WEBVTT 00:00:00.000 --> 00:00:02.460 align:middle line:90% [SQUEAKING] 00:00:02.460 --> 00:00:04.420 align:middle line:90% [RUSTLING] 00:00:04.420 --> 00:00:08.340 align:middle line:90% [CLICKING] 00:00:08.340 --> 00:00:16.865 align:middle line:90% 00:00:16.865 --> 00:00:17.740 align:middle line:90% JUEJUN HU: All right. 00:00:17.740 --> 00:00:20.840 align:middle line:90% So let's get started. 00:00:20.840 --> 00:00:22.720 align:middle line:84% So in this lecture, we are going to talk 00:00:22.720 --> 00:00:24.760 align:middle line:84% about, essentially, optical waveguides 00:00:24.760 --> 00:00:29.120 align:middle line:84% and why glass has such unique properties that 00:00:29.120 --> 00:00:31.200 align:middle line:84% make them so useful for waveguide applications. 00:00:31.200 --> 00:00:32.439 align:middle line:84% Now, speaking of waveguides, we'll 00:00:32.439 --> 00:00:34.689 align:middle line:84% be talking about basically what we call the integrated 00:00:34.689 --> 00:00:35.960 align:middle line:90% photonics. 00:00:35.960 --> 00:00:38.080 align:middle line:84% By way of introduction, I always like 00:00:38.080 --> 00:00:41.720 align:middle line:84% to draw an analogy with it's a close analog electronic 00:00:41.720 --> 00:00:43.700 align:middle line:84% integrated circuits or integrated electronics. 00:00:43.700 --> 00:00:46.920 align:middle line:84% So you all know that you have this thing called 00:00:46.920 --> 00:00:48.720 align:middle line:84% Moore's law, which basically says 00:00:48.720 --> 00:00:52.400 align:middle line:84% that, essentially, the function of electronic circuits 00:00:52.400 --> 00:00:54.680 align:middle line:84% would grow exponentially over time. 00:00:54.680 --> 00:00:56.740 align:middle line:84% And this applies not just to CPUs. 00:00:56.740 --> 00:00:58.560 align:middle line:84% This also applies to storage units, 00:00:58.560 --> 00:01:00.320 align:middle line:84% to other electronic components that 00:01:00.320 --> 00:01:03.840 align:middle line:84% similarly leverage the electronic CMOS microfabrication 00:01:03.840 --> 00:01:04.420 align:middle line:90% techniques. 00:01:04.420 --> 00:01:08.360 align:middle line:84% So just to show you the power of this kind of exponential scaling 00:01:08.360 --> 00:01:11.200 align:middle line:84% law, I always like to present this picture here. 00:01:11.200 --> 00:01:14.120 align:middle line:84% You can make a guess about what is this big piece of equipment 00:01:14.120 --> 00:01:16.880 align:middle line:84% that they are trying to carry onto the aircraft. 00:01:16.880 --> 00:01:19.800 align:middle line:90% AUDIENCE: A GPS? 00:01:19.800 --> 00:01:23.740 align:middle line:84% JUEJUN HU: It's a hard drive with a 5-megabyte capacity. 00:01:23.740 --> 00:01:25.920 align:middle line:90% This is back in 1956. 00:01:25.920 --> 00:01:29.640 align:middle line:84% So you can see, with exponential scaling, 00:01:29.640 --> 00:01:32.040 align:middle line:84% how the capacity of a hard drive has actually 00:01:32.040 --> 00:01:34.880 align:middle line:90% evolved towards today. 00:01:34.880 --> 00:01:37.440 align:middle line:84% What I'm trying to argue here is that people 00:01:37.440 --> 00:01:40.520 align:middle line:84% believe that the same kind or similar kind of paradigm shift 00:01:40.520 --> 00:01:42.980 align:middle line:84% is going to take place for optics as well. 00:01:42.980 --> 00:01:44.480 align:middle line:84% What I'm showing you in this picture 00:01:44.480 --> 00:01:47.000 align:middle line:84% here is what an optics lab traditionally 00:01:47.000 --> 00:01:48.478 align:middle line:90% would be looking like. 00:01:48.478 --> 00:01:51.020 align:middle line:84% I don't know if you've ever been to any of these optics labs, 00:01:51.020 --> 00:01:52.600 align:middle line:84% but it's going to be a dark room. 00:01:52.600 --> 00:01:55.340 align:middle line:84% You have a bunch of discrete elements. 00:01:55.340 --> 00:01:58.250 align:middle line:84% You have lenses, prisms, gratings. 00:01:58.250 --> 00:02:01.690 align:middle line:84% One of the big part of the job for a PhD student 00:02:01.690 --> 00:02:04.370 align:middle line:84% specializing in optics is to do alignment. 00:02:04.370 --> 00:02:06.676 align:middle line:84% You have to carefully align all the pieces together 00:02:06.676 --> 00:02:08.509 align:middle line:84% so that you can actually do your experiment. 00:02:08.509 --> 00:02:11.178 align:middle line:84% And depending on how complicated the experiment is, 00:02:11.178 --> 00:02:12.970 align:middle line:84% the alignment may be the step that actually 00:02:12.970 --> 00:02:14.610 align:middle line:90% takes the most of the time. 00:02:14.610 --> 00:02:16.850 align:middle line:84% So you can imagine that this is something that's 00:02:16.850 --> 00:02:19.038 align:middle line:90% very complicated, not robust. 00:02:19.038 --> 00:02:21.330 align:middle line:84% It's also very expensive because there's a large number 00:02:21.330 --> 00:02:23.130 align:middle line:90% of elements involved. 00:02:23.130 --> 00:02:28.150 align:middle line:84% So what people envision is that they can essentially leverage, 00:02:28.150 --> 00:02:30.210 align:middle line:84% similarly, the microfabrication techniques, 00:02:30.210 --> 00:02:34.770 align:middle line:84% try to essentially create microscale counterparts 00:02:34.770 --> 00:02:38.830 align:middle line:84% of the discrete elements using microfabrication methods, 00:02:38.830 --> 00:02:42.480 align:middle line:84% and therefore put all of them together on a common substrate 00:02:42.480 --> 00:02:44.730 align:middle line:84% to create what are called photonic integrated circuits 00:02:44.730 --> 00:02:49.050 align:middle line:90% or integrated photonics. 00:02:49.050 --> 00:02:53.730 align:middle line:84% So the value of this kind of technology 00:02:53.730 --> 00:02:55.490 align:middle line:90% is pretty clear there. 