YT YouTube2Text

YouTube2Text → Transcripts → Digital Twin and IC Packaging 20261026

Digital Twin and IC Packaging 20261026 — Transcript

by Chi-Hua Yu · 12,740 words · 2,384 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Can you guys see this one?
0:03The weekday this week?
0:06>> Yes.
0:07>> Okay, good. Good. Good. Thank you so
0:09much, guys. I'm sorry about that. There
0:11are some technical issues. I haven't
0:12used this one for so long. So, yeah,
0:14it's going to take a while. Yeah. And
0:17how's your long weekend? I I hope you
0:19guys have a wonderful long weekend this
0:22uh this long long period of time. Uh
0:26Yeah, we got an extra day due to the
0:28typhoon. The the the typhoon, the
0:30hurricane the typhoon's here. So, hope
0:33everyone's well getting well. And
0:37And so, yeah, today we're going to talk
0:38about something that is important
0:40recently in several different panel
0:43aspect.
0:45So, it's about a seat.
0:48Uh but I cannot use this.
0:50I cannot draw it here, right? Do you Do
0:52you know how can I draw it?
0:56Uh do you know?
0:59This is a slide.
1:02Yeah, I don't think I can draw it on
1:04here.
1:06Okay, I'll try it later. So, so let's
1:08see what's happening here. Okay.
1:13So, yeah,
1:15from previous week week we talking about
1:18uh the the packaging.
1:22Yeah, are you recording this?
1:25>> Yes.
1:26>> Okay, good. So,
1:28we are working on the packaging stuff.
1:32So, start from the beginning.
1:35It's just
1:36typical
1:40typical one.
1:42And the typical one is wire bonding.
1:48So, so no matter what we do is that we
1:51try to buy this die
1:55onto
1:57the substrate.
2:00Right? We will try to build a put a die
2:03onto the substrate.
2:05So,
2:06so it it's important to to see this
2:09onto
2:11we mount the die onto the the substrate.
2:15So, so what is substrate?
2:17Substrate is
2:19an an organic
2:23organic plate.
2:27What's the daughter yell?
2:38What's the daughter yell?
2:40Okay.
2:42And there is very some something that is
2:46not familiar or or may not be
2:49a familiar with you guys is the
2:52how how to monitor, right?
2:56So,
2:57there are there
2:59there are several different kind of
3:00things that you can do. The the first
3:02thing is you can try
3:05you can try
3:07uh just monitor using glue, right?
3:10Why?
3:12We use glue.
3:15Okay, or try with epoxy.
3:20Sorry, epoxy.
3:22Glue or it is epoxy.
3:25Yeah, it should be epoxy.
3:30So, if you try those things together,
3:32you you can just get the uh things that
3:35is that you can you you can binding them
3:38together.
3:39So,
3:40or it should something looks like this.
3:42This is bonding
3:49your substrate.
3:50So,
3:52I'll put a die
3:54the die
3:56on top of it.
4:02It's a die on top of it.
4:04Okay, so the signal
4:06signal out is on top it's on the top or
4:09surface.
4:10It's on the top surface of the die.
4:13So you have so the the one you can get
4:16the signal out is there.
4:18You can do one thing is wire bonding so
4:21you can get the signal out from top of
4:23it and then go to the substrate.
4:27So this is a way we can
4:29easily get the signal out.
4:32And
4:34So there is a power there's a Do you
4:36remember what we do uh last week? We
4:38talked about that the power there's a
4:40signal out.
4:42So once you have current inside
4:44then the current with the the thermal
4:47the those electronic resistance that I
4:49get you
4:50something that important
4:53is the heat. They're going to They're
4:54going to generate heat so the
4:56temperature will rise.
4:58So once you have the heat
5:01once the T increased
5:04then for copper
5:07copper is the the material for those
5:09kind of layout
5:11for for the interconnections.
5:14So the copper
5:17So copper going to be something
5:20that we need to taking care of it. So
5:22one thing is the epsilon
5:25the resistance going to increase as
5:27well.
5:29Why? Because
5:31Why is that?
5:34So we The the reason is we need to ask
5:37why.
5:39Why the resistance is going to getting
5:41higher increase with the temperature?
5:44So
5:45it's need to go back to the electronic
5:48system the atomic systems.
5:50So ladies and gentlemen, let's go back
5:51to So the copper is FCC material.
5:55So what is FCC is
5:57in this kind of a sense in this kind of
6:00a cubic system.
6:03So you have the system we call it cubic
6:06system.
6:07And you can have it
6:11like this way. Okay.
6:13So, the FCC is that we have every atoms
6:17here, here, here, here, here, here. So,
6:21every corners of this
6:24uh it's a
6:29Every corner you have it. And also the
6:31face.
6:33So, you have the face.
6:35Every face you have the
6:38So, so on So, we've got a six face.
6:43Got a six face.
6:48And eight corners.
6:57Uh with all the atoms and put inside.
7:00So, so it's like a face it would call it
7:02FCC.
7:04And FCC is like the
7:07uh close packed.
7:10We call it close packed.
7:15Sorry.
7:16Close packed.
7:19Which means the density
7:21uh of how you do it it is the most most
7:23efficient way you can occupy your system
7:26uh in the in the unit volume.
7:30So, we've got a close packed.
7:31So, the atom here Usually, this is at 0
7:36K. The system is at 0 K.
7:39And
7:40in coppers that uh just think about
7:42there is a pure copper is 99%
7:45of the copper that knows So, so
7:48how does
7:49how does one can be conducted can be a
7:52conductor is so electron can be moved
7:55along this this crystal line structures.
7:57We call it FCC structure. The electron
8:00can be moved.
8:02So, the electron can move along this
8:04kind of wire. So, we copper wire we put
8:07a copper wire.
8:09Let's move along this copper wire. So,
8:12this is a wire that the key can be
8:15wire, they can be transferred inside.
8:18So, the electron can be transferred by
8:20the those because the electron in the
8:23in this metals of the FCC structure is
8:26sharing those kind of a electrons. We
8:29call it free electron because share uh
8:31there's no exactly the position.
8:34So,
8:35we call it electronic cloud. Uh you you
8:38you can use it to to transfer electron
8:42current.
8:43You can transfer. So, you can transfer
8:47electron current.
8:51Electron current.
8:54And heat.
8:57And how do you transfer it with the
8:59heat?
9:03Okay. But, something going to change.
9:05So,
9:06the electron is still moving. Yeah, it's
9:08like driving in China city. So, you are
9:12a
9:13car driver. You you are a driver. So,
9:15there are a lot of scooters.
9:18Right? There are lots of scooters
9:19scooter like electrons. So, the electron
9:21are going to moving away. But, if you
9:24owe this kind of a owe this line is is
9:27aligned perfectly. So, the electron will
9:30find a path to crossing all this kind of
9:32traffic, right?
9:33But, if someone that is blocked. So,
9:36it's like this.
9:37This is a street. This is two lines.
9:40But, one car is here, but the other car
9:42is blocking here.
9:44So,
9:46the electron is difficult to move along
9:49this line. So, they have to crossing
9:52different lines.
9:54Right? So, this is how the resistance be
9:57made be made because the the electron
9:59cannot jump cannot bump into those
10:02protons.
10:04Right? So,
10:05in the this is the original
10:07uh causes of the the the electron
10:10resistance.
10:11And how can you So, can we imagine that
10:14this is from 0 K? This is not okay. This
10:17is a 0 K, sorry. 0
10:20K.
10:21Okay. So, from 0 K the atoms been
10:24certain and everything been been set to
10:27the point. So, it it's all
10:29deterministic. But, once you have
10:31temperature rise,
10:33so according to Boltzmann Boltzmann
10:35distributions, the everything is moving
10:39in a randomness way and also the
10:42randomness way just following a
10:43distribution we call it Boltzmann
10:45distribution.
10:46And this is a the the the system we
10:49describe in a thermodynamic system.
10:52So,
10:53that's how we can say that everything
10:56the entropy the delta S
10:59should be
11:01non-negative.
11:03So, everything system
11:06all the things in this system is
11:08is not to go back to the
11:11ordinary system because the the trend
11:14should be the gradient of the entropy
11:16needs to be non-negative.
11:18Okay. So, once you have a temperature is
11:20higher than it's getting higher, so the
11:22entropy is getting bigger.
11:25So, if you like to go back to the
11:27original system, the entropy need to be
11:30negative. So, somehow it's impossible.
11:33But, we have the temperatures getting
11:35higher, so
11:38Actually, this atom here this atom here
11:42this atom should be moving in a certain
11:46way within the past. So, they're going
11:48to vibrate. We call it vibration.
11:53So, this is
11:54this explain something. So, once you
11:56heat up something,
11:58if you if you heat up
12:00a a matter,
12:02what's the volumetric changes? It's
12:04going to be swell, right? The the the
12:06volume going to expand and swell, going
12:09to get bigger because
12:10the the the the vibration of the atoms.
12:13and the assumption of that is the atom
12:15cannot collide to each other.
12:18So, they're going to occupy more more
12:20space, right? So,
12:23the the volume they they're going to
12:25swell like an expansion. That's why we
12:27have the thermal expansion record
12:29efficient of that. We call it CTE.
12:32Because of the vibration.
12:34So, in metals, everything going to
12:36change it.
12:37And And this is very crucial
12:40to what we're going to talk about today.
12:43So,
12:44in the system of the atomic and the
12:48metal system,
12:50that that this kind of vibration
12:53is depends on the temperature, depends
12:57on the initial system. So, so,
13:00that's going to make you elongations and
13:02volumetric swelling,
13:05or somehow you get the volumetric shrink
13:08to If you If you uh
13:10If you decrease the temperature.
13:13So, once you have the vibration,
13:15those metals going to expand it.
13:18So, you going to expansion, you can have
13:20a expansion.
13:25We call it CTE, the coefficient of
13:27thermal expansion, CTE.
13:30Okay. So, it it going to expansion
13:32because of the the nature of the atoms
13:34that are going to vibrate it.
