WEBVTT 00:00:00.000 --> 00:00:04.991 align:middle line:84% [SQUEAKING] [RUSTLING] [CLICKING] 00:00:04.991 --> 00:00:16.780 align:middle line:90% 00:00:16.780 --> 00:00:19.072 align:middle line:84% PROFESSOR: OK so let's start here. 00:00:19.072 --> 00:00:21.280 align:middle line:84% So what I will do here is I will discuss a little bit 00:00:21.280 --> 00:00:23.160 align:middle line:84% about the devices people have actually 00:00:23.160 --> 00:00:26.160 align:middle line:84% been able to make using this kind of phase change materials. 00:00:26.160 --> 00:00:29.180 align:middle line:84% The first example here obviously is a bit outdated. 00:00:29.180 --> 00:00:32.560 align:middle line:84% I don't know if any of you still use CDs, DVDs. 00:00:32.560 --> 00:00:35.040 align:middle line:84% But the technology actually behind this kind 00:00:35.040 --> 00:00:37.360 align:middle line:84% of rewritable CD and DVD is exactly this kind 00:00:37.360 --> 00:00:38.920 align:middle line:90% of phase change alloys. 00:00:38.920 --> 00:00:42.040 align:middle line:84% So what I'm showing here is like a laminate structure 00:00:42.040 --> 00:00:45.460 align:middle line:84% of if you actually look into what rewritable DVDs, 00:00:45.460 --> 00:00:46.858 align:middle line:90% this is what it looks like. 00:00:46.858 --> 00:00:49.400 align:middle line:84% They always have a substrate made of polycarbonate, basically 00:00:49.400 --> 00:00:50.220 align:middle line:90% just a plastic. 00:00:50.220 --> 00:00:53.480 align:middle line:84% That's the material you use to actually make the disk. 00:00:53.480 --> 00:00:55.800 align:middle line:84% On top of that, you then have a layer 00:00:55.800 --> 00:00:59.340 align:middle line:84% of these phase-change materials sandwiched between the two 00:00:59.340 --> 00:01:02.820 align:middle line:84% dielectric layers that provide essentially a protection 00:01:02.820 --> 00:01:04.680 align:middle line:84% to prevent essentially oxidation, 00:01:04.680 --> 00:01:06.900 align:middle line:84% and also prevent any kind of harmful interactions 00:01:06.900 --> 00:01:10.020 align:middle line:84% of the [INAUDIBLE] materials with the plastic substrate. 00:01:10.020 --> 00:01:13.080 align:middle line:84% Because keep in mind, when you actually switch these materials, 00:01:13.080 --> 00:01:14.713 align:middle line:84% especially when you amortize them, 00:01:14.713 --> 00:01:16.880 align:middle line:84% you have to heat them up to pretty high temperature, 00:01:16.880 --> 00:01:20.220 align:middle line:84% usually around 600 degrees Celsius to melt it. 00:01:20.220 --> 00:01:22.720 align:middle line:84% And then you quickly quench down to get the amorphous state. 00:01:22.720 --> 00:01:24.740 align:middle line:84% So you don't want that process to actually harm 00:01:24.740 --> 00:01:27.140 align:middle line:90% your plastic substrate. 00:01:27.140 --> 00:01:28.700 align:middle line:84% Now one thing I want to notice here 00:01:28.700 --> 00:01:32.580 align:middle line:84% is that for this particular application, 00:01:32.580 --> 00:01:36.540 align:middle line:84% the material we're using here is not a typical GST. 00:01:36.540 --> 00:01:40.540 align:middle line:84% It's a little bit modified composition over indium 00:01:40.540 --> 00:01:43.880 align:middle line:84% antimony telluride, or what people call AIST. 00:01:43.880 --> 00:01:47.740 align:middle line:84% So you can think of the AIST essentially 00:01:47.740 --> 00:01:51.943 align:middle line:84% as a variant of essentially antimony telluride. 00:01:51.943 --> 00:01:54.360 align:middle line:84% It's the base composition is basically antimony telluride. 00:01:54.360 --> 00:01:57.710 align:middle line:84% People add still over and indium to basically modulate 00:01:57.710 --> 00:01:59.270 align:middle line:90% the crystallization kinetics. 00:01:59.270 --> 00:02:02.110 align:middle line:84% It really doesn't play that much of a role. 00:02:02.110 --> 00:02:06.910 align:middle line:84% So it's basically indium telluride-based materials. 00:02:06.910 --> 00:02:08.789 align:middle line:84% And then on top of that, you then actually 00:02:08.789 --> 00:02:10.949 align:middle line:90% have a metal reflective layer. 00:02:10.949 --> 00:02:13.030 align:middle line:84% The reflective layer serves to essentially try 00:02:13.030 --> 00:02:16.270 align:middle line:84% to capture as much optical power as possible into the phase 00:02:16.270 --> 00:02:17.550 align:middle line:90% change material layer. 00:02:17.550 --> 00:02:20.310 align:middle line:84% So it literally lowers the power requirement for writing. 00:02:20.310 --> 00:02:22.088 align:middle line:84% And then finally, on top, there's 00:02:22.088 --> 00:02:23.130 align:middle line:90% another layer of raising. 00:02:23.130 --> 00:02:25.510 align:middle line:84% And then there's actually another polycarbonate. 00:02:25.510 --> 00:02:28.030 align:middle line:84% So that's what the disk structure looks like. 00:02:28.030 --> 00:02:30.330 align:middle line:84% So when you're actually doing a data recording, 00:02:30.330 --> 00:02:32.990 align:middle line:84% you then use a focused laser beam 00:02:32.990 --> 00:02:36.470 align:middle line:84% to deliver the power that's needed to either crystallize it 00:02:36.470 --> 00:02:38.070 align:middle line:90% or to amortize it. 00:02:38.070 --> 00:02:40.950 align:middle line:84% And you change the duration of the laser pulse 00:02:40.950 --> 00:02:43.590 align:middle line:84% so it's a longer pulse, lower power. 00:02:43.590 --> 00:02:46.430 align:middle line:84% This is how you actually induce crystallization, very short 00:02:46.430 --> 00:02:48.050 align:middle line:90% pulse, but very intense. 00:02:48.050 --> 00:02:48.830 align:middle line:90% You then melt it. 00:02:48.830 --> 00:02:50.390 align:middle line:90% And you amortize it. 00:02:50.390 --> 00:02:54.310 align:middle line:84% So when you actually switch the crystalline structure 00:02:54.310 --> 00:02:56.570 align:middle line:84% of the material, as we have mentioned before, 00:02:56.570 --> 00:02:59.570 align:middle line:84% there is a very large change in terms of the refractive index, 00:02:59.570 --> 00:03:01.450 align:middle line:90% and hence the reflectance. 00:03:01.450 --> 00:03:03.490 align:middle line:84% Crystalline state is more reflective. 00:03:03.490 --> 00:03:05.550 align:middle line:84% And more first state is less reflective. 00:03:05.550 --> 00:03:07.850 align:middle line:84% So then based on that, you can then actually 00:03:07.850 --> 00:03:09.963 align:middle line:84% impact laser intensity writing and then 00:03:09.963 --> 00:03:12.130 align:middle line:84% use that to actually record digital information ones 00:03:12.130 --> 00:03:13.250 align:middle line:90% and zeros. 00:03:13.250 --> 00:03:16.610 align:middle line:84% Now this is the first technology where 00:03:16.610 --> 00:03:19.050 align:middle line:84% this fancy new material was applied in the data storage 00:03:19.050 --> 00:03:20.330 align:middle line:90% domain. 00:03:20.330 --> 00:03:22.963 align:middle line:84% Obviously, as I said, this is an outdated technology. 00:03:22.963 --> 00:03:24.630 align:middle line:84% I don't think anyone really even use it. 00:03:24.630 --> 00:03:27.490 align:middle line:84% I don't even have a CD-DVD slot on my laptop. 00:03:27.490 --> 00:03:30.970 align:middle line:84% So, however, it doesn't mean that this phase change material 00:03:30.970 --> 00:03:32.920 align:middle line:90% is completely out of favor. 00:03:32.920 --> 00:03:35.170 align:middle line:84% So the other technology where they have been used here 00:03:35.170 --> 00:03:39.530 align:middle line:90% is called phase change memory. 00:03:39.530 --> 00:03:41.030 align:middle line:90% For short, it's also PCM. 00:03:41.030 --> 00:03:43.447 align:middle line:84% So it can be either Phase Change Materials or Phase Change 00:03:43.447 --> 00:03:43.970 align:middle line:90% Memory. 00:03:43.970 --> 00:03:45.428 align:middle line:84% What I'm showing on this slide here 00:03:45.428 --> 00:03:47.190 align:middle line:84% is just a little bit of historic context, 00:03:47.190 --> 00:03:48.570 align:middle line:84% how this technology was developed 00:03:48.570 --> 00:03:50.930 align:middle line:90% over the past few decades. 00:03:50.930 --> 00:03:54.270 align:middle line:84% It all started back in about in the 1960s, 00:03:54.270 --> 00:03:58.950 align:middle line:84% where the scientist Ovshinsky, he discovered a unique 00:03:58.950 --> 00:04:01.430 align:middle line:84% phenomenon called threshold switching in [INAUDIBLE] 00:04:01.430 --> 00:04:02.190 align:middle line:90% compounds. 00:04:02.190 --> 00:04:04.030 align:middle line:84% We are going to talk about it in the next slide about what 00:04:04.030 --> 00:04:04.510 align:middle line:90% this is. 00:04:04.510 --> 00:04:05.790 align:middle line:84% But it's basically a phenomenon that 00:04:05.790 --> 00:04:08.165 align:middle line:84% really enables the operation of this kind of phase change 00:04:08.165 --> 00:04:09.250 align:middle line:90% material devices. 00:04:09.250 --> 00:04:14.190 align:middle line:84% And that's why a lot of these cautionary-based phase 00:04:14.190 --> 00:04:18.630 align:middle line:84% change devices, sometimes they are also called ovonic devices, 00:04:18.630 --> 00:04:21.829 align:middle line:90% oh-von-ic. 00:04:21.829 --> 00:04:24.630 align:middle line:84% This is really just named after the person 00:04:24.630 --> 00:04:28.390 align:middle line:84% who discovered this phenomenon here, Ovshinsky. 00:04:28.390 --> 00:04:28.890 align:middle line:90% All right. 00:04:28.890 --> 00:04:32.790 align:middle line:90% Then a couple of years later-- 00:04:32.790 --> 00:04:37.750 align:middle line:84% actually, Gordon Moore-- so he's the same person from Intel 00:04:37.750 --> 00:04:39.963 align:middle line:90% that proposes Moore's law. 00:04:39.963 --> 00:04:41.630 align:middle line:84% So he was actually the one that actually 00:04:41.630 --> 00:04:45.710 align:middle line:84% developed the first phase change memory device. 00:04:45.710 --> 00:04:51.410 align:middle line:84% The first device can storage up to 256 bits of information. 00:04:51.410 --> 00:04:53.030 align:middle line:90% Now you wonder how much that is. 00:04:53.030 --> 00:04:55.730 align:middle line:90% Each English letter is 8-bit. 00:04:55.730 --> 00:04:58.250 align:middle line:84% So basically what I'm putting in this box here, 00:04:58.250 --> 00:05:00.930 align:middle line:84% 256-bit phase change memory demo. 00:05:00.930 --> 00:05:03.572 align:middle line:90% That's right, about 256-bit. 00:05:03.572 --> 00:05:05.530 align:middle line:84% That's about how much information you can store 00:05:05.530 --> 00:05:08.130 align:middle line:90% in that entire memory unit. 00:05:08.130 --> 00:05:10.490 align:middle line:84% Now at the time, probably you will wonder, 00:05:10.490 --> 00:05:14.650 align:middle line:84% why is that even useful to store just a few words? 00:05:14.650 --> 00:05:17.450 align:middle line:84% But fortunately, what we know is that a lot 00:05:17.450 --> 00:05:20.170 align:middle line:84% of these electronic devices, semiconductor devices, 00:05:20.170 --> 00:05:23.130 align:middle line:84% they are density follows exponential growth, known 00:05:23.130 --> 00:05:24.330 align:middle line:90% as Moore's law. 00:05:24.330 --> 00:05:26.930 align:middle line:84% Same thing happens, actually, with memory units. 00:05:26.930 --> 00:05:29.010 align:middle line:84% And I don't know who said it, but I 00:05:29.010 --> 00:05:31.490 align:middle line:84% think they said exponential growth is the most 00:05:31.490 --> 00:05:33.410 align:middle line:90% powerful force in the universe. 00:05:33.410 --> 00:05:35.850 align:middle line:84% So after a few decades, you can imagine that you actually 00:05:35.850 --> 00:05:38.610 align:middle line:84% get devices that can store lots of information, hundreds 00:05:38.610 --> 00:05:40.850 align:middle line:90% of gigabytes of information. 00:05:40.850 --> 00:05:41.350 align:middle line:90% All right. 00:05:41.350 --> 00:05:43.990 align:middle line:84% So the next milestone comes in the 1980s. 00:05:43.990 --> 00:05:46.610 align:middle line:84% This is from a group in Japan, Yamada's group. 00:05:46.610 --> 00:05:49.402 align:middle line:84% They first discovered this GST competition decomposition 00:05:49.402 --> 00:05:51.860 align:middle line:84% along the pseudo boundary line, essentially pseudo boundary 00:05:51.860 --> 00:05:54.500 align:middle line:84% between germanium telluride, and [INAUDIBLE]. 00:05:54.500 --> 00:05:57.820 align:middle line:84% And they discovered that this combination uniquely exhibit 00:05:57.820 --> 00:05:59.780 align:middle line:84% this very fast and also reversible 00:05:59.780 --> 00:06:02.820 align:middle line:84% switching behavior that enabled the whole field application. 00:06:02.820 --> 00:06:06.820 align:middle line:84% So starting from about 1990s, early 2000, 00:06:06.820 --> 00:06:09.620 align:middle line:84% several big companies started to deliver essentially commercial 00:06:09.620 --> 00:06:10.740 align:middle line:90% modules. 00:06:10.740 --> 00:06:14.300 align:middle line:84% They actually offer memory operation. 00:06:14.300 --> 00:06:17.340 align:middle line:84% So now in this next slide, I'm showing you 00:06:17.340 --> 00:06:21.860 align:middle line:84% a quick schematic about the essential structure of this kind 00:06:21.860 --> 00:06:22.880 align:middle line:90% of phase change memory. 00:06:22.880 --> 00:06:26.540 align:middle line:84% So the core of it is really pretty simple. 00:06:26.540 --> 00:06:28.660 align:middle line:84% The most simple version basically 00:06:28.660 --> 00:06:32.740 align:middle line:84% has two metal electrodes with a layer of this phase change 00:06:32.740 --> 00:06:34.460 align:middle line:90% material sandwiched in between. 00:06:34.460 --> 00:06:37.980 align:middle line:84% So you can use those metal electrodes to essentially apply 00:06:37.980 --> 00:06:40.460 align:middle line:84% current, pass it directly through a phase change 00:06:40.460 --> 00:06:43.140 align:middle line:84% material-- which then generate joule heating-- 00:06:43.140 --> 00:06:44.520 align:middle line:90% and use the joule heating. 00:06:44.520 --> 00:06:46.980 align:middle line:84% You can use that to actually change the structural state 00:06:46.980 --> 00:06:47.760 align:middle line:90% of the material. 00:06:47.760 --> 00:06:50.600 align:middle line:90% 00:06:50.600 --> 00:06:52.720 align:middle line:84% At the same time, you also want to be 00:06:52.720 --> 00:06:55.400 align:middle line:84% able to read out the structural information, all 00:06:55.400 --> 00:06:58.800 align:middle line:84% the digital information encoded in the structure. 00:06:58.800 --> 00:07:00.600 align:middle line:84% So for this kind of memory, this is 00:07:00.600 --> 00:07:05.240 align:middle line:84% done through measuring the electrical conductivity. 00:07:05.240 --> 00:07:08.200 align:middle line:84% So the crystalline state, because of the delocalization 00:07:08.200 --> 00:07:11.520 align:middle line:84% of the electrons in this kind of resonance bonding structure, 00:07:11.520 --> 00:07:13.760 align:middle line:84% it tends to be a lot more conductive compared 00:07:13.760 --> 00:07:15.520 align:middle line:90% to the amorphous state. 00:07:15.520 --> 00:07:17.960 align:middle line:84% It's several orders of magnitude more conductive. 00:07:17.960 --> 00:07:22.600 align:middle line:84% So then these two's the same pair of electrodes 00:07:22.600 --> 00:07:24.720 align:middle line:84% that you use to inject current can also 00:07:24.720 --> 00:07:28.237 align:middle line:84% be used to essentially read out the resistance by essentially 00:07:28.237 --> 00:07:30.320 align:middle line:84% setting a much lower voltage so it doesn't disrupt 00:07:30.320 --> 00:07:31.580 align:middle line:90% the structural material. 00:07:31.580 --> 00:07:33.140 align:middle line:84% But then you measure the current through it. 00:07:33.140 --> 00:07:34.640 align:middle line:84% So then from that, you actually know 00:07:34.640 --> 00:07:37.400 align:middle line:84% whether it's 0 or 1 or it's amorphous or crystalline. 00:07:37.400 --> 00:07:40.820 align:middle line:90% 00:07:40.820 --> 00:07:42.500 align:middle line:84% So there's a pretty simple design here. 00:07:42.500 --> 00:07:44.700 align:middle line:84% I mean, the more state of the art efficient memory, 00:07:44.700 --> 00:07:46.440 align:middle line:84% they have different kind of architect architectures, 00:07:46.440 --> 00:07:48.840 align:middle line:84% more like a mush-- what they call mushroom design. 00:07:48.840 --> 00:07:50.268 align:middle line:90% But the basic idea is the same. 00:07:50.268 --> 00:07:52.060 align:middle line:84% Essentially, you still have two electrodes. 00:07:52.060 --> 00:07:53.600 align:middle line:84% And paths come directly through it. 00:07:53.600 --> 00:07:58.483 align:middle line:84% Now one challenge, however, is that in order to be able to pass 00:07:58.483 --> 00:08:00.900 align:middle line:84% current through it-- so to generate enough power to switch 00:08:00.900 --> 00:08:02.540 align:middle line:90% the face-- 00:08:02.540 --> 00:08:05.940 align:middle line:84% if you think about the-- look at the IV diagram here, 00:08:05.940 --> 00:08:07.120 align:middle line:90% We have two lines here. 00:08:07.120 --> 00:08:11.260 align:middle line:84% If you focus on this two solid lines, this line, 00:08:11.260 --> 00:08:14.820 align:middle line:84% this is basically the crystalline state. 00:08:14.820 --> 00:08:18.980 align:middle line:84% The line at the bottom here, this one is amorphous state. 00:08:18.980 --> 00:08:21.020 align:middle line:84% Now, of course, amorphous state is electrically 00:08:21.020 --> 00:08:22.860 align:middle line:90% almost insulating. 00:08:22.860 --> 00:08:24.613 align:middle line:84% So that's why it has a very small slope. 00:08:24.613 --> 00:08:26.780 align:middle line:84% But then one of the challenge to immediately come up 00:08:26.780 --> 00:08:28.660 align:middle line:84% is that you still need to pass a lot of current 00:08:28.660 --> 00:08:30.410 align:middle line:84% through amorphous state to crystallize it. 00:08:30.410 --> 00:08:32.980 align:middle line:84% If it's insulating, how do you even actually pass current 00:08:32.980 --> 00:08:34.500 align:middle line:90% through it? 00:08:34.500 --> 00:08:37.140 align:middle line:84% So that's where the threshold switching behavior comes in. 00:08:37.140 --> 00:08:38.620 align:middle line:84% So in these [INAUDIBLE] materials 00:08:38.620 --> 00:08:40.880 align:middle line:84% exhibit this kind of very unique behavior. 00:08:40.880 --> 00:08:44.800 align:middle line:84% If you apply an amorphous state, high voltage, 00:08:44.800 --> 00:08:47.640 align:middle line:84% once you pass some kind of threshold voltage, 00:08:47.640 --> 00:08:51.800 align:middle line:84% the material suddenly becomes electrically conductive. 00:08:51.800 --> 00:08:55.960 align:middle line:84% Now this electrical increase of electrical conductivity 00:08:55.960 --> 00:08:59.040 align:middle line:84% is different from crystallization 00:08:59.040 --> 00:09:01.820 align:middle line:84% in that the microstructure, if you look at atomic arrangement, 00:09:01.820 --> 00:09:03.680 align:middle line:90% it still remains the same. 00:09:03.680 --> 00:09:07.083 align:middle line:84% In fact, when it goes to a conductive state, that's 00:09:07.083 --> 00:09:08.500 align:middle line:84% why you see voltage suddenly drop, 00:09:08.500 --> 00:09:12.840 align:middle line:84% current suddenly increases, if you decrease the voltage again, 00:09:12.840 --> 00:09:15.980 align:middle line:84% it goes back to the initial amorphous structure. 00:09:15.980 --> 00:09:18.960 align:middle line:84% As long as you don't go into this very high current regime 00:09:18.960 --> 00:09:21.840 align:middle line:84% where you start introducing material modification. 00:09:21.840 --> 00:09:25.520 align:middle line:84% So fundamentally, it basically says 00:09:25.520 --> 00:09:28.720 align:middle line:84% that this kind of so-called threshold switching behavior, 00:09:28.720 --> 00:09:31.180 align:middle line:84% it's purely electronic in nature. 00:09:31.180 --> 00:09:34.040 align:middle line:84% It doesn't involve atomic rearrangement. 00:09:34.040 --> 00:09:39.160 align:middle line:84% So to explain what happens, we can take a look 00:09:39.160 --> 00:09:42.090 align:middle line:84% at this, a very schematic electronic tectonic structure 00:09:42.090 --> 00:09:42.750 align:middle line:90% diagram there. 00:09:42.750 --> 00:09:44.930 align:middle line:84% So what I'm showing you here, this 00:09:44.930 --> 00:09:47.950 align:middle line:84% will be conduction band, valence band. 00:09:47.950 --> 00:09:52.313 align:middle line:84% This is the band gap in the amorphous phase change material. 00:09:52.313 --> 00:09:53.730 align:middle line:84% So later on in the semester, we're 00:09:53.730 --> 00:09:57.370 align:middle line:84% going to discuss more in details about the electronic structures 00:09:57.370 --> 00:09:59.010 align:middle line:90% of amorphous semiconductors. 