00:02:55.490 --> 00:03:01.410 align:middle line:84% One, of course, as I mentioned, in terms of cost, 00:03:01.410 --> 00:03:04.170 align:middle line:84% it costs much less to actually fabricate 00:03:04.170 --> 00:03:08.530 align:middle line:84% a chip compared to having all the large assembly 00:03:08.530 --> 00:03:10.150 align:middle line:90% of discrete elements. 00:03:10.150 --> 00:03:13.490 align:middle line:84% What's also more important is about scalability. 00:03:13.490 --> 00:03:14.907 align:middle line:84% As I mentioned, optical alignment 00:03:14.907 --> 00:03:16.490 align:middle line:84% is a big challenge when you're dealing 00:03:16.490 --> 00:03:19.130 align:middle line:90% with discrete optical systems. 00:03:19.130 --> 00:03:23.050 align:middle line:84% But in photonic integrated circuits, all the elements, 00:03:23.050 --> 00:03:25.410 align:middle line:84% essentially, they are precisely aligned together 00:03:25.410 --> 00:03:28.170 align:middle line:84% with lithographic precision, because everything is defined 00:03:28.170 --> 00:03:29.530 align:middle line:90% basically by lithography. 00:03:29.530 --> 00:03:32.130 align:middle line:84% So that way, you can actually create much more scalable 00:03:32.130 --> 00:03:33.723 align:middle line:90% optical systems. 00:03:33.723 --> 00:03:35.890 align:middle line:84% The small footprint also means that you can actually 00:03:35.890 --> 00:03:38.370 align:middle line:84% deploy them for many different applications that previously 00:03:38.370 --> 00:03:40.650 align:middle line:90% is not possible. 00:03:40.650 --> 00:03:45.270 align:middle line:84% It can range from, for example, point-of-care diagnosis, 00:03:45.270 --> 00:03:48.330 align:middle line:84% if you're using them for sensing, to essentially all type 00:03:48.330 --> 00:03:50.370 align:middle line:90% of, for example, sensors. 00:03:50.370 --> 00:03:57.340 align:middle line:84% You can deploy onto vehicles, robotics, UAVs, whatever. 00:03:57.340 --> 00:03:59.300 align:middle line:90% Last part, power. 00:03:59.300 --> 00:04:01.180 align:middle line:84% Turns out that the power that's consumed 00:04:01.180 --> 00:04:03.760 align:middle line:84% by a lot of these devices actually scales with the volume, 00:04:03.760 --> 00:04:04.740 align:middle line:90% in lots of cases. 00:04:04.740 --> 00:04:06.772 align:middle line:84% So by making them small, you also 00:04:06.772 --> 00:04:08.980 align:middle line:84% make them a lot more power efficient at the same time 00:04:08.980 --> 00:04:09.780 align:middle line:90% as well. 00:04:09.780 --> 00:04:11.420 align:middle line:90% OK. 00:04:11.420 --> 00:04:11.960 align:middle line:90% All right. 00:04:11.960 --> 00:04:14.160 align:middle line:84% So this is actually not a new concept. 00:04:14.160 --> 00:04:16.300 align:middle line:84% And what you will see is that glasses actually 00:04:16.300 --> 00:04:19.380 align:middle line:84% play a very important role from the very beginning of the idea 00:04:19.380 --> 00:04:20.820 align:middle line:90% of integrated photonics. 00:04:20.820 --> 00:04:23.140 align:middle line:84% So what I'm showing you here, this is actually-- 00:04:23.140 --> 00:04:28.840 align:middle line:84% I'm quoting-- this is a paper by Stewart Miller back in 1969, 00:04:28.840 --> 00:04:30.700 align:middle line:84% where the concept of integrated optics 00:04:30.700 --> 00:04:32.500 align:middle line:90% was actually first proposed. 00:04:32.500 --> 00:04:34.700 align:middle line:84% What he has proposed is that you can actually 00:04:34.700 --> 00:04:37.240 align:middle line:84% combine these different elements here, 00:04:37.240 --> 00:04:39.900 align:middle line:84% like laser cavities, switch, detectors, 00:04:39.900 --> 00:04:42.860 align:middle line:90% on one common glass plate. 00:04:42.860 --> 00:04:44.360 align:middle line:84% And then these different components, 00:04:44.360 --> 00:04:46.620 align:middle line:84% they are linked together by essentially glass 00:04:46.620 --> 00:04:49.020 align:middle line:90% optical waveguides. 00:04:49.020 --> 00:04:52.800 align:middle line:84% So that's what I'm trying to argue here, 00:04:52.800 --> 00:04:54.357 align:middle line:84% that glass materials is basically 00:04:54.357 --> 00:04:56.940 align:middle line:84% the first material people have actually investigated to create 00:04:56.940 --> 00:04:58.960 align:middle line:84% this kind of integrated optics platforms, 00:04:58.960 --> 00:05:02.625 align:middle line:84% both as a substrate material and also leveraging its transparency 00:05:02.625 --> 00:05:04.500 align:middle line:84% for essentially a low-loss light transmission 00:05:04.500 --> 00:05:06.020 align:middle line:84% between the different components. 00:05:06.020 --> 00:05:08.768 align:middle line:90% 00:05:08.768 --> 00:05:12.580 align:middle line:84% Now, more than 50 years later, we 00:05:12.580 --> 00:05:16.740 align:middle line:84% are obviously looking at much more advanced and complex 00:05:16.740 --> 00:05:19.060 align:middle line:90% integrated systems. 00:05:19.060 --> 00:05:22.700 align:middle line:84% But despite that, I would want to argue that glass materials 00:05:22.700 --> 00:05:26.100 align:middle line:84% still play a pivotal role in state-of-the-art integrated 00:05:26.100 --> 00:05:26.920 align:middle line:90% photonics systems. 00:05:26.920 --> 00:05:32.940 align:middle line:84% So for a typical waveguide, this is the configuration 00:05:32.940 --> 00:05:34.420 align:middle line:90% we are talking about here. 00:05:34.420 --> 00:05:36.500 align:middle line:84% You essentially have a substrate, 00:05:36.500 --> 00:05:39.580 align:middle line:84% and then you grow a low-index material 00:05:39.580 --> 00:05:41.180 align:middle line:90% that serves as a cladding. 00:05:41.180 --> 00:05:45.700 align:middle line:84% You then structure a high-index core on top of substrate here. 00:05:45.700 --> 00:05:48.380 align:middle line:84% So this gives rise to many different possible variations 00:05:48.380 --> 00:05:50.750 align:middle line:84% of materials that can be used for these kind 00:05:50.750 --> 00:05:52.110 align:middle line:90% of waveguide systems. 