13:36But, something different from that. So,
13:39not just the vibrating, it's also a
13:41randomness in moving. So, which can make
13:44the electron moving through a
13:47transportation of electron getting
13:49difficult even you have the voltage.
13:52So, voltage is a is a driving force. We
13:54get a higher voltage, which means the
13:56atom Those electrons have a higher
13:58potential
13:59energies that can moving through
14:01traveling through different kind of
14:03direction.
14:05But, somehow it causes another
14:07another thing from that is the expansion
14:10can also cause it with different kind of
14:12a directions.
14:14So, it may not be different. It's may be
14:16different maybe maybe the same maybe
14:19identical, but it's depends on the
14:21electronic structures.
14:23Okay.
14:24So,
14:26once it back to the back to the real
14:29bottom of the the physics or the
14:31crystallizations,
14:33it's it's crucial that you can
14:35understand how it can happen. So,
14:38uh the the electron getting difficult to
14:40moving through the system, so you can
14:43get a higher resistant electronic
14:46electronic electronic resistant.
14:48So, the omega
14:50what we call the the electronic
14:51resistant getting higher.
14:53So,
14:54as you get higher,
14:56what's the energy goes? It's because the
14:58electron we got a potential energy right
15:00from the voltage. So, we got electron
15:04So, we got the electron
15:06with a high
15:09potential energy.
15:14High potential energy here. But,
15:17the moving the the velocity So, usually
15:20they were getting into
15:22uh
15:23transferred into the the the
15:26energies of the
15:28kinetic energy.
15:30So, the electron should So, so the
15:32voltage should make the electron moving
15:34faster.
15:36But, somehow it's not.
15:38Because of the the vibrations going to
15:41be obstacle for the the electron to
15:44moving, so there may be some energies
15:47being transferred into
15:48different kind of way.
15:50So,
15:51how does energy goes? So, usually
15:54they'll be into 100%
15:57from there to kinetic energy.
16:04So, the kinetic energy is the the
16:07velocity
16:09is increasing to
16:11mv squared.
16:15Okay. So, let's start
16:18This is what we call conservative field,
16:20conservative energy.
16:22But not exactly this because we I I what
16:25I mentioned is with a Boltzmann
16:26distribution. Boltzmann is that
16:29the atom is not certain at the
16:32atom structure is vibrating.
16:35So,
16:36it's making it difficult to move. So,
16:39there are some and the So, what do you
16:41make it difficult to move? The velocity
16:44going to change. Right? Some electron
16:47going to be obstacle, some electron are
16:50moving through this. So, the the the
16:53electron goes through that direction
16:55getting
16:56the the number of overall electron is is
17:00less than we expected.
17:03So, what's those energies go? Well, it's
17:06a problem. So, the energy will transfer
17:09into those energy will transfer into
17:11heat.
17:14So, what do you mean transfer into heat?
17:18So, it's create more vibrations. So, the
17:21electron going to
17:23bump into those those kind of
17:25uh protons. They were that's a forbidden
17:28by the quantum physics.
17:31So, once you have electron facing
17:34through the atoms, so they going to
17:37change exchange energies.
17:40So, some power of energy goes
17:42with the electron, but some power of
17:44energy goes to those atoms. So, the
17:47atoms will have more energy, so which
17:50means it's caused more velocity of the
17:52vibrations. So, once you have much
17:56higher vibration, which means the
17:59temperature going to get higher. So,
18:01also they can increase the temperature.
18:04And increase the temperature which means
18:07those electron are getting much more
18:10difficult to move again, right? So, it's
18:14going to cause iteratively.
18:17So, the electron difficult to move and
18:20those four electrons get more density
18:22and more power density. So that not more
18:24more more vibrations. So that so the
18:26temperature getting higher.
18:29And the electron get
18:31much more difficult to move. So much
18:33more energy is transferred
18:35to the atoms and the atom vibrates more
18:39and the temperature increased and
18:42getting uh as a result the signal gets
18:45bad and the electron difficult to move
18:48and everything is in a bad loop of a
18:50this cycle.
18:52Okay. So, this is what we have now. So,
18:56e- you got the power of that and and we
19:00try to reduce the resistance.
19:03But the temperature higher because all
19:05these things that they all the silicon
19:07sink together and
19:10and the high power density.
19:13So, we try to get it more more low low
19:15voltages.
19:17So, if you have low voltages and but the
19:19power is getting bigger. So, you can you
19:21can think about that as a lot of the
19:25the current is flowing through inside.
19:28But it's very low velocity because the
19:31thin resistance is high.
19:34And then you get more and more difficult
19:36to move with this kind of a system.
19:40And then you get more heat. So, it's a
19:42dead end.
19:44So,
19:45this is what we need to solve now. So,
19:47this is a fundamental
19:49uh we call it first principle of why we
19:53have to why we got a heat
19:56uh from electronic device and from the
19:58electronics
20:00uh why is difficult why it is important
20:03we need
20:04to control all this kind of a a current
20:07heat power map stuff.
20:09And the way find a way and we need to
20:12find a way engineer a way that you can
20:14have
20:16all the systems in a controllable
20:18thermal budget or the power budget.
20:22Okay. So, so it's important.
20:25Uh it's something that we need to think
20:26about.
20:28Okay? So guys, any questions from here
20:31from this point of view?
20:34Anyone?
20:39Okay, if no, I will move further. Okay?
20:44I will move further.
20:46Back to the slide. So let's go back to
20:48the slide. So here is
20:50the heat.
20:52So the heat heat up things is that as
20:55you can see here from this
20:59from here from this part of the the
21:02images.
21:03So what's here?
21:05So the bottom one So
21:07I think I can
21:10Uh what should I do? Okay.
21:17Okay, yeah.
21:18I should copy this.
21:21Yes. Okay, great. Uh wonderful.
21:25Uh
21:28I should try this first.
21:30Okay.
21:32Okay, guys.
21:38Okay, guys. Yeah.
21:40So
21:42there are several This is your
21:43substrate, right? This is the substrate.
21:48The substrate you got the signal out.
21:50The heat in here.
21:53And this is the die.
21:59This is what we call die.
22:07Okay, it's purely functional also to
22:09logic.
22:11Logic unit.
22:13So you can see the heat map here. So
22:16what's in what's on top? So this is some
22:19bump.
22:21This is maybe the PGA or C4 bump. So
22:25this may be the C4 bomb.
22:28Maybe the C4 bomb.
22:30And here is the via.
22:37Okay, here is the via. Oh, sorry. Let me
22:40get this out.
22:47This is the via.
22:51And this part
22:54this part
22:56is the lead.
23:00And this part is we call
23:04the cold plate.
23:11So, Leah.
23:12Leah.
23:13Are you there?
23:15>> Yes, Professor.
23:17>> Could you please prepare the demo for
23:19cold plate microchannel?
23:22>> Cold plate microchannel? Uh yes, but
23:24maybe later.
23:26>> Uh-huh.
23:31Okay.
23:35Yeah, so later is okay. Yeah, later is
23:37okay.
23:53Okay.
23:54So,
23:56yeah, this is how we
23:58uh this is thermal So, the this is the
24:01cold plate is the one we call the
24:03thermal management device.
24:06This is for thermal management.
24:13Okay, and there is the one thing we
24:14missed, right?
24:16So, that's the two thing. One is this is
24:18the epoxy. This is for inner compound.
24:24We're just collecting them together.
24:27And what's this one inside
24:30is the thermal interface material.
24:34So, we call this is a thermal
24:38TIM is the
24:40thermal
24:44interface
24:47material.
24:52Thermal interface materials. For the
24:54first one that is on top of the die and
24:58between the die and the lead, we call it
25:00TIM 1. The first
25:03This is the TIM 1.
25:07And also
25:09in between this core play and the lead,
25:12we have another layer is called TIM 2.
25:15Usually, we have a TIM 1 and TIM 2. The
25:18different kind of a material uh it could
25:20be a gel, it could be a glue, it could
25:22be a metal, it could be something we can
25:26use. Some people are using diamonds
25:28uh for the the TIM material as as as
25:32possible as we can for now.
25:35Okay.
25:36So, this is first one we need to talk
25:38about is that
25:41So,
25:42the reason why we have it the die
25:44convert electronic electronic power into
25:47heat. And the reason I just mentioned
25:50that it's from the elect from the atomic
25:55atomic vibration.
26:01Right, the reason from the atomic
26:03vibration.
26:07Mhm.
26:32Okay. So, any questions?
26:43Okay.
26:45So, this is the how we do
26:47with this system. So, we we we need
26:49Today, we're going to we're going to
26:51talk about this. Okay. So,
26:56is the
27:04For the next page, let's go.
27:16Okay.
27:21I need to copy it.
27:37Okay.
27:38Good.
27:44So, today we're going to show you
27:46uh
27:47we're going to
27:49we're going to show you how to do it.
27:50It's we call it one page thermal record.
27:54And we got to take we we need to get
27:56this packaging here. This It's like the
27:59This is the power map.
28:02We're starting with the power map
28:05and the layout design
28:12and get to this
28:15this work. Okay. So,
28:18and for the reliability, this is for the
28:20reliability, not for manufacturing
28:22itself. It's for reliability. So,
28:24reliability is the one that
28:27uh how this company can make money for
28:28it.
28:30So, is it cost too much time for that?
28:33Is it cost too much
28:35uh lot to keep it?
28:37Okay.
28:39So, this is how we can do and how we
28:42should do for the next round.
28:45So, yeah, let's get to the next next
28:48slide, guys. For the next slide.
28:54Okay, let's see.
29:16Okay.
29:17So,
29:19So, we got a
29:20Uh so, this is two things.
29:23So, this two dies, take a look at this.
29:26This is two different die. Right? And
29:28this is what I call it wire bonding.
29:30So, even though they have the overall
29:32power map, the the the overall average
29:35the average power map is the same.
29:38So, which one is going to fail first?
29:41So, so we've got a A and B.
29:44So, I will do the vote. So, you can use
29:48this one here as a vote. Yeah, we can
29:52use
29:53uh this.