00:09:59.010 --> 00:10:01.610 align:middle line:84% But for now, it suffices to say that 00:10:01.610 --> 00:10:04.530 align:middle line:84% unlike crystalline materials, where in the band gap 00:10:04.530 --> 00:10:07.090 align:middle line:90% you have zero density of states. 00:10:07.090 --> 00:10:10.970 align:middle line:84% For amorphous materials, there are still some localized states 00:10:10.970 --> 00:10:13.210 align:middle line:90% within the band gap. 00:10:13.210 --> 00:10:15.050 align:middle line:90% So you have electronic states. 00:10:15.050 --> 00:10:18.730 align:middle line:84% And the Fermi level typically will still populate somewhere 00:10:18.730 --> 00:10:21.170 align:middle line:90% within the band gap. 00:10:21.170 --> 00:10:24.090 align:middle line:84% And below the Fermi level, obviously, the states 00:10:24.090 --> 00:10:24.730 align:middle line:90% are filled. 00:10:24.730 --> 00:10:27.970 align:middle line:84% Above the Fermi level, the states are empty. 00:10:27.970 --> 00:10:29.790 align:middle line:84% So from your solid-state physics, 00:10:29.790 --> 00:10:32.127 align:middle line:84% you may have learned that if it's off your state, 00:10:32.127 --> 00:10:34.710 align:middle line:84% it doesn't conduct electricity because electrons are fermions. 00:10:34.710 --> 00:10:35.668 align:middle line:90% You cannot move around. 00:10:35.668 --> 00:10:37.793 align:middle line:84% If it's empty, obviously, there are no [INAUDIBLE]. 00:10:37.793 --> 00:10:39.130 align:middle line:90% You cannot buy electricity. 00:10:39.130 --> 00:10:42.230 align:middle line:84% So this kind of well-defined Fermi surface 00:10:42.230 --> 00:10:44.790 align:middle line:84% essentially says that this system is electrically 00:10:44.790 --> 00:10:45.478 align:middle line:90% insulating. 00:10:45.478 --> 00:10:47.270 align:middle line:84% It's very difficult to conduct electricity. 00:10:47.270 --> 00:10:49.790 align:middle line:84% So what happens when you actually 00:10:49.790 --> 00:10:51.590 align:middle line:84% start to apply a large voltage across it 00:10:51.590 --> 00:10:55.150 align:middle line:90% to trigger threshold switching? 00:10:55.150 --> 00:10:59.030 align:middle line:84% Once the voltage reaches a certain level, the electrons 00:10:59.030 --> 00:11:01.190 align:middle line:84% and holes, they start to get enough energy 00:11:01.190 --> 00:11:03.550 align:middle line:84% so that you can start to cause intermixing of the states 00:11:03.550 --> 00:11:05.550 align:middle line:90% around the Fermi surface. 00:11:05.550 --> 00:11:08.790 align:middle line:84% And as soon as you have intermixing of these states, 00:11:08.790 --> 00:11:10.990 align:middle line:84% then you essentially create essentially still 00:11:10.990 --> 00:11:12.790 align:middle line:84% empty states below the Fermi level 00:11:12.790 --> 00:11:14.670 align:middle line:84% and fewer states above the Fermi level. 00:11:14.670 --> 00:11:16.270 align:middle line:84% And those are the states that now 00:11:16.270 --> 00:11:18.970 align:middle line:84% account for this increase of electrical conductivity. 00:11:18.970 --> 00:11:22.310 align:middle line:90% 00:11:22.310 --> 00:11:25.790 align:middle line:84% This is, as I said, this is purely electronic in nature. 00:11:25.790 --> 00:11:27.610 align:middle line:84% So that's why it's fully reversible. 00:11:27.610 --> 00:11:30.950 align:middle line:84% If you now remove the field, then it essentially 00:11:30.950 --> 00:11:32.810 align:middle line:84% goes back to the same amorphous structure, 00:11:32.810 --> 00:11:34.230 align:middle line:90% same insulating state. 00:11:34.230 --> 00:11:38.970 align:middle line:84% But practically what this allows you to do 00:11:38.970 --> 00:11:42.050 align:middle line:84% is now it allows you to apply large current through this 00:11:42.050 --> 00:11:45.170 align:middle line:84% normally insulating amorphous structure. 00:11:45.170 --> 00:11:47.650 align:middle line:84% So if you continue to increase the current, 00:11:47.650 --> 00:11:50.330 align:middle line:84% then you can go to the set regime, 00:11:50.330 --> 00:11:53.470 align:middle line:84% where essentially now, there's large enough joule heating, 00:11:53.470 --> 00:11:55.767 align:middle line:84% the material is above its crystallization temperature. 00:11:55.767 --> 00:11:58.350 align:middle line:84% The atoms start to move around, start to rearrange themselves. 00:11:58.350 --> 00:11:59.550 align:middle line:90% You then crystallize it. 00:11:59.550 --> 00:12:02.370 align:middle line:84% So now if you go to this kind of gray zone here, 00:12:02.370 --> 00:12:04.530 align:middle line:84% and then you start to decrease the voltage again, 00:12:04.530 --> 00:12:07.510 align:middle line:84% instead of retracing this kind of amorphous curve here, 00:12:07.510 --> 00:12:10.330 align:middle line:84% it now goes into this crystalline curve. 00:12:10.330 --> 00:12:12.450 align:middle line:84% So you basically start from amorphous here. 00:12:12.450 --> 00:12:13.630 align:middle line:90% You apply the voltage. 00:12:13.630 --> 00:12:16.305 align:middle line:84% You cause threshold switching, goes over here. 00:12:16.305 --> 00:12:18.930 align:middle line:84% You stay here for a little bit of time to allow crystallization 00:12:18.930 --> 00:12:19.510 align:middle line:90% to happen. 00:12:19.510 --> 00:12:20.968 align:middle line:84% And then when you come back, it now 00:12:20.968 --> 00:12:23.050 align:middle line:84% follows this kind of open square symbols. 00:12:23.050 --> 00:12:26.430 align:middle line:84% On the other hand, if it's in the crystalline state, 00:12:26.430 --> 00:12:27.430 align:middle line:90% you want to amortize it. 00:12:27.430 --> 00:12:28.430 align:middle line:90% That's quite straightforward. 00:12:28.430 --> 00:12:30.222 align:middle line:84% You just directly apply current through it. 00:12:30.222 --> 00:12:30.990 align:middle line:90% It's conductive. 00:12:30.990 --> 00:12:32.890 align:middle line:84% You bring up to this regime here. 00:12:32.890 --> 00:12:34.370 align:middle line:90% And you quickly cool it down. 00:12:34.370 --> 00:12:37.310 align:middle line:84% Then you go to the amorphous state. 00:12:37.310 --> 00:12:44.390 align:middle line:84% So this is just basically a mechanism 00:12:44.390 --> 00:12:46.510 align:middle line:84% that enables the operation of these devices 00:12:46.510 --> 00:12:49.230 align:middle line:84% without having to use excessively large bias. 00:12:49.230 --> 00:12:51.750 align:middle line:90% 00:12:51.750 --> 00:12:52.250 align:middle line:90% All right. 00:12:52.250 --> 00:12:56.190 align:middle line:84% So that's basically how this efficient memory actually works. 00:12:56.190 --> 00:12:59.190 align:middle line:84% Before I go into some specific examples, 00:12:59.190 --> 00:13:03.190 align:middle line:84% I want to show you this table here that basically compares 00:13:03.190 --> 00:13:05.030 align:middle line:84% these phase change memory against 00:13:05.030 --> 00:13:10.070 align:middle line:84% some other so-called non-volatile or emerging memory 00:13:10.070 --> 00:13:11.470 align:middle line:90% techniques. 00:13:11.470 --> 00:13:14.202 align:middle line:84% Now when I'm talking about this non-volatility, 00:13:14.202 --> 00:13:15.910 align:middle line:84% this is one very important characteristic 00:13:15.910 --> 00:13:17.910 align:middle line:84% of this kind of phase change material. 00:13:17.910 --> 00:13:20.990 align:middle line:84% So non-volatility essentially means 00:13:20.990 --> 00:13:25.390 align:middle line:84% that after you remove the power, the data is still maintained. 00:13:25.390 --> 00:13:28.470 align:middle line:84% This is different from DRAM in your computer. 00:13:28.470 --> 00:13:31.305 align:middle line:84% The DRAM in your computer is basically semiconductor based. 00:13:31.305 --> 00:13:33.430 align:middle line:84% So you need to essentially plug into power in order 00:13:33.430 --> 00:13:34.920 align:middle line:90% to maintain the data. 00:13:34.920 --> 00:13:37.440 align:middle line:84% As long as you unplug it, data is gone. 00:13:37.440 --> 00:13:39.460 align:middle line:84% In phase change materials, on the other hand, 00:13:39.460 --> 00:13:41.720 align:middle line:84% because the data is actually encoded in the material's 00:13:41.720 --> 00:13:46.600 align:middle line:84% microstructure and you don't need to continuously supply 00:13:46.600 --> 00:13:49.320 align:middle line:84% energy to maintain that microstructure state, 00:13:49.320 --> 00:13:51.620 align:middle line:84% so that's why it's called non-volatile. 00:13:51.620 --> 00:13:56.440 align:middle line:90% 00:13:56.440 --> 00:13:59.020 align:middle line:84% A lot of the memory technologies I'm comparing here, 00:13:59.020 --> 00:14:01.360 align:middle line:84% these are essentially considered non-volatile memory 00:14:01.360 --> 00:14:02.500 align:middle line:90% technologies. 00:14:02.500 --> 00:14:04.180 align:middle line:84% So some examples I'm listing here-- 00:14:04.180 --> 00:14:08.120 align:middle line:84% Memristor, this essentially uses a conductance change 00:14:08.120 --> 00:14:11.440 align:middle line:84% in oxide materials, essentially the movement of ions 00:14:11.440 --> 00:14:13.893 align:middle line:84% to form essentially a conductive path 00:14:13.893 --> 00:14:16.060 align:middle line:84% and then use that to essentially recall information. 00:14:16.060 --> 00:14:20.220 align:middle line:84% You have a spin transfer torque random access memory. 00:14:20.220 --> 00:14:22.720 align:middle line:84% This is essentially using spin in magnetic materials. 00:14:22.720 --> 00:14:25.040 align:middle line:84% And the three columns on the right-hand side, these 00:14:25.040 --> 00:14:26.680 align:middle line:90% are more traditional memory. 00:14:26.680 --> 00:14:28.600 align:middle line:84% DRAM, this is the internal memory 00:14:28.600 --> 00:14:30.040 align:middle line:90% you have on your computer. 00:14:30.040 --> 00:14:33.340 align:middle line:84% Flash, this is basically a solid-state drive, and then 00:14:33.340 --> 00:14:34.960 align:middle line:84% hard drive, essentially-- hard disk. 00:14:34.960 --> 00:14:39.800 align:middle line:84% So I like to keep this slide here. 00:14:39.800 --> 00:14:41.800 align:middle line:84% This chart is actually a bit outdated. 00:14:41.800 --> 00:14:43.540 align:middle line:90% It's more than 10 years old. 00:14:43.540 --> 00:14:46.500 align:middle line:84% So if you actually download a slide from Canvas 00:14:46.500 --> 00:14:48.580 align:middle line:84% a few days ago, I had actually replaced it 00:14:48.580 --> 00:14:50.240 align:middle line:90% with a more recent version. 00:14:50.240 --> 00:14:52.620 align:middle line:84% But then I decided to revert back to the old version 00:14:52.620 --> 00:14:54.380 align:middle line:84% because this was actually a good example 00:14:54.380 --> 00:14:58.620 align:middle line:84% to show how this table sometimes can be misleading. 00:14:58.620 --> 00:15:00.640 align:middle line:84% OK, so if you look at this table, 00:15:00.640 --> 00:15:04.400 align:middle line:84% first question for you, who do you think made the table? 00:15:04.400 --> 00:15:06.420 align:middle line:84% I mean, I'm not asking about specific names. 00:15:06.420 --> 00:15:09.420 align:middle line:84% But what would be the technical background of the person that 00:15:09.420 --> 00:15:12.660 align:middle line:90% actually made the table? 00:15:12.660 --> 00:15:13.845 align:middle line:90% AUDIENCE: The memristors? 00:15:13.845 --> 00:15:14.720 align:middle line:90% PROFESSOR: Obviously. 00:15:14.720 --> 00:15:18.260 align:middle line:84% That's why everything is green for memristors. 00:15:18.260 --> 00:15:20.300 align:middle line:84% Well, this try is actually made by HP. 00:15:20.300 --> 00:15:25.340 align:middle line:84% So HP, 10 years ago, it was a big proponent of memristors. 00:15:25.340 --> 00:15:29.400 align:middle line:90% They give it up later on. 00:15:29.400 --> 00:15:31.570 align:middle line:84% So I think this is one of the things you 00:15:31.570 --> 00:15:32.820 align:middle line:90% have to be very, very careful. 00:15:32.820 --> 00:15:34.653 align:middle line:84% You often see these kind of tables comparing 00:15:34.653 --> 00:15:38.020 align:middle line:84% different competing technologies and papers, 00:15:38.020 --> 00:15:40.440 align:middle line:90% whatever publications. 00:15:40.440 --> 00:15:45.160 align:middle line:84% You have to be careful here because number one, 00:15:45.160 --> 00:15:47.967 align:middle line:84% you tend to be more familiar with the technology you're 00:15:47.967 --> 00:15:48.800 align:middle line:90% actually working on. 00:15:48.800 --> 00:15:50.883 align:middle line:84% You know about the latest and greatest, about what 00:15:50.883 --> 00:15:51.760 align:middle line:90% you actually develop. 00:15:51.760 --> 00:15:53.320 align:middle line:84% But when you're trying to compare 00:15:53.320 --> 00:15:57.040 align:middle line:84% against other technologies, you should always keep in mind 00:15:57.040 --> 00:16:00.480 align:middle line:84% that you are actually comparing against moving targets 00:16:00.480 --> 00:16:03.240 align:middle line:84% in most cases, other technologies that are evolving, 00:16:03.240 --> 00:16:04.240 align:middle line:90% as well. 00:16:04.240 --> 00:16:07.280 align:middle line:84% So you should be aware about also 00:16:07.280 --> 00:16:09.720 align:middle line:84% what's the latest advances in other fields 00:16:09.720 --> 00:16:11.600 align:middle line:90% to draw a fair comparison. 00:16:11.600 --> 00:16:14.500 align:middle line:84% You should also understand where you're seeing, for example, 00:16:14.500 --> 00:16:18.320 align:middle line:84% red boxes here, is it because some fundamental limitation that 00:16:18.320 --> 00:16:20.720 align:middle line:84% the technology simply just cannot overcome? 00:16:20.720 --> 00:16:24.560 align:middle line:84% Or is it just because people just haven't done it yet? 00:16:24.560 --> 00:16:27.600 align:middle line:84% So those are the things that may actually impact eventually 00:16:27.600 --> 00:16:29.700 align:middle line:90% the selection of the technology. 00:16:29.700 --> 00:16:32.460 align:middle line:84% So, for example, let's specifically look at this table 00:16:32.460 --> 00:16:32.960 align:middle line:90% here. 00:16:32.960 --> 00:16:35.100 align:middle line:84% If you look at this column with respect 00:16:35.100 --> 00:16:37.665 align:middle line:84% to phase change materials, there are several figures of merit 00:16:37.665 --> 00:16:39.040 align:middle line:84% that are essentially quoted here. 00:16:39.040 --> 00:16:43.780 align:middle line:84% So density, this actually increases. 00:16:43.780 --> 00:16:46.040 align:middle line:84% Every generation of memory actually increases. 00:16:46.040 --> 00:16:48.820 align:middle line:84% So it's not really a fundamental limitation. 00:16:48.820 --> 00:16:51.500 align:middle line:84% If you look at energy per bit, this actually 00:16:51.500 --> 00:16:53.560 align:middle line:84% dictates essentially the energy efficiency. 00:16:53.560 --> 00:16:54.940 align:middle line:90% So it's quite important. 00:16:54.940 --> 00:16:58.040 align:middle line:84% For PCM, as we mentioned, it does have this drawback. 00:16:58.040 --> 00:17:01.420 align:middle line:84% It tends to have very high energy consumption 00:17:01.420 --> 00:17:03.660 align:middle line:84% because you need to heat up the material to fairly 00:17:03.660 --> 00:17:05.859 align:middle line:84% high temperature to crystallize it or amorphize it. 00:17:05.859 --> 00:17:08.780 align:middle line:84% So that's one limitation of PCM technology. 00:17:08.780 --> 00:17:13.980 align:middle line:84% Read time-- this quoted number is quite outdated. 00:17:13.980 --> 00:17:17.060 align:middle line:84% State of the art phase change memories, 00:17:17.060 --> 00:17:19.500 align:middle line:84% they usually go with single digit, like two nanoseconds, 00:17:19.500 --> 00:17:20.480 align:middle line:90% for example, real time. 00:17:20.480 --> 00:17:22.619 align:middle line:84% So it can be very, very fast because all you need to do 00:17:22.619 --> 00:17:24.244 align:middle line:84% is to send a very short electric pulse, 00:17:24.244 --> 00:17:26.552 align:middle line:84% measure electrical conductance. very low voltage pulse. 00:17:26.552 --> 00:17:28.010 align:middle line:84% So it's very, very straightforward. 00:17:28.010 --> 00:17:29.427 align:middle line:84% There's no fundamental limitation. 00:17:29.427 --> 00:17:30.670 align:middle line:90% It cannot be fast. 00:17:30.670 --> 00:17:34.910 align:middle line:84% Write time, I would say it's more close to the shorter end. 00:17:34.910 --> 00:17:39.230 align:middle line:84% For GST, the kinetics is limited by crystallization. 00:17:39.230 --> 00:17:40.670 align:middle line:90% Amplification is faster. 00:17:40.670 --> 00:17:43.390 align:middle line:84% So for crystallization, the crystallization time 00:17:43.390 --> 00:17:45.490 align:middle line:84% is of the order of 50 nanoseconds. 00:17:45.490 --> 00:17:47.830 align:middle line:90% So it's off the lower end. 00:17:47.830 --> 00:17:49.990 align:middle line:84% But people have come up with other alloys 00:17:49.990 --> 00:17:53.070 align:middle line:84% that can actually crystallize a lot faster than GST. 00:17:53.070 --> 00:17:54.870 align:middle line:84% So it's potentially possible to get down 00:17:54.870 --> 00:17:58.790 align:middle line:84% to a even sub nanosecond operation time. 00:17:58.790 --> 00:18:02.110 align:middle line:84% Retention, it can go about above 10 years. 00:18:02.110 --> 00:18:05.530 align:middle line:84% Endurance, at least in literature, 00:18:05.530 --> 00:18:11.430 align:middle line:84% people are reporting endurance up to 10 to 11, 10 to 12 cycles. 00:18:11.430 --> 00:18:13.230 align:middle line:84% So this also highlights one thing, though. 00:18:13.230 --> 00:18:15.670 align:middle line:84% When you're comparing some of these parameters, 00:18:15.670 --> 00:18:18.590 align:middle line:84% are you comparing a lab prototype with something 00:18:18.590 --> 00:18:20.590 align:middle line:84% that people is already commercially making? 00:18:20.590 --> 00:18:23.190 align:middle line:84% Or should you be comparing, essentially, 00:18:23.190 --> 00:18:25.832 align:middle line:84% a more mature technology with a mature technology? 00:18:25.832 --> 00:18:27.290 align:middle line:84% So making sure that you're actually 00:18:27.290 --> 00:18:30.250 align:middle line:84% comparing apples with apples, not apples with oranges. 00:18:30.250 --> 00:18:32.450 align:middle line:84% This is one of the pitfalls that lots of people 00:18:32.450 --> 00:18:34.690 align:middle line:84% actually falls in when they're doing this kind of comparison 00:18:34.690 --> 00:18:35.190 align:middle line:90% there. 00:18:35.190 --> 00:18:37.810 align:middle line:84% And that's why I want to list of as an example. 00:18:37.810 --> 00:18:41.290 align:middle line:90% Now last slide, 3D capabilities. 00:18:41.290 --> 00:18:43.790 align:middle line:84% This is important for also increasing the memory density. 00:18:43.790 --> 00:18:45.890 align:middle line:84% Essentially, you want to stack many layers 00:18:45.890 --> 00:18:47.570 align:middle line:84% of this kind of memory together so 00:18:47.570 --> 00:18:49.130 align:middle line:84% that you can pack more information 00:18:49.130 --> 00:18:50.670 align:middle line:90% density in one given area. 00:18:50.670 --> 00:18:54.273 align:middle line:90% So you put a node here for PCM. 00:18:54.273 --> 00:18:56.690 align:middle line:84% Well, I'm going to show you an example how people actually 00:18:56.690 --> 00:19:00.570 align:middle line:90% stack PCM memories. 00:19:00.570 --> 00:19:01.110 align:middle line:90% All right. 00:19:01.110 --> 00:19:03.450 align:middle line:84% So this is a technology that was actually 00:19:03.450 --> 00:19:06.190 align:middle line:84% developed by Intel about 10 years ago, 00:19:06.190 --> 00:19:07.770 align:middle line:90% exactly 10 years ago, 2015. 00:19:07.770 --> 00:19:10.270 align:middle line:84% Five years later, they actually developed second generation. 00:19:10.270 --> 00:19:13.490 align:middle line:84% They actually realized this 3D stacking of the memory there, 00:19:13.490 --> 00:19:15.290 align:middle line:90% it's called 3D XPoint. 00:19:15.290 --> 00:19:17.310 align:middle line:84% Now before I go into more details, 00:19:17.310 --> 00:19:20.690 align:middle line:84% I actually want to show you a little video here. 00:19:20.690 --> 00:19:23.070 align:middle line:90% And you can now watch the video. 00:19:23.070 --> 00:19:27.310 align:middle line:84% And keep in mind, they are also comparing this technology memory 00:19:27.310 --> 00:19:30.850 align:middle line:84% technology with some existing storage or memory technologies. 00:19:30.850 --> 00:19:32.610 align:middle line:84% So when they are talking about comparison, 00:19:32.610 --> 00:19:34.050 align:middle line:90% they are 1,000 times better. 00:19:34.050 --> 00:19:37.722 align:middle line:84% Be aware of what they're comparing against. 00:19:37.722 --> 00:19:39.550 align:middle line:90% [VIDEO PLAYBACK] 00:19:39.550 --> 00:19:40.850 align:middle line:90% - Forget all the big buzzwords. 