00:05:52.110 --> 00:05:54.850 align:middle line:84% For substrate, the commonly used substrate platforms, 00:05:54.850 --> 00:05:57.630 align:middle line:84% you can use crystalline silicon or III-V semiconductor, 00:05:57.630 --> 00:06:00.550 align:middle line:84% but amorphous substrates, such as glass and polymer, 00:06:00.550 --> 00:06:03.390 align:middle line:84% are often being used for various applications. 00:06:03.390 --> 00:06:05.950 align:middle line:84% They offer the advantage that they 00:06:05.950 --> 00:06:08.790 align:middle line:84% tend to be a lower cost compared to their single crystalline 00:06:08.790 --> 00:06:09.570 align:middle line:90% counterpart. 00:06:09.570 --> 00:06:12.190 align:middle line:84% Also, if you're using polymers, they also 00:06:12.190 --> 00:06:16.710 align:middle line:84% offer additional properties, such as mechanical flexibility 00:06:16.710 --> 00:06:18.950 align:middle line:90% as well as biocompatibility. 00:06:18.950 --> 00:06:22.590 align:middle line:84% The waveguide core materials, I'm 00:06:22.590 --> 00:06:24.830 align:middle line:84% again listing some of the common materials here. 00:06:24.830 --> 00:06:29.230 align:middle line:84% Besides crystalline silicon, a lot of amorphous materials 00:06:29.230 --> 00:06:32.550 align:middle line:84% are considered viable waveguide materials. 00:06:32.550 --> 00:06:36.350 align:middle line:84% These include amorphous silicon, silica glass, silicon nitride, 00:06:36.350 --> 00:06:38.590 align:middle line:90% as well as polymers. 00:06:38.590 --> 00:06:43.330 align:middle line:84% So again, compared to a crystalline material, 00:06:43.330 --> 00:06:46.070 align:middle line:84% the main advantage people want to use this material 00:06:46.070 --> 00:06:48.850 align:middle line:90% for waveguides-- a few things. 00:06:48.850 --> 00:06:52.070 align:middle line:84% One is that they are a lot more easier to process. 00:06:52.070 --> 00:06:54.110 align:middle line:84% If it's a single crystalline material, 00:06:54.110 --> 00:06:57.830 align:middle line:84% you need to grow them either through a bulk single crystal 00:06:57.830 --> 00:07:00.030 align:middle line:84% growth process, or you can grow them 00:07:00.030 --> 00:07:02.630 align:middle line:84% in thin-film form on a lattice-matched substrate 00:07:02.630 --> 00:07:04.190 align:middle line:90% through epitaxy. 00:07:04.190 --> 00:07:07.150 align:middle line:84% This is actually a pretty severe constraint 00:07:07.150 --> 00:07:08.945 align:middle line:84% when it comes to integrating devices. 00:07:08.945 --> 00:07:10.570 align:middle line:84% You want to be able to have waveguides, 00:07:10.570 --> 00:07:14.270 align:middle line:84% for example, in different layers of photonic integrated circuits. 00:07:14.270 --> 00:07:16.390 align:middle line:84% But if you use crystalline material, 00:07:16.390 --> 00:07:19.470 align:middle line:84% you basically restrict to either you 00:07:19.470 --> 00:07:25.470 align:middle line:84% have to use a substrate with a lattice constant 00:07:25.470 --> 00:07:26.910 align:middle line:90% matching your material. 00:07:26.910 --> 00:07:30.855 align:middle line:84% Or you can use wafer bonding to bond essentially 00:07:30.855 --> 00:07:33.230 align:middle line:84% a single crystalline material onto a different substrate. 00:07:33.230 --> 00:07:35.770 align:middle line:84% But again, that's also a relatively complicated process. 00:07:35.770 --> 00:07:39.830 align:middle line:84% And you can usually only do that for one time. 00:07:39.830 --> 00:07:41.490 align:middle line:84% Amorphous material, on the other hand, 00:07:41.490 --> 00:07:43.830 align:middle line:84% you can basically deposit it on practically any kind 00:07:43.830 --> 00:07:45.710 align:middle line:84% of substrate without having to worry 00:07:45.710 --> 00:07:47.390 align:middle line:84% about lattice-matching constraints. 00:07:47.390 --> 00:07:51.210 align:middle line:84% So that gives rise to a lot of process flexibility. 00:07:51.210 --> 00:07:53.270 align:middle line:84% You can essentially place those waveguides 00:07:53.270 --> 00:07:55.557 align:middle line:90% at different layers in a stack. 00:07:55.557 --> 00:07:56.890 align:middle line:90% You can put it on semiconductor. 00:07:56.890 --> 00:07:58.530 align:middle line:84% You can put it on dielectric materials. 00:07:58.530 --> 00:08:00.790 align:middle line:84% So it's really making integration 00:08:00.790 --> 00:08:04.490 align:middle line:90% a lot more convenient. 00:08:04.490 --> 00:08:08.430 align:middle line:84% So now, before I go ahead and talk about how you actually 00:08:08.430 --> 00:08:11.390 align:middle line:84% fabricate waveguides out of glass materials, 00:08:11.390 --> 00:08:13.670 align:middle line:84% I do want to highlight some important differences 00:08:13.670 --> 00:08:16.970 align:middle line:84% between optical fibers, which we discussed in the last lecture, 00:08:16.970 --> 00:08:19.690 align:middle line:84% as on-chip waveguides or planar waveguides. 00:08:19.690 --> 00:08:21.903 align:middle line:84% One of the big thing that's different 00:08:21.903 --> 00:08:24.070 align:middle line:84% here is essentially loss mechanism and the magnitude 00:08:24.070 --> 00:08:25.470 align:middle line:90% of optical loss. 00:08:25.470 --> 00:08:27.010 align:middle line:84% So I mentioned about optical fibers. 00:08:27.010 --> 00:08:30.510 align:middle line:84% You can get your loss down to 0.2 dB per kilometer in silicon 00:08:30.510 --> 00:08:32.850 align:middle line:90% fiber at the 1,550 nanometer. 00:08:32.850 --> 00:08:34.669 align:middle line:90% Very, very low loss. 00:08:34.669 --> 00:08:37.470 align:middle line:84% Now, in waveguides on a chip, the loss 00:08:37.470 --> 00:08:39.870 align:middle line:90% tends to be a lot higher. 00:08:39.870 --> 00:08:41.669 align:middle line:84% The typical loss we are talking about here 00:08:41.669 --> 00:08:46.400 align:middle line:84% is of the order of dB per centimeter. 00:08:46.400 --> 00:08:50.917 align:middle line:84% So a million times higher compared to optical fiber there. 