29:56So, if you vote for A, give me a love
30:00sign.
30:02Give me a
30:04a thumbs up for B, okay? So, which one
30:07going to fail first?
30:09Guys,
30:11let's do voting now, okay?
30:14We let's see.
30:16So, vote for heart, A for heart, and B
30:19for thumbs. Okay. Starting now.
30:24Everybody? Yeah. Everyone, go.
30:29Good.
30:32Everybody Everyone should do it now if
30:35you're in the room, yes.
30:37Good job.
30:41Which one going to fail first?
30:43A or B?
30:44Okay, let me know.
30:52Okay, so which one going to fail first?
30:55It depends on uh it's just
30:58it it depends on the peak temperature.
31:04So, this is average. So, the overall
31:06power map is the same. So, this is
31:08averagely.
31:10So, the heat
31:11and the the the energy being dissipated
31:14dissipated and distributed equally. So,
31:17the we can expect the peak temperature
31:19should be lower.
31:20But this one, you can see that all the
31:23power
31:24all the heat is concentrated, right? The
31:27singular we call it singularity here.
31:30So, which means something happens and
31:33some bad things happens here.
31:35So, the peak temperature should be
31:36higher.
31:38And this one should be
31:40easily to break because the temperature
31:42is higher and the wiring should be
31:45uh
31:46easier to be wired up to to be break.
31:49Right? So, the short the shortage here
31:52is is likely to to have a shortage.
31:56So, this one is fail from
31:59we call it fail from.
32:04So, they're going to fail easier.
32:06So, [clears throat]
32:08it's not just the power map.
32:11The power density
32:13the power density matters because the
32:14concentrated source causes a steeper
32:17local temperature. So, the the the local
32:20temperature see we can we can call the
32:23conjunction.
32:25Congestion, sorry. The congestion.
32:32Congestion of signal.
32:35And they're going to cause a fail of a
32:36shortage.
32:38Right?
32:40So, this is uh difficult to get this
32:42power map out.
32:44Okay?
32:49So, how can you imagine this one? So, so
32:51the pitch maybe those maybe go together
32:54or the power being located in the single
32:56source.
32:58And and somehow it's causes kind of a
33:01uh we call it power density this
33:03mismatch.
33:05Okay.
33:15So, let's go here.
33:24Here is the
33:27the next one.
33:30Okay.
33:32So, we need to define the thermal
33:33system.
33:35So, how do we define how to predict the
33:37temperature? It's need the source.
33:40What is the source?
33:43The source is the power map.
33:51The source is the power map.
33:57And also the power density.
34:00It's more
34:03power density to more importantly.
34:06Okay.
34:19And the materials. Materials needs
34:21several things. One is
34:24uh for the electric
34:26the Young's modulus
34:29with
34:31the Young's modulus
34:33with temperature
34:35with T.
34:36The Young's modulus with T.
34:39And the K thermal
34:42uh the the the symbol of transmission
34:45the conductivity of the material.
34:48And also you need to CTE.
34:51And some material with the viscoelastic
34:54property.
34:57It's for the polymer.
34:59And also the interface interface
35:02property.
35:03Somehow that the interface that can see
35:05two materials sometimes it's like two
35:07material. It's like two surfaces and the
35:10geometry. What is the geometry? For
35:13that. It's the layout everything and the
35:16boundary conditions. But it's the heat
35:19source been been been processed been
35:22taken care properly.
35:24Okay.
35:28So there's several things the material
35:30property the power map the interface the
35:33geometry and the boundaries.
35:35Okay.
35:46Let's go to the next one.
35:50It's here.
35:57Okay.
36:00So.
36:02The the thermal like a the thermal heat
36:05transfer problem.
36:08We're talking about heat transfer.
36:12There are two stage. One is transient.
36:19Transient step.
36:20So the temperature is still evolving
36:22that the all the things of heat like
36:24electron is moving.
36:26And the once we have long enough time
36:30that may be a few seconds.
36:32It reaches a stable stable status we
36:36call it steady state.
36:40We call it steady state.
36:43So.
36:45Let's just. uh this is usually somehow
36:48we evaluate eventually how this during
36:51the elevated temperature, during the
36:52elevated working working voltages,
36:56how does GPU CPU or chips
36:58uh is working under the supply power.
37:01And so, there is a balance between those
37:05kind of power consumption
37:07and the thermal management system.
37:10So, the balance between these two is
37:14this is microchannel.
37:19This is a fin
37:21fin channel.
37:23And the the the bottom is the power
37:25consumption.
37:27So, you have both of them. You need to
37:29considering how to balance
37:32the the the the consumption from the the
37:35the chip and the thermal management
37:38system you can provide.
37:40But, usually this is difficult because
37:43all the heat, only way you can get it
37:45out is from the top surface of the die,
37:48right?
37:49And the the surface is smaller than you
37:52can imagine. So, uh the overall the
37:55density area is much higher than you can
37:58imagine and you can do.
38:00So, it's not a good way to just control
38:02it afterward. But, there's no way you
38:05can do before you decide the thing. So,
38:08the ones we do with the reliability
38:10check that the you figure it out, okay,
38:13the the the wrong design.
38:15Oh, the yield design, the poor design
38:18can cause a lot of trouble because of
38:20this kind of a
38:22uh e- in equal or or very distributed uh
38:27power map distribution,
38:29power density, then like like a
38:32heterogeneous power density.
38:35And it cause a lot of problem afterwards
38:37like that like like you can imagine.
38:40So,
38:42here's how we can think is can we
38:45prevent it from before from before.
38:47That's why we have this lectures. We're
38:49not just talking about how to do, how
38:53can we do, and how we not to do.
38:56If we talk about if you have the
38:58opportunity, if you have the the ability
39:02to prevent all the things happens, what
39:04can you do? So, we're starting from the
39:06layout. You need to find a better way to
39:08do the layout.
39:09So, try to decrease the power map
39:12heterogeneity.
39:15And then you can move
39:17those kind of a risk factor to the left
39:20hand side.
39:22And there will be uh something
39:25easier to management, easier to control,
39:28and get things done
39:30by yourself.
39:31Okay.
39:34So, here's the the the reason and the
39:37and the philosophy that we do here.
39:40And let's get back to to the next one,
39:43next slides. Okay.
39:50Let's go back to here. Okay. So,
39:53we go to the
39:55the Fourier's law.
39:59So, guys,
40:05the the one dimension we think of this
40:08one is the one dimensional
40:10conduction problem of a continuous lab.
40:14So, the the thermal the thermal flux Q
40:19is equal to cup K is the thermal
40:22conductivity times the area
40:25times the delta T
40:26minus all lines.
40:28So,
40:30so this is the the transmission the the
40:32transmission or constant. If the
40:35conductivity is the K is constant,
40:38then we the only thing we can do is
40:40increase the area
40:42or decrease the the inter connection
40:44length, right? We try to decrease it.
40:47So, we can get higher those kind of a
40:51uh conductions. The the rate the the
40:54heat heat conduction rate.
40:57Q.
40:58Okay.
41:00So, one thing is we can There are
41:03several things if you'd like to increase
41:05those heat flux.
41:08What can you do?
41:10Like to increase the Q.
41:13If you'd like to increase Q.
41:17The first thing is
41:18we can change the material, right? We we
41:21find a high K material.
41:22One is to find a high K material.
41:26But some people are working on that.
41:28So, you can have a high K material.
41:34Like diamond or graphene.
41:39Like diamond or graphene.
41:43The first thing is if you'd like to
41:44increase the the flux.
41:48And
41:49the other one is we try to The second
41:52one is
41:54we increase the the cross-section area,
41:56right? We get the cross-section area get
42:00more the in the contact face.
42:03But like what I say is depends on the
42:06chip you can design. It's It's the So,
42:08so you can have a even you have a higher
42:10core plate
42:11and the bigger core plate the bigger
42:13surface of the core plate you cannot
42:15guarantee that you get the contact
42:16surface cross-section is getting larger
42:19because we are making a like a thin
42:21slide.
42:23So, it's difficult. Uh it's in in
42:26realistic.
42:28The The second thing you can The The
42:30third thing is we try to decrease the
42:32interconnection, right? We try to
42:34decrease the interconnection size. We
42:36try to decrease uh the the the the path
42:39of those kind of flux.
42:42So, it's a trade-off.
42:45How can you find a the best combination
42:47to find a a high flux
42:50of for the heat transmission? Um
42:53it's a it's I have to say it's
42:55it's a typical in art.
42:58It's it's like an art. There's no
43:01straight forward
43:02or do or dies or
43:05wrong or
43:06right answers. It depends on the
43:08trade-off. It It depends on the
43:10compromise. Depends on how many costs
43:13you like to pay.
43:15Right? Also there has a cost.
43:17Everything has a cost. Is it difficult
43:19to make? What's the ERA?
43:22So you need to think of more and more.
43:25Like what I mentioned last week
43:27and not last uh yes last week on
43:29Thursday on Wednesday
43:31the ERA is a single single pass of the
43:34ERA.
43:36But overall ERA is depends on
43:38every step for the single pass. You can
43:40combine them together and
43:42and get the overall resultant ERA.
43:46Oh.
43:48So uh this is the whole of Fourier's
43:50rate and that's why we need to design.
44:07So
44:09So the the thermal resistance we got a
44:12thermal resistance R
44:16thermal resistance get the
44:19but the
44:20this
44:22is
44:23your thermal resistance is defined by
44:27L
44:28is the length and the kappa is A. So
44:32your delta T, remember what we have now?
44:35Remember what we have.
44:37The flux equal to K A delta T over L,
44:43right?
44:44So we now have this one L over KA this
44:47one.
44:48So the delta T is equal to
44:53flux times resistance.
45:02So, if you like to
45:04uh like to get this this temperature out
45:06easily, get a delta T higher. And that's
45:09why we need a delta T higher.
45:12Because we would like to get the
45:13temperature out, right? We get the
45:14thermal out, we get the heat out. So,
45:16you need a delta T to decrease uh
45:20significantly. So, we need to drop the
45:22temperature. So, you need to decrease
45:25uh you need to get it this down. Uh this
45:28is a trade-off.