00:19:40.850 --> 00:19:42.630 align:middle line:90% [INAUDIBLE] is so slow. 00:19:42.630 --> 00:19:46.350 align:middle line:84% Chances are your storage device is the problem, 00:19:46.350 --> 00:19:48.550 align:middle line:84% --to your processor's data requests 00:19:48.550 --> 00:19:52.670 align:middle line:84% and causing it to work at a fraction of its potential. 00:19:52.670 --> 00:19:56.630 align:middle line:84% Eliminating that bottleneck requires memory that is fast, 00:19:56.630 --> 00:20:00.790 align:middle line:84% dense, and non-volatile, so your data doesn't go away 00:20:00.790 --> 00:20:03.190 align:middle line:90% when your power does. 00:20:03.190 --> 00:20:06.250 align:middle line:84% DRAM memory is great because it's very fast. 00:20:06.250 --> 00:20:07.630 align:middle line:90% But it's volatile. 00:20:07.630 --> 00:20:09.770 align:middle line:84% And unlike many silicon components, 00:20:09.770 --> 00:20:12.930 align:middle line:84% DRAM density isn't doubling every two years. 00:20:12.930 --> 00:20:15.150 align:middle line:90% So it's hitting a wall. 00:20:15.150 --> 00:20:18.950 align:middle line:84% On the other hand, NAND memory found in today's SSDs, 00:20:18.950 --> 00:20:22.530 align:middle line:90% is both dense and non-volatile. 00:20:22.530 --> 00:20:25.670 align:middle line:84% And while it's fast compared to other storage technologies, 00:20:25.670 --> 00:20:28.570 align:middle line:84% it's still much slower than DRAM. 00:20:28.570 --> 00:20:29.990 align:middle line:90% But get ready. 00:20:29.990 --> 00:20:34.410 align:middle line:84% We're about to shatter today's rules in a very big way. 00:20:34.410 --> 00:20:35.930 align:middle line:90% [LAUGHTER] 00:20:35.930 --> 00:20:37.650 align:middle line:90% - --we've been working on. 00:20:37.650 --> 00:20:40.090 align:middle line:84% Now for the first time ever, we'll 00:20:40.090 --> 00:20:43.930 align:middle line:84% be able to equip computers with memory that's faster than NAND 00:20:43.930 --> 00:20:48.370 align:middle line:84% and more dense than DRAM and non-volatile. 00:20:48.370 --> 00:20:52.530 align:middle line:84% Intel Optane technology combines a next generation non-volatile 00:20:52.530 --> 00:20:56.730 align:middle line:84% memory called 3D XPoint Memory with Intel's advanced system 00:20:56.730 --> 00:20:59.810 align:middle line:84% memory controllers and interface hardware and software 00:20:59.810 --> 00:21:03.450 align:middle line:84% enhancements to bring amazing new computing possibilities 00:21:03.450 --> 00:21:06.010 align:middle line:90% to a variety of markets. 00:21:06.010 --> 00:21:07.970 align:middle line:90% So how do we do it? 00:21:07.970 --> 00:21:12.250 align:middle line:84% To make 3D XPoint Memory dense, we packed lots of capacity 00:21:12.250 --> 00:21:14.530 align:middle line:90% into a tiny footprint. 00:21:14.530 --> 00:21:17.410 align:middle line:84% We started by slicing submicroscopic layers 00:21:17.410 --> 00:21:19.637 align:middle line:90% of materials into columns. 00:21:19.637 --> 00:21:21.220 align:middle line:84% PROFESSOR: By lithography and etching. 00:21:21.220 --> 00:21:24.620 align:middle line:84% - And then we connected those columns 00:21:24.620 --> 00:21:27.500 align:middle line:84% using an innovative XPoint structure consisting 00:21:27.500 --> 00:21:30.700 align:middle line:84% of perpendicular wires that enables memory cells to be 00:21:30.700 --> 00:21:34.220 align:middle line:84% individually addressed by selecting one wire on top 00:21:34.220 --> 00:21:36.420 align:middle line:90% and another at the bottom. 00:21:36.420 --> 00:21:40.060 align:middle line:84% We can stack these memory grids three dimensionally to maximize 00:21:40.060 --> 00:21:41.620 align:middle line:90% density. 00:21:41.620 --> 00:21:45.020 align:middle line:84% And whereas DRAM requires a transistor at each memory cell 00:21:45.020 --> 00:21:49.740 align:middle line:84% to access or modify the cell, making DRAM big and expensive, 00:21:49.740 --> 00:21:51.580 align:middle line:84% each memory cell can be written to 00:21:51.580 --> 00:21:55.640 align:middle line:84% or read by simply varying the voltage sent to its selector, 00:21:55.640 --> 00:21:59.340 align:middle line:84% completely eliminating the need for transistors. 00:21:59.340 --> 00:22:02.340 align:middle line:84% This simple, stackable transistorless design 00:22:02.340 --> 00:22:04.580 align:middle line:84% enables us to deliver greater capacity 00:22:04.580 --> 00:22:07.620 align:middle line:90% than DRAM in the same space. 00:22:07.620 --> 00:22:09.580 align:middle line:84% When it comes to memory performance, 00:22:09.580 --> 00:22:12.380 align:middle line:84% low latency is the key, especially for today's 00:22:12.380 --> 00:22:14.260 align:middle line:90% demanding workloads. 00:22:14.260 --> 00:22:16.620 align:middle line:84% Intel Optane technology's latency 00:22:16.620 --> 00:22:18.900 align:middle line:84% is much shorter than NAND memory, 00:22:18.900 --> 00:22:21.280 align:middle line:84% eliminating a significant portion of the performance 00:22:21.280 --> 00:22:23.800 align:middle line:90% penalty for each data request. 00:22:23.800 --> 00:22:26.520 align:middle line:84% This makes Intel Optane technology ideal 00:22:26.520 --> 00:22:29.440 align:middle line:84% for all devices, applications, or services 00:22:29.440 --> 00:22:33.080 align:middle line:84% requiring fast access to large sets of data. 00:22:33.080 --> 00:22:36.400 align:middle line:84% There you have it, a truly disruptive memory technology 00:22:36.400 --> 00:22:38.680 align:middle line:84% designed to change the face of computing 00:22:38.680 --> 00:22:42.420 align:middle line:84% with an unparalleled combination of speed, density, 00:22:42.420 --> 00:22:44.200 align:middle line:90% and non-volatility. 00:22:44.200 --> 00:22:47.200 align:middle line:84% Get ready to unleash your processor's true potential 00:22:47.200 --> 00:22:51.320 align:middle line:84% and realize possibilities we thought were decades away. 00:22:51.320 --> 00:22:53.880 align:middle line:84% So be sure to stay tuned for upcoming announcements 00:22:53.880 --> 00:22:55.360 align:middle line:90% on Intel Optane technology. 00:22:55.360 --> 00:22:56.143 align:middle line:90% [END PLAYBACK] 00:22:56.143 --> 00:22:58.560 align:middle line:84% PROFESSOR: OK, so now you can count how many big buzzwords 00:22:58.560 --> 00:22:59.660 align:middle line:90% they use in there. 00:22:59.660 --> 00:23:03.160 align:middle line:90% 00:23:03.160 --> 00:23:04.320 align:middle line:90% OK, comes back here. 00:23:04.320 --> 00:23:07.040 align:middle line:90% 00:23:07.040 --> 00:23:09.400 align:middle line:84% Now tell me, they talk about a few different performance 00:23:09.400 --> 00:23:09.820 align:middle line:90% metrics. 00:23:09.820 --> 00:23:12.403 align:middle line:84% When they talk about latency, what are they comparing against? 00:23:12.403 --> 00:23:15.460 align:middle line:90% 00:23:15.460 --> 00:23:17.308 align:middle line:90% Or speed, essentially? 00:23:17.308 --> 00:23:18.600 align:middle line:90% There are basically two things. 00:23:18.600 --> 00:23:20.340 align:middle line:90% There's speed. 00:23:20.340 --> 00:23:22.860 align:middle line:90% There's also capacity. 00:23:22.860 --> 00:23:25.562 align:middle line:84% AUDIENCE: Speed requires [INAUDIBLE]. 00:23:25.562 --> 00:23:27.020 align:middle line:84% PROFESSOR: The other one, actually. 00:23:27.020 --> 00:23:29.060 align:middle line:90% It's the solid-state drive, SSD. 00:23:29.060 --> 00:23:31.900 align:middle line:84% So NAND, NAND is basically like flash, essentially. 00:23:31.900 --> 00:23:33.980 align:middle line:84% So flash has very large capacity. 00:23:33.980 --> 00:23:35.400 align:middle line:90% All your flash drives are huge. 00:23:35.400 --> 00:23:37.740 align:middle line:84% But the bottleneck that is relatively slow in terms of 00:23:37.740 --> 00:23:38.860 align:middle line:90% read and write. 00:23:38.860 --> 00:23:41.723 align:middle line:84% On the other hand, DRAM is what's 00:23:41.723 --> 00:23:43.640 align:middle line:84% actually the internal memory of your computer. 00:23:43.640 --> 00:23:45.100 align:middle line:90% Those are extremely fast. 00:23:45.100 --> 00:23:46.760 align:middle line:90% But they are volatile. 00:23:46.760 --> 00:23:49.980 align:middle line:84% And also they do not store nearly as much information 00:23:49.980 --> 00:23:52.100 align:middle line:90% as your flash drive. 00:23:52.100 --> 00:23:54.765 align:middle line:84% So the initial point of this Optane technology 00:23:54.765 --> 00:23:57.140 align:middle line:84% was that they tried to put something actually essentially 00:23:57.140 --> 00:23:59.620 align:middle line:84% in between, combine the best of both worlds. 00:23:59.620 --> 00:24:03.220 align:middle line:84% You have the capacity of flash or SSD. 00:24:03.220 --> 00:24:07.420 align:middle line:84% But at the same time, you also have the speed benefit of DRAM. 00:24:07.420 --> 00:24:11.300 align:middle line:84% So we'll get into a bit later and today actually really 00:24:11.300 --> 00:24:12.440 align:middle line:90% fulfill that promise. 00:24:12.440 --> 00:24:14.080 align:middle line:84% But for now, let's look at some technical details 00:24:14.080 --> 00:24:15.300 align:middle line:90% about how they realize it. 00:24:15.300 --> 00:24:18.880 align:middle line:84% Now, first thing you notice here is that I put a "likely" 00:24:18.880 --> 00:24:24.560 align:middle line:84% with a question mark there, because Intel actually never 00:24:24.560 --> 00:24:28.080 align:middle line:84% officially admitted that this is made out of PCM. 00:24:28.080 --> 00:24:31.800 align:middle line:84% In fact, a couple of cases, they essentially 00:24:31.800 --> 00:24:34.440 align:middle line:90% denied that it is made of PCM. 00:24:34.440 --> 00:24:37.040 align:middle line:84% But there are actually people going there 00:24:37.040 --> 00:24:39.095 align:middle line:90% and do reverse engineering. 00:24:39.095 --> 00:24:41.220 align:middle line:84% So this group of people, they actually go in there. 00:24:41.220 --> 00:24:42.500 align:middle line:90% They actually get the memory. 00:24:42.500 --> 00:24:44.522 align:middle line:84% They slice it open using focused ion beam. 00:24:44.522 --> 00:24:46.480 align:middle line:84% And then they did elemental composition looking 00:24:46.480 --> 00:24:47.760 align:middle line:90% at the structures there. 00:24:47.760 --> 00:24:50.380 align:middle line:84% And they find, of course, germanium, antimony, tellurium. 00:24:50.380 --> 00:24:52.400 align:middle line:90% So what do you expect? 00:24:52.400 --> 00:24:54.080 align:middle line:84% It's a phase change memory technology. 00:24:54.080 --> 00:24:55.400 align:middle line:90% That's the consensus. 00:24:55.400 --> 00:24:57.100 align:middle line:84% Now if you look at the structure here, 00:24:57.100 --> 00:24:59.532 align:middle line:84% this cross-section of memory, the bottom here, these 00:24:59.532 --> 00:25:00.740 align:middle line:90% are actually the metal wires. 00:25:00.740 --> 00:25:02.720 align:middle line:84% You see essentially a forest of the metal wire 00:25:02.720 --> 00:25:04.160 align:middle line:90% that provides interconnect. 00:25:04.160 --> 00:25:06.680 align:middle line:84% On top, this very thin layer there, this 00:25:06.680 --> 00:25:08.720 align:middle line:84% is actually the memory unit cell. 00:25:08.720 --> 00:25:12.050 align:middle line:84% In this particular module you see two stacked layers. 00:25:12.050 --> 00:25:14.450 align:middle line:84% In both layers you have a memory. 00:25:14.450 --> 00:25:18.450 align:middle line:84% And the memory is sitting on top of a selector. 00:25:18.450 --> 00:25:22.050 align:middle line:84% And then this is actually the XPoint structure across it. 00:25:22.050 --> 00:25:27.410 align:middle line:84% So now, about why you actually need a selector. 00:25:27.410 --> 00:25:32.790 align:middle line:84% So if you look at this what they call XPoint technology, 00:25:32.790 --> 00:25:35.810 align:middle line:84% I mean, this is really just essentially-- in the community, 00:25:35.810 --> 00:25:37.370 align:middle line:90% people call it crossbar. 00:25:37.370 --> 00:25:39.630 align:middle line:84% So this is not really anything new. 00:25:39.630 --> 00:25:43.290 align:middle line:84% People know about this architecture for a long time. 00:25:43.290 --> 00:25:45.930 align:middle line:84% But the idea is that you have two sets of metal wires. 00:25:45.930 --> 00:25:49.090 align:middle line:84% They are residing in two separate planes. 00:25:49.090 --> 00:25:51.150 align:middle line:84% You have the horizontal wires at the bottom, 00:25:51.150 --> 00:25:52.810 align:middle line:90% the vertical wires on top. 00:25:52.810 --> 00:25:54.790 align:middle line:84% And then at each cross point of wire, 00:25:54.790 --> 00:25:56.563 align:middle line:84% you have essentially a memory cell there 00:25:56.563 --> 00:25:58.230 align:middle line:84% where you can then pass current through. 00:25:58.230 --> 00:26:01.075 align:middle line:84% So normally it seems pretty straightforward. 00:26:01.075 --> 00:26:02.450 align:middle line:84% For example, if I were to operate 00:26:02.450 --> 00:26:05.370 align:middle line:84% this cell-- for example-- then I essentially 00:26:05.370 --> 00:26:08.370 align:middle line:84% apply a bias across this horizontal wire 00:26:08.370 --> 00:26:10.562 align:middle line:90% and across this vertical wire. 00:26:10.562 --> 00:26:12.270 align:middle line:84% Then you have current passing through it. 00:26:12.270 --> 00:26:14.230 align:middle line:84% That seems straight and simple forward. 00:26:14.230 --> 00:26:16.190 align:middle line:90% However, there's one caveat. 00:26:16.190 --> 00:26:20.950 align:middle line:84% There is a so-called sneak path current you have to overcome-- 00:26:20.950 --> 00:26:25.150 align:middle line:90% sneak path. 00:26:25.150 --> 00:26:27.950 align:middle line:84% So if we apply bias across these two wires 00:26:27.950 --> 00:26:31.270 align:middle line:84% here, instead of going through this particular memory cell, 00:26:31.270 --> 00:26:32.870 align:middle line:84% the current can also go this way. 00:26:32.870 --> 00:26:35.610 align:middle line:84% We can go from here, for example. 00:26:35.610 --> 00:26:37.470 align:middle line:90% We can come to this one. 00:26:37.470 --> 00:26:42.830 align:middle line:84% It then goes up, move up, then goes into vertical line here. 00:26:42.830 --> 00:26:47.030 align:middle line:84% Once you get to this point, it goes down again. 00:26:47.030 --> 00:26:51.350 align:middle line:84% And when it's coming over here, it goes back up again. 00:26:51.350 --> 00:26:52.530 align:middle line:90% And it goes up. 00:26:52.530 --> 00:26:56.990 align:middle line:90% 00:26:56.990 --> 00:26:59.150 align:middle line:84% So what you're seeing here is that instead 00:26:59.150 --> 00:27:02.130 align:middle line:84% of passing through one memory cell, in this case 00:27:02.130 --> 00:27:04.870 align:middle line:84% you pass through a total of three memory cells. 00:27:04.870 --> 00:27:06.630 align:middle line:90% So it's lower voltage. 00:27:06.630 --> 00:27:07.510 align:middle line:90% But it adds up. 00:27:07.510 --> 00:27:10.410 align:middle line:84% If you doing anything, if you think about huge network, 00:27:10.410 --> 00:27:12.750 align:middle line:84% there are tons of this three paths current. 00:27:12.750 --> 00:27:16.930 align:middle line:84% So they add up together will consume all the power. 00:27:16.930 --> 00:27:19.610 align:middle line:84% So that's why to solve it, one way to do 00:27:19.610 --> 00:27:21.810 align:middle line:84% it is you put a little transistor next 00:27:21.810 --> 00:27:24.010 align:middle line:90% to each of these memory cells. 00:27:24.010 --> 00:27:25.810 align:middle line:84% A transistor is basically a switch. 00:27:25.810 --> 00:27:27.610 align:middle line:84% You can individually control each switch 00:27:27.610 --> 00:27:31.010 align:middle line:84% to ensure that you're actually reading the right memory cell. 00:27:31.010 --> 00:27:32.610 align:middle line:84% Downside is that transistor obviously 00:27:32.610 --> 00:27:34.690 align:middle line:84% is more complicated, takes more footprint. 00:27:34.690 --> 00:27:37.530 align:middle line:84% So that's what the video is actually alluding to. 00:27:37.530 --> 00:27:39.850 align:middle line:90% Now what about a selector? 00:27:39.850 --> 00:27:42.690 align:middle line:84% A selector, in this particular case, 00:27:42.690 --> 00:27:45.290 align:middle line:84% is just a piece of [INAUDIBLE] compound. 00:27:45.290 --> 00:27:48.170 align:middle line:84% So, again, these people, they are doing reverse engineering. 00:27:48.170 --> 00:27:48.872 align:middle line:90% They look at it. 00:27:48.872 --> 00:27:50.330 align:middle line:84% They found essentially the selector 00:27:50.330 --> 00:27:54.590 align:middle line:84% is made of an alloy of germanium, arsenic, selenium. 00:27:54.590 --> 00:27:57.570 align:middle line:84% So it's a [INAUDIBLE] compound that also 00:27:57.570 --> 00:28:00.290 align:middle line:90% exhibits threshold switching. 00:28:00.290 --> 00:28:02.650 align:middle line:84% So why is threshold switching good? 00:28:02.650 --> 00:28:05.910 align:middle line:84% So think about what you want for selector. 00:28:05.910 --> 00:28:10.510 align:middle line:84% So if you think about the IV characteristics for selector, 00:28:10.510 --> 00:28:15.670 align:middle line:84% IV, current voltage, ideally you want 00:28:15.670 --> 00:28:17.830 align:middle line:84% a selector to turn on if you are directly 00:28:17.830 --> 00:28:19.670 align:middle line:90% going through one memory cell. 00:28:19.670 --> 00:28:21.670 align:middle line:84% But it will remain off if you have 00:28:21.670 --> 00:28:24.230 align:middle line:90% three memory cells in series. 00:28:24.230 --> 00:28:26.290 align:middle line:84% So the difference here is the voltage. 00:28:26.290 --> 00:28:27.990 align:middle line:84% For this one, this is the whole bias 00:28:27.990 --> 00:28:29.290 align:middle line:90% against essentially one cell. 00:28:29.290 --> 00:28:32.910 align:middle line:84% This is the bias distributed over three cells. 00:28:32.910 --> 00:28:38.470 align:middle line:84% So if you have an ideal selector that exhibit this kind of IV 00:28:38.470 --> 00:28:42.230 align:middle line:84% characteristic, it's like a step function. 00:28:42.230 --> 00:28:44.510 align:middle line:84% So if your voltage exceeds a certain value, 00:28:44.510 --> 00:28:46.370 align:middle line:90% it suddenly turns on. 00:28:46.370 --> 00:28:49.470 align:middle line:90% Otherwise, it remains off. 00:28:49.470 --> 00:28:52.670 align:middle line:84% So as long as you set your operating voltage 00:28:52.670 --> 00:28:58.750 align:middle line:84% to be larger than this threshold here but at the same time, 00:28:58.750 --> 00:29:01.430 align:middle line:84% smaller than three times the threshold, then 00:29:01.430 --> 00:29:03.190 align:middle line:84% you will be able to turn on selectively 00:29:03.190 --> 00:29:05.040 align:middle line:84% only this particular cell, not all 00:29:05.040 --> 00:29:07.420 align:middle line:90% the other in the current path. 00:29:07.420 --> 00:29:11.600 align:middle line:90% 00:29:11.600 --> 00:29:13.638 align:middle line:84% So now you already see this is what, exactly, 00:29:13.638 --> 00:29:14.680 align:middle line:90% threshold switching does. 00:29:14.680 --> 00:29:16.720 align:middle line:84% So threshold switching exhibit this kind 00:29:16.720 --> 00:29:18.480 align:middle line:84% of voltage characteristic, essentially. 00:29:18.480 --> 00:29:20.600 align:middle line:84% Beyond some threshold voltage, it suddenly 00:29:20.600 --> 00:29:22.560 align:middle line:84% turns on, allows you to go to large current. 00:29:22.560 --> 00:29:26.320 align:middle line:90% 00:29:26.320 --> 00:29:29.080 align:middle line:84% So as long as you set your threshold voltage correctly 00:29:29.080 --> 00:29:31.280 align:middle line:84% with respect to your operating voltage, 00:29:31.280 --> 00:29:33.560 align:middle line:84% that selector that allows you to pick out 00:29:33.560 --> 00:29:35.580 align:middle line:84% the right cell that actually is under operation. 00:29:35.580 --> 00:29:38.240 align:middle line:90% 00:29:38.240 --> 00:29:40.520 align:middle line:84% And this is actually quite a neat concept here, 00:29:40.520 --> 00:29:43.360 align:middle line:84% because if you look at how they actually make it together, 00:29:43.360 --> 00:29:45.880 align:middle line:84% they just simply stack the memory cell directly 00:29:45.880 --> 00:29:48.360 align:middle line:90% in series with the selector. 00:29:48.360 --> 00:29:50.020 align:middle line:84% So they are just connected in series. 00:29:50.020 --> 00:29:51.980 align:middle line:84% And because they are all [INAUDIBLE] compounds, 00:29:51.980 --> 00:29:53.640 align:middle line:84% they are both [INAUDIBLE] compounds, 00:29:53.640 --> 00:29:54.920 align:middle line:84% it's quite straightforward to make them 00:29:54.920 --> 00:29:56.337 align:middle line:84% in the same manufacturing process. 