00:08:50.917 --> 00:08:52.500 align:middle line:84% I mean, there are some recent reports. 00:08:52.500 --> 00:08:54.760 align:middle line:84% People are essentially demonstrating very low-loss 00:08:54.760 --> 00:09:01.480 align:middle line:84% waveguides with loss of the order of 0.0-something dB per 00:09:01.480 --> 00:09:02.440 align:middle line:90% meter. 00:09:02.440 --> 00:09:05.280 align:middle line:90% Still, it's higher than fibers. 00:09:05.280 --> 00:09:08.000 align:middle line:84% But usually those require very special process. 00:09:08.000 --> 00:09:11.080 align:middle line:84% So bottom line, the losses tend to be a lot higher 00:09:11.080 --> 00:09:13.680 align:middle line:90% in on-chip waveguides. 00:09:13.680 --> 00:09:18.280 align:middle line:84% This is usually not considered as a showstopper 00:09:18.280 --> 00:09:21.440 align:middle line:84% because, for fiber, we have to transmit signals essentially 00:09:21.440 --> 00:09:24.260 align:middle line:84% over very long distances, tens of hundreds of kilometers. 00:09:24.260 --> 00:09:26.980 align:middle line:84% But for a lot of the on-chip applications, 00:09:26.980 --> 00:09:29.680 align:middle line:84% you really only need to send signal 00:09:29.680 --> 00:09:33.000 align:middle line:84% across essentially a chip, which is like a centimeter across. 00:09:33.000 --> 00:09:35.572 align:middle line:84% So this level of loss is considered acceptable. 00:09:35.572 --> 00:09:37.280 align:middle line:84% I mean, it's not great, but it's actually 00:09:37.280 --> 00:09:40.040 align:middle line:84% acceptable for many applications that involve 00:09:40.040 --> 00:09:42.240 align:middle line:90% on-chip integrated photonics. 00:09:42.240 --> 00:09:44.700 align:middle line:84% Now, the reason why on-chip waveguide has such a high loss, 00:09:44.700 --> 00:09:47.340 align:middle line:84% it goes back to essentially the loss mechanisms here. 00:09:47.340 --> 00:09:51.580 align:middle line:84% So there are essentially two kinds of major loss mechanisms. 00:09:51.580 --> 00:09:53.960 align:middle line:84% You have loss coming from the material itself, material 00:09:53.960 --> 00:09:58.880 align:middle line:84% attenuation, and then also coming from scattering, Now, 00:09:58.880 --> 00:10:02.480 align:middle line:84% for scattering, a lot comes from, essentially, 00:10:02.480 --> 00:10:04.040 align:middle line:90% any kind of surface roughness. 00:10:04.040 --> 00:10:06.735 align:middle line:84% So when you prepare a fiber or when you prepare a waveguide, 00:10:06.735 --> 00:10:08.860 align:middle line:84% you're not going to get a perfectly smooth surface. 00:10:08.860 --> 00:10:11.280 align:middle line:84% There's always some interface or surface roughness 00:10:11.280 --> 00:10:13.280 align:middle line:84% that can then actually divert light propagation, 00:10:13.280 --> 00:10:14.840 align:middle line:90% leading to scattering loss. 00:10:14.840 --> 00:10:18.080 align:middle line:84% So a simple equation I'm listing here essentially 00:10:18.080 --> 00:10:22.800 align:middle line:84% writes this roughness scattering loss, alpha s. 00:10:22.800 --> 00:10:26.960 align:middle line:84% It should be scales with the index difference 00:10:26.960 --> 00:10:31.120 align:middle line:84% between the core and the cladding squared and also 00:10:31.120 --> 00:10:33.940 align:middle line:90% a square of the RMS roughness. 00:10:33.940 --> 00:10:36.800 align:middle line:84% So basically, if you have a waveguide that 00:10:36.800 --> 00:10:39.240 align:middle line:84% has a large surface roughness, then the loss 00:10:39.240 --> 00:10:42.130 align:middle line:90% increases quadratically. 00:10:42.130 --> 00:10:45.930 align:middle line:84% Also, if a waveguide has a core and cladding material 00:10:45.930 --> 00:10:48.588 align:middle line:84% with large index differences, then effectively, you 00:10:48.588 --> 00:10:50.130 align:middle line:84% have a much larger index perturbation 00:10:50.130 --> 00:10:52.088 align:middle line:84% for the same roughness, and therefore, you also 00:10:52.088 --> 00:10:54.690 align:middle line:84% have a much higher scattering loss. 00:10:54.690 --> 00:10:57.130 align:middle line:84% So now we can take a look at why optical fiber can 00:10:57.130 --> 00:10:58.650 align:middle line:90% have such a low loss. 00:10:58.650 --> 00:11:02.450 align:middle line:84% So for optical fibers, normal optical fiber, the core 00:11:02.450 --> 00:11:05.013 align:middle line:84% and cladding typically has a very small index difference. 00:11:05.013 --> 00:11:06.930 align:middle line:84% Because the core and cladding index difference 00:11:06.930 --> 00:11:08.763 align:middle line:84% is generated by essentially doping the glass 00:11:08.763 --> 00:11:11.330 align:middle line:84% core or the cladding with some dopant materials, 00:11:11.330 --> 00:11:12.810 align:middle line:90% like germanium oxide. 00:11:12.810 --> 00:11:14.990 align:middle line:84% So typically, the index difference 00:11:14.990 --> 00:11:16.770 align:middle line:84% we're talking about for fibers is 00:11:16.770 --> 00:11:21.090 align:middle line:90% of the order of 0.01 or 0.001. 00:11:21.090 --> 00:11:23.010 align:middle line:84% So very, very small index difference. 00:11:23.010 --> 00:11:26.970 align:middle line:84% So in that regard, it's inherently less sensitive 00:11:26.970 --> 00:11:28.410 align:middle line:90% to any kind of roughness. 00:11:28.410 --> 00:11:29.730 align:middle line:90% That's the first thing. 00:11:29.730 --> 00:11:32.450 align:middle line:84% Second thing is if you think about the fiber manufacturing 00:11:32.450 --> 00:11:35.330 align:middle line:84% process that we showed in the last lecture, 00:11:35.330 --> 00:11:37.178 align:middle line:84% you see that we essentially have a preform. 00:11:37.178 --> 00:11:38.970 align:middle line:84% You have to heat it up to basically turn it 00:11:38.970 --> 00:11:41.310 align:middle line:84% into a viscous liquid, more or less. 00:11:41.310 --> 00:11:43.810 align:middle line:84% And then you apply axial stress to deform it 00:11:43.810 --> 00:11:45.250 align:middle line:90% into an optical fiber. 