45:32Okay.
45:33So, here's the you can do.
45:38And the this is a unit, the raise per
45:41the the temperature raised by by per by
45:44power.
45:46So, it's it's the the ratio of this page
45:49this this R
45:51is the temperature raised per watt.
45:54We call it one-dimensional path. This is
45:57ideally a theoretical way you can change
46:00it and and think about how can you
46:02do the first round of judgment call.
46:05Okay, it's not a overall estimation
46:08rule, but it's one thing that we use.
46:11Okay?
46:15So, for the next one,
46:23let's get this one done.
46:25First.
46:32Okay.
46:35So, let's try a 10 W
46:37effective 2 K W pass and 20 K steady
46:40state
46:42the boundary. Okay, so 20 K is which is
46:4520°.
46:4720° C also, but it's 20 K.
46:50So, 2 K W volume or precise entire
46:53effective pass. So,
46:56it's like a 20 K.
46:58So, it's 2 K.
47:00The delta T, this is delta T is 2 K, 20
47:04K, right?
47:08So, 20 K is like a
47:102 *
47:12R.
47:18So, this is your R. This is your R. So,
47:21this is 2
47:23and this is 10 W.
47:26So, you can get the power, the flux.
47:29So, the delta T is
47:3110 * 2 is 20 K.
47:36So, it's like this.
47:39So, if you have the initial boundary
47:41from 25°, with this kind of calculation,
47:43you can imagine, okay, so the the the
47:45temperature should be rise to maybe 45
47:49after this kind of a thermal management
47:51system.
47:53Okay.
47:54So,
47:55let's all let's all for the first round.
47:58So, I think we'll take a break here.
48:00Now, it's yeah, perfect timing. So, now
48:03it's around 7:00. I'll take a break and
48:05let's
48:06talk later, okay?
48:08Good. Thank you so much. And so, Liam,
48:11I'll be back around
48:13around 7:50, okay?
48:16Just take a break, guys. Thank you.
48:19Okay, guys. Welcome back. I'm
48:22so
48:23Yeah, this also the thermal management
48:26problem is is
48:27is getting more and more crucial
48:29recently.
48:32So, let's get back to what we have to to
48:35take care of it now.
48:37So,
48:41>> [clears throat]
48:43>> so
48:44this is 20 K raise.
48:46And it's it's not a the temperature
48:50raise
48:51and absolutely the temperature answer is
48:53different questions. The temperature
48:55raise
48:56uh is 20 K.
48:58Um but the you don't have the boundary
49:01conditions, right? So, so you you need
49:03to get the boundary condition first.
49:06And then you can you can answer it like
49:08what I say.
49:10Uh like what I say in the previous
49:12couple of minutes. So,
49:14we got a series element
49:16here.
49:18And along to the path. So, there are
49:21several one-dimensional series paths
49:23carrying the same Q.
49:29So, the R total can be here.
49:33Let's copy this
49:34and back to here.
49:38Okay, so we can have this here.
49:52Oh, maybe there are several components
49:54here. So, every component
49:58So, we have the every component with the
50:00same Q.
50:03So, the R total can be the the the
50:05summation of overall the thermal
50:07resistance. Do you remember what a
50:09thermal resistance is?
50:12Thermal resistance A
50:14kappa A, right? So, so there are several
50:17uh several factors here.
50:20One is the material property.
50:23The other one is the geometry property.
50:26And this is also for the geometry.
50:28Also, one is cross section, one is the
50:31length, inter connection length. And one
50:33is from the material. So, it combines
50:35two things. One is
50:37one part from material,
50:40one part from the geometry.
50:42So,
50:43if you have several different component
50:45together, then you can combine them with
50:47the resistance
50:48and the overall overall they have the
50:51same heat flux.
50:53So,
50:54the
50:55the the thing that we call it ideal one
50:59uh one pass network. This is the one we
51:02can evaluate this.
51:04And this is a theoretic theoretical
51:06value.
51:10The theoretical value.
51:15Okay?
51:16Same.
51:29So, how many do we have done?
51:32So, how many do we have for this kind of
51:35for this car?
51:44The resistance of the heat transfer from
51:46this direction. There is a die, huh?
51:49Die.
51:51And thermal interface material.
51:56And the serious one here is the co-play.
52:02Or maybe the copper copper co-play.
52:05This is a copper.
52:09See you played.
52:12Or it's a co-play.
52:13Right?
52:15So, it's like a one dimensional
52:18one dimensional system going one
52:20direction.
52:22So, this is the way we can do even even
52:25theoretically to evaluate how the
52:27thermal resistance is.
52:30And this is somehow that we we should
52:33we can provide here.
52:36So, it's also
52:37also can be contributed differently. The
52:39bulk of the tin layer.
52:42And that may maybe the void pressure
52:45sickness and surface shape and
52:47effectively
52:49So, here
52:50this region like you can see here
52:52through this region.
53:07So, just give you an example for that.
53:28So, the pass So, we got this one. This
53:30is synthetic. The 0.4 6.5 kW.
53:35Seriously?
53:36Uh gives a 1.8 K per watt. Uh this is a
53:41pass.
53:42So, from here to here, you can see this
53:44Oh, sorry.
53:46From here to here, this one dimension.
53:51So, how is the raise?
53:54The word raise is 18 K, right?
53:57So, this is just a simple case. So,
53:59usually how many this is This is can we
54:02do?
54:05Well, usually we can get this one to to
54:09maybe this level, but usually
54:12this value is incredibly high.
54:16So, the the heat flux the power density
54:18maybe sometimes is
54:20more than four
54:22more than 450
54:24W.
54:26Somehow it's could be 900 W.
54:31Okay, you do the math. If you do the
54:33math, and this is times it's going to
54:35burn down everything. So, the the
54:37temperature decreases impossibly very
54:40high.
54:41Right? So, so this is the one we need to
54:43decrease just the the temperature and
54:46the power consumption of overall the
54:48flux.
54:49So, we need to separate everything. The
54:52Oh, this is the overall
54:54overall ideally cases we've been taking
54:57We like to tackle.
54:59And you can see that you do the math,
55:00it's impossibly high temperature to
55:02going to melt down everything even with
55:04the silicon.
55:06So,
55:07there are some trade-off we need to
55:08make.
55:10Okay?
55:11So, so how to do this is try to decrease
55:13the the power
55:15or sometimes you need to decrease the
55:17the the thermal thermal resistance
55:19value.
55:27So, with a common hot node and common
55:30core reference,
55:32the heat is inversely related to the the
55:36branches of resistance. If you have
55:38several different branches,
55:40and the resistance the most bottom neck
55:42going to control everything. The
55:44interface going to have
55:45uh going to cause the heat to pass. We
55:48call it heat resistant here. So, that
55:51the heat going to be blocked
55:53and
55:54and being being stored inside the
55:56interface.
55:59And so, we can go next.
56:02So,
56:04So, if you send uh let's see.
56:07Uh
56:08copy this.
56:10So, let's do the math. So, just very
56:12simple math. You you see that it's
56:13really simple math that you if you like
56:16to you don't have to use calculus. Just
56:18a very simple mathematics
56:20uh
56:21consumption here. So,
56:23and the the [clears throat]
56:23conceptually, we got a branch split
56:27to to do the different branches.
56:30But, we say that it 10 W with 2 kW.
56:33So, the drop is 12 the temperature decay
56:36is 22.
56:39But, two and six branches.
56:42Okay?
56:44So, the two branches is not serially.
56:47So,
56:48the R equal
56:52The R equal can be
56:542.1 + 6.1 over 1.
57:01X 6 over 4.
57:043 over 2.
57:07I did Right.
57:09R over 1 so R equal
57:13is 2 over 3.
57:15It's 3 over 2, sorry.
57:20This 1.5.
57:23A for what?
57:27So once you do if you do the math, you
57:30you can decrease it, but then you
57:32increase more pass.
57:34So which means you can increase more
57:36pass, so that's why we need those are
57:38multiple pass, multiple layer because
57:41you you see the top carrier now, the
57:44races it only
57:46uh only 16k.
57:48Right, the the the total what?
57:51I just raised 15k and it's okay.
57:54So
57:55there are several ways that you can do
57:56theoretically to to decrease the
57:58temperature. One is to serial but then
58:00now if you like to do the temperature,
58:03you'd like to decrease the temperatures
58:05or or or make the temperature uh
58:08increase less than than than the race is
58:11less than than the one thing you need to
58:13think about is how to get a branch of
58:15the spirit the temperature pass. Get
58:17more passes, you get more theoretical
58:20value and it it will be easier to
58:22design.
58:25So there going to be a problem, there
58:27going to be a question in the midterm.
58:30I will show you that
58:32uh you need to define which one's
58:34serial, which one's branches.
58:36And then you do the math to calculate
58:39it. It's a simple calculation.
58:41Okay, so this going to happens in you
58:43during the in your midterm exam. Okay?
58:49Good. And
58:51for next one.
58:54So
58:55it got several branches, right?
58:58So you got the code play, you got the
59:00same team material, and you got several
59:02different layout for the branches you
59:05can do.
59:06So the sink
59:09uh it's one branch from the simplified
59:11circuit and you can have branches.
59:14You can have hopefully you can have two
59:15faces or two several different make
59:17mechanisms. You combine them together
59:19like a cold plate.
59:21With a vapor chamber.
59:24So,
59:27Also, heat this one So, so also you can
59:29increase the temperature and the the
59:31area, right? Like you can you can
59:33decrease the thermal resistance by
59:35increase the area.
59:37So,
59:38remember how we define it?
59:41The R is equal to what? The R is equal
59:44to
59:47Remember the R is equal to L over K
59:51times A, right? So,
59:55So, if you increase the area, increase
59:57the cross section area, increase the
59:59surface area, then then one is
1:00:01important. Then you can get to this
1:00:03point is try to reduce it.
1:00:13Okay? Great job.
1:00:18So, let me ask one thing. Leia,
1:00:22Leia, are you there?