00:29:56.337 --> 00:29:58.240 align:middle line:84% It's basically a multilayer deposition 00:29:58.240 --> 00:29:59.840 align:middle line:90% followed by pattern etching. 00:29:59.840 --> 00:30:03.340 align:middle line:84% So that's one quite neat technology actually 00:30:03.340 --> 00:30:06.420 align:middle line:84% apply in this process here to enable this technology. 00:30:06.420 --> 00:30:13.660 align:middle line:84% So that's about the basics of this kind of a 3D XPoint 00:30:13.660 --> 00:30:14.940 align:middle line:90% technology. 00:30:14.940 --> 00:30:20.340 align:middle line:84% Now as I mentioned, the originally entails intention 00:30:20.340 --> 00:30:23.020 align:middle line:84% was they will fulfill, essentially, the need. 00:30:23.020 --> 00:30:25.540 align:middle line:84% They fill in the gap between typical computer 00:30:25.540 --> 00:30:28.840 align:middle line:84% memories with respect to solid-state storage devices. 00:30:28.840 --> 00:30:31.460 align:middle line:90% 00:30:31.460 --> 00:30:34.780 align:middle line:84% In terms of technology wise, this 00:30:34.780 --> 00:30:37.760 align:middle line:84% is a huge feat, lots of exciting advances. 00:30:37.760 --> 00:30:39.980 align:middle line:84% But from a market perspective, it's 00:30:39.980 --> 00:30:43.980 align:middle line:84% very hard to say that this is actually a successful product. 00:30:43.980 --> 00:30:47.352 align:middle line:84% In fact, Intel-- so they started in 2015, 00:30:47.352 --> 00:30:49.060 align:middle line:84% when they first developed the technology. 00:30:49.060 --> 00:30:50.240 align:middle line:90% They partnered with Micron. 00:30:50.240 --> 00:30:52.532 align:middle line:84% Micron is actually the company that actually developed, 00:30:52.532 --> 00:30:54.580 align:middle line:84% originally, the phase change memory technology. 00:30:54.580 --> 00:30:56.740 align:middle line:84% In 2020, they developed second generation 00:30:56.740 --> 00:30:58.500 align:middle line:90% with increased density. 00:30:58.500 --> 00:31:00.960 align:middle line:84% The market reception has always been lukewarm. 00:31:00.960 --> 00:31:04.485 align:middle line:84% So they decided to wind down the business, starting 2022. 00:31:04.485 --> 00:31:05.860 align:middle line:84% You can still buy these memories. 00:31:05.860 --> 00:31:08.200 align:middle line:84% But they are not in active production anymore. 00:31:08.200 --> 00:31:10.720 align:middle line:84% Some other companies continued the trajectory 00:31:10.720 --> 00:31:13.280 align:middle line:84% to develop technology like Micron, maybe Samsung. 00:31:13.280 --> 00:31:16.680 align:middle line:90% But it is not anymore Intel. 00:31:16.680 --> 00:31:18.880 align:middle line:90% So why is that? 00:31:18.880 --> 00:31:20.160 align:middle line:90% A few reasons here. 00:31:20.160 --> 00:31:24.560 align:middle line:84% So you remember when I was talking about this technology 00:31:24.560 --> 00:31:25.720 align:middle line:90% comparison chart? 00:31:25.720 --> 00:31:28.640 align:middle line:84% I said that you should always be aware that your competitors are 00:31:28.640 --> 00:31:29.617 align:middle line:90% moving targets. 00:31:29.617 --> 00:31:31.700 align:middle line:84% They don't sit still and wait for you to catch up. 00:31:31.700 --> 00:31:35.400 align:middle line:90% So same thing applies here. 00:31:35.400 --> 00:31:40.240 align:middle line:84% When they're comparing against DRAM and SSD, what 00:31:40.240 --> 00:31:44.080 align:middle line:84% happens over the past few years is that DRAM becomes cheaper. 00:31:44.080 --> 00:31:47.440 align:middle line:84% So it's easier to get high-capacity DRAM. 00:31:47.440 --> 00:31:50.640 align:middle line:84% At the same time SSD, the bottleneck for SSD 00:31:50.640 --> 00:31:52.400 align:middle line:90% has always been speed. 00:31:52.400 --> 00:31:54.880 align:middle line:84% Latest version of SSD is a lot faster 00:31:54.880 --> 00:31:56.360 align:middle line:90% compared to previous versions. 00:31:56.360 --> 00:32:00.300 align:middle line:84% So this cuts the space where the [INAUDIBLE] technology 00:32:00.300 --> 00:32:01.780 align:middle line:90% can actually thrive. 00:32:01.780 --> 00:32:05.100 align:middle line:84% And on top of that, there are also other limitations there. 00:32:05.100 --> 00:32:07.100 align:middle line:84% I mean, their manufacturing never really reached 00:32:07.100 --> 00:32:08.202 align:middle line:90% the capacity they want. 00:32:08.202 --> 00:32:09.660 align:middle line:84% So it's actually the product always 00:32:09.660 --> 00:32:14.222 align:middle line:84% has been too expensive, compared to other memory technologies. 00:32:14.222 --> 00:32:16.180 align:middle line:84% So it eventually become question about how much 00:32:16.180 --> 00:32:18.360 align:middle line:84% people want to pay for the added performance. 00:32:18.360 --> 00:32:22.060 align:middle line:90% 00:32:22.060 --> 00:32:24.700 align:middle line:84% The other problem here is actually ecosystems. 00:32:24.700 --> 00:32:26.580 align:middle line:84% So when Intel designed these memories, 00:32:26.580 --> 00:32:29.940 align:middle line:84% they have to work with specific chipsets and specific hardware 00:32:29.940 --> 00:32:30.580 align:middle line:90% frames. 00:32:30.580 --> 00:32:32.180 align:middle line:84% And, again, that would obviously limit 00:32:32.180 --> 00:32:34.840 align:middle line:90% the adoption of technology. 00:32:34.840 --> 00:32:37.870 align:middle line:84% People also never really picked up, at least 00:32:37.870 --> 00:32:39.780 align:middle line:90% passing like 10 years ago. 00:32:39.780 --> 00:32:43.060 align:middle line:84% There were not much software that was specifically optimized 00:32:43.060 --> 00:32:44.660 align:middle line:90% for non-volatile memory. 00:32:44.660 --> 00:32:45.720 align:middle line:90% It's a great feature. 00:32:45.720 --> 00:32:48.500 align:middle line:84% But ecosystem simply wasn't there. 00:32:48.500 --> 00:32:50.780 align:middle line:84% So it doesn't mean that non-volatile memory is out 00:32:50.780 --> 00:32:51.580 align:middle line:90% of favor. 00:32:51.580 --> 00:32:53.160 align:middle line:90% People are still developing it. 00:32:53.160 --> 00:32:55.980 align:middle line:84% And now people are starting to take advantage of that 00:32:55.980 --> 00:32:57.410 align:middle line:90% on the software side, as well. 00:32:57.410 --> 00:33:00.290 align:middle line:84% But it was a bit too early for this technology. 00:33:00.290 --> 00:33:02.710 align:middle line:84% And finally, there are also some business decisions. 00:33:02.710 --> 00:33:04.293 align:middle line:84% You know about all the turbulence that 00:33:04.293 --> 00:33:06.630 align:middle line:90% happened around Intel recently. 00:33:06.630 --> 00:33:09.670 align:middle line:84% They have been trying to focus on their foundry business 00:33:09.670 --> 00:33:10.970 align:middle line:90% over the past few years. 00:33:10.970 --> 00:33:12.950 align:middle line:84% They cut down a lot of other businesses. 00:33:12.950 --> 00:33:14.350 align:middle line:90% This is one of them. 00:33:14.350 --> 00:33:17.230 align:middle line:90% So that's why. 00:33:17.230 --> 00:33:20.630 align:middle line:84% So takeaway message, you always should be very, very careful 00:33:20.630 --> 00:33:23.370 align:middle line:84% when deciding on what technology you want to work on. 00:33:23.370 --> 00:33:26.470 align:middle line:84% You should always be an expert, not just only technology 00:33:26.470 --> 00:33:28.470 align:middle line:84% you are focusing on, but also at least be 00:33:28.470 --> 00:33:31.790 align:middle line:84% aware of the latest and greatest of your competitors, 00:33:31.790 --> 00:33:35.350 align:middle line:84% as well, in order to make a fair assessment. 00:33:35.350 --> 00:33:35.970 align:middle line:90% All right. 00:33:35.970 --> 00:33:37.990 align:middle line:84% So that's about electronic memories. 00:33:37.990 --> 00:33:40.150 align:middle line:84% Now, I also want to mention the other area 00:33:40.150 --> 00:33:41.870 align:middle line:84% where these phase change materials is 00:33:41.870 --> 00:33:45.050 align:middle line:84% getting a lot of traction is in the area of optics. 00:33:45.050 --> 00:33:49.570 align:middle line:84% No, I'm not talking about the old rewritable CD-DVDs. 00:33:49.570 --> 00:33:51.870 align:middle line:84% I'm more talking about essentially optics that 00:33:51.870 --> 00:33:53.730 align:middle line:84% can be integrated using state of the art 00:33:53.730 --> 00:33:56.770 align:middle line:84% semiconductor technologies for essentially photonics. 00:33:56.770 --> 00:34:00.730 align:middle line:84% So the basic property that all this optical application 00:34:00.730 --> 00:34:03.403 align:middle line:84% builds on is the fact that there's a huge optical property 00:34:03.403 --> 00:34:05.570 align:middle line:84% contrast when you switch the material from amorphous 00:34:05.570 --> 00:34:07.010 align:middle line:90% to crystalline state. 00:34:07.010 --> 00:34:09.889 align:middle line:84% And this is actually very unique. 00:34:09.889 --> 00:34:11.550 align:middle line:90% I'm giving you here one example. 00:34:11.550 --> 00:34:14.409 align:middle line:84% This is basically silicon, very common material. 00:34:14.409 --> 00:34:16.810 align:middle line:84% I'm comparing here, again, now three different states 00:34:16.810 --> 00:34:17.690 align:middle line:90% of silicon. 00:34:17.690 --> 00:34:19.330 align:middle line:84% You have single crystalline silicon. 00:34:19.330 --> 00:34:20.395 align:middle line:90% You have polysilicon. 00:34:20.395 --> 00:34:21.770 align:middle line:84% This is actually, in this case, I 00:34:21.770 --> 00:34:25.110 align:middle line:84% believe it's more like nanocrystalline silicon. 00:34:25.110 --> 00:34:27.770 align:middle line:84% And then the last one, this is amorphous silicon. 00:34:27.770 --> 00:34:30.512 align:middle line:84% Now what you notice here, if you look at the refractive index, 00:34:30.512 --> 00:34:32.929 align:middle line:84% especially in this region-- this is where silicon actually 00:34:32.929 --> 00:34:36.969 align:middle line:84% become transparent-- it's basically the same. 00:34:36.969 --> 00:34:41.090 align:middle line:84% It differs by maybe up to 0.1 or 0.0 or something. 00:34:41.090 --> 00:34:45.650 align:middle line:84% So for most materials, this is actually the case. 00:34:45.650 --> 00:34:48.050 align:middle line:84% If you change the arrangement of atoms, 00:34:48.050 --> 00:34:49.610 align:middle line:90% it's still the same atoms. 00:34:49.610 --> 00:34:52.409 align:middle line:84% So you expect more or less the same electronic polarizability 00:34:52.409 --> 00:34:54.110 align:middle line:90% ability for the material. 00:34:54.110 --> 00:34:57.790 align:middle line:84% So this is what really makes phase change material so unique. 00:34:57.790 --> 00:34:59.618 align:middle line:84% As I mentioned in previous lectures, 00:34:59.618 --> 00:35:01.410 align:middle line:84% because when you change atomic arrangement, 00:35:01.410 --> 00:35:03.170 align:middle line:84% you also change the bonding configuration. 00:35:03.170 --> 00:35:05.830 align:middle line:84% You go from covalent bonding to this kind of resonant or method 00:35:05.830 --> 00:35:06.750 align:middle line:90% of bonding. 00:35:06.750 --> 00:35:09.670 align:middle line:84% As a result, this increases the polarizability 00:35:09.670 --> 00:35:13.310 align:middle line:84% and therefore significantly enhances the refractive index. 00:35:13.310 --> 00:35:16.510 align:middle line:84% So if you look at this classical GST 225 alloy, 00:35:16.510 --> 00:35:19.510 align:middle line:84% this is refractive index, amorphous crystalline. 00:35:19.510 --> 00:35:21.190 align:middle line:90% There's a huge jump. 00:35:21.190 --> 00:35:23.710 align:middle line:90% It's like index change by 3. 00:35:23.710 --> 00:35:28.590 align:middle line:84% It's like totally two totally different optical materials. 00:35:28.590 --> 00:35:30.990 align:middle line:84% At the same time, you also notice here 00:35:30.990 --> 00:35:32.950 align:middle line:84% that the absorption-- so extinction coefficient 00:35:32.950 --> 00:35:34.450 align:middle line:84% essentially gives you the absorption 00:35:34.450 --> 00:35:35.710 align:middle line:90% as a function of wavelength. 00:35:35.710 --> 00:35:37.168 align:middle line:84% You notice that the absorption also 00:35:37.168 --> 00:35:40.870 align:middle line:84% increases, going from the amorphous to crystalline phase. 00:35:40.870 --> 00:35:43.910 align:middle line:84% Now this is also understandable because the crystalline phase 00:35:43.910 --> 00:35:45.370 align:middle line:90% is electrically conductive. 00:35:45.370 --> 00:35:48.070 align:middle line:84% Anything that's very electrically conductive 00:35:48.070 --> 00:35:50.810 align:middle line:84% is always optically absorbing, because free electrons 00:35:50.810 --> 00:35:53.450 align:middle line:84% can respond to essentially an incoming oscillating 00:35:53.450 --> 00:35:54.770 align:middle line:90% electromagnetic field. 00:35:54.770 --> 00:35:55.930 align:middle line:90% They can polarize. 00:35:55.930 --> 00:35:58.340 align:middle line:84% And in that process, they actually dissipate energy. 00:35:58.340 --> 00:36:00.090 align:middle line:84% This is what's causing optical absorption. 00:36:00.090 --> 00:36:03.010 align:middle line:84% So that's why, going from amorphous crystalline, 00:36:03.010 --> 00:36:04.770 align:middle line:84% you increase the refractive index. 00:36:04.770 --> 00:36:08.530 align:middle line:84% And you also increase the optical absorption. 00:36:08.530 --> 00:36:11.130 align:middle line:84% So this is the behavior that people use now to develop, 00:36:11.130 --> 00:36:13.490 align:middle line:90% essentially, optical memories. 00:36:13.490 --> 00:36:15.790 align:middle line:84% Again, I'm not talking about rewritable CD or DVDs. 00:36:15.790 --> 00:36:17.707 align:middle line:84% These are actually optical memories integrated 00:36:17.707 --> 00:36:19.210 align:middle line:90% on the photonic circuits. 00:36:19.210 --> 00:36:21.530 align:middle line:84% So what I'm showing you here-- this is an example-- 00:36:21.530 --> 00:36:24.490 align:middle line:84% well, you basically have optical waveguide. 00:36:24.490 --> 00:36:26.770 align:middle line:84% Optical waveguide is basically photonic wire. 00:36:26.770 --> 00:36:30.570 align:middle line:84% It's basically a strip of high-index material embedded 00:36:30.570 --> 00:36:33.330 align:middle line:84% or surrounded by a low-index material. 00:36:33.330 --> 00:36:35.250 align:middle line:90% So the light can get trapped. 00:36:35.250 --> 00:36:37.410 align:middle line:84% And they propagate along the high-index material 00:36:37.410 --> 00:36:38.830 align:middle line:84% through total internal reflection, 00:36:38.830 --> 00:36:40.290 align:middle line:90% similar to optical fibers. 00:36:40.290 --> 00:36:43.170 align:middle line:84% But this is made essentially on silicon wafer. 00:36:43.170 --> 00:36:47.410 align:middle line:84% Now you can put a little chip of this GST material 00:36:47.410 --> 00:36:48.662 align:middle line:90% on top of this waveguide here. 00:36:48.662 --> 00:36:50.120 align:middle line:84% So this is actually your SEM image. 00:36:50.120 --> 00:36:51.940 align:middle line:90% You see a little patch of GST. 00:36:51.940 --> 00:36:54.420 align:middle line:84% When you switch this GST between this amorphous 00:36:54.420 --> 00:36:56.300 align:middle line:90% and crystalline state-- 00:36:56.300 --> 00:36:59.140 align:middle line:84% again, crystalline state is absorbing. 00:36:59.140 --> 00:37:01.180 align:middle line:84% Amorphous state is more transparent. 00:37:01.180 --> 00:37:03.660 align:middle line:84% So then this actually gives rise to a difference 00:37:03.660 --> 00:37:05.980 align:middle line:84% of the transmitted light intensity. 00:37:05.980 --> 00:37:08.660 align:middle line:84% You can then encode this digital information 00:37:08.660 --> 00:37:12.540 align:middle line:84% in the transmittance of the cell here. 00:37:12.540 --> 00:37:14.402 align:middle line:84% So there's actually some data here. 00:37:14.402 --> 00:37:16.860 align:middle line:84% What they were able to realize, this actually, essentially, 00:37:16.860 --> 00:37:18.660 align:middle line:84% vertical axis is basically the intensity 00:37:18.660 --> 00:37:19.895 align:middle line:90% transmitted through a device. 00:37:19.895 --> 00:37:22.020 align:middle line:84% And you can see that a horizontal axis is basically 00:37:22.020 --> 00:37:22.760 align:middle line:90% like time. 00:37:22.760 --> 00:37:25.340 align:middle line:84% You can see they can switch between several discrete levels, 00:37:25.340 --> 00:37:28.180 align:middle line:84% in this case eight different levels, so 3-bit. 00:37:28.180 --> 00:37:31.460 align:middle line:84% So you can store 3-bit of information in one cell 00:37:31.460 --> 00:37:34.540 align:middle line:84% by essentially changing the fraction of crystalline versus 00:37:34.540 --> 00:37:37.312 align:middle line:90% amorphous state. 00:37:37.312 --> 00:37:39.020 align:middle line:84% I mean, a few years later, the same group 00:37:39.020 --> 00:37:41.560 align:middle line:84% were able to actually push it to even more levels. 00:37:41.560 --> 00:37:43.940 align:middle line:84% In this case, this one is actually 32 levels. 00:37:43.940 --> 00:37:47.480 align:middle line:84% I believe they were also able to do 64 levels, so up 00:37:47.480 --> 00:37:51.000 align:middle line:84% to six bits within just one single unit cell. 00:37:51.000 --> 00:37:51.500 align:middle line:90% All right. 00:37:51.500 --> 00:37:53.120 align:middle line:84% So now you may wonder, why do you 00:37:53.120 --> 00:37:55.360 align:middle line:84% care about this optical memory here? 00:37:55.360 --> 00:37:59.400 align:middle line:84% One of the applications that actually attract 00:37:59.400 --> 00:38:02.120 align:middle line:84% lots of attention is actually in so-called optical analog 00:38:02.120 --> 00:38:02.860 align:middle line:90% computing. 00:38:02.860 --> 00:38:06.240 align:middle line:84% So the big background there is that everyone 00:38:06.240 --> 00:38:08.120 align:middle line:90% is talking about AI. 00:38:08.120 --> 00:38:10.440 align:middle line:84% And for AI, it turns out the conventional computer 00:38:10.440 --> 00:38:12.303 align:middle line:84% architecture is very inefficient. 00:38:12.303 --> 00:38:13.720 align:middle line:84% Conventional computer architecture 00:38:13.720 --> 00:38:16.187 align:middle line:84% builds on so-called Von Neumann architecture there. 00:38:16.187 --> 00:38:18.520 align:middle line:84% In Von Neumann architecture, you have two separate units 00:38:18.520 --> 00:38:19.140 align:middle line:90% in computer. 00:38:19.140 --> 00:38:21.200 align:middle line:84% You have your central processing unit, or CPU. 00:38:21.200 --> 00:38:22.300 align:middle line:90% You have your memory. 00:38:22.300 --> 00:38:24.855 align:middle line:84% When you do this kind of computing, 00:38:24.855 --> 00:38:25.980 align:middle line:90% they're like deep learning. 00:38:25.980 --> 00:38:27.760 align:middle line:84% It requires lots of data exchange 00:38:27.760 --> 00:38:29.680 align:middle line:90% between CPU and the memory. 00:38:29.680 --> 00:38:32.280 align:middle line:84% So the connection, the interconnect 00:38:32.280 --> 00:38:34.660 align:middle line:84% between these two units, is actually the bottleneck. 00:38:34.660 --> 00:38:36.980 align:middle line:84% It's not so much about CPU or the memory itself. 00:38:36.980 --> 00:38:40.480 align:middle line:84% It's about the data transfer in between them. 00:38:40.480 --> 00:38:42.880 align:middle line:84% So to resolve this kind of bottleneck, 00:38:42.880 --> 00:38:45.480 align:middle line:84% people come up with lots of paradigms. 00:38:45.480 --> 00:38:50.100 align:middle line:84% And in particular, one promising pathway 00:38:50.100 --> 00:38:52.460 align:middle line:84% is to the so-called in-memory computing. 00:38:52.460 --> 00:38:57.100 align:middle line:84% You essentially have the memory itself as the computing unit. 00:38:57.100 --> 00:38:58.780 align:middle line:84% By doing so, you can avoid a lot of data 00:38:58.780 --> 00:39:00.905 align:middle line:84% transfer because you essentially-- data all resides 00:39:00.905 --> 00:39:02.420 align:middle line:84% in nowadays' computing memory there. 00:39:02.420 --> 00:39:05.960 align:middle line:84% And this kind of what I have referred there, 00:39:05.960 --> 00:39:09.060 align:middle line:84% this optical memory is one way to actually realize 00:39:09.060 --> 00:39:11.340 align:middle line:84% this kind of in-memory computing. 00:39:11.340 --> 00:39:14.020 align:middle line:84% So just to give you one example how this in-memory computing can 00:39:14.020 --> 00:39:17.040 align:middle line:84% be executed, let's think about one very simple example here, 00:39:17.040 --> 00:39:22.100 align:middle line:84% which is basically matrix multiplication Now matrix 00:39:22.100 --> 00:39:24.400 align:middle line:84% multiplication, while this looks very straightforward, 00:39:24.400 --> 00:39:27.240 align:middle line:84% it's actually very relevant for AI applications. 