00:11:45.250 --> 00:11:48.170 align:middle line:84% So in this whole process, because it's 00:11:48.170 --> 00:11:51.010 align:middle line:84% sort of like a viscous liquid, surface tension 00:11:51.010 --> 00:11:53.650 align:middle line:84% actually plays a very big role in dictating the surface 00:11:53.650 --> 00:11:54.770 align:middle line:90% morphology. 00:11:54.770 --> 00:11:57.130 align:middle line:84% Now, surface tension will always try 00:11:57.130 --> 00:12:00.570 align:middle line:84% to smooth the surface, therefore minimizing surface area 00:12:00.570 --> 00:12:03.290 align:middle line:84% and lowering the Gibbs free energy of the overall system. 00:12:03.290 --> 00:12:06.830 align:middle line:84% So because of presence of surface tension, 00:12:06.830 --> 00:12:09.570 align:middle line:84% it means that the roughness, residual roughness, 00:12:09.570 --> 00:12:13.090 align:middle line:84% on the fiber surface is minuscule. 00:12:13.090 --> 00:12:15.090 align:middle line:84% So these two factors combine together. 00:12:15.090 --> 00:12:16.990 align:middle line:84% One, it has low surface roughness. 00:12:16.990 --> 00:12:18.950 align:middle line:84% Two, it also has very low index contrast. 00:12:18.950 --> 00:12:22.210 align:middle line:84% And that's why, in a typical optical fiber, 00:12:22.210 --> 00:12:23.890 align:middle line:84% this kind of surface scattering loss 00:12:23.890 --> 00:12:26.703 align:middle line:84% is usually negligible compared to material loss. 00:12:26.703 --> 00:12:29.370 align:middle line:84% And that's why we are getting to essentially material loss limit 00:12:29.370 --> 00:12:32.510 align:middle line:84% down to 0.2 dB per kilometer for silica, for example. 00:12:32.510 --> 00:12:40.410 align:middle line:90% 00:12:40.410 --> 00:12:43.170 align:middle line:84% Now, however, there's actually one case 00:12:43.170 --> 00:12:46.410 align:middle line:84% where the surface roughness loss can become significant 00:12:46.410 --> 00:12:48.770 align:middle line:90% for optical fibers-- 00:12:48.770 --> 00:12:51.470 align:middle line:84% the microstructured optical fibers. 00:12:51.470 --> 00:12:52.970 align:middle line:84% You may remember in the last lecture 00:12:52.970 --> 00:12:54.550 align:middle line:84% when we talk about microstructured fibers, 00:12:54.550 --> 00:12:56.170 align:middle line:84% these can actually be a hollow core. 00:12:56.170 --> 00:12:57.710 align:middle line:90% Essentially, the core is hollow. 00:12:57.710 --> 00:12:59.850 align:middle line:84% You have this kind of holes arranged 00:12:59.850 --> 00:13:01.830 align:middle line:84% in hexagonal or whatever lattice to create 00:13:01.830 --> 00:13:03.330 align:middle line:84% essentially photonic bandgap effect, 00:13:03.330 --> 00:13:04.790 align:middle line:84% to trap light in the hollow core. 00:13:04.790 --> 00:13:08.072 align:middle line:84% Now the difference here is that for the hollow core, 00:13:08.072 --> 00:13:10.530 align:middle line:84% the index difference is actually a lot higher, effectively, 00:13:10.530 --> 00:13:13.650 align:middle line:84% because the core is basically air, index 1. 00:13:13.650 --> 00:13:17.222 align:middle line:84% The cladding effectively now is essentially glass, which is 1.5. 00:13:17.222 --> 00:13:19.930 align:middle line:84% So the index difference, instead of being like 10 to the minus 3, 00:13:19.930 --> 00:13:22.250 align:middle line:90% this is like 0.5. 00:13:22.250 --> 00:13:26.050 align:middle line:84% So in a hollow-core or microstructured optical fiber, 00:13:26.050 --> 00:13:28.610 align:middle line:84% the roughness scattering can become the dominant loss 00:13:28.610 --> 00:13:31.490 align:middle line:90% mechanism. 00:13:31.490 --> 00:13:34.250 align:middle line:84% And in this case, if you think about-- 00:13:34.250 --> 00:13:39.020 align:middle line:84% let's think about a simple model of just a glass rod, a glass 00:13:39.020 --> 00:13:39.720 align:middle line:90% cylinder. 00:13:39.720 --> 00:13:42.380 align:middle line:90% 00:13:42.380 --> 00:13:45.140 align:middle line:84% You are going to have some-- despite that you have surface 00:13:45.140 --> 00:13:49.580 align:middle line:84% tension, you are still going to have some essentially roughness. 00:13:49.580 --> 00:13:53.740 align:middle line:84% The roughness takes the form of this kind of surface waviness. 00:13:53.740 --> 00:13:57.140 align:middle line:84% So it creates basically a periodic modulation 00:13:57.140 --> 00:14:00.580 align:middle line:90% of the diameter of the rod. 00:14:00.580 --> 00:14:04.380 align:middle line:84% This is called a surface capillary wave. 00:14:04.380 --> 00:14:09.780 align:middle line:90% 00:14:09.780 --> 00:14:13.020 align:middle line:84% Now, surface capillary wave, there 00:14:13.020 --> 00:14:16.100 align:middle line:84% is some energy associated with this surface capillary wave 00:14:16.100 --> 00:14:17.580 align:middle line:84% because it's basically associated 00:14:17.580 --> 00:14:20.140 align:middle line:84% with some kind of a degree of freedom for the system. 00:14:20.140 --> 00:14:20.640 align:middle line:90% OK. 00:14:20.640 --> 00:14:24.460 align:middle line:90% 00:14:24.460 --> 00:14:27.380 align:middle line:84% From statistical mechanics, you know 00:14:27.380 --> 00:14:29.500 align:middle line:84% that the Hamiltonian, for the Hamiltonian 00:14:29.500 --> 00:14:32.380 align:middle line:84% of the system, every quadratic term in the Hamiltonian 00:14:32.380 --> 00:14:35.040 align:middle line:84% corresponds to half kT of energy. 00:14:35.040 --> 00:14:37.580 align:middle line:90% 00:14:37.580 --> 00:14:40.220 align:middle line:84% I don't know if that's how you learn the energy equipartition 00:14:40.220 --> 00:14:40.800 align:middle line:90% theorem. 