1:00:26Leia,
1:00:26>> Yes, professor. Yes.
1:00:28>> And Leia, thank you so much. Uh the So,
1:00:30are you ready for the cold plate
1:00:32microchannel?
1:00:33>> Yeah, microchannel is ready.
1:00:35>> Okay. So, how about that? Because now
1:00:38we're almost done with the theoretical
1:00:40part. I will give you 5 minutes to warm
1:00:43up with the microchannel stuff.
1:00:47>> Okay, then let me show my screen.
1:00:50>> Absolutely. Thank you so much. So, guys,
1:00:53we'll move on to the microchannel first.
1:00:56And then use and that Leia to to to show
1:01:00you the cases.
1:01:01Uh Leia, please
1:01:03Uh it's a little bit small. I think that
1:01:05the screen is Yeah.
1:01:06>> Yeah, sure.
1:01:06>> I want to
1:01:07>> Yeah.
1:01:08>> So, could you please uh explain us a
1:01:11little bit of what we are doing with
1:01:13this microchannel? Uh What's this What's
1:01:16this color means and what are you doing?
1:01:19And can you use agent
1:01:22to talk to this?
1:01:24>> Okay, sure. And everyone
1:01:28First of all, let me introduce the
1:01:30Uh for one for we are doing the thermal
1:01:33dispense, we will first of all build up
1:01:35a heat map.
1:01:37For example, this should be a chip and
1:01:39there are some dies on it.
1:01:41And so it will be a heat source.
1:01:44And for example, if this part is much
1:01:47more temperature higher, so maybe from
1:01:5025 W, we can set it as maybe 80 W.
1:01:54So it will be much more heat.
1:01:57And therefore, we will try to make the
1:02:02the heat to disperse to everywhere and
1:02:05by by the cold plate.
1:02:08So our AI prediction will do the will
1:02:11help us to do the design.
1:02:13And this is our material that can This
1:02:16is a model that can to
1:02:19do some can do for us.
1:02:22It will automatically run many
1:02:25They will run the AI prediction. So the
1:02:27these three are the
1:02:30best option that AI recommend us to use.
1:02:34And then we will make try to use these
1:02:37three kind of the AI prediction.
1:02:40And so
1:02:41maybe I'll use a a new one because I
1:02:43just changed the design.
1:02:45So be here and see if I can
1:02:49bring it to a brand new line.
1:02:53Yeah, it is.
1:02:54And then we will There will be three
1:02:56kind candidates. For example, I can see
1:02:58here.
1:03:00And it will automatically help us to
1:03:02design what the microfluid will go
1:03:05through and the shape will be like this.
1:03:08So we will press the button to generate
1:03:11the microfluid channel.
1:03:14And it might take a while, maybe a few
1:03:16seconds.
1:03:19And here, here's the design.
1:03:21And so, the pattern will be like just
1:03:23like what we
1:03:25draw here on the left-hand side.
1:03:29And then,
1:03:31we can do do the simulation here.
1:03:34And we can set the flow rate and maybe
1:03:38you what kind of fluid we use. For
1:03:40example, in this case, it's water or
1:03:42maybe we can use PG25 or anything you
1:03:45need, maybe air if you need some.
1:03:49And also the what kind of solid you use.
1:03:51For example, here is the copper.
1:03:55And after a few seconds of the
1:03:58simulation, it will show the result
1:04:00here.
1:04:07And here, here's the result.
1:04:11And it seems weird because it will be a
1:04:13solid part and fluid part. If once I
1:04:17hide the solid part, it will be the only
1:04:19the liquid part here.
1:04:21So, we can see that this part is will be
1:04:23much more higher
1:04:24of temperature is because the heat
1:04:26source is based on this place, just like
1:04:28the design we made.
1:04:30And here, yes, you can you can see
1:04:34this is a hot spot.
1:04:36So, we are trying to make the
1:04:38temperature lower and in this place and
1:04:41in every place. So, this is our first
1:04:44design.
1:04:45And then, we can try to do another two
1:04:49simulation.
1:04:51So, I am maybe not try not to do this in
1:04:53same time.
1:04:55And this is done, so I can run the
1:04:56second one.
1:05:04And of course, if you made a flow rate
1:05:06higher, the temperature will be lower.
1:05:08But the But you would you will
1:05:12if your material is not that stable,
1:05:14maybe the higher flow rate will break
1:05:17the microchannel. So, you have to set
1:05:20this flow rate carefully.
1:05:24Also, the third one.
1:06:05Okay, then there will be a compare of
1:06:07the results with three kind of design.
1:06:11And we can choose the result which which
1:06:13one is what we prefer.
1:06:16And the professor says about our AI
1:06:18agent, we also had a sub agent for our
1:06:22model, and maybe I will make a new
1:06:25new one to ask the agent help us do it.
1:06:29And so, this might be small for you. So,
1:06:31there's a
1:06:33some buttons the the default buttons.
1:06:35So, for example, I can say
1:06:38beautiful work work flow for me.
1:06:43And you will say which product would
1:06:46So, for now I will use microfluidic
1:06:49and
1:06:50I'll save this building.
1:06:53Oh, maybe I didn't
1:06:55make a name first.
1:06:56Sure.
1:06:58Okay,
1:06:59I'll put here. Sorry, a little bit okay.
1:07:03Okay.
1:07:04Demo one.
1:07:06See.
1:07:09And then
1:07:10it's already saved. Okay, sorry.
1:07:14But
1:07:18Okay, so we
1:07:19we are so lazy to and a new name. But
1:07:21anyway, I will I will
1:07:23keep going.
1:07:24And so we can say we use AI predict.
1:07:28And how do you want to set the heat
1:07:29power map? We can draw by ourselves or
1:07:31maybe we can use our file to upload.
1:07:35For here, I will say draw a power map.
1:07:38And also and with the default power map.
1:07:44Do you like any tiny agent note that can
1:07:48uh write the requirement so we can
1:07:50conclude the workflow.
1:07:55So it'll be here.
1:07:57So as you can see, just just like we
1:08:01previous one, and this will be the
1:08:03default heat map.
1:08:06And also it can just automatically help
1:08:08us to build a
1:08:11workflow for us. And you once we use the
1:08:13run pipeline, it will automatically run
1:08:15everything and we can just compare the
1:08:18result here.
1:08:20Or also we can upload uh
1:08:23file in here. It just depends on what
1:08:25you need.
1:08:26So
1:08:27upload will be right here.
1:08:50Yeah, for example, this is my is a is a
1:08:55the heat map I randomly draw, so it'll
1:08:57be look hot. Also you it can help help
1:09:01us to see what the result will be like.
1:09:11And then
1:09:14So the AI particular is uh based on the
1:09:16previous previous, so maybe I will make
1:09:19it a new one.
1:09:22Yeah.
1:09:23Predict.
1:09:36Okay, and these are my brand new
1:09:40uh prediction for three kind of design.
1:09:43And I can just run the pipeline and
1:09:44we'll see the result.
1:09:46And it would take some time, so maybe
1:09:50I think that's all for the demo here.
1:09:55Any question for the students?
1:09:58>> Uh thank you, Leon. So, so could you
1:10:00please go back to the and explain for
1:10:04the power map?
1:10:06Let's go back to the power map.
1:10:08>> Here?
1:10:10>> Okay, yeah, guys. So,
1:10:11this is actually from the die.
1:10:15And if somehow there is some discrete
1:10:18discretization for the other maybe very
1:10:21high power density
1:10:23some place and very
1:10:25uh discrete discretized. So, so somehow
1:10:29phantom may not be so converted
1:10:32together.
1:10:33And
1:10:34the
1:10:35the the this is like the boundary
1:10:36condition
1:10:38or we have now.
1:10:40So, usually we we're thinking about the
1:10:43the overall geometry is like this. So,
1:10:46Leon, could you please go to
1:10:48go back to that. So, there is something
1:10:50that we can change, right?
1:10:51For the power map, not just for the
1:10:53power map, also for the power density
1:10:56and the geometry. Like you can you can
1:10:58you can change the geometry.
1:11:01>> Sure, I can change the place and also I
1:11:03can make it more tiny if you want.
1:11:08>> You can change the size, right? You can
1:11:10you can get the size.
1:11:12>> And the center will usually be the ASIC
1:11:15and this might be the HBM and this might
1:11:18be the in and out.
1:11:21Right, oh.
1:11:22Let's say we can change the layout here
1:11:25just what we need.
1:11:29And maybe you can change the width and
1:11:30length if you need.
1:11:35>> So so they are to
1:11:37Uh so can they work on this one
1:11:42by themselves as a homework?
1:11:45>> Uh you may let them to use our website
1:11:49and do something
1:11:50>> Yeah, using the online version. Is it
1:11:53possible? So so do we have this kind of
1:11:55a
1:11:56capa- capability for them to work on
1:11:59this as a homework?
1:12:00>> Uh I think we can
1:12:02minor uh
1:12:04let everyone to do it at same time.
1:12:06Maybe it will
1:12:08be stuck.
1:12:10But I think I can ask the IT to let
1:12:12everyone get a
1:12:14permission to log in.
1:12:17>> I see. I see. But
1:12:19that would be great. So
1:12:21uh I did guys, I'd like you to play with
1:12:24this tool. And
1:12:26and let us know is there anything that
1:12:29do you think it's it's
1:12:30it's it's useful or it's not or even
1:12:33it's not useful. Okay?
1:12:36So
1:12:38So they are good Just do the next one,
1:12:40please.
1:12:41For the next step.
1:12:43>> Next step, you mean the
1:12:46Yeah, here's a Go back.
1:12:48>> no no not a result. Not a result. Oh,
1:12:50here.
1:12:51>> You mean the layout?
1:12:53>> Yeah, they are here. Okay. So for this
1:12:55layout, ladies and gentlemen,
1:12:57so uh do you know what does this
1:12:59meaning? What's this a meaning for?
1:13:02Uh this is the the the chance you can
1:13:04create a channel. So usually people like
1:13:06to create a channel.