00:39:27.240 --> 00:39:28.620 align:middle line:84% If you think about deep learning, 00:39:28.620 --> 00:39:31.540 align:middle line:84% it's all about big matrices, huge matrices multiplied 00:39:31.540 --> 00:39:32.040 align:middle line:90% together. 00:39:32.040 --> 00:39:33.748 align:middle line:84% This is actually how you do your training 00:39:33.748 --> 00:39:35.020 align:middle line:90% of your neural networks. 00:39:35.020 --> 00:39:38.120 align:middle line:84% So big matrix multiplication is quite inefficient 00:39:38.120 --> 00:39:39.620 align:middle line:84% if you use computer-computer, again, 00:39:39.620 --> 00:39:41.162 align:middle line:84% because there's lots of data transfer 00:39:41.162 --> 00:39:43.190 align:middle line:84% that's needed So that's why people are looking 00:39:43.190 --> 00:39:44.990 align:middle line:84% for ways to do this so-called matrix 00:39:44.990 --> 00:39:47.510 align:middle line:84% multiplication and accumulation, MAC. 00:39:47.510 --> 00:39:49.230 align:middle line:84% So what I'm showing here is basically 00:39:49.230 --> 00:39:51.652 align:middle line:84% a simple vector-by-vector multiplication. 00:39:51.652 --> 00:39:53.110 align:middle line:84% A matrix multiplication essentially 00:39:53.110 --> 00:39:56.790 align:middle line:84% is a duplication of this task many, many times. 00:39:56.790 --> 00:39:59.870 align:middle line:84% So to do it optically, this is one scheme 00:39:59.870 --> 00:40:00.930 align:middle line:90% that people can do it. 00:40:00.930 --> 00:40:02.150 align:middle line:90% There are other schemes. 00:40:02.150 --> 00:40:04.150 align:middle line:84% So you start out by essentially sending 00:40:04.150 --> 00:40:07.270 align:middle line:84% light of different wavelengths, Initially, 00:40:07.270 --> 00:40:08.770 align:middle line:84% they all have the same intensity. 00:40:08.770 --> 00:40:10.510 align:middle line:84% You combine all the light into one 00:40:10.510 --> 00:40:12.630 align:middle line:84% waveguide, and so that each wavelength is 00:40:12.630 --> 00:40:15.030 align:middle line:84% essentially more or less propagate independently. 00:40:15.030 --> 00:40:18.690 align:middle line:84% You then pass through these memory cells here. 00:40:18.690 --> 00:40:20.550 align:middle line:84% There's a cascade of this memory cell 00:40:20.550 --> 00:40:23.030 align:middle line:84% here that I'm labeling with different colors. 00:40:23.030 --> 00:40:25.270 align:middle line:84% Each of these memory cells essentially 00:40:25.270 --> 00:40:29.150 align:middle line:84% interacts with one particular wavelength So I 00:40:29.150 --> 00:40:32.550 align:middle line:84% have labeled the number here W. So W is basically 00:40:32.550 --> 00:40:36.130 align:middle line:84% the transmission coefficient of each of the optical memory 00:40:36.130 --> 00:40:39.230 align:middle line:84% So, for example, W1, suppose you come in here. 00:40:39.230 --> 00:40:41.410 align:middle line:84% The light intensity is actually x1. 00:40:41.410 --> 00:40:43.570 align:middle line:84% You encode essentially the input vector 00:40:43.570 --> 00:40:47.250 align:middle line:84% using the intensity in the input light at different wavelengths. 00:40:47.250 --> 00:40:49.950 align:middle line:84% So for the first wavelength of light, it goes in here. 00:40:49.950 --> 00:40:54.963 align:middle line:84% And x1, when it transmits through this little memory here, 00:40:54.963 --> 00:40:57.130 align:middle line:84% it light will then come out from the other waveguide 00:40:57.130 --> 00:40:58.750 align:middle line:90% that we labeled as drop. 00:40:58.750 --> 00:41:00.708 align:middle line:84% But the transmission coefficient is W1. 00:41:00.708 --> 00:41:02.250 align:middle line:84% So the light intensity coming through 00:41:02.250 --> 00:41:05.390 align:middle line:84% is going to be x1 times W1, same thing. 00:41:05.390 --> 00:41:07.670 align:middle line:84% The next one, this is a second wavelength. 00:41:07.670 --> 00:41:10.690 align:middle line:84% The intensity is going to be W2 times x2. 00:41:10.690 --> 00:41:12.450 align:middle line:84% So then all the wavelengths then again 00:41:12.450 --> 00:41:15.250 align:middle line:84% get combined in this second drop waveguide here. 00:41:15.250 --> 00:41:16.810 align:middle line:84% If you have detected at the end, you 00:41:16.810 --> 00:41:19.770 align:middle line:84% can detect all the intensity adds up together. 00:41:19.770 --> 00:41:22.610 align:middle line:84% Then that will give you essentially immediately 00:41:22.610 --> 00:41:27.490 align:middle line:84% the sum of the product or the product of two vectors there. 00:41:27.490 --> 00:41:32.250 align:middle line:84% So what we are doing here, this is called analog computing 00:41:32.250 --> 00:41:36.227 align:middle line:84% because that's actually, you are not using digital processors. 00:41:36.227 --> 00:41:38.310 align:middle line:84% You're not using transistors to perform computing. 00:41:38.310 --> 00:41:40.270 align:middle line:84% You're essentially leveraging the fact 00:41:40.270 --> 00:41:43.750 align:middle line:84% that you can encode this analog information in terms 00:41:43.750 --> 00:41:46.290 align:middle line:84% of the memory state, optical transmittance state, 00:41:46.290 --> 00:41:48.890 align:middle line:84% and use that to directly perform the summation. 00:41:48.890 --> 00:41:52.990 align:middle line:90% 00:41:52.990 --> 00:41:55.850 align:middle line:84% So the advantage here is that this can be extremely fast, 00:41:55.850 --> 00:41:58.390 align:middle line:84% especially when you go to very large matrices 00:41:58.390 --> 00:42:02.510 align:middle line:84% because the technology is, in essence, scalable, as 00:42:02.510 --> 00:42:05.030 align:middle line:84% long as you can provide a large number of wavelength. 00:42:05.030 --> 00:42:08.870 align:middle line:84% And then you can generate all of these filters 00:42:08.870 --> 00:42:12.670 align:middle line:84% and all of these memories and program them appropriately. 00:42:12.670 --> 00:42:14.607 align:middle line:84% The addition process is done automatically. 00:42:14.607 --> 00:42:16.190 align:middle line:84% You just have one single photodetector 00:42:16.190 --> 00:42:18.790 align:middle line:84% collect all the lights, essentially does the summation 00:42:18.790 --> 00:42:19.990 align:middle line:90% automatically. 00:42:19.990 --> 00:42:24.280 align:middle line:90% 00:42:24.280 --> 00:42:27.350 align:middle line:84% And that's actually what the people have done. 00:42:27.350 --> 00:42:28.590 align:middle line:90% Let me see. 00:42:28.590 --> 00:42:31.110 align:middle line:90% Oh, OK. 00:42:31.110 --> 00:42:32.732 align:middle line:84% I'm actually missing a slide there. 00:42:32.732 --> 00:42:34.190 align:middle line:84% But basically, people have actually 00:42:34.190 --> 00:42:37.410 align:middle line:84% implemented, essentially, using these cells, based 00:42:37.410 --> 00:42:39.450 align:middle line:84% on essentially using the phase change 00:42:39.450 --> 00:42:41.830 align:middle line:84% memory elements I have mentioned in the previous slide. 00:42:41.830 --> 00:42:43.788 align:middle line:84% And then that allows you to essentially perform 00:42:43.788 --> 00:42:44.990 align:middle line:90% this kind of a computing. 00:42:44.990 --> 00:42:46.490 align:middle line:84% People have demonstrated prototypes 00:42:46.490 --> 00:42:49.770 align:middle line:84% of analog optical computers using this kind of phase change 00:42:49.770 --> 00:42:51.970 align:middle line:90% memory materials. 00:42:51.970 --> 00:42:54.850 align:middle line:84% So that's one application where this kind of material 00:42:54.850 --> 00:42:56.770 align:middle line:84% has found a lot of traction there. 00:42:56.770 --> 00:43:00.330 align:middle line:84% Now the other useful aspect where this could also be used 00:43:00.330 --> 00:43:02.650 align:middle line:84% is what we have now, essentially, 00:43:02.650 --> 00:43:05.292 align:middle line:84% is a reconfigurable optical media. 00:43:05.292 --> 00:43:07.250 align:middle line:84% If you think about conventional optical devices 00:43:07.250 --> 00:43:11.250 align:middle line:84% like lens, prism, or whatever, once it's made, it's made. 00:43:11.250 --> 00:43:13.110 align:middle line:84% If you have a lens, if it's fabricated, 00:43:13.110 --> 00:43:14.930 align:middle line:90% the focal length is fixed. 00:43:14.930 --> 00:43:17.490 align:middle line:84% So there's nothing you can do to really change it. 00:43:17.490 --> 00:43:19.430 align:middle line:84% But now with this kind of new material, 00:43:19.430 --> 00:43:21.250 align:middle line:84% it gives you essentially additional knob 00:43:21.250 --> 00:43:24.450 align:middle line:84% that allows you to tune the properties of optical systems 00:43:24.450 --> 00:43:26.450 align:middle line:90% post fabrication. 00:43:26.450 --> 00:43:27.828 align:middle line:84% So that's quite useful to realize 00:43:27.828 --> 00:43:29.870 align:middle line:84% different kind of reconfigurable optical devices. 00:43:29.870 --> 00:43:31.090 align:middle line:84% For example, optical switch, you can 00:43:31.090 --> 00:43:32.130 align:middle line:84% use that to essentially steer light 00:43:32.130 --> 00:43:33.490 align:middle line:90% towards different directions. 00:43:33.490 --> 00:43:35.980 align:middle line:84% Now this is another interesting example here. 00:43:35.980 --> 00:43:38.380 align:middle line:84% What people have shown in this case 00:43:38.380 --> 00:43:40.420 align:middle line:84% is that if you deposit a very thin layer 00:43:40.420 --> 00:43:43.980 align:middle line:84% of this phase-changing material on top of some dielectric stack, 00:43:43.980 --> 00:43:45.740 align:middle line:84% by controlling the light interference 00:43:45.740 --> 00:43:47.940 align:middle line:84% at different wavelengths you can actually 00:43:47.940 --> 00:43:49.860 align:middle line:84% get different colors when you switch 00:43:49.860 --> 00:43:51.940 align:middle line:90% from amorphous to crystalline. 00:43:51.940 --> 00:43:55.180 align:middle line:90% Amorphous is like pink. 00:43:55.180 --> 00:43:58.020 align:middle line:84% Crystalline becomes more like yellowish. 00:43:58.020 --> 00:44:01.660 align:middle line:84% So what can you use this technology for? 00:44:01.660 --> 00:44:04.960 align:middle line:84% You can actually change the color of a thin film media. 00:44:04.960 --> 00:44:08.620 align:middle line:90% 00:44:08.620 --> 00:44:11.210 align:middle line:90% You can do displays. 00:44:11.210 --> 00:44:13.460 align:middle line:84% So this is actually a demonstration first demonstrated 00:44:13.460 --> 00:44:15.100 align:middle line:90% by a group in Oxford. 00:44:15.100 --> 00:44:17.737 align:middle line:84% What they were able to show is that by essentially switching 00:44:17.737 --> 00:44:19.320 align:middle line:84% the material between these two states, 00:44:19.320 --> 00:44:23.020 align:middle line:84% they can essentially realize color displays. 00:44:23.020 --> 00:44:25.020 align:middle line:84% What they would do is they deposited 00:44:25.020 --> 00:44:27.860 align:middle line:90% the material, a thin film. 00:44:27.860 --> 00:44:30.380 align:middle line:84% They then define small pixels, where 00:44:30.380 --> 00:44:32.860 align:middle line:84% each pixel can be individually switched between amorphous 00:44:32.860 --> 00:44:35.120 align:middle line:90% and crystalline states. 00:44:35.120 --> 00:44:38.720 align:middle line:84% And by essentially interlacing different kind 00:44:38.720 --> 00:44:41.220 align:middle line:84% of pixels corresponding to different colors together, 00:44:41.220 --> 00:44:42.320 align:middle line:84% they were then able to demonstrate 00:44:42.320 --> 00:44:44.700 align:middle line:84% that they can actually realize color displays in this case. 00:44:44.700 --> 00:44:46.680 align:middle line:84% OK, so these are actually some actual patterns 00:44:46.680 --> 00:44:48.760 align:middle line:90% like this logo of Oxford. 00:44:48.760 --> 00:44:51.840 align:middle line:84% This is actually written in this kind of phase change materials. 00:44:51.840 --> 00:44:55.360 align:middle line:84% So the advantage of this kind of display 00:44:55.360 --> 00:44:58.080 align:middle line:84% is that this is a little bit different from the display 00:44:58.080 --> 00:45:00.080 align:middle line:90% you have on your laptop. 00:45:00.080 --> 00:45:03.920 align:middle line:84% It relies on light reflection rather than the light emission. 00:45:03.920 --> 00:45:06.380 align:middle line:84% Normal displays essentially emits light of different color. 00:45:06.380 --> 00:45:08.520 align:middle line:84% But this is actually purely based on reflection. 00:45:08.520 --> 00:45:11.520 align:middle line:84% So this is more like a electronic paper. 00:45:11.520 --> 00:45:14.360 align:middle line:84% The more exact analog of this kind of display 00:45:14.360 --> 00:45:18.320 align:middle line:84% is not your laptop, but rather Kindle. 00:45:18.320 --> 00:45:20.175 align:middle line:84% Have you used Kindle from Amazon? 00:45:20.175 --> 00:45:20.800 align:middle line:90% AUDIENCE: Yeah. 00:45:20.800 --> 00:45:22.680 align:middle line:90% PROFESSOR: Yeah, like e-readers. 00:45:22.680 --> 00:45:25.413 align:middle line:84% So the way Kindles work is also reflective display. 00:45:25.413 --> 00:45:27.580 align:middle line:84% They have essentially little spheres with two sides, 00:45:27.580 --> 00:45:28.902 align:middle line:90% black and white. 00:45:28.902 --> 00:45:30.360 align:middle line:84% By electric field they can actually 00:45:30.360 --> 00:45:32.000 align:middle line:84% control orientation of those spheres, 00:45:32.000 --> 00:45:34.500 align:middle line:84% whether it's black or white, and then train the reflectance. 00:45:34.500 --> 00:45:38.340 align:middle line:84% The advantage of this kind of electronic paper display, 00:45:38.340 --> 00:45:40.540 align:middle line:84% like Kindle, is that they are extremely 00:45:40.540 --> 00:45:43.820 align:middle line:84% power efficient because they are non-volatile. 00:45:43.820 --> 00:45:46.120 align:middle line:84% Once you have absorb the state of the sphere, 00:45:46.120 --> 00:45:48.600 align:middle line:84% you do need continuous power supply to maintain it. 00:45:48.600 --> 00:45:51.300 align:middle line:84% So that's why your Kindle can last 00:45:51.300 --> 00:45:53.500 align:middle line:84% way longer than your laptop lasts, 00:45:53.500 --> 00:45:57.940 align:middle line:84% for days, months, rather than a laptop just a few hours. 00:45:57.940 --> 00:45:59.498 align:middle line:90% OK, so same idea here. 00:45:59.498 --> 00:46:01.540 align:middle line:84% Again, this feature, here, they are non-volatile. 00:46:01.540 --> 00:46:03.460 align:middle line:84% So the hope there is that they can utilize 00:46:03.460 --> 00:46:05.780 align:middle line:84% this non-volatile reflected display, 00:46:05.780 --> 00:46:08.640 align:middle line:84% but in color rather than just in black and white, 00:46:08.640 --> 00:46:10.340 align:middle line:90% just like in Kindle. 00:46:10.340 --> 00:46:11.480 align:middle line:90% OK, yes. 00:46:11.480 --> 00:46:14.640 align:middle line:84% AUDIENCE: Is it still restricted to what colors 00:46:14.640 --> 00:46:18.220 align:middle line:84% it can use because the amorphous and crystalline states 00:46:18.220 --> 00:46:20.883 align:middle line:90% aren't specific? 00:46:20.883 --> 00:46:21.800 align:middle line:90% PROFESSOR: Not really. 00:46:21.800 --> 00:46:24.900 align:middle line:84% Basically, the idea there is that even though you 00:46:24.900 --> 00:46:27.160 align:middle line:84% have the same material with fixed optical constants, 00:46:27.160 --> 00:46:29.280 align:middle line:84% you can train the layer structures. 00:46:29.280 --> 00:46:31.280 align:middle line:84% If you put it around different dielectric layers 00:46:31.280 --> 00:46:33.320 align:middle line:84% with different thicknesses, it changes the interference 00:46:33.320 --> 00:46:34.812 align:middle line:84% of light at different wavelengths. 00:46:34.812 --> 00:46:37.020 align:middle line:84% So you can encode essentially one pixel, for example, 00:46:37.020 --> 00:46:39.160 align:middle line:84% a switch between, say, black and red. 00:46:39.160 --> 00:46:41.880 align:middle line:84% The other ones switch between black and gray, the other one, 00:46:41.880 --> 00:46:43.038 align:middle line:90% black and blue. 00:46:43.038 --> 00:46:44.580 align:middle line:84% Then by combining the three together, 00:46:44.580 --> 00:46:47.180 align:middle line:84% you can actually get essentially [AUDIO OUT]. 00:46:47.180 --> 00:46:51.320 align:middle line:84% AUDIENCE: So each pixel can be layered with different-- 00:46:51.320 --> 00:46:52.807 align:middle line:84% PROFESSOR: In principle, at least. 00:46:52.807 --> 00:46:54.640 align:middle line:84% Yeah, you can design it so that you actually 00:46:54.640 --> 00:46:55.640 align:middle line:90% get full-color display. 00:46:55.640 --> 00:46:59.080 align:middle line:90% AUDIENCE: In color. 00:46:59.080 --> 00:47:01.220 align:middle line:90% Those are really expensive. 00:47:01.220 --> 00:47:05.160 align:middle line:90% 00:47:05.160 --> 00:47:07.960 align:middle line:90% PROFESSOR: So, no. 00:47:07.960 --> 00:47:10.360 align:middle line:84% Last example here, I just want to mention 00:47:10.360 --> 00:47:13.308 align:middle line:84% that for the efficient mirror I mentioned before, GST, you 00:47:13.308 --> 00:47:15.600 align:middle line:84% notice that when you switch from amorphous crystalline, 00:47:15.600 --> 00:47:17.580 align:middle line:84% you introduce larger refraction exchange. 00:47:17.580 --> 00:47:20.240 align:middle line:84% You also cause a very large absorption change, as well. 00:47:20.240 --> 00:47:23.280 align:middle line:84% The [INAUDIBLE] is optically absorbing. 00:47:23.280 --> 00:47:25.853 align:middle line:84% This is considered as a potential drawback because 00:47:25.853 --> 00:47:27.520 align:middle line:84% of a lot of optical devices, you imagine 00:47:27.520 --> 00:47:29.650 align:middle line:84% that you want something that's transparent, 00:47:29.650 --> 00:47:32.330 align:middle line:90% rather than absorbing. 00:47:32.330 --> 00:47:34.710 align:middle line:84% So a few years back-- this is actually from my group-- 00:47:34.710 --> 00:47:36.250 align:middle line:84% we actually discovered that you can actually 00:47:36.250 --> 00:47:38.500 align:middle line:84% make a phase-changing material that switch between two 00:47:38.500 --> 00:47:42.090 align:middle line:84% transparent states, rather than one absorbing and one 00:47:42.090 --> 00:47:43.370 align:middle line:90% transparent state. 00:47:43.370 --> 00:47:45.590 align:middle line:84% So the material, this is actually the material. 00:47:45.590 --> 00:47:48.150 align:middle line:84% Again, I'm not going to go through details, JSSD. 00:47:48.150 --> 00:47:49.710 align:middle line:84% But if you look at optical constants, 00:47:49.710 --> 00:47:52.630 align:middle line:84% it still shows a very large change in refractive index. 00:47:52.630 --> 00:47:55.330 align:middle line:84% But the absorption, which is labeled by extinction 00:47:55.330 --> 00:47:58.810 align:middle line:84% coefficient here, is mostly zero across the entire infrared range 00:47:58.810 --> 00:48:01.610 align:middle line:84% for both crystalline and amorphous state. 00:48:01.610 --> 00:48:05.210 align:middle line:84% So now you can imagine there's a whole lot of applications 00:48:05.210 --> 00:48:07.410 align:middle line:84% we can actually now explore in this kind 00:48:07.410 --> 00:48:09.570 align:middle line:84% of bi-state transparent phase change materials. 00:48:09.570 --> 00:48:12.410 align:middle line:84% This is actually something we have done a few years back. 00:48:12.410 --> 00:48:13.970 align:middle line:84% We demonstrate that by essentially 00:48:13.970 --> 00:48:18.370 align:middle line:84% patterning this material into these little subwavelength 00:48:18.370 --> 00:48:19.610 align:middle line:90% structures. 00:48:19.610 --> 00:48:21.832 align:middle line:84% And by designing that appropriately when you actually 00:48:21.832 --> 00:48:23.790 align:middle line:84% switch between amorphous and crystalline state, 00:48:23.790 --> 00:48:27.350 align:middle line:84% you can change the phase of light. 00:48:27.350 --> 00:48:29.530 align:middle line:84% Not phase, not in terms of material structure. 00:48:29.530 --> 00:48:33.050 align:middle line:84% I'm talking about the oscillation of the phase, sine, 00:48:33.050 --> 00:48:34.490 align:middle line:90% cosine phase of the light. 00:48:34.490 --> 00:48:37.310 align:middle line:84% So you can change the phase of light at different locations. 00:48:37.310 --> 00:48:38.410 align:middle line:84% And therefore, you can actually change. 00:48:38.410 --> 00:48:40.118 align:middle line:84% For example, in this case, this is a lens 00:48:40.118 --> 00:48:43.550 align:middle line:84% that actually can shift the focus by just simply switching 00:48:43.550 --> 00:48:45.690 align:middle line:84% between amorphous and crystalline states. 