thermal expansion and stuff like that and how these boards are sewn. They're designed to be so dense that they really have to take and they get so hot of course. Um, and that they have to take thermal expansion into account. So that includes the PCB materials and the uh, you know, the copper and the solder joints and everything right? You have to consider thermal expansion because you can get cracked solder joints you can get.

You know all sorts of things. You can get cracked chips and dyes and all sorts of stuff, right? So what they started to get here were intermittent fires in the Big Y compute modules. You know, they've got thousands, tens of thousands of these bricks, right? And each one of them is powered by one of those individual things. And they're starting to get these intermittent fires.

and they, you know, had to do some analysis to figure out why and they tracked it down to the compute module here. What was happening is that they were losing the clock from the built-in mems oscillator on here and that was clocking. presumably the you know, the the tile or whatever the process is on the tile, but that uh, that oscillator due to density. you know, packaging fit to envelope requirements.

They had to actually, uh, put that on the power supply board and they've come a gutsy here. Um, and and they admit it and they show it as a learning example. Um, it's absolutely so yeah. they're getting no no clock outputs from the oscillators here.

Okay, and then they modeled it. It's like thermally like like a vibrational wise. So you put accelerometers. I've done videos on this.

You can get little micro accelerometers. In fact, I've got one. so I've shown this before. This is a PCB Piezotronics uh accelerometer.

which it's just that's the name of the company. It's not to do with Pcbs, although I used to use these on Pcbs all the time and they're tiny little accelerometers. They've got B and C outputs on there tiny little uh accelerometers. You'll calibrate these things and you would stick them.

You'd super glue them onto your PCB like this and in in often in multiple locations and then and different orientations as well. You know you get different axes like this so anyway, you can see that they they modeled this so they've you know, really put this into like visual, uh, you know, processing and and to get like a 3D map of what's happening. So I'm not sure exactly what they're you know showing here. but anyway, um yeah, they've gone to a lot of trouble to figure out why they're oscillators.

While they're getting, you know, failures in their oscillators on here and you guessed it, it turns out that the Mems oscillator right? It members these little micro machined uh, oscillators right? They're physical like if you look inside them. maybe I can include a photo up here of one inside. but they've actually got a photo there. So they were getting cracks.

So they're physically little micro machine oscillators that like oscillate in there, right? They're physically right oscillating. Well, all oscillators are like, you know, physical like quartz, quartz, uh Crystal or whatever. These are mems oscillators and they were physically cracking inside here. So what they were getting was out of plane resonance.

A mechanical resonance we're talking about actually caused by the vibration of the capacitors due to the switching frequency and the high currents in there. It just so happened to be at the right frequency and you get like a resonant mode and you can get harmonics and you can get resonant. You can get mechanical resonances of there and you know Murphy bit them on the ass and said I'm gonna crack those oscillators So they got out of plane. This is like inside the physical you know thing here.

they're representing this and you can actually play it and it it goes. You know. and it was actually vibrating and it would reach a point where it would make resonance and then it'd just crack. It'd crack the little machined micro machined arms on the mems oscillator.

they were that delicate caused by vibration in the capacitors due to the switching frequency. and then of course I You know the these. Once they discovered that there's obviously uh, several Solutions one would be a soft uh terminal capacitor. I've mentioned these before.

these are anti-vibration uh capacitors solve ceramic capacitor cracking. I've done entire videos on this. I'll uh link this in so you can get ones with like soft terminals on the end. you can buy them.

they're you know, fairly expensive. they're usually use them in automotive applications. cars uh, for example, uh, have lots of vibration, lots of vibe, unknown vibrational modes. um that you wanna? Yeah, you don't want your capacitors.

In this case, you don't want your capacitors cracking and then losing capacitance or shorting out or doing whatever in your car, stops or blows up or does whatever right. and um, yeah, these things can catch on fire. which I've also done videos on right of the ceramic capacitors crack. They can go short circuit and they can catch on fire because caps are usually across power rails.