1:13:08Already to hand draw it. Uh it's rather
1:13:11difficult. Some somehow it's very
1:13:13difficult to create this patterns
1:13:15according to that.
1:13:17So so it can be
1:13:19adjusted by the micro channel or
1:13:22adjusted by the by the power power power
1:13:26map. Oh, yeah, I remember we can do this
1:13:28one.
1:13:29Using AI to create automatically, right?
1:13:33>> As these results are made by AI.
1:13:37>> Right. So, so the AI generate the
1:13:39pattern.
1:13:40>> Yes, these patterns are generated by AI.
1:13:43>> It's Yeah. So,
1:13:45the idea that we can do is we got the
1:13:48very powerful model.
1:13:51And this model can just figure it out
1:13:54uh what is the base temperature once you
1:13:57have this cold plate and with the the
1:14:00fluid channel.
1:14:02Um
1:14:03also you can change it with this So,
1:14:05this is how many How many branches do we
1:14:07have, guys?
1:14:09To this design, how many branches do we
1:14:12have?
1:14:13Uh
1:14:15The
1:14:16As you can see, with the two branches,
1:14:18right?
1:14:19So, one one branch is for single
1:14:21channel. So, you got two branches.
1:14:24The which means we got
1:14:26the overall one-dimensional
1:14:29branches from one to two, so you can
1:14:31decrease
1:14:33uh the thermal resistance.
1:14:35So, this is ideally designed.
1:14:37And then
1:14:39you can have one more mechanism
1:14:41to to branches
1:14:43uh the heat pass.
1:14:45Then you can have more power black at
1:14:47the If you have phase change, then it's
1:14:50another pass.
1:14:52Okay?
1:14:53So, it's one thing It's something like
1:14:55that.
1:14:57Uh let's move on. Move on to the next
1:14:59one. So, so Leah, please go to
1:15:02uh the design. So, go to the AI design.
1:15:06>> AI design? You mean AI prediction?
1:15:09>> Yeah, so AI prediction.
1:15:11Here, uh you can choose the temperature.
1:15:15The first one is the somehow we decide
1:15:17the temperature.
1:15:19Uh somehow we'll decide uh not just for
1:15:23the temperature, but the
1:15:25but the velocity also the pressure drop.
1:15:28So, so you can have the the in the
1:15:30temperature and so the boundary
1:15:32temperatures can be here. And the agent
1:15:34here, we got the agent
1:15:36to tell you what is where running now.
1:15:40So the engine can can can see you see
1:15:42this is real agents that can talk and
1:15:45think what is going to have what's
1:15:47happening here.
1:15:49So
1:15:51for this microchannel design, you can
1:15:53you also send it to 3D printing if you
1:15:56have the chance.
1:15:58Okay.
1:15:59So guys,
1:16:00uh
1:16:01uh this is how we do and and and go to
1:16:03the the result.
1:16:05Yeah.
1:16:06The CFD result. So after the AI
1:16:09calculation, you can just send them
1:16:11send all the results
1:16:13uh to a AI
1:16:15to to for AI results for verification.
1:16:19So we can see the verification now.
1:16:21Yeah.
1:16:27Yeah, yes.
1:16:29Yeah, thank you so much to see this one.
1:16:31Yeah. Yeah.
1:16:33So
1:16:34Leo Wu, please explain what's happening
1:16:37here.
1:16:38Uh this is a temperature field, right?
1:16:41>> Yes, it's temperature field and this is
1:16:42here's the inlet and here's the outlet.
1:16:45So the water will go through this tube
1:16:47and go
1:16:49here and go out.
1:16:51Uh we can show you the temperature and
1:16:54also we can show the velocity and also
1:16:56also the pressure.
1:16:58So you can see here is the inlet so the
1:17:01pressure might is the largest.
1:17:05So the pressure drop is important
1:17:07because we want to know about what
1:17:10if the pressure drop too much, maybe
1:17:13the flow will stuck in some place and
1:17:15maybe broke the microchannel.
1:17:18And here's the velocity. You may want to
1:17:20know to see if the
1:17:22also if the flow will stuck or maybe
1:17:25make a vortex inside and it's bad for
1:17:28our simulation.
1:17:32And also here we can see the temperature
1:17:34T max and T min
1:17:37and also the for the solid and for the
1:17:39liquid
1:17:41the fluid and also the summer resistance
1:17:43you can calculate here.
1:17:45And also we can make a probe to see
1:17:49check the for example here is the
1:17:51most higher highest temperature might be
1:17:53in this area we can check.
1:17:56And we if you want to check the lower
1:17:58place you can check make a probe here.
1:18:02Once you want to see the result.
1:18:15>> Yeah, also we have velocity, right?
1:18:18Pressure and velocity.
1:18:19>> Yes, velocity is over here.
1:18:22>> And do we have the the third the
1:18:24particle tracing is not ready, right?
1:18:26>> Yes, it's not ready yet.
1:18:29>> Uh but the the pressure and velocity
1:18:33and this velocity is for the micro
1:18:35flowing
1:18:36the most comfortable liquid
1:18:39or is up flowing through?
1:18:46And they are this is some problem. They
1:18:47said that the engine never settle or
1:18:50converges the solution.
1:18:56For this one that you see there's
1:18:58warning.
1:19:00>> Yes, the solver did not converge.
1:19:02>> Yeah.
1:19:03Send this to send this to send this case
1:19:06so for this case and send this to
1:19:09the formula.
1:19:10>> Okay.
1:19:11>> Yeah, we need to figure out what
1:19:13happened here.
1:19:15Yes, so so sometimes this happens here.
1:19:17So the AI may have the result but the
1:19:21the CFD
1:19:23the computational fluid dynamics may not
1:19:25be able to find a result.
1:19:27Also the pressure, yeah.
1:19:30Go to the pressure.
1:19:31>> The pressure.
1:19:34>> Yeah, also that this is some point
1:19:36there's
1:19:37there are some critical points seems to
1:19:39be uh seems odd.
1:19:42Okay, but yeah, go to next one. We we
1:19:44got several several different results.
1:19:46This one may not may need to run and
1:19:48check the
1:19:49to check the data first. Let's go to the
1:19:52second solution.
1:19:54>> Yes, here and first I have to comment
1:19:57that this this model is the previous one
1:20:00we set on the internet and
1:20:04maybe it's not update for the latest
1:20:06version. So
1:20:07>> Okay, maybe there is some maybe there is
1:20:10probably something wrong. So maybe
1:20:11something wrong with that.
1:20:13>> Yes.
1:20:14Yeah, this one is also wrong.
1:20:15>> see this is also something wrong with
1:20:18the results. Yeah. Anyone knows why this
1:20:20is odd?
1:20:23Oh, because of the temperature, right?
1:20:24You see the temperature here at the
1:20:26boundary.
1:20:28It's not working. So yeah, something
1:20:30happens here. Okay, go to the next one.
1:20:33So Leon, this one maybe the
1:20:35I think maybe the the Reynolds number.
1:20:37Did you check the Reynolds number?
1:20:40>> Not yet and
1:20:42>> I think the Reynolds number is too high
1:20:45for this one.
1:20:46Yeah, but talk to him.
1:20:48>> Yeah, in our latest version the Reynolds
1:20:51the Reynolds number is supposed to in
1:20:53the model, but it did not is the much
1:20:56earlier one. So there we can see here.
1:20:59>> I see. I see. So so this is not the
1:21:01latest version.
1:21:03>> Yes.
1:21:04>> Okay. Yeah.
1:21:06No worries. Uh so so yeah, thank you so
1:21:08much Leon. And so guys, this is a
1:21:11microfluidic channel. So this version
1:21:13may not we have some maybe issues for
1:21:16the the simulation, but the AI result is
1:21:19I think it's okay because we tried it
1:21:21with some our customer and partners
1:21:24before.
1:21:26So yeah, I think we're going to stop
1:21:28here for a while and then back in 10
1:21:32we'll get a recession for 10 minutes and
1:21:34then we'll be back.
1:21:36So
1:21:37uh after we get back,
1:21:40we will show you the auto routing we
1:21:43talked to you last week, right? The auto
1:21:45routing. Starting with the auto routing.
1:21:48And then we'll move on to the lectures.
1:21:50And then after the lectures, we'll end
1:21:53around 8:30.
1:21:55Okay?
1:21:57So, any questions?
1:22:00No, then we'll get 10-minute break. So,
1:22:02Liam talk to Wei-Hao and ask him to demo
1:22:06>> Okay.
1:22:06>> at the 8:00 p.m. shop. Okay?
1:22:08>> Okay. Okay.
1:22:10>> Thank you so much.
1:22:12>> Hey guys, we are back.
1:22:13>> Thank you so much. Liam.
1:22:17>> Yes.
1:22:18>> Are you ready?
1:22:20>> Uh is Wei-Hao Wei-Hao is ready, I think.
1:22:26>> Liam, are you there?
1:22:27>> Yes, I'm here. And Wei-Hao, are you
1:22:29here?
1:22:31>> Hey, sorry. I think that my my earphone
1:22:34is getting
1:22:37weird.
1:22:38Let's get it down here.
1:22:40I'm back.
1:22:42So, do you hear me?
1:22:43>> Yes.
1:22:45>> Okay, good. Liam, uh so did you get
1:22:47Wei-Hao ready?
1:22:49>> Uh yes, Wei-Hao, can you answer me?
1:22:52>> Wei-Hao, are you there?
1:22:52>> Wei-Hao.
1:22:53He's online now.
1:22:55>> Yeah, Wei-Hao. Yeah.
1:22:56So, we we So, ladies and gentlemen, we
1:22:59have
1:23:00we had introduced
1:23:02the auto routing last week.
1:23:05And we'd like to show you what to do
1:23:07with the auto routing.
1:23:08So, Wei-Hao, could you please
1:23:14So, Liam, please explain
1:23:16uh the the the stuff I'm going to teach
1:23:19you.
1:23:20Wei-Hao.
1:23:23Yes.
1:23:24You [clears throat] can share your
1:23:25screen.
1:23:28What happens?