00:48:45.690 --> 00:48:48.830 align:middle line:84% So there are quite some interesting applications 00:48:48.830 --> 00:48:52.190 align:middle line:84% you can now explore using these new materials there, 00:48:52.190 --> 00:48:55.150 align:middle line:84% now they are actually transparent. 00:48:55.150 --> 00:48:58.110 align:middle line:84% So just to summarize what I have discussed here, 00:48:58.110 --> 00:49:00.050 align:middle line:84% this is a list of further readings 00:49:00.050 --> 00:49:02.670 align:middle line:84% if you're interested in this topic. 00:49:02.670 --> 00:49:07.350 align:middle line:84% I have already included here the Google Book link and this one. 00:49:07.350 --> 00:49:09.550 align:middle line:84% If you have an MIT certificate, you can actually 00:49:09.550 --> 00:49:12.375 align:middle line:90% download this book for free. 00:49:12.375 --> 00:49:14.250 align:middle line:84% I have also included a couple of papers here. 00:49:14.250 --> 00:49:17.490 align:middle line:84% This is a very nice review that discusses about, essentially, 00:49:17.490 --> 00:49:18.990 align:middle line:84% the phase change memory specifically 00:49:18.990 --> 00:49:20.550 align:middle line:90% for electronic applications. 00:49:20.550 --> 00:49:24.890 align:middle line:84% And then the last one here, this is the more recent publication. 00:49:24.890 --> 00:49:28.770 align:middle line:84% I consider it as a very nice paper 00:49:28.770 --> 00:49:30.770 align:middle line:84% that talks about the different strategies people 00:49:30.770 --> 00:49:33.312 align:middle line:84% use to engineer the material, how to combine material science 00:49:33.312 --> 00:49:35.810 align:middle line:84% and use that to actually make informed decisions about how 00:49:35.810 --> 00:49:37.210 align:middle line:84% to actually make better devices, how 00:49:37.210 --> 00:49:39.043 align:middle line:84% to tailor material competition, for example, 00:49:39.043 --> 00:49:41.450 align:middle line:84% to get faster crystallization, faster speed, low energy 00:49:41.450 --> 00:49:42.150 align:middle line:90% consumption. 00:49:42.150 --> 00:49:44.400 align:middle line:84% They talk about all the material design considerations 00:49:44.400 --> 00:49:46.210 align:middle line:90% in that particular paper. 00:49:46.210 --> 00:49:47.290 align:middle line:90% All right. 00:49:47.290 --> 00:49:52.930 align:middle line:84% So any more questions about phase change materials 00:49:52.930 --> 00:49:57.570 align:middle line:84% before we proceed to talk about the next topic? 00:49:57.570 --> 00:50:05.270 align:middle line:90% 00:50:05.270 --> 00:50:09.410 align:middle line:84% So we move on to the next, our fundamental science topic here. 00:50:09.410 --> 00:50:12.610 align:middle line:84% The next topic here is about viscosity of glass. 00:50:12.610 --> 00:50:14.990 align:middle line:84% Now specifically for this particular topic, 00:50:14.990 --> 00:50:17.050 align:middle line:84% we are going to try to address this question. 00:50:17.050 --> 00:50:20.090 align:middle line:90% Is glass considered a liquid? 00:50:20.090 --> 00:50:22.910 align:middle line:90% Or is it considered a solid? 00:50:22.910 --> 00:50:26.070 align:middle line:84% Now, you may already have answered from the first lecture. 00:50:26.070 --> 00:50:27.750 align:middle line:84% You know that glass is very different 00:50:27.750 --> 00:50:29.210 align:middle line:90% from supercooled liquid. 00:50:29.210 --> 00:50:30.610 align:middle line:84% They have different thermal expansion coefficient. 00:50:30.610 --> 00:50:31.650 align:middle line:84% They have different heat capacity. 00:50:31.650 --> 00:50:33.050 align:middle line:84% So fundamentally, they are very different, 00:50:33.050 --> 00:50:34.130 align:middle line:90% two different states of matter. 00:50:34.130 --> 00:50:35.990 align:middle line:84% Well, that's not the answer I'm looking for. 00:50:35.990 --> 00:50:39.390 align:middle line:84% I'm looking more from rheological perspective. 00:50:39.390 --> 00:50:44.510 align:middle line:90% Does glass flow, basically? 00:50:44.510 --> 00:50:46.630 align:middle line:84% Now your immediate intuitive answer 00:50:46.630 --> 00:50:49.190 align:middle line:84% may be that, well, glass doesn't really flow. 00:50:49.190 --> 00:50:51.110 align:middle line:90% You don't see a flow. 00:50:51.110 --> 00:50:55.590 align:middle line:84% Now let me give you a counter example just to confuse you 00:50:55.590 --> 00:50:56.830 align:middle line:90% a little bit. 00:50:56.830 --> 00:51:01.930 align:middle line:84% Think about pitch, or asphalt. It will be considered as solid. 00:51:01.930 --> 00:51:04.310 align:middle line:84% That's why people use them to pave roads. 00:51:04.310 --> 00:51:07.390 align:middle line:84% However, it's actually a liquid, just 00:51:07.390 --> 00:51:10.110 align:middle line:90% with very, very high viscosity. 00:51:10.110 --> 00:51:12.630 align:middle line:84% So in this long-term experiment there, what they did is 00:51:12.630 --> 00:51:14.950 align:middle line:84% they actually put some pitch there in this funnel. 00:51:14.950 --> 00:51:17.510 align:middle line:84% And then they just basically wait. 00:51:17.510 --> 00:51:21.240 align:middle line:84% And over time, this pitch slowly flows. 00:51:21.240 --> 00:51:24.840 align:middle line:84% And eventually, you can collect the droplets at the bottom here. 00:51:24.840 --> 00:51:27.680 align:middle line:84% So the most famous version of this experiment 00:51:27.680 --> 00:51:30.400 align:middle line:84% was started-- this is almost 100 years ago. 00:51:30.400 --> 00:51:32.600 align:middle line:90% And now they have how many jobs? 00:51:32.600 --> 00:51:33.100 align:middle line:90% Let's see. 00:51:33.100 --> 00:51:36.297 align:middle line:84% I think they only have something like eight to nine jobs fail. 00:51:36.297 --> 00:51:37.380 align:middle line:90% I don't recall the number. 00:51:37.380 --> 00:51:38.968 align:middle line:90% But you can go on Wikipedia. 00:51:38.968 --> 00:51:40.260 align:middle line:90% They make it a pretty big deal. 00:51:40.260 --> 00:51:42.440 align:middle line:84% Every time one job falls, it's like there's 00:51:42.440 --> 00:51:44.600 align:middle line:90% a celebration around it. 00:51:44.600 --> 00:51:45.760 align:middle line:90% All right. 00:51:45.760 --> 00:51:47.565 align:middle line:84% So takeaway message number one, if you're 00:51:47.565 --> 00:51:49.440 align:middle line:84% considering going to grad school where you're 00:51:49.440 --> 00:51:51.200 align:middle line:84% looking for a PhD thesis topic, this 00:51:51.200 --> 00:51:54.600 align:middle line:90% is probably a terrible choice. 00:51:54.600 --> 00:51:55.860 align:middle line:90% Well, to be real. 00:51:55.860 --> 00:51:57.720 align:middle line:84% Now the main takeaway message here 00:51:57.720 --> 00:52:00.280 align:middle line:84% is that the boundary between solid and liquid 00:52:00.280 --> 00:52:02.560 align:middle line:90% is not so well defined. 00:52:02.560 --> 00:52:04.920 align:middle line:84% How do you tell whether something is actually solid, 00:52:04.920 --> 00:52:08.960 align:middle line:84% or something which is extremely high viscosity? 00:52:08.960 --> 00:52:14.760 align:middle line:84% Now coming to the specific case of glass, some of you 00:52:14.760 --> 00:52:17.400 align:middle line:84% may have heard this urban legend. 00:52:17.400 --> 00:52:24.700 align:middle line:84% If you look at the color windows in medieval churches, 00:52:24.700 --> 00:52:26.580 align:middle line:84% if you look at them closely, you probably 00:52:26.580 --> 00:52:28.322 align:middle line:84% notice that a lot of these window panes, 00:52:28.322 --> 00:52:30.780 align:middle line:84% they tend to be a little bit thicker at the bottom compared 00:52:30.780 --> 00:52:32.307 align:middle line:90% on the top. 00:52:32.307 --> 00:52:34.140 align:middle line:84% And that's why some people claim that that's 00:52:34.140 --> 00:52:37.440 align:middle line:84% because over centuries of time, glass flows slowly. 00:52:37.440 --> 00:52:40.480 align:middle line:84% It's just a liquid, basically, with high viscosity. 00:52:40.480 --> 00:52:42.620 align:middle line:84% So that's why it gets thicker at the bottom. 00:52:42.620 --> 00:52:44.810 align:middle line:90% Is it true? 00:52:44.810 --> 00:52:46.560 align:middle line:84% Some of you are already shaking your head. 00:52:46.560 --> 00:52:48.160 align:middle line:90% Well, that's a correct answer. 00:52:48.160 --> 00:52:50.980 align:middle line:84% Well now, just to confuse it a little bit more, 00:52:50.980 --> 00:52:52.780 align:middle line:84% let's think about this press release. 00:52:52.780 --> 00:52:53.787 align:middle line:90% So this is from Hitachi. 00:52:53.787 --> 00:52:55.620 align:middle line:84% What they were able to develop, essentially, 00:52:55.620 --> 00:52:56.840 align:middle line:90% is a new memory technology. 00:52:56.840 --> 00:52:59.425 align:middle line:84% It's somewhat similar to optical disks. 00:52:59.425 --> 00:53:00.800 align:middle line:84% But it's a more advanced version. 00:53:00.800 --> 00:53:02.340 align:middle line:90% You basically have silica glass. 00:53:02.340 --> 00:53:04.900 align:middle line:84% You can use laser to essentially locally modify the material 00:53:04.900 --> 00:53:08.118 align:middle line:84% structure and change optical properties so that you can then 00:53:08.118 --> 00:53:08.660 align:middle line:90% later on use. 00:53:08.660 --> 00:53:10.520 align:middle line:84% Essentially, you can read it out by looking at optical property 00:53:10.520 --> 00:53:11.308 align:middle line:90% contrast. 00:53:11.308 --> 00:53:13.100 align:middle line:84% The advantage there is that you can read it 00:53:13.100 --> 00:53:15.160 align:middle line:84% at very high density, unlike optical disk. 00:53:15.160 --> 00:53:17.500 align:middle line:84% It's basically one layer that can do many, many layers. 00:53:17.500 --> 00:53:21.200 align:middle line:84% And that's why they have high information storage density. 00:53:21.200 --> 00:53:22.362 align:middle line:90% And it's also very stable. 00:53:22.362 --> 00:53:23.820 align:middle line:84% That's why it's getting attractive. 00:53:23.820 --> 00:53:25.695 align:middle line:84% People are trying to use it as a backup media 00:53:25.695 --> 00:53:28.260 align:middle line:84% to backup everything you have on the cloud. 00:53:28.260 --> 00:53:30.560 align:middle line:84% Microsoft is also working on the technologies. 00:53:30.560 --> 00:53:34.280 align:middle line:84% But what they claim is that you can now 00:53:34.280 --> 00:53:41.280 align:middle line:84% store the information for 300 million years. 00:53:41.280 --> 00:53:43.083 align:middle line:84% So if you search for that press release 00:53:43.083 --> 00:53:45.500 align:middle line:84% and see the comments people post under that press release, 00:53:45.500 --> 00:53:47.760 align:middle line:84% people were like, well, this is-- 00:53:47.760 --> 00:53:49.880 align:middle line:90% OK, yes. 00:53:49.880 --> 00:53:52.540 align:middle line:84% Over millions of years, glass is going to flow like liquid. 00:53:52.540 --> 00:53:53.940 align:middle line:90% Nothing is going to be kept. 00:53:53.940 --> 00:53:56.220 align:middle line:90% Is that true? 00:53:56.220 --> 00:53:58.220 align:middle line:84% Now we're not talking about a few hundred years. 00:53:58.220 --> 00:54:02.240 align:middle line:84% There's millions of years, million years. 00:54:02.240 --> 00:54:05.400 align:middle line:84% OK, well, that's what we're trying to address in this class 00:54:05.400 --> 00:54:06.060 align:middle line:90% here. 00:54:06.060 --> 00:54:08.200 align:middle line:90% Let's see if it's true. 00:54:08.200 --> 00:54:10.498 align:middle line:84% Before we go into more specific details, 00:54:10.498 --> 00:54:12.040 align:middle line:84% now we need to first essentially look 00:54:12.040 --> 00:54:14.193 align:middle line:84% at some basic properties about viscosity. 00:54:14.193 --> 00:54:15.860 align:middle line:84% And I'm trying to draw a comparison here 00:54:15.860 --> 00:54:18.340 align:middle line:84% between viscosity and elasticity. 00:54:18.340 --> 00:54:21.480 align:middle line:84% So you are all aware of elasticity. 00:54:21.480 --> 00:54:23.580 align:middle line:84% That's very simple for mechanics. 00:54:23.580 --> 00:54:25.740 align:middle line:84% You can approximate lots of materials essentially 00:54:25.740 --> 00:54:26.980 align:middle line:90% as a who can media. 00:54:26.980 --> 00:54:29.700 align:middle line:84% So the strain and stress, they follow this very simple 00:54:29.700 --> 00:54:31.080 align:middle line:90% nice linear relation. 00:54:31.080 --> 00:54:33.520 align:middle line:84% This is for normal strain and stress. 00:54:33.520 --> 00:54:35.940 align:middle line:84% This is for sheer strain and stress. 00:54:35.940 --> 00:54:42.220 align:middle line:84% But essentially, strain is proportional to stress. 00:54:42.220 --> 00:54:43.320 align:middle line:90% And I put it here. 00:54:43.320 --> 00:54:45.780 align:middle line:84% They are instantaneous and transient. 00:54:45.780 --> 00:54:48.220 align:middle line:84% What that means is that if you have elastic media, 00:54:48.220 --> 00:54:51.660 align:middle line:84% ideal elastic media, you try to deform it, apply a force. 00:54:51.660 --> 00:54:54.620 align:middle line:84% You immediately see elastic deformation. 00:54:54.620 --> 00:54:59.260 align:middle line:84% If you remove the force, it immediately recovers. 00:54:59.260 --> 00:55:02.140 align:middle line:84% That's why this trend is directly proportional 00:55:02.140 --> 00:55:03.020 align:middle line:90% to stress here. 00:55:03.020 --> 00:55:04.880 align:middle line:90% There's no time dependence. 00:55:04.880 --> 00:55:08.740 align:middle line:90% Everything is instantaneous. 00:55:08.740 --> 00:55:12.180 align:middle line:90% Now viscous is the opposite. 00:55:12.180 --> 00:55:14.250 align:middle line:84% This viscous flow is always time dependent. 00:55:14.250 --> 00:55:16.110 align:middle line:84% If you look at the governing law here, 00:55:16.110 --> 00:55:17.910 align:middle line:90% this is more Newtonian fluid. 00:55:17.910 --> 00:55:22.070 align:middle line:84% What it basically says is that the shear stress 00:55:22.070 --> 00:55:26.670 align:middle line:84% is proportional to the shear strain rate. 00:55:26.670 --> 00:55:29.850 align:middle line:84% It's not proportional to shear strain, but shear strain rate. 00:55:29.850 --> 00:55:32.880 align:middle line:90% 00:55:32.880 --> 00:55:35.550 align:middle line:84% So first thing I want to mention there is therefore essentially 00:55:35.550 --> 00:55:39.110 align:middle line:84% for viscosity, you don't have normal stress. 00:55:39.110 --> 00:55:42.150 align:middle line:84% Normal stress is exclusive for elastics. 00:55:42.150 --> 00:55:44.047 align:middle line:84% For liquid, if you try to compress it, 00:55:44.047 --> 00:55:45.130 align:middle line:90% you get the same behavior. 00:55:45.130 --> 00:55:47.575 align:middle line:84% You get elastic deformation, similar to solids. 00:55:47.575 --> 00:55:48.950 align:middle line:84% When you talk about viscous flow, 00:55:48.950 --> 00:55:51.250 align:middle line:84% this always happens due to shear motion. 00:55:51.250 --> 00:55:52.490 align:middle line:90% You have one pass of liquid. 00:55:52.490 --> 00:55:54.230 align:middle line:84% They try to flow past the other part, 00:55:54.230 --> 00:55:57.430 align:middle line:90% try to move past the other part. 00:55:57.430 --> 00:56:00.030 align:middle line:84% So only shear motion is involved. 00:56:00.030 --> 00:56:04.750 align:middle line:84% Two is that because this is shear rate, what it means 00:56:04.750 --> 00:56:08.110 align:middle line:84% is that you don't need a constant force to sustain 00:56:08.110 --> 00:56:10.890 align:middle line:90% the shape deformation. 00:56:10.890 --> 00:56:14.290 align:middle line:84% So suppose you already have some sheer deformation, here epsilon. 00:56:14.290 --> 00:56:15.910 align:middle line:90% But it's not changing with time. 00:56:15.910 --> 00:56:20.530 align:middle line:84% The corresponding shear force is 0. 00:56:20.530 --> 00:56:22.170 align:middle line:84% And that's intuitive because liquid, 00:56:22.170 --> 00:56:23.790 align:middle line:90% it can hold any kind of shape. 00:56:23.790 --> 00:56:25.190 align:middle line:84% So as long as definition happens, 00:56:25.190 --> 00:56:26.857 align:middle line:84% as long as you don't confuse, deform it, 00:56:26.857 --> 00:56:29.210 align:middle line:84% you don't need to apply any force. 00:56:29.210 --> 00:56:34.730 align:middle line:84% So in this regard, viscosity is sort of like internal friction. 00:56:34.730 --> 00:56:37.850 align:middle line:84% Only when you're trying to deform the liquid actively, 00:56:37.850 --> 00:56:40.710 align:middle line:84% that's when you actually see the viscous force coming into play. 00:56:40.710 --> 00:56:43.570 align:middle line:90% 00:56:43.570 --> 00:56:45.950 align:middle line:84% So the first thing you try to deform it, 00:56:45.950 --> 00:56:47.610 align:middle line:84% the more this friction is going to be. 00:56:47.610 --> 00:56:50.890 align:middle line:90% 00:56:50.890 --> 00:56:51.390 align:middle line:90% All right. 00:56:51.390 --> 00:56:55.770 align:middle line:84% So now in terms of glass, the answer about 00:56:55.770 --> 00:56:57.330 align:middle line:84% whether glass is liquid or solid, 00:56:57.330 --> 00:56:59.050 align:middle line:84% it really depends on lots of factors, 00:56:59.050 --> 00:57:00.758 align:middle line:90% most dominantly temperature. 00:57:00.758 --> 00:57:03.050 align:middle line:84% So we know that obviously if you go to high temperature 00:57:03.050 --> 00:57:05.550 align:middle line:84% when you go to supercooled liquid, it is liquid, 00:57:05.550 --> 00:57:06.730 align:middle line:90% no argument. 00:57:06.730 --> 00:57:10.750 align:middle line:84% If you go to a low temperature, like room temperature, 00:57:10.750 --> 00:57:12.830 align:middle line:90% you can go to the glassy state. 00:57:12.830 --> 00:57:15.090 align:middle line:84% And as I will show you later, real logically, 00:57:15.090 --> 00:57:17.950 align:middle line:84% a lot of this glassy state is practically solid. 00:57:17.950 --> 00:57:18.690 align:middle line:90% It doesn't flow. 00:57:18.690 --> 00:57:21.810 align:middle line:84% The viscosity is so low that it's negligible, even over 00:57:21.810 --> 00:57:24.030 align:middle line:84% like millions of years of timescales. 00:57:24.030 --> 00:57:29.870 align:middle line:84% But then in between, you have the glass transition regime, 00:57:29.870 --> 00:57:31.990 align:middle line:84% where this material actually exhibits 00:57:31.990 --> 00:57:35.830 align:middle line:84% a combination of properties from both solid and liquid. 00:57:35.830 --> 00:57:39.290 align:middle line:84% So later on, we'll have one chapter dedicated to it, 00:57:39.290 --> 00:57:41.510 align:middle line:84% just talk about so-called viscoelasticity of glass 00:57:41.510 --> 00:57:43.770 align:middle line:84% materials, which, judging from its name, 00:57:43.770 --> 00:57:47.750 align:middle line:84% is a combination of viscous properties and elastic behavior. 00:57:47.750 --> 00:57:48.250 align:middle line:90% All right. 00:57:48.250 --> 00:57:50.710 align:middle line:84% So now let's start by taking a look at some 00:57:50.710 --> 00:57:53.830 align:middle line:84% of the microscopic origins of elasticity and viscosity, 00:57:53.830 --> 00:57:55.910 align:middle line:84% for elasticity is quite straightforward. 00:57:55.910 --> 00:57:58.310 align:middle line:84% When you're trying to compress or stretch a material, 00:57:58.310 --> 00:58:00.470 align:middle line:84% you're changing it into atomic spacing. 00:58:00.470 --> 00:58:02.850 align:middle line:84% So when you change it into atomic spacing, 00:58:02.850 --> 00:58:05.030 align:middle line:84% you're changing the relative magnitude 00:58:05.030 --> 00:58:07.610 align:middle line:84% of the attractive and repulsive forces. 00:58:07.610 --> 00:58:10.470 align:middle line:84% If you try to compress it, repulsive force becomes larger. 00:58:10.470 --> 00:58:12.730 align:middle line:84% So that's why you see essentially a resistive force 00:58:12.730 --> 00:58:17.930 align:middle line:84% resisting this kind of deformation, and vice versa. 00:58:17.930 --> 00:58:20.930 align:middle line:84% Viscosity, on the other hand, this 00:58:20.930 --> 00:58:27.412 align:middle line:84% comes from internal friction or relaxation, as I call it. 00:58:27.412 --> 00:58:29.370 align:middle line:84% Essentially, when you have viscous deformation, 00:58:29.370 --> 00:58:31.730 align:middle line:84% you're changing geometric configuration of the material. 00:58:31.730 --> 00:58:37.030 align:middle line:84% You need to move some molecules or atoms past the others. 00:58:37.030 --> 00:58:38.290 align:middle line:90% That's the whole idea. 00:58:38.290 --> 00:58:41.290 align:middle line:84% And this can be realized with different kind 00:58:41.290 --> 00:58:43.370 align:middle line:84% of microstructure mechanisms, depending 00:58:43.370 --> 00:58:44.690 align:middle line:90% on different kind of materials. 