00:14:40.800 --> 00:14:43.300 align:middle line:84% The other way to present it is that essentially every degree 00:14:43.300 --> 00:14:48.460 align:middle line:84% of freedom you have a 1/2 kT associated with every degree 00:14:48.460 --> 00:14:50.260 align:middle line:90% of freedom 00:14:50.260 --> 00:14:53.420 align:middle line:84% I mentioned that this kind of surface modulation of surface 00:14:53.420 --> 00:14:56.500 align:middle line:84% capillary wave is associated essentially 00:14:56.500 --> 00:14:58.700 align:middle line:84% with the degree of freedom for the system. 00:14:58.700 --> 00:15:01.900 align:middle line:84% So it means that whenever temperature is not 0, 00:15:01.900 --> 00:15:04.740 align:middle line:84% there's always some finite energy associated 00:15:04.740 --> 00:15:09.020 align:middle line:84% with every wavelength of the surface capillary wave. 00:15:09.020 --> 00:15:15.100 align:middle line:84% So the wavelength essentially means the axial period for it. 00:15:15.100 --> 00:15:19.220 align:middle line:84% So when we are gauging the residual roughness in a glass 00:15:19.220 --> 00:15:21.640 align:middle line:84% fiber, we have to look at-- for example, 00:15:21.640 --> 00:15:24.740 align:middle line:84% this has to be somewhere around Tf. 00:15:24.740 --> 00:15:26.860 align:middle line:84% Essentially, you have a glass rod. 00:15:26.860 --> 00:15:29.420 align:middle line:84% As it cools down, when it goes through glass transition 00:15:29.420 --> 00:15:31.700 align:middle line:84% temperature or goes through effective temperature, 00:15:31.700 --> 00:15:33.990 align:middle line:84% essentially any kind of roughness 00:15:33.990 --> 00:15:36.470 align:middle line:84% due to this kind of energy equipartition 00:15:36.470 --> 00:15:39.110 align:middle line:90% is going to get frozen. 00:15:39.110 --> 00:15:43.190 align:middle line:84% And those are going to be the loss limit for-- 00:15:43.190 --> 00:15:44.830 align:middle line:84% that's going to be a surface roughness 00:15:44.830 --> 00:15:49.030 align:middle line:84% limit for a fiber that's prepared using the process. 00:15:49.030 --> 00:15:54.350 align:middle line:84% Now, fortunately, this kind of roughness 00:15:54.350 --> 00:15:57.430 align:middle line:84% tends to be pretty small because k, the Boltzmann constant, 00:15:57.430 --> 00:15:58.230 align:middle line:90% is very small. 00:15:58.230 --> 00:15:59.790 align:middle line:90% 10 to the minus 23. 00:15:59.790 --> 00:16:02.230 align:middle line:84% So if you convert that into essentially the surface 00:16:02.230 --> 00:16:05.690 align:middle line:90% roughness, it's quite small. 00:16:05.690 --> 00:16:09.070 align:middle line:84% But it can still play a significant role 00:16:09.070 --> 00:16:14.430 align:middle line:84% when it comes to devices like photonic crystal fibers. 00:16:14.430 --> 00:16:18.750 align:middle line:84% So that's where the roughness comes in for optical fibers. 00:16:18.750 --> 00:16:21.210 align:middle line:84% Now, for planar waveguide, on the other hand, 00:16:21.210 --> 00:16:23.090 align:middle line:84% the roughness usually comes from, 00:16:23.090 --> 00:16:24.887 align:middle line:84% essentially, the manufacturing process So 00:16:24.887 --> 00:16:27.470 align:middle line:84% in the next few slides, I will talk about a few different ways 00:16:27.470 --> 00:16:29.720 align:middle line:84% how you actually prepare planar optical waveguides out 00:16:29.720 --> 00:16:30.990 align:middle line:90% of glass materials. 00:16:30.990 --> 00:16:34.630 align:middle line:84% But one common feature is that none of the fabrication process 00:16:34.630 --> 00:16:36.110 align:middle line:90% is perfect. 00:16:36.110 --> 00:16:41.270 align:middle line:84% They always leave behind some residual roughness, 00:16:41.270 --> 00:16:44.207 align:middle line:84% especially on the sidewalls of the waveguide. 00:16:44.207 --> 00:16:45.790 align:middle line:84% A typical fabrication process that I'm 00:16:45.790 --> 00:16:47.790 align:middle line:84% going to talk about uses essentially lithography 00:16:47.790 --> 00:16:48.370 align:middle line:90% and etching. 00:16:48.370 --> 00:16:51.870 align:middle line:84% You lithographically define the waveguide pattern, 00:16:51.870 --> 00:16:53.350 align:middle line:84% and then you actually etch it down 00:16:53.350 --> 00:16:55.490 align:middle line:84% to actually create the waveguide structure. 00:16:55.490 --> 00:16:57.790 align:middle line:84% So during lithographic patterning, 00:16:57.790 --> 00:16:59.710 align:middle line:84% when you're trying to pattern the line, 00:16:59.710 --> 00:17:01.210 align:middle line:84% it's not going to be a perfect line. 00:17:01.210 --> 00:17:04.790 align:middle line:84% There's always a little bit of roughness at the edges, 00:17:04.790 --> 00:17:07.310 align:middle line:84% just due to intrinsic imperfection of the photomask 00:17:07.310 --> 00:17:09.430 align:middle line:84% creation, due to some kind of variation 00:17:09.430 --> 00:17:12.447 align:middle line:84% in the mass fabrication process, as well as 00:17:12.447 --> 00:17:14.530 align:middle line:84% variation in the etching process for the material. 00:17:14.530 --> 00:17:17.829 align:middle line:84% So those then get translated into the sidewalls 00:17:17.829 --> 00:17:19.069 align:middle line:90% of the waveguides. 00:17:19.069 --> 00:17:22.030 align:middle line:84% So the sidewalls of waveguides usually 00:17:22.030 --> 00:17:23.430 align:middle line:84% tend to be a lot rougher compared 00:17:23.430 --> 00:17:24.930 align:middle line:90% to the case of optical fibers. 00:17:24.930 --> 00:17:26.670 align:middle line:84% We are talking about RMS roughness 00:17:26.670 --> 00:17:30.960 align:middle line:90% of the order of a nanometer. 00:17:30.960 --> 00:17:33.200 align:middle line:84% So this is actually significantly larger 00:17:33.200 --> 00:17:34.800 align:middle line:90% compared to optical fibers. 00:17:34.800 --> 00:17:38.825 align:middle line:84% The top surface can be pretty smooth if it's a deposited film. 00:17:38.825 --> 00:17:40.200 align:middle line:84% So most of the roughness actually 00:17:40.200 --> 00:17:41.742 align:middle line:84% comes from essentially the sidewalls. 00:17:41.742 --> 00:17:44.240 align:middle line:90% Sidewall scattering roughness. 