1:23:31>> Okay.
1:23:32>> Good.
1:23:36Wei-Hao, please go ahead and explain
1:23:39what we are doing.
1:23:40>> So, so so first upload the
1:23:44the file magnet
1:23:45>> As you are yeah. Quite helpful, please.
1:23:47>> Yeah.
1:23:48>> So, starting from what we are doing,
1:23:51uh what we what we we like to share.
1:23:54Uh what's
1:23:56what's this interface? Okay. So, yeah.
1:24:01>> Oh, so this is
1:24:03the best for our auto loading.
1:24:07So, so so fast. First, we upload the
1:24:13the bar magnet uh the data of the bar
1:24:15magnet, then
1:24:19Okay.
1:24:20In the
1:24:21in the
1:24:22camera, then
1:24:27I think this is for our bar magnet.
1:24:31And here, our software can do the auto
1:24:36Mhm.
1:24:39grouping and
1:24:43I need one file to open.
1:25:02>> Okay, go ahead.
1:25:03>> Wait.
1:25:05>> Me, too. So, what's this?
1:25:07>> Oh, so this is the auto grouping of the
1:25:11bar magnet.
1:25:13And
1:25:19And next,
1:25:23next, we go to the
1:25:26routing.
1:25:29So, here we do the
1:25:32the auto loading from bar to die S like
1:25:38Hey,
1:25:39wait.
1:25:40Hey.
1:25:41Wait a minute.
1:25:43Maybe I just
1:25:47Run.
1:25:49Okay.
1:26:19No.
1:26:22Please wait for the auto loading now.
1:27:27>> Wait, how do what's what's here now?
1:27:29>> Oh, the the loading is not loading now.
1:27:31Oh.
1:27:32Almost done.
1:27:40Okay, then.
1:27:42It's the third part of the loading, so
1:27:47you can see
1:27:49the loading is only two
1:27:52from the die
1:27:54for the bump to the die edge.
1:27:57So,
1:27:58after this, we can do the
1:28:02next one.
1:28:05So, the die die edge to the ball is
1:28:08the full full map loading
1:28:12from the die edge
1:28:14to the ball outside.
1:28:32So, back to the
1:28:34bump to die edge.
1:28:37So, you can see our software can do the
1:28:40order lane like we got layer one, layer
1:28:44layer two.
1:28:46Layer four.
1:29:44>> So, how so what's what's happening here
1:29:46now?
1:29:47>> Oh.
1:29:47>> You're waiting
1:29:48>> the
1:29:49Yeah, yeah, yeah.
1:29:50>> You're waiting for the run. So,
1:29:51So, you guys So, let's us overall
1:29:55ideally that that you can run
1:29:58the files those kind of file
1:30:01originated from Cadence
1:30:03or some EDA tool EDA EDA software.
1:30:07Those software help you to do the IC
1:30:09design.
1:30:11Once you have a physical design you can
1:30:12do auto routing. They help you to get
1:30:15those things done
1:30:17by by just automatically get the the
1:30:21first priority first run of the test. So
1:30:25this one could going to cause maybe from
1:30:27from days to months is
1:30:30for the first design. So sometimes it's
1:30:32very difficult to work with.
1:30:35So in our tool that you can you can say
1:30:38that just just clicking and AI going to
1:30:40do the rest of the trick.
1:30:43It's quite quite significant and quite
1:30:45remarkable.
1:30:52So now it's around 70%. So that we have
1:30:55this one is getting smart they are
1:30:56getting faster, right?
1:30:59I remember previous one is taking a lot
1:31:01of time.
1:31:03Yeah, yeah.
1:31:05This
1:31:07small case
1:31:08maybe for only only
1:31:12200 100 bump.
1:31:46Most time maybe 2 minutes more. Okay.
1:32:09>> Done.
1:32:12Yeah, it's fully
1:32:15from from to ball routing.
1:32:28>> So guys, you can see this auto routing
1:32:30with different layer. So you can see
1:32:32that from the from the signal to the
1:32:35diode. So so where is it now? It's just
1:32:38from to the diode to to to the ball.
1:32:43Yeah.
1:32:43>> From the From the ball to the to the
1:32:46ball.
1:32:47>> Diode to the ball. The ball to the ball.
1:32:49>> this is a fully
1:32:51fully routing. And here.
1:32:55We into
1:32:56from to diode.
1:32:57>> From to diode. Just only. Yeah, this is
1:32:59your only to the
1:33:01>> to the diode.
1:33:03So you need to go to the diode first to
1:33:06see how the conjunctions is and then we
1:33:08So So this is what we have now. So so
1:33:12the also also the different color means
1:33:14the different type of the signal, right?
1:33:18>> Uh yeah, for for the
1:33:20ball. I mean the the the the one
1:33:24the
1:33:25the different signal. And
1:33:27for the light, this
1:33:30show the different layer.
1:33:33Yeah.
1:33:42>> Okay. Good.
1:33:44So any questions from the audience?
1:33:46Anyone?
1:33:48So the two Actually, only 10 minutes you
1:33:50can you can run this. So So you can do
1:33:53it. So I don't I'm not sure if anyone
1:33:55you're working with this before. But but
1:33:58this one you you should take a
1:34:00even not just for days, months, or weeks
1:34:03to finish.
1:34:04So, it's quite remarkable you can use a
1:34:07to do things that is Oh, now we keep it
1:34:09to 10 to to a couple minutes.
1:34:12Okay?
1:34:14So, yeah, thank you for your help. It's
1:34:15great great job.
1:34:18Thank you.
1:34:19And so, yeah, we we keep moving. Thank
1:34:21you.
1:34:22And any questions from the audience?
1:34:26If it's not, then we'll move on to
1:34:30uh the next part.
1:34:32Okay?
1:34:36So,
1:34:50you don't look at the things
1:34:54Thank you for your help. Thank Thank you
1:34:56so much.
1:35:16Okay.
1:35:17Guys, so we'll move on to
1:35:20uh to the the rest of the the lectures.
1:35:24I will try to get this one
1:35:26uh done as soon as you can as we can.
1:35:30Okay.
1:35:31So, here's the
1:35:36Oh, the iPad. It's not working.
1:35:52Okay, good. Now it's getting better. We
1:35:54get that.
1:35:55Be here. Okay.
1:35:57So,
1:35:58yeah, let's get back to it.
1:36:10So, that like that This is what we call
1:36:12the share row share row. So, you can
1:36:15have different passes.
1:36:18Okay.
1:36:23And so, you can So, so, look here. This
1:36:26is a spreading the plate.
1:36:29Okay. So, there's some some some
1:36:31strategies you can do.
1:36:33And
1:36:37So, this is two dies. Uh if you couple
1:36:40them together, like what we have at the
1:36:42advanced packaging,
1:36:43so uh this is two dies that can be a
1:36:45similar similar couple.
1:36:48So, that which means the heat's going to
1:36:49change
1:36:51uh the effect the could effect each
1:36:53other.
1:36:54So,
1:36:55So, power which means power in one die
1:36:59can therefore shift the temperature to
1:37:01another. So, there has a heat and
1:37:04temperature conduction from this one to
1:37:06another one.
1:37:08And
1:37:10So, with the boundary conditions,
1:37:12sometimes it's different. So, so, this
1:37:13is maybe the cooling test.
1:37:16And and the air cooling temp test and
1:37:19this is the CT TCT thermal cycle
1:37:21testing.
1:37:23Okay. So, there's several different
1:37:24ways.
1:37:25So, here is the fixed temperature. The
1:37:28first one is a fixed temperature. This
1:37:30one fixed temperature. We can we can
1:37:32keep the temperature in a certain way.
1:37:35And the prescribed heat flux.
1:37:41So, we get the ambient over the
1:37:42temperature, so the heat flux out, and
1:37:45the convection impulse
1:37:47or the convection impulse.
1:37:49So, now, uh we as one we have the
1:37:52SpaceX. I'm not sure if you know the
1:37:54SpaceX.
1:37:57The SpaceX. That is a radiation. Those
1:38:01for the boundary condition we see we
1:38:02think about radiation more and more
1:38:04often.
1:38:08And
1:38:09so this is some some we have conduction
1:38:12before, right? Fourier one. But also we
1:38:14can have the convections.
1:38:17So
1:38:18our convections is another term that is
1:38:21uh air ambient or something that is cool
1:38:25to to to try to get the heat out also.
1:38:29So the T uh
1:38:32absolute very infinite is the ambient
1:38:35temperature. This is ambient
1:38:36temperature. Usually it's 25, but it's
1:38:39depends.
1:38:41And TS is the surface temperature. H
1:38:44is
1:38:46uh H is the the the the height
1:38:49of the an area.
1:38:51Okay.
1:38:52The H is
1:38:55effective H effective height at exposed
1:38:58area.
1:39:00Okay.
1:39:04So the H depends on the fin.
1:39:08This is what we call the fin. The fin
1:39:10height. The H depend on the fin height.
1:39:13So there is a highest temperature you
1:39:16can define and the lowest temperature
1:39:18you like to achieve, but the higher
1:39:19there is a limited high that you can you
1:39:23can achieve by the manufacturing. So
1:39:25somehow it's going to be a trade-off.
1:39:33And the thermal capacity capacitance
1:39:36thermal capacitance like a thermal
1:39:38reservoir.
1:39:40So before the steady state, which means
1:39:42something the input power raises store
1:39:45the the energy temperature can be stored
1:39:47inside the system like a heat like a
1:39:50like a we get a capacitor.
1:39:52So we can store those electronic current
1:39:56like a reservoir, like a water
1:39:58reservoir.
1:39:59And so the part of the input power can
1:40:02be transferred into the energy. So, the
1:40:05thermal energy. And it's it's it's very
1:40:08difficult
1:40:10to get it back
1:40:11to the real electronic power electronic
1:40:14energies.
1:40:16Once it get into the thermal because it
1:40:18it what I mentioned that it's a it's a
1:40:20system vibration, right?
1:40:22So, you got the system is system
1:40:24vibration.
1:40:31You've got a system vibration.