00:58:44.690 --> 00:58:46.650 align:middle line:84% If you think about polymers, if you 00:58:46.650 --> 00:58:48.530 align:middle line:84% think of long polymer chains, they 00:58:48.530 --> 00:58:49.910 align:middle line:84% can have different conformations. 00:58:49.910 --> 00:58:50.590 align:middle line:90% They can be twisted. 00:58:50.590 --> 00:58:51.430 align:middle line:90% They can be folded. 00:58:51.430 --> 00:58:53.270 align:middle line:84% They can be stretched to a straight line. 00:58:53.270 --> 00:58:55.353 align:middle line:84% So when you're changing in between these different 00:58:55.353 --> 00:58:58.690 align:middle line:84% conformations, you change the overall shape of the material. 00:58:58.690 --> 00:59:01.530 align:middle line:84% So that's a kind of viscous flow. 00:59:01.530 --> 00:59:05.500 align:middle line:84% If it's a material like glass, or maybe even like ceramics 00:59:05.500 --> 00:59:09.580 align:middle line:84% or metal under a high temperature, they can creep. 00:59:09.580 --> 00:59:11.940 align:middle line:84% Essentially, they deform plastically. 00:59:11.940 --> 00:59:14.820 align:middle line:84% In that process, you have essentially viscous information 00:59:14.820 --> 00:59:15.640 align:middle line:90% coming in. 00:59:15.640 --> 00:59:17.420 align:middle line:84% And the underlying mechanism would 00:59:17.420 --> 00:59:20.952 align:middle line:84% be for things like vacancy diffusion or dislocation motion. 00:59:20.952 --> 00:59:22.660 align:middle line:84% That's how you induce plastic deformation 00:59:22.660 --> 00:59:23.760 align:middle line:90% in disordered materials. 00:59:23.760 --> 00:59:27.825 align:middle line:90% 00:59:27.825 --> 00:59:31.020 align:middle line:84% So one key takeaway point here is 00:59:31.020 --> 00:59:33.340 align:middle line:84% that these different kind of mechanisms, they 00:59:33.340 --> 00:59:35.597 align:middle line:84% may have very different relaxation time scales. 00:59:35.597 --> 00:59:37.180 align:middle line:84% So they may or may not come into play, 00:59:37.180 --> 00:59:38.480 align:middle line:90% depending on the strain rate. 00:59:38.480 --> 00:59:42.460 align:middle line:84% Faster strain rate, you may have faster deformation mechanism 00:59:42.460 --> 00:59:44.420 align:middle line:84% coming into play, and vice versa. 00:59:44.420 --> 00:59:47.860 align:middle line:84% Also the rule of thumb is that if you 00:59:47.860 --> 00:59:50.660 align:middle line:84% have a stronger intermolecular interactions 00:59:50.660 --> 00:59:52.900 align:middle line:84% or inter atomic interactions, the materials 00:59:52.900 --> 00:59:57.240 align:middle line:84% tend to be a stronger resistance to deformation. 00:59:57.240 --> 00:59:59.880 align:middle line:84% And therefore, they will usually exhibit a higher viscosity. 00:59:59.880 --> 01:00:01.780 align:middle line:90% Again, quite intuitive there. 01:00:01.780 --> 01:00:04.480 align:middle line:84% So what I'm showing on this plot here, 01:00:04.480 --> 01:00:07.760 align:middle line:84% because viscosity is actually a parameter that is very less 01:00:07.760 --> 01:00:09.400 align:middle line:84% intuitive, I'm actually showing you 01:00:09.400 --> 01:00:10.960 align:middle line:84% a little chart that should compare 01:00:10.960 --> 01:00:14.320 align:middle line:84% some of the common liquid with on the viscosity scale. 01:00:14.320 --> 01:00:22.760 align:middle line:84% So the unit of viscosity is in a Pascal second The other unit 01:00:22.760 --> 01:00:25.100 align:middle line:84% people also use often use here is called Poise. 01:00:25.100 --> 01:00:30.640 align:middle line:90% 01:00:30.640 --> 01:00:35.760 align:middle line:84% So the Si unit is Pascal's second. 01:00:35.760 --> 01:00:42.080 align:middle line:90% And then you can also use Poise. 01:00:42.080 --> 01:00:48.080 align:middle line:84% I think it's a 0.1 Pascal second equal to 1 or equal to 1 poise, 01:00:48.080 --> 01:00:49.060 align:middle line:90% or maybe vice versa. 01:00:49.060 --> 01:00:51.625 align:middle line:84% You can Google up on wiki to find out exactly. 01:00:51.625 --> 01:00:53.000 align:middle line:84% But basically, they are trivially 01:00:53.000 --> 01:00:54.680 align:middle line:90% connected by a factor of 10. 01:00:54.680 --> 01:00:58.760 align:middle line:84% So what we see here is that for some liquid 01:00:58.760 --> 01:01:01.680 align:middle line:84% like water or acetone, they just flow easily. 01:01:01.680 --> 01:01:04.980 align:middle line:84% So that's why they tend to have low, essentially, 01:01:04.980 --> 01:01:06.140 align:middle line:90% viscosity here. 01:01:06.140 --> 01:01:09.700 align:middle line:84% But then if you go to something like ketchup or syrup 01:01:09.700 --> 01:01:13.100 align:middle line:84% or even like toothpaste, they're a lot more difficult to deform. 01:01:13.100 --> 01:01:14.800 align:middle line:90% So they have a higher viscosity. 01:01:14.800 --> 01:01:17.395 align:middle line:84% The glasses actually has a viscosity 01:01:17.395 --> 01:01:19.020 align:middle line:84% spending much wider scale than what I'm 01:01:19.020 --> 01:01:20.270 align:middle line:90% showing on the previous slide. 01:01:20.270 --> 01:01:22.107 align:middle line:90% OK I'm going to show that later. 01:01:22.107 --> 01:01:24.440 align:middle line:84% So now the next thing I also want to discuss about here, 01:01:24.440 --> 01:01:27.023 align:middle line:84% this is actually quite important is the temperature dependence 01:01:27.023 --> 01:01:28.460 align:middle line:90% of this differential phenomena. 01:01:28.460 --> 01:01:30.880 align:middle line:84% This is quite important for glass processing, 01:01:30.880 --> 01:01:33.560 align:middle line:90% especially so for elasticity. 01:01:33.560 --> 01:01:35.700 align:middle line:84% This is actually relatively straightforward. 01:01:35.700 --> 01:01:37.420 align:middle line:84% In this plot, I'm showing you how 01:01:37.420 --> 01:01:40.540 align:middle line:84% the elastic constant of silica glass 01:01:40.540 --> 01:01:44.420 align:middle line:84% is going to change as the temperature. 01:01:44.420 --> 01:01:48.540 align:middle line:84% Now clearly over this extremely wide temperature range 01:01:48.540 --> 01:01:51.580 align:middle line:84% this is like over 1000 degrees Celsius huge span 01:01:51.580 --> 01:01:52.700 align:middle line:90% of temperature. 01:01:52.700 --> 01:01:55.100 align:middle line:84% The compressibility or less constant 01:01:55.100 --> 01:01:58.540 align:middle line:90% change like 30%, barely. 01:01:58.540 --> 01:02:02.240 align:middle line:84% So it's a very, very small, very, very weak dependence 01:02:02.240 --> 01:02:03.160 align:middle line:90% on temperature. 01:02:03.160 --> 01:02:06.120 align:middle line:84% Now this is also understandable because if you 01:02:06.120 --> 01:02:08.340 align:middle line:84% think about the nature of this kind of elasticity, 01:02:08.340 --> 01:02:10.960 align:middle line:84% it comes from, essentially interatomic interactions. 01:02:10.960 --> 01:02:14.080 align:middle line:84% What this tells you is that this interatomic interactions 01:02:14.080 --> 01:02:16.740 align:middle line:84% in terms of attractive and repulsive forces, 01:02:16.740 --> 01:02:22.477 align:middle line:84% they are just simply not very strong function of temperature. 01:02:22.477 --> 01:02:25.060 align:middle line:84% As you heat up, you may get a little bit of thermal expansion. 01:02:25.060 --> 01:02:26.980 align:middle line:84% You change the spacing of atoms a little bit, 01:02:26.980 --> 01:02:29.220 align:middle line:84% but you are not changing the nature of the force. 01:02:29.220 --> 01:02:30.960 align:middle line:90% They interact in between them. 01:02:30.960 --> 01:02:34.880 align:middle line:84% So that's why elasticity tend to exhibit a very weak dependence 01:02:34.880 --> 01:02:37.400 align:middle line:90% on temperature. 01:02:37.400 --> 01:02:41.000 align:middle line:84% Now viscosity is different because viscosity, 01:02:41.000 --> 01:02:45.180 align:middle line:84% as we mentioned, it involves the relative motion of molecules. 01:02:45.180 --> 01:02:48.080 align:middle line:84% Some part of it has to move past the other parts. 01:02:48.080 --> 01:02:52.000 align:middle line:84% So any mechanism that involves molecular motion 01:02:52.000 --> 01:02:54.800 align:middle line:90% tend to be thermally activated. 01:02:54.800 --> 01:02:57.330 align:middle line:84% High temperature usually allows you to expect expedite 01:02:57.330 --> 01:02:59.870 align:middle line:90% the process exponentially. 01:02:59.870 --> 01:03:02.930 align:middle line:84% So we can formulate logically, essentially 01:03:02.930 --> 01:03:06.610 align:middle line:84% describe it using the Arrhenius equation there. 01:03:06.610 --> 01:03:08.648 align:middle line:84% And I have shown you at the bottom here, 01:03:08.648 --> 01:03:10.690 align:middle line:84% this is essentially some activation energy values 01:03:10.690 --> 01:03:11.890 align:middle line:90% for different materials. 01:03:11.890 --> 01:03:15.410 align:middle line:84% So you notice here for silica, it 01:03:15.410 --> 01:03:17.810 align:middle line:84% tends to have a pretty high activation energy. 01:03:17.810 --> 01:03:21.262 align:middle line:84% Germania is a little bit lower because silica and germania, 01:03:21.262 --> 01:03:22.470 align:middle line:90% they have the same structure. 01:03:22.470 --> 01:03:24.270 align:middle line:90% Germania is basically GO2. 01:03:24.270 --> 01:03:25.450 align:middle line:90% It's the same structure. 01:03:25.450 --> 01:03:27.690 align:middle line:90% But germanium is a larger atom. 01:03:27.690 --> 01:03:30.730 align:middle line:84% Larger atom has a weaker interaction between the nucleus 01:03:30.730 --> 01:03:32.030 align:middle line:90% and outer shell electrons. 01:03:32.030 --> 01:03:33.950 align:middle line:84% So they tend to form weaker chemical bonds. 01:03:33.950 --> 01:03:36.150 align:middle line:84% Weaker chemical bonds means weaker interactions. 01:03:36.150 --> 01:03:39.290 align:middle line:84% And therefore you tend to have a less activation energy barrier 01:03:39.290 --> 01:03:41.170 align:middle line:90% for this kind of process there. 01:03:41.170 --> 01:03:43.010 align:middle line:84% So if you go to a other kind of solids 01:03:43.010 --> 01:03:45.230 align:middle line:84% there with even weaker chemical bonds, 01:03:45.230 --> 01:03:49.530 align:middle line:84% you see that the activation energy continue to decrease. 01:03:49.530 --> 01:03:52.170 align:middle line:84% Now the other thing you probably also 01:03:52.170 --> 01:03:55.030 align:middle line:84% have noticed here for silica, I have actually 01:03:55.030 --> 01:03:58.470 align:middle line:84% quoted two different activation energy 01:03:58.470 --> 01:04:00.830 align:middle line:84% values for different temperature ranges. 01:04:00.830 --> 01:04:04.470 align:middle line:84% So that means that this activation energy, in general, 01:04:04.470 --> 01:04:07.350 align:middle line:90% is temperature dependent. 01:04:07.350 --> 01:04:12.110 align:middle line:84% So the original law is not perfectly obeyed in a glass melt 01:04:12.110 --> 01:04:16.030 align:middle line:90% or in this kind of liquid here. 01:04:16.030 --> 01:04:19.750 align:middle line:84% So you can discard this kind of Arrhenius behavior 01:04:19.750 --> 01:04:22.430 align:middle line:84% of glass materials in terms of viscosity, 01:04:22.430 --> 01:04:25.470 align:middle line:84% which this is a well-known so-called angel plot. 01:04:25.470 --> 01:04:28.150 align:middle line:84% So let me explain what angel plot is. 01:04:28.150 --> 01:04:33.630 align:middle line:84% So for an angel plot, the horizontal axis 01:04:33.630 --> 01:04:37.030 align:middle line:84% is basically inverse temperature It's 01:04:37.030 --> 01:04:39.830 align:middle line:90% normalized with respect to TG. 01:04:39.830 --> 01:04:45.630 align:middle line:84% So again, TG, TG is glass transition temperature. 01:04:45.630 --> 01:04:48.150 align:middle line:84% And in the context of glass viscosity, 01:04:48.150 --> 01:04:52.750 align:middle line:84% we are always defining it as corresponding to a viscosity 01:04:52.750 --> 01:04:55.670 align:middle line:84% value of 10 to the 12th Pascal second. 01:04:55.670 --> 01:05:06.010 align:middle line:90% 01:05:06.010 --> 01:05:08.850 align:middle line:84% That's why, if you look at this horizontal axis here, 01:05:08.850 --> 01:05:13.330 align:middle line:84% this point is basically glass transition temperature. 01:05:13.330 --> 01:05:15.410 align:middle line:84% And that's why if you look at it here-- 01:05:15.410 --> 01:05:19.690 align:middle line:84% all the data-- by definition, all of these different curves 01:05:19.690 --> 01:05:21.567 align:middle line:84% corresponding to different glass competition, 01:05:21.567 --> 01:05:23.150 align:middle line:84% they all converge at one single point. 01:05:23.150 --> 01:05:24.570 align:middle line:90% This is by definition. 01:05:24.570 --> 01:05:27.290 align:middle line:84% A glass transition temperature, the viscosity should be 10 01:05:27.290 --> 01:05:30.970 align:middle line:84% to the 12th Pascal second, or 10 to the 13th poise. 01:05:30.970 --> 01:05:32.510 align:middle line:84% So now you know which one is which. 01:05:32.510 --> 01:05:35.410 align:middle line:84% So 1 plus a second equals to 10 poise. 01:05:35.410 --> 01:05:38.610 align:middle line:90% We have conclusive answer here. 01:05:38.610 --> 01:05:41.450 align:middle line:84% So that's why all the curves converge at this point. 01:05:41.450 --> 01:05:46.050 align:middle line:84% Now the vertical axis is basically viscosity here 01:05:46.050 --> 01:05:47.370 align:middle line:90% in logarithmic scale. 01:05:47.370 --> 01:05:50.470 align:middle line:84% So if you see a straight line here, 01:05:50.470 --> 01:05:53.990 align:middle line:84% it essentially indicates that the viscosity 01:05:53.990 --> 01:05:56.830 align:middle line:84% follows exponential dependence with respect to temperature. 01:05:56.830 --> 01:05:59.950 align:middle line:84% So any straight line is basically Arrhenius behavior. 01:05:59.950 --> 01:06:01.950 align:middle line:84% Now if you look at this plot, clearly there 01:06:01.950 --> 01:06:05.130 align:middle line:84% are two big categories of this glass melts. 01:06:05.130 --> 01:06:10.030 align:middle line:84% You get something like silica or germania. 01:06:10.030 --> 01:06:12.710 align:middle line:84% Again, we see that the activation 01:06:12.710 --> 01:06:15.490 align:middle line:84% energy changes a little bit, but not too dramatically. 01:06:15.490 --> 01:06:18.630 align:middle line:84% They still more or less follow the straight line here. 01:06:18.630 --> 01:06:20.510 align:middle line:90% You also get some other solids. 01:06:20.510 --> 01:06:22.530 align:middle line:84% You have this kind of ionic compounds. 01:06:22.530 --> 01:06:24.890 align:middle line:84% You get this organic or molecular solids. 01:06:24.890 --> 01:06:28.470 align:middle line:84% You see that they have this kind of very large change 01:06:28.470 --> 01:06:32.510 align:middle line:84% of the activation energy with respect to temperature. 01:06:32.510 --> 01:06:34.590 align:middle line:84% So we call these two kinds of behavior-- 01:06:34.590 --> 01:06:38.910 align:middle line:84% this is called a strong liquid for silicon, germanium, 01:06:38.910 --> 01:06:40.790 align:middle line:84% where it follows Arrhenius behavior. 01:06:40.790 --> 01:06:44.290 align:middle line:84% No Arrhenius behavior, this is called the fragile liquid. 01:06:44.290 --> 01:06:47.390 align:middle line:90% 01:06:47.390 --> 01:06:50.360 align:middle line:84% Now first thing you want to know is that this strong and fragile 01:06:50.360 --> 01:06:55.880 align:middle line:84% have nothing to do with mechanical properties of liquid. 01:06:55.880 --> 01:06:57.320 align:middle line:90% Again, liquid is liquid. 01:06:57.320 --> 01:07:01.132 align:middle line:84% You cannot describe, basically, like the strong or fragile. 01:07:01.132 --> 01:07:02.840 align:middle line:84% You cannot even use that to label liquid. 01:07:02.840 --> 01:07:03.980 align:middle line:90% You can use that to label solid. 01:07:03.980 --> 01:07:05.140 align:middle line:84% You cannot use that to label liquid. 01:07:05.140 --> 01:07:07.182 align:middle line:84% So it has nothing to do with mechanical behavior. 01:07:07.182 --> 01:07:10.473 align:middle line:84% All it says is how the viscosity depends on temperature, 01:07:10.473 --> 01:07:12.140 align:middle line:84% whether or not it follows Reynolds' law. 01:07:12.140 --> 01:07:13.320 align:middle line:90% That's what's shown. 01:07:13.320 --> 01:07:14.360 align:middle line:90% Fragile says. 01:07:14.360 --> 01:07:14.860 align:middle line:90% OK. 01:07:14.860 --> 01:07:17.400 align:middle line:90% 01:07:17.400 --> 01:07:19.760 align:middle line:84% Now they can be engaged quantitatively 01:07:19.760 --> 01:07:21.880 align:middle line:84% by this so-called fragility parameter here. 01:07:21.880 --> 01:07:23.260 align:middle line:90% It looks complicated. 01:07:23.260 --> 01:07:24.880 align:middle line:84% But basically all it says is that it's 01:07:24.880 --> 01:07:28.148 align:middle line:84% the slope of the curve in your plot, right at this point here. 01:07:28.148 --> 01:07:29.440 align:middle line:90% And it's quite straightforward. 01:07:29.440 --> 01:07:31.818 align:middle line:84% If you have a smaller slope, it goes along this line. 01:07:31.818 --> 01:07:32.860 align:middle line:90% This is basically strong. 01:07:32.860 --> 01:07:35.220 align:middle line:90% Your large slope, it goes here. 01:07:35.220 --> 01:07:37.020 align:middle line:90% This is basically fragile. 01:07:37.020 --> 01:07:39.220 align:middle line:84% So fragility parameter, just gauges there. 01:07:39.220 --> 01:07:42.240 align:middle line:90% 01:07:42.240 --> 01:07:44.560 align:middle line:90% Now I wonder why that matters? 01:07:44.560 --> 01:07:50.500 align:middle line:84% So from a practical perspective, fragility is quite important 01:07:50.500 --> 01:07:54.260 align:middle line:84% when it comes to glass processing. 01:07:54.260 --> 01:07:57.415 align:middle line:84% So question for you, which one do you think 01:07:57.415 --> 01:07:59.040 align:middle line:84% is more preferred for glass processing? 01:07:59.040 --> 01:08:03.817 align:middle line:84% Do you wants fragile liquid or do you wants strong liquid? 01:08:03.817 --> 01:08:04.900 align:middle line:90% AUDIENCE: Stronger liquid. 01:08:04.900 --> 01:08:08.953 align:middle line:90% AUDIENCE: What [INAUDIBLE]? 01:08:08.953 --> 01:08:11.120 align:middle line:84% PROFESSOR: Oh, that's a great way to get the answer. 01:08:11.120 --> 01:08:12.460 align:middle line:90% Which is correct? 01:08:12.460 --> 01:08:14.900 align:middle line:84% What happens there is that later on we're 01:08:14.900 --> 01:08:17.020 align:middle line:84% going to talk about specific viscosity windows 01:08:17.020 --> 01:08:18.700 align:middle line:84% where we actually process the materials. 01:08:18.700 --> 01:08:21.117 align:middle line:84% The viscosity window actually falls somewhere around here, 01:08:21.117 --> 01:08:22.380 align:middle line:90% like this regime here. 01:08:22.380 --> 01:08:24.340 align:middle line:84% So in this regime, what you notice 01:08:24.340 --> 01:08:26.160 align:middle line:84% here is that for fragile liquids, 01:08:26.160 --> 01:08:29.060 align:middle line:84% its viscosity changes a lot more dramatically with respect 01:08:29.060 --> 01:08:30.380 align:middle line:90% to temperature. 01:08:30.380 --> 01:08:32.740 align:middle line:84% And viscosity is a key process parameter. 01:08:32.740 --> 01:08:35.540 align:middle line:84% It controls how the liquid flows and how you deform it. 01:08:35.540 --> 01:08:38.300 align:middle line:84% So if you have a liquid where its property becomes 01:08:38.300 --> 01:08:40.340 align:middle line:84% very sensitive to temperature, it 01:08:40.340 --> 01:08:42.460 align:middle line:84% means that your process control has 01:08:42.460 --> 01:08:44.540 align:middle line:84% to be a lot more precise in order 01:08:44.540 --> 01:08:46.040 align:middle line:90% to get the desired results. 01:08:46.040 --> 01:08:48.220 align:middle line:84% So it has a much higher process sensitivity. 01:08:48.220 --> 01:08:49.640 align:middle line:84% Any temperature excursion is going 01:08:49.640 --> 01:08:51.560 align:middle line:84% to lead to much larger deviation. 01:08:51.560 --> 01:08:54.160 align:middle line:84% So practically, that's why it's important. 01:08:54.160 --> 01:08:56.859 align:middle line:84% Having a strong liquid can be beneficial in that case. 01:08:56.859 --> 01:09:00.200 align:middle line:90% 01:09:00.200 --> 01:09:03.560 align:middle line:84% Now fundamentally, from a microscopic perspective, 01:09:03.560 --> 01:09:07.204 align:middle line:84% the fact that you have the activation barrier 01:09:07.204 --> 01:09:08.579 align:middle line:84% can be a function of temperature. 01:09:08.579 --> 01:09:10.840 align:middle line:84% It essentially means that at different temperature 01:09:10.840 --> 01:09:16.620 align:middle line:84% regimes, the deformation, or the flow mechanism is different. 