00:17:44.240 --> 00:17:47.760 align:middle line:84% So in addition to the much larger magnitude of roughness, 00:17:47.760 --> 00:17:50.920 align:middle line:84% on-chip waveguides also tend to have a much larger index 00:17:50.920 --> 00:17:53.440 align:middle line:84% difference between core and cladding. 00:17:53.440 --> 00:17:54.820 align:middle line:90% So this depends on the material. 00:17:54.820 --> 00:17:57.780 align:middle line:84% So if it's, say, a silicon waveguide, 00:17:57.780 --> 00:18:00.280 align:middle line:84% then the silicon has an index of 3.45. 00:18:00.280 --> 00:18:02.160 align:middle line:84% The core is, say, silicon dioxide. 00:18:02.160 --> 00:18:03.340 align:middle line:90% It's 1.45. 00:18:03.340 --> 00:18:05.160 align:middle line:90% So index difference is like 2. 00:18:05.160 --> 00:18:07.080 align:middle line:84% So much, much higher compared to fibers. 00:18:07.080 --> 00:18:08.720 align:middle line:84% Silicon nitride, index difference 00:18:08.720 --> 00:18:10.400 align:middle line:90% is something like 0.5. 00:18:10.400 --> 00:18:12.120 align:middle line:84% For other glass materials, depending 00:18:12.120 --> 00:18:14.200 align:middle line:84% on fabrication process, the index function 00:18:14.200 --> 00:18:18.200 align:middle line:84% varies between somewhere from 0.5 to 0.1. 00:18:18.200 --> 00:18:21.720 align:middle line:84% So usually it tends to be a lot larger compared to index 00:18:21.720 --> 00:18:23.840 align:middle line:90% difference of optical fibers. 00:18:23.840 --> 00:18:28.480 align:middle line:84% And that's why, because of the much larger surface roughness 00:18:28.480 --> 00:18:30.820 align:middle line:84% magnitude and the high index difference, 00:18:30.820 --> 00:18:34.320 align:middle line:84% this is what leads to the dominance of surface 00:18:34.320 --> 00:18:38.300 align:middle line:84% roughness scattering in waveguide loss contributions. 00:18:38.300 --> 00:18:39.680 align:middle line:84% So typically, when you're talking 00:18:39.680 --> 00:18:42.080 align:middle line:84% about loss of planar waveguides, a large part of it 00:18:42.080 --> 00:18:44.863 align:middle line:84% is basically coming from this kind of a roughness scattering. 00:18:44.863 --> 00:18:46.780 align:middle line:84% So this is quite different compared to fibers. 00:18:46.780 --> 00:18:50.680 align:middle line:84% Fibers is, in most cases, intrinsic material loss. 00:18:50.680 --> 00:18:52.120 align:middle line:84% Waveguides, it's basically coming 00:18:52.120 --> 00:18:55.560 align:middle line:84% from this extrinsic scattering-induced attenuation. 00:18:55.560 --> 00:19:00.480 align:middle line:90% 00:19:00.480 --> 00:19:01.020 align:middle line:90% All right. 00:19:01.020 --> 00:19:04.920 align:middle line:90% 00:19:04.920 --> 00:19:06.520 align:middle line:84% So now, in the next few slides, I 00:19:06.520 --> 00:19:09.320 align:middle line:84% will just give you some example processes 00:19:09.320 --> 00:19:11.000 align:middle line:84% of how people actually can pattern 00:19:11.000 --> 00:19:13.320 align:middle line:90% this kind of waveguide devices. 00:19:13.320 --> 00:19:16.700 align:middle line:84% So the most standard one, the most commonly adopted process, 00:19:16.700 --> 00:19:17.620 align:middle line:90% this is quite generic. 00:19:17.620 --> 00:19:19.660 align:middle line:84% This is essentially lithographic patterning. 00:19:19.660 --> 00:19:21.000 align:middle line:90% You start from a substrate. 00:19:21.000 --> 00:19:24.080 align:middle line:84% You do a deposition to form this low-index cladding layer. 00:19:24.080 --> 00:19:25.800 align:middle line:84% You then, again, do another deposition 00:19:25.800 --> 00:19:28.160 align:middle line:84% to form the high-index core layer. 00:19:28.160 --> 00:19:29.873 align:middle line:84% You can also do a bonding, wafer bonding, 00:19:29.873 --> 00:19:31.540 align:middle line:84% if it's for single crystalline material, 00:19:31.540 --> 00:19:33.082 align:middle line:84% and then you get rid of the substrate 00:19:33.082 --> 00:19:35.760 align:middle line:84% to get a thin layer of crystalline material. 00:19:35.760 --> 00:19:37.020 align:middle line:90% For glass, it's a lot easier. 00:19:37.020 --> 00:19:39.240 align:middle line:84% You usually just do a simple deposition. 00:19:39.240 --> 00:19:41.660 align:middle line:84% You then go through a lithographic process. 00:19:41.660 --> 00:19:45.020 align:middle line:84% You first pattern a photoresist mask through photolithography. 00:19:45.020 --> 00:19:47.520 align:middle line:84% You then do an etching process to essentially transfer 00:19:47.520 --> 00:19:50.820 align:middle line:84% the pattern from the resist into the waveguide layer. 00:19:50.820 --> 00:19:54.840 align:middle line:84% And in the end, it's customary that people 00:19:54.840 --> 00:19:57.640 align:middle line:84% deposit another layer of cladding 00:19:57.640 --> 00:20:00.740 align:middle line:90% to encapsulate the waveguide. 00:20:00.740 --> 00:20:04.600 align:middle line:84% People typically do this just to avoid any kind of a surface 00:20:04.600 --> 00:20:07.600 align:middle line:84% damage and make the waveguide a lot more robust against any kind 00:20:07.600 --> 00:20:08.860 align:middle line:90% of surface contamination. 00:20:08.860 --> 00:20:11.910 align:middle line:84% If you want to continue to do the integration, 00:20:11.910 --> 00:20:14.160 align:middle line:84% it's also a lot easier because now it creates actually 00:20:14.160 --> 00:20:16.240 align:middle line:84% small flat surfaces where you can 00:20:16.240 --> 00:20:17.860 align:middle line:90% fabricate other devices on top. 00:20:17.860 --> 00:20:21.444 align:middle line:90% 00:20:21.444 --> 00:20:23.640 align:middle line:84% So amorphous materials, again, this 00:20:23.640 --> 00:20:27.510 align:middle line:84% is why people still like them for waveguide fabrication. 00:20:27.510 --> 00:20:28.970 align:middle line:84% For reasons I've already mentioned, 00:20:28.970 --> 00:20:33.822 align:middle line:84% it's a lot easier to deposit these amorphous materials. 