1:40:35So, then some power was stored in a
1:40:37thermal management in die and packaging.
1:40:40And then it's also the one you need to
1:40:43think about.
1:40:47Okay. So, RC time constant.
1:40:51Uh so, we call it lump first order
1:40:53response. T
1:40:55This is tau equal to the resistance
1:40:58and the
1:41:00capacity.
1:41:03The R times C
1:41:05is the characteristic characteristic
1:41:07time line determinants.
1:41:10Okay. So, if it's just one effective
1:41:13node
1:41:15in a single packaging. So, like in this
1:41:17directions.
1:41:27So,
1:41:28somehow uh during our design especially
1:41:31for reliability
1:41:33we need to check are you designing for
1:41:35the peak value
1:41:37or the average value?
1:41:40If you design for the peak value,
1:41:42something is correlated to the the
1:41:44things is different.
1:41:46And then we will say, "Okay, this is
1:41:50uh this is somehow we need to change it
1:41:52with uh we change it from the average
1:41:54value to the peak value because somehow
1:41:56peak value going to going to break the
1:41:59whole system."
1:42:00So, by average temperature can heat hide
1:42:04a local aspect.
1:42:06So, usually this is the concentration.
1:42:12Concentration that we'd like to avoid.
1:42:15So, usually, the the design the
1:42:17mechanism before people use it like
1:42:20average. But, now more and more people,
1:42:22especially when getting to
1:42:24the new era for AI PCs and the it's the
1:42:27high power density
1:42:30high power consumption system
1:42:32architecture. So, we try to avoid the
1:42:34die average or a design parameter.
1:42:38We try to use is
1:42:41a a maximum. What is the peak
1:42:44temperature?
1:42:45What is highest temperature during the
1:42:47work step of the process?
1:42:51So,
1:42:52also,
1:42:53uh we need to think about that the
1:42:55temperature only with the power.
1:42:58We need to compare the peak
1:43:01temperature alone
1:43:03with power state.
1:43:06What's the power map
1:43:07and the boundary condition?
1:43:12Uh where?
1:43:13Where is the measurement and the time
1:43:15definition?
1:43:20So, the candidate temperature must refer
1:43:22to the same power, cooling time. Uh so
1:43:25so the candidate temperature, what do
1:43:27you mean? The candidate temperatures is
1:43:30the design temperatures. Refer to the
1:43:32same power.
1:43:34If you talking about the temperature is
1:43:36around
1:43:38around what? It should be under the same
1:43:40power map, power density.
1:43:42And also for the same cooling system.
1:43:45The time, the extraction location, where
1:43:48the thermal couple be measured.
1:43:51So, there'll be different different
1:43:52design design design mechanism, but it
1:43:55it's overall good to the So,
1:43:59So,
1:44:00after that, like what we do? After the
1:44:02design, you need to check is the
1:44:04numerical convergence. So, unlikely our
1:44:07model I the have only some problems
1:44:10finding out the numerical convergence.
1:44:11It's happens all the time. So, that that
1:44:13the model going to warn you that you
1:44:15need to check the setup for the
1:44:17simulation.
1:44:18But, overall for here,
1:44:20the the the fine mesh and find a better
1:44:23time step,
1:44:24it's a numerical stability
1:44:27uh uh not a physical accuracy. You you
1:44:30can get the better result, but you
1:44:31cannot guarantee the accuracy for
1:44:33physics. It's different. So, running a
1:44:36simulations
1:44:37is try to get the simulation down
1:44:40without no numerical errors one thing.
1:44:44But, try to guarantee
1:44:46those result of physics is another
1:44:49thing. Another thing which you
1:44:51guarantee.
1:44:52>> [snorts]
1:44:52>> So, the numerical convergence and
1:44:54everything and and and the rest of the
1:44:57thing
1:44:58uh is different.
1:44:59So,
1:45:01So, the most So, you guys are engineers.
1:45:05So, you will be engineers. You need to
1:45:07measure your things. So, we'd like to
1:45:09measure where the decision is sensible
1:45:12sensitive.
1:45:13The very crucial part of this working
1:45:18is to find out where to measure
1:45:20and what decision is sensitive to. So,
1:45:23you can have a design map. Those design
1:45:26map is to find out which a physical
1:45:28quantity
1:45:30can in or or the interface or the peak
1:45:33value could
1:45:34inference can reverse the candidate can
1:45:38try to find out okay, this this is the
1:45:40best. So, like what I mentioned,
1:45:43if you change the code code rate,
1:45:46the main change to the the peak
1:45:48temperature or your design temperature.
1:45:51But, if you change to the IC design and
1:45:54the physical design, they're going to
1:45:55significantly change something.
1:45:58But, it's going to take a lot of time.
1:46:01So,
1:46:02measure the decision is what to make it
1:46:05sensitive.
1:46:06Uh it's it's also important to that.
1:46:11So, there's is 2K so a 2K nominal leads
1:46:16can be reversed from A along to the
1:46:19interface sensitivity. So,
1:46:22what do you mean that a 2K nominal
1:46:25nominal lead?
1:46:27So, this is just example. We have a
1:46:31board. So, all our assumptions
1:46:34theoretically everything should be
1:46:35contained and everything should be works
1:46:37through perfectly.
1:46:39But the predict here and it's it's an
1:46:42interface
1:46:44because of the uncertainty, it's about
1:46:460.7 kilowatts. So, it's about three
1:46:48different degrees. Somehow, this could
1:46:51be very sensitive
1:46:53to your power map.
1:46:57Okay.
1:46:58So,
1:47:01So,
1:47:02once we done
1:47:04with the measurement with everything
1:47:06that you have already tested, you check
1:47:09all the design parameters and send it to
1:47:11manufacturing.
1:47:13The the next step is to test the images,
1:47:17test the results in shadow marine
1:47:20using infrared.
1:47:22So, use the infrared camera can see the
1:47:25radiation.
1:47:28And from the accessible surfaces, we
1:47:31have to find out, okay, which angle is
1:47:33good, which the temperature distribution
1:47:35is close to what I measured.
1:47:39So,
1:47:40So, you you can do it with the impact
1:47:42AI. So, so like a like a what we show
1:47:46you here, the platform record must be
1:47:47reserved exact model.
1:47:49And then you can have the model to run
1:47:51the simulations and do do a digital twin
1:47:54like what we do
1:47:55here.
1:47:56Okay. So, and also for the the the heat
1:48:00paths,
1:48:01so in our impact AI can have the heat
1:48:04paths, right, during the different
1:48:05passes to see what's happening, what's
1:48:08inside the microchannel.
1:48:10So,
1:48:11so we do have all everything done for
1:48:15the summer and what did you think for
1:48:17this week? So, yeah, I think overall
1:48:20it's great. And it's a wrong time. So,
1:48:24yeah, so sorry about that. I have to I
1:48:26have to drag you to here
1:48:29for another time for for the lecture,
1:48:32but I'm
1:48:33I'm sorry that I have to meet with some
1:48:35important important partners here. So,
1:48:39yeah.
1:48:40And thank you for your consideration and
1:48:42and cooperation.
1:48:46But before we answer it,
1:48:48so I'd like to mention one thing is that
1:48:51next week I'll be in United States. So,
1:48:56I would like you guys maybe
1:48:59I would see if we can do it online
1:49:02in the United States during the time,
1:49:03but maybe it's like a different time
1:49:05zone. So, maybe
1:49:08different So, we will also use the
1:49:10online versions with a chat and I'll
1:49:13find a place maybe inside Stanford
1:49:15University
1:49:17to to do have the lectures. So, yeah, we
1:49:20would we will announce it just just
1:49:22close just stay close to what we have
1:49:25and try to
1:49:27try to get the things that we can down
1:49:30with you guys
1:49:31and to to try to get you to get for more
1:49:34and more latest cutting edge stuff in
1:49:38advanced packaging.
1:49:39Okay?
1:49:40So, any questions for the audience?
1:49:45And if it's not, I think we'll stop here
1:49:48and let you
1:49:50all for for dinners or supper and hope
1:49:54you guys have a wonderful week this
1:49:56week. And then we'll have we'll have
1:49:59announcement for the what what what
1:50:01going to do for the next week. I would
1:50:03say that it would be great if we can
1:50:06live this lectures
1:50:08inside the the campus of Stanford.
1:50:11Because it's beautiful to really
1:50:12beautiful. I love that place.
1:50:15And also I can show you what's Silicon
1:50:17Valley inside. I can maybe we can find a
1:50:19office from Google
1:50:21and for for maybe we can settle down it
1:50:23for we maybe several different hours.
1:50:26Uh I I can I can go to I can I can live
1:50:30broadcasting you with him inside meta
1:50:32inside
1:50:33uh inside inside the Google Google
1:50:36building and also inside Nvidia and
1:50:38Apple Park. Right? How about that? And
1:50:41so I will let you know. It is so if it's
1:50:43going to be a a
1:50:45a an hourly, that would be easier. Maybe
1:50:47we can all arrange it
1:50:49uh during the time that you you're all
1:50:52all all feasible and easy access.
1:50:55Uh so uh the TA going to going to
1:50:58announce it uh on the Moodle and also
1:51:01you can you can download Now you can
1:51:03download the the slides. People some
1:51:05some students that let me know that
1:51:07you'd like to download the slides and
1:51:08then so the slides now can be
1:51:10downloadable.
1:51:12Okay?
1:51:13So, any questions from from the
1:51:15audience? Final Final remark?
1:51:18No?
1:51:19Uh if it's not, then thank you so much
1:51:21and thank you for your cooperation to to
1:51:25I'm sorry for the short notice.
1:51:27But I appreciate your time and
1:51:29appreciate your effort. Thank you so
1:51:31much and we'll I will meet you next week
1:51:33in in North America. Okay? Have a good
1:51:36one. Bye. See you guys.

About this transcript

This page contains the full transcript of Digital Twin and IC Packaging 20261026 by Chi-Hua Yu, generated from the public captions YouTube serves with the video. The transcript has 12,740 words across 2,384 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.