01:09:16.620 --> 01:09:19.279 align:middle line:90% 01:09:19.279 --> 01:09:21.399 align:middle line:84% So for a strong liquid like silica, 01:09:21.399 --> 01:09:22.880 align:middle line:90% you have strong chemical bonds. 01:09:22.880 --> 01:09:25.479 align:middle line:84% Those chemical bonds essentially retain this connection, 01:09:25.479 --> 01:09:28.120 align:middle line:90% even at high temperature. 01:09:28.120 --> 01:09:30.359 align:middle line:84% So that's why its activation energy doesn't really 01:09:30.359 --> 01:09:32.800 align:middle line:84% change with temperature, because the microscopic deformation is 01:09:32.800 --> 01:09:33.300 align:middle line:90% the same. 01:09:33.300 --> 01:09:36.040 align:middle line:84% You still get the same essentially atomic network. 01:09:36.040 --> 01:09:39.880 align:middle line:84% But for this kind of fragile liquid, basically what it says 01:09:39.880 --> 01:09:43.560 align:middle line:84% is that as you go to high temperature, the cluster 01:09:43.560 --> 01:09:45.220 align:middle line:84% of the molecules that essentially 01:09:45.220 --> 01:09:48.540 align:middle line:84% move past each other, the size is changing. 01:09:48.540 --> 01:09:50.300 align:middle line:84% As you go to high temperature, you 01:09:50.300 --> 01:09:52.120 align:middle line:90% tend to have smaller clusters. 01:09:52.120 --> 01:09:54.360 align:middle line:84% And smaller clusters are a bit more mobile. 01:09:54.360 --> 01:09:56.900 align:middle line:84% And that's why the activation energy barrier essentially 01:09:56.900 --> 01:09:58.540 align:middle line:90% becomes lower. 01:09:58.540 --> 01:10:02.078 align:middle line:84% The specific mechanism depends on the specific type of material 01:10:02.078 --> 01:10:03.120 align:middle line:90% we're talking about here. 01:10:03.120 --> 01:10:04.620 align:middle line:84% But the general trend is the same. 01:10:04.620 --> 01:10:06.060 align:middle line:84% As you go to high temperature, you 01:10:06.060 --> 01:10:07.900 align:middle line:84% have essentially smaller cluster, 01:10:07.900 --> 01:10:09.440 align:middle line:90% more flexible structures. 01:10:09.440 --> 01:10:11.220 align:middle line:84% And that's why you get the low viscosity. 01:10:11.220 --> 01:10:13.678 align:middle line:84% That's why you get this kind of strong and fragile behavior 01:10:13.678 --> 01:10:16.220 align:middle line:90% difference. 01:10:16.220 --> 01:10:19.980 align:middle line:84% To describe this kind of behavior quantitatively, 01:10:19.980 --> 01:10:22.803 align:middle line:84% there are a few equations people use here. 01:10:22.803 --> 01:10:24.720 align:middle line:84% For strong liquid, it's quite straightforward. 01:10:24.720 --> 01:10:26.200 align:middle line:84% You just use a Arrhenius equation. 01:10:26.200 --> 01:10:28.020 align:middle line:84% Maybe you add a little modification here 01:10:28.020 --> 01:10:30.760 align:middle line:84% so this delta EA is slightly temperature dependent. 01:10:30.760 --> 01:10:32.220 align:middle line:90% But that's about it. 01:10:32.220 --> 01:10:34.400 align:middle line:84% And as I've shown you in previous slides, 01:10:34.400 --> 01:10:39.660 align:middle line:84% this usually applies to covalent glass-forming liquids that 01:10:39.660 --> 01:10:42.590 align:middle line:84% has strong chemical bonds, like silicon or germanium. 01:10:42.590 --> 01:10:46.550 align:middle line:84% So the connection mechanism, or microscopic motion mechanism, 01:10:46.550 --> 01:10:48.390 align:middle line:84% doesn't change as a function of temperature. 01:10:48.390 --> 01:10:50.348 align:middle line:84% That's when you get this kind of strong liquid. 01:10:50.348 --> 01:10:53.738 align:middle line:90% 01:10:53.738 --> 01:10:56.030 align:middle line:84% Now, one thing I also want to just very quickly mention 01:10:56.030 --> 01:10:57.810 align:middle line:84% there is that the fragility, as I said, 01:10:57.810 --> 01:11:01.190 align:middle line:84% often exhibit a minimum at the rigidity percolation 01:11:01.190 --> 01:11:03.550 align:middle line:84% threshold, which, again highlights the fact that 01:11:03.550 --> 01:11:07.270 align:middle line:84% topology has very important impact on the glass property. 01:11:07.270 --> 01:11:09.910 align:middle line:90% OK, just a quick point there. 01:11:09.910 --> 01:11:10.410 align:middle line:90% All right. 01:11:10.410 --> 01:11:14.212 align:middle line:84% Now fragile liquid, so you cannot use Arrhenius equation. 01:11:14.212 --> 01:11:15.670 align:middle line:84% So what people do is they introduce 01:11:15.670 --> 01:11:18.350 align:middle line:90% a new set of equations here. 01:11:18.350 --> 01:11:20.470 align:middle line:84% And depending on who you talk to, 01:11:20.470 --> 01:11:23.710 align:middle line:84% this equation, same equation, has two different names. 01:11:23.710 --> 01:11:29.090 align:middle line:84% Glass scientists call it a vocal fracture equation, VFT equation. 01:11:29.090 --> 01:11:30.810 align:middle line:84% If you talk to a polymer scientist, 01:11:30.810 --> 01:11:32.185 align:middle line:84% they will tell you this is called 01:11:32.185 --> 01:11:34.430 align:middle line:90% Williams-Landel-Ferry equation. 01:11:34.430 --> 01:11:36.670 align:middle line:90% They are exactly the same. 01:11:36.670 --> 01:11:39.037 align:middle line:84% It just basically means that there are two communities. 01:11:39.037 --> 01:11:40.870 align:middle line:84% Historically, they don't talk to each other. 01:11:40.870 --> 01:11:42.570 align:middle line:84% They independently develop these equations 01:11:42.570 --> 01:11:44.070 align:middle line:84% without knowing that someone else is 01:11:44.070 --> 01:11:45.888 align:middle line:90% doing exactly the same job. 01:11:45.888 --> 01:11:47.430 align:middle line:84% But if you look at the equation here, 01:11:47.430 --> 01:11:49.850 align:middle line:84% it's a modification of the Arrhenius equation. 01:11:49.850 --> 01:11:52.970 align:middle line:84% You add, essentially, this additional term here, TVFT. 01:11:52.970 --> 01:11:56.290 align:middle line:90% 01:11:56.290 --> 01:11:58.250 align:middle line:84% Then that allows you, this additional parameter 01:11:58.250 --> 01:12:00.000 align:middle line:84% now allows you to capture this temperature 01:12:00.000 --> 01:12:05.130 align:middle line:84% dependence of the activation energy barrier. 01:12:05.130 --> 01:12:07.650 align:middle line:84% So this applies mostly to liquids 01:12:07.650 --> 01:12:09.570 align:middle line:84% with relatively weak atomic interactions 01:12:09.570 --> 01:12:11.530 align:middle line:84% like ionic liquid or molecular liquids, 01:12:11.530 --> 01:12:14.587 align:middle line:84% sometimes some glasses, like, if you have, say, 01:12:14.587 --> 01:12:17.170 align:middle line:84% a lot of network modifiers, you put a lot of network modifiers 01:12:17.170 --> 01:12:18.770 align:middle line:84% into glass network, they also tend 01:12:18.770 --> 01:12:21.890 align:middle line:84% to become a little more fragile when they go into liquid state. 01:12:21.890 --> 01:12:26.370 align:middle line:90% 01:12:26.370 --> 01:12:30.910 align:middle line:84% So if you look at this equation, it's actually obviously flawed. 01:12:30.910 --> 01:12:35.010 align:middle line:84% It's more of a semi-empirical equation. 01:12:35.010 --> 01:12:37.490 align:middle line:84% It's quite easy to see why it's flawed. 01:12:37.490 --> 01:12:40.390 align:middle line:84% If you go to low temperature, when the temperature starts 01:12:40.390 --> 01:12:43.310 align:middle line:84% to approach this VFT temperature there, 01:12:43.310 --> 01:12:48.830 align:middle line:84% the whole equation diverges, which is unphysical. 01:12:48.830 --> 01:12:50.910 align:middle line:84% So that's why for a long time, people have always 01:12:50.910 --> 01:12:55.230 align:middle line:84% been trying to look for a better equation to replace VF equation, 01:12:55.230 --> 01:12:58.430 align:middle line:84% and also to yield more insight into the microscopic mechanism 01:12:58.430 --> 01:13:00.430 align:middle line:90% of viscous flow. 01:13:00.430 --> 01:13:02.670 align:middle line:84% Well, that equation actually exists. 01:13:02.670 --> 01:13:06.430 align:middle line:84% So this was about 15 years ago, when a group of scientists 01:13:06.430 --> 01:13:08.470 align:middle line:84% actually developed this new equation here, 01:13:08.470 --> 01:13:11.190 align:middle line:84% derived from first principles, using statistical mechanics 01:13:11.190 --> 01:13:12.850 align:middle line:84% principles, which is quite remarkable. 01:13:12.850 --> 01:13:13.830 align:middle line:84% This is one of the few examples where 01:13:13.830 --> 01:13:15.653 align:middle line:84% you start from very basic first principle 01:13:15.653 --> 01:13:18.070 align:middle line:84% all the way to equation that can vary the precise view you 01:13:18.070 --> 01:13:19.790 align:middle line:90% can experiment. 01:13:19.790 --> 01:13:21.330 align:middle line:90% This equation is called a-- 01:13:21.330 --> 01:13:23.030 align:middle line:84% I'm not going to read all the names-- 01:13:23.030 --> 01:13:26.030 align:middle line:90% it's called MYEGA equation. 01:13:26.030 --> 01:13:27.162 align:middle line:90% Now you have five letters. 01:13:27.162 --> 01:13:29.870 align:middle line:84% This is showing the trend in the scientific community that author 01:13:29.870 --> 01:13:32.790 align:middle line:90% names is increasing over time. 01:13:32.790 --> 01:13:35.310 align:middle line:84% We'll see what happens in a few decades, how long we 01:13:35.310 --> 01:13:37.570 align:middle line:90% have to name equations. 01:13:37.570 --> 01:13:39.490 align:middle line:84% But what's remarkable about this equation 01:13:39.490 --> 01:13:41.930 align:middle line:84% in itself is that it only contains 01:13:41.930 --> 01:13:44.930 align:middle line:84% three fitting parameters, exactly the same as the VFT 01:13:44.930 --> 01:13:45.610 align:middle line:90% equation. 01:13:45.610 --> 01:13:48.550 align:middle line:84% If you look at VFT, you have this pre-commercial vector. 01:13:48.550 --> 01:13:51.530 align:middle line:84% You have this, basically, the activation energy barrier. 01:13:51.530 --> 01:13:55.250 align:middle line:84% And then you also get this VFT temperature, three parameters. 01:13:55.250 --> 01:13:58.330 align:middle line:84% This new equation also contains three parameters here, 01:13:58.330 --> 01:14:03.890 align:middle line:84% the pre-exponential factor, K, and C. And that's it. 01:14:03.890 --> 01:14:05.810 align:middle line:84% Now it's quite remarkable that they also 01:14:05.810 --> 01:14:07.030 align:middle line:90% get excellent agreement. 01:14:07.030 --> 01:14:09.650 align:middle line:84% They are trying to fit essentially this equation 01:14:09.650 --> 01:14:12.730 align:middle line:84% to the viscosity of different glass-forming liquids. 01:14:12.730 --> 01:14:14.655 align:middle line:84% And the solid line is basically what's 01:14:14.655 --> 01:14:16.030 align:middle line:84% predicted by this equation there. 01:14:16.030 --> 01:14:18.322 align:middle line:84% And the dotted line obviously is the experimental data. 01:14:18.322 --> 01:14:19.970 align:middle line:84% So you see remarkable agreement that 01:14:19.970 --> 01:14:22.770 align:middle line:84% can explain the liquid viscosity behavior 01:14:22.770 --> 01:14:26.450 align:middle line:90% across different fragility. 01:14:26.450 --> 01:14:31.130 align:middle line:84% So the specific derivation of this equation 01:14:31.130 --> 01:14:33.370 align:middle line:84% is obviously outside the scope of this class. 01:14:33.370 --> 01:14:36.960 align:middle line:84% But I have listed this very remarkable landmark paper here, 01:14:36.960 --> 01:14:37.877 align:middle line:90% if you're interested. 01:14:37.877 --> 01:14:38.960 align:middle line:90% You can take a look at it. 01:14:38.960 --> 01:14:41.660 align:middle line:90% 01:14:41.660 --> 01:14:43.040 align:middle line:90% So we are almost out of time. 01:14:43.040 --> 01:14:44.680 align:middle line:90% We still have three minutes. 01:14:44.680 --> 01:14:47.460 align:middle line:84% I guess I will just quickly talk about one last topic here. 01:14:47.460 --> 01:14:49.400 align:middle line:84% So we have discussed about this difference. 01:14:49.400 --> 01:14:50.380 align:middle line:90% We have fragile. 01:14:50.380 --> 01:14:52.100 align:middle line:90% You have strong liquid. 01:14:52.100 --> 01:14:59.980 align:middle line:84% Turns out there are actually some anomalous compounds 01:14:59.980 --> 01:15:02.820 align:middle line:84% that can actually switch between these two behaviors, 01:15:02.820 --> 01:15:05.720 align:middle line:84% with water being one of the examples. 01:15:05.720 --> 01:15:07.500 align:middle line:84% So the most common liquid actually 01:15:07.500 --> 01:15:10.140 align:middle line:84% exhibit an underlying liquid-to-liquid phase 01:15:10.140 --> 01:15:11.780 align:middle line:90% transformation. 01:15:11.780 --> 01:15:14.720 align:middle line:84% So if you look at, for example, let's see which one is water. 01:15:14.720 --> 01:15:15.820 align:middle line:90% So this one is water. 01:15:15.820 --> 01:15:20.820 align:middle line:84% So the open red circles, if you look at the viscosity behavior 01:15:20.820 --> 01:15:23.720 align:middle line:84% of water, initially it seems to be residing somewhere here, 01:15:23.720 --> 01:15:25.220 align:middle line:90% photo fragile behavior. 01:15:25.220 --> 01:15:27.180 align:middle line:84% But as you go to lower temperature, 01:15:27.180 --> 01:15:28.520 align:middle line:90% it starts to actually come up. 01:15:28.520 --> 01:15:32.680 align:middle line:84% It actually follows more like a strong liquid. 01:15:32.680 --> 01:15:35.820 align:middle line:84% So people are actually looking into this a little bit further. 01:15:35.820 --> 01:15:37.320 align:middle line:84% They were investigating essentially 01:15:37.320 --> 01:15:39.800 align:middle line:84% how the property actually changes upon this inflection 01:15:39.800 --> 01:15:40.420 align:middle line:90% point here. 01:15:40.420 --> 01:15:42.320 align:middle line:84% They noticed that a lot of the properties 01:15:42.320 --> 01:15:44.240 align:middle line:84% actually go to divergence when they 01:15:44.240 --> 01:15:45.840 align:middle line:90% approach this specific point. 01:15:45.840 --> 01:15:49.760 align:middle line:84% This is a clear signature of a high-order phase transition. 01:15:49.760 --> 01:15:51.960 align:middle line:84% A lot of the properties should diverge. 01:15:51.960 --> 01:15:54.880 align:middle line:84% So without going into too much detail, 01:15:54.880 --> 01:15:57.360 align:middle line:84% suffice it to say that some liquids actually exhibit 01:15:57.360 --> 01:15:59.742 align:middle line:90% a liquid phase transformation. 01:15:59.742 --> 01:16:01.700 align:middle line:84% In the liquid phase, at different temperatures, 01:16:01.700 --> 01:16:06.200 align:middle line:84% they can switch from a fragile liquid to a to a strong liquid. 01:16:06.200 --> 01:16:08.700 align:middle line:84% It's a high-order phase transition. 01:16:08.700 --> 01:16:12.160 align:middle line:84% So you don't see an abrupt jump of extensive thermodynamic 01:16:12.160 --> 01:16:12.900 align:middle line:90% parameters. 01:16:12.900 --> 01:16:16.280 align:middle line:84% But you do see the divergence for some of this, 01:16:16.280 --> 01:16:17.680 align:middle line:90% for these parameters there. 01:16:17.680 --> 01:16:23.445 align:middle line:84% And besides water, the other example 01:16:23.445 --> 01:16:25.320 align:middle line:84% where you have this kind of transition there, 01:16:25.320 --> 01:16:28.048 align:middle line:90% is this phase change material. 01:16:28.048 --> 01:16:30.340 align:middle line:84% Germanium telluride is actually one of the phase change 01:16:30.340 --> 01:16:31.460 align:middle line:90% materials. 01:16:31.460 --> 01:16:33.300 align:middle line:84% You see this very clear signature. 01:16:33.300 --> 01:16:34.820 align:middle line:84% You see this inflection point here 01:16:34.820 --> 01:16:37.740 align:middle line:84% that indicates a fragile two-stroke liquid 01:16:37.740 --> 01:16:38.660 align:middle line:90% transformation. 01:16:38.660 --> 01:16:40.660 align:middle line:84% And that's actually quite important practically, 01:16:40.660 --> 01:16:42.703 align:middle line:84% it turns out, because now, remember, 01:16:42.703 --> 01:16:45.120 align:middle line:84% we were talking about these kind of phase-change memories. 01:16:45.120 --> 01:16:47.100 align:middle line:84% We said that they can switch extremely fast 01:16:47.100 --> 01:16:48.300 align:middle line:90% at high temperature. 01:16:48.300 --> 01:16:51.680 align:middle line:84% But the data can be kept for decades at room temperature. 01:16:51.680 --> 01:16:52.180 align:middle line:90% Why? 01:16:52.180 --> 01:16:54.140 align:middle line:84% Because of the high activation energy. 01:16:54.140 --> 01:16:57.520 align:middle line:84% I said that activation should have a quotation mark there, 01:16:57.520 --> 01:17:00.060 align:middle line:84% because what I'm really referring to is 01:17:00.060 --> 01:17:01.887 align:middle line:90% this part, the strong part. 01:17:01.887 --> 01:17:04.220 align:middle line:84% The fragile part, obviously you cannot describe it using 01:17:04.220 --> 01:17:05.240 align:middle line:90% Arrhenius equation. 01:17:05.240 --> 01:17:07.980 align:middle line:84% So it doesn't make sense to describe it using essentially 01:17:07.980 --> 01:17:09.080 align:middle line:90% activation energy barrier. 01:17:09.080 --> 01:17:13.300 align:middle line:84% But for the strong part, it does exhibit very high activation 01:17:13.300 --> 01:17:13.840 align:middle line:90% energy. 01:17:13.840 --> 01:17:15.860 align:middle line:90% You have a very steep slope. 01:17:15.860 --> 01:17:17.740 align:middle line:84% So that's why you can have essentially 01:17:17.740 --> 01:17:19.560 align:middle line:90% a fast crystallization. 01:17:19.560 --> 01:17:22.347 align:middle line:90% 01:17:22.347 --> 01:17:24.180 align:middle line:84% That's why you can have fast crystallization 01:17:24.180 --> 01:17:27.220 align:middle line:84% at high temperature and also, essentially, good data 01:17:27.220 --> 01:17:29.360 align:middle line:90% retention at room temperature. 01:17:29.360 --> 01:17:33.360 align:middle line:90% Also, it has a fragile behavior. 01:17:33.360 --> 01:17:35.240 align:middle line:84% This is useful at high temperature 01:17:35.240 --> 01:17:38.080 align:middle line:84% because we want to promote viscous flow to ensure you have 01:17:38.080 --> 01:17:39.820 align:middle line:90% high crystallization speed. 01:17:39.820 --> 01:17:43.680 align:middle line:84% But you don't want the fragile behavior at room temperature, 01:17:43.680 --> 01:17:46.560 align:middle line:84% because it turns out you can correlate fragile behavior 01:17:46.560 --> 01:17:49.000 align:middle line:90% into structural relaxation. 01:17:49.000 --> 01:17:51.760 align:middle line:84% A fragile liquid tend to have more structural relaxation 01:17:51.760 --> 01:17:53.200 align:middle line:90% at low temperature. 01:17:53.200 --> 01:17:55.320 align:middle line:84% For memories, that's problematic because it 01:17:55.320 --> 01:17:57.600 align:middle line:90% leads to essentially a drift. 01:17:57.600 --> 01:17:59.220 align:middle line:90% You switch to some state. 01:17:59.220 --> 01:18:01.360 align:middle line:84% But its conductivity actually changes over time. 01:18:01.360 --> 01:18:03.240 align:middle line:84% So at some point, you lose the ability 01:18:03.240 --> 01:18:06.480 align:middle line:84% to tell the 1 bits from the 0's bits. 01:18:06.480 --> 01:18:09.080 align:middle line:84% So the fact that it's transitioned 01:18:09.080 --> 01:18:12.160 align:middle line:84% into a strong liquid at low temperature 01:18:12.160 --> 01:18:14.240 align:middle line:84% means that this phase change material can also 01:18:14.240 --> 01:18:16.840 align:middle line:84% have very good data retention without drift. 01:18:16.840 --> 01:18:20.180 align:middle line:84% So anyway, that's the basic idea there 01:18:20.180 --> 01:18:22.953 align:middle line:84% why this kind of transition can be useful. 01:18:22.953 --> 01:18:24.620 align:middle line:84% I have also included the reference here, 01:18:24.620 --> 01:18:26.400 align:middle line:84% if you want to look for more information specifically 01:18:26.400 --> 01:18:28.890 align:middle line:84% about phase change materials and other liquids that 01:18:28.890 --> 01:18:31.170 align:middle line:90% exhibit this peculiar behavior. 01:18:31.170 --> 01:18:32.290 align:middle line:90% All right. 01:18:32.290 --> 01:18:33.510 align:middle line:90% We are right on time here. 01:18:33.510 --> 01:18:35.410 align:middle line:90% Any last minute questions? 01:18:35.410 --> 01:18:39.090 align:middle line:90% 01:18:39.090 --> 01:18:40.410 align:middle line:90% All good? 01:18:40.410 --> 01:18:41.030 align:middle line:90% All right. 01:18:41.030 --> 01:18:43.520 align:middle line:90% Thank you, and see you tomorrow. 01:18:43.520 --> 01:19:26.000 align:middle line:90%