In this case, they can dump a thousand amps into these suckers and um, yeah, the magic Smoke's going to escape real quick. So yeah, not an I Mean if a capacitor fails open, it often wouldn't ruin your day. But if they fail short, you're in trouble. But in this particular case, they were mechanically vibrating and that vibration goes travels through the rigid PCB which also has its own uh vibrational mode.

If you don't Mount your PC If you mount your PCB on the corners like this, then the board is as small as it might do, it might Flex in the middle. So actually, uh, you know, physically choosing your mounting points for your PCB can matter a lot. and that's why you often use these sorts of accelerometers to actually model these types of vibrant physical vibrational modes. And this is the kind of stuff that I've done in the industry quite extensively.

and it's really important stuff. And it turns out that yeah, so they can solve it by uh, having the so you know buying better quality soft capacitors like they probably use in the Teslas um anyway, because they're all Automotive grade anti-vibration Caps or you can choose a different mems oscillator with a like. In this particular case, they say 10 times lower out of playing Q factor which is the resonance uh, the physical resonance Q factor on there so they can you know or they can often mount it on a different board or something like that I've done that. I've mounted oscillators physically on separate boards in their own anti-vibration um, you know, like foam compartment um that that takes all the stress out or they can change the switch in frequency although they can do any combination of those sorts of things.

And and here's the Um Advanced Computing rack actually developed and at the end of this, I might actually include the full um Talk of this guy sorry, I don't know his name I should uh, check it out, but he's the one who's you know this dojo cabinet thing and how they've developed all these tiles and then they've developed these interface cards and these racks and this all this processing all this power stuff. This is all for their in-house Learning Systems To try and you know, learn like they pull the data from all the presumably all the Tesla cars I Don't know where they get the data from Um does it automatically download from your car? Can you opt in or something to uh like to include like training data or something and it gets sent back to Tesla and it gets all processed in these Um racks. And that's how they developed the AI So yeah, these are all the Um the solution that they've got. It's absolutely incredible.

but uh, and they've developed these massive compute racks 1.3 terabytes of high speed SRAM that 1.1 extra flops um 13 terabytes of high bandwidth dram and it's just absolutely amazing. But I thought anyway I thought that was a real interesting bit of the Um AI day was this mechanic that they just talked about the fire of that they had in this and I guess the engineers were so excited that they discovered that this failure mode I mean engineers get really, really giddy when they finally solve this sort of uh issue right and you discover it. and they obviously went to a lot of trouble to um, test. You know, lots of, uh, expensive accelerometers used to try and um, find these vibrational mode issues I'm sure.

But anyway, absolutely fascinating little look to come out of Um AI day here, which you may have missed because it's buried in some, you know, super Computing part of it or something I Just found it fascinating and I hope you enjoyed me sharing that with you. Anyway, the uh Tesla bot was um I was very impressed by the Tesla bot. Uh, like not so much that okay, you're going to be buying one next year or even this decade. okay to help around the home and do your humanoid um things.

no I've seen the movie okay I know it happens. Um and yeah, but like, no, it's come on, you've got it. Like you've got to give them props for what they've done in like under 12 months. Like nine months or something.

Um, it's it's really remarkable stuff. So anyway, um yeah, that's all I'll say on the there it is actually doing a thing in the Tesla Factory this is actually their internal camera or whatever. so it was actually doing something physical. They are here.

it is inside the Tesla Factory here and it's actually doing a task really slowly. But all this has been done from scratch from nine months. So like yeah, absolutely. hats off.

Um, but yeah I think I uh to solve the humanoid robot AI thing. The whole AI solution needs to be solved. Same with full self-driving as well which I've been a Critic of um and but yeah, come on. Hats off, hats off to them and hats off for sharing that fire mode I Love that Beautiful nerds couldn't help themselves.

The engineering nerds at Tesla they couldn't help themselves. Look at this fire mode we had. and let's put it in the big presentation, you know I it's it's just good stuff and they really went to town. Fantastic! Anyway, thoroughly impressed.

Catch you next time.