00:20:33.822 --> 00:20:36.530 align:middle line:84% In addition to the fact that you can deposit essentially any kind 00:20:36.530 --> 00:20:38.030 align:middle line:84% of substrate without having to worry 00:20:38.030 --> 00:20:39.610 align:middle line:84% about lattice-matching constraints, 00:20:39.610 --> 00:20:42.570 align:middle line:84% the deposition rate for amorphous materials 00:20:42.570 --> 00:20:44.250 align:middle line:84% also tends to be a lot higher compared 00:20:44.250 --> 00:20:46.417 align:middle line:90% to crystalline materials. 00:20:46.417 --> 00:20:48.750 align:middle line:84% So the reason for that is because if you think about it, 00:20:48.750 --> 00:20:50.850 align:middle line:84% you want to deposit a single crystalline material. 00:20:50.850 --> 00:20:54.050 align:middle line:84% Now what you have to do is that the process 00:20:54.050 --> 00:20:56.430 align:middle line:84% goes like-- you have atoms coming from gas space. 00:20:56.430 --> 00:20:58.450 align:middle line:84% They essentially adsorb to the surface. 00:20:58.450 --> 00:21:00.250 align:middle line:84% They then go through surface diffusion 00:21:00.250 --> 00:21:03.170 align:middle line:84% to try to find essentially lattice sites where they 00:21:03.170 --> 00:21:05.170 align:middle line:84% essentially can attach to and then actually grow 00:21:05.170 --> 00:21:06.290 align:middle line:90% the single crystal. 00:21:06.290 --> 00:21:09.770 align:middle line:84% So in this process, because they rely on this kind of 2D 00:21:09.770 --> 00:21:13.330 align:middle line:84% diffusion for the atoms to find the correct lattice sites, 00:21:13.330 --> 00:21:15.850 align:middle line:84% you need to have sufficient diffusion length for the surface 00:21:15.850 --> 00:21:16.890 align:middle line:90% atoms. 00:21:16.890 --> 00:21:19.330 align:middle line:84% Now if you have a very high diffusion flux, 00:21:19.330 --> 00:21:22.150 align:middle line:84% then what happens is that these atoms, 00:21:22.150 --> 00:21:24.650 align:middle line:84% before they were able to diffuse sufficiently long distances 00:21:24.650 --> 00:21:26.450 align:middle line:84% to find the lattice sites, it's going 00:21:26.450 --> 00:21:30.170 align:middle line:84% to get disrupted by the incident atoms due to high atomic flux. 00:21:30.170 --> 00:21:32.170 align:middle line:84% It's going to get scattered because of very high 00:21:32.170 --> 00:21:33.210 align:middle line:90% deposition flux. 00:21:33.210 --> 00:21:37.330 align:middle line:84% And that's why, as you increase the deposition flux, 00:21:37.330 --> 00:21:39.930 align:middle line:84% your structure tends to become more defective. 00:21:39.930 --> 00:21:42.570 align:middle line:84% And eventually you transition from single crystal 00:21:42.570 --> 00:21:46.290 align:middle line:84% to polycrystalline and then to amorphous with an increasing 00:21:46.290 --> 00:21:47.750 align:middle line:90% deposition flux. 00:21:47.750 --> 00:21:51.315 align:middle line:90% 00:21:51.315 --> 00:21:53.490 align:middle line:84% And high deposition rate is actually 00:21:53.490 --> 00:21:55.450 align:middle line:84% beneficial from a manufacturing perspective 00:21:55.450 --> 00:21:56.890 align:middle line:84% because it means that you can actually make things 00:21:56.890 --> 00:21:57.930 align:middle line:90% with high throughput. 00:21:57.930 --> 00:22:00.150 align:middle line:84% It takes less time to do deposition. 00:22:00.150 --> 00:22:03.210 align:middle line:84% So that's why amorphous materials 00:22:03.210 --> 00:22:05.250 align:middle line:84% tend to be a lot easier in this case. 00:22:05.250 --> 00:22:08.868 align:middle line:84% Also, you can actually do the deposition-- 00:22:08.868 --> 00:22:10.910 align:middle line:84% you can do the deposition at a lower temperature. 00:22:10.910 --> 00:22:12.470 align:middle line:90% Again, for the same reason. 00:22:12.470 --> 00:22:14.710 align:middle line:84% High temperature, due to Arrhenius law, 00:22:14.710 --> 00:22:17.090 align:middle line:84% it means the surface diffusion is more effective. 00:22:17.090 --> 00:22:19.170 align:middle line:84% So the atoms will be a lot more efficient 00:22:19.170 --> 00:22:21.960 align:middle line:84% in finding the low-energy sites attached to the lattice. 00:22:21.960 --> 00:22:25.340 align:middle line:84% At low temperature, on the other hand, diffusivity is limited. 00:22:25.340 --> 00:22:28.020 align:middle line:84% Then the atom has to more or less settle where they actually 00:22:28.020 --> 00:22:31.380 align:middle line:84% land on the substrate, and you tend to get amorphous structure. 00:22:31.380 --> 00:22:33.620 align:middle line:84% Again, low temperature is actually beneficial 00:22:33.620 --> 00:22:35.780 align:middle line:84% because, number one, it prevents you to disrupt 00:22:35.780 --> 00:22:37.500 align:middle line:90% any kind of on-chip structures. 00:22:37.500 --> 00:22:40.760 align:middle line:84% Two is that it's also more energy efficient. 00:22:40.760 --> 00:22:43.180 align:middle line:90% It's also a faster process. 00:22:43.180 --> 00:22:46.300 align:middle line:84% So for all those reasons, that's why amorphous materials is, 00:22:46.300 --> 00:22:48.660 align:middle line:84% in lots of cases, considered preferred 00:22:48.660 --> 00:22:50.000 align:middle line:90% for waveguide fabrication. 00:22:50.000 --> 00:22:53.720 align:middle line:90% 00:22:53.720 --> 00:22:56.380 align:middle line:84% And in fact, if you look at some of the dominant waveguide 00:22:56.380 --> 00:22:58.660 align:middle line:84% materials, one of them is actually silicon nitride. 00:22:58.660 --> 00:23:01.060 align:middle line:84% You deposit through PECVD, and then you pattern. 00:23:01.060 --> 00:23:03.560 align:middle line:84% So that's a very common waveguide material 00:23:03.560 --> 00:23:04.620 align:middle line:90% that's amorphous. 00:23:04.620 --> 00:23:07.540 align:middle line:84% And also, when I'm talking about the lowest waveguide loss 00:23:07.540 --> 00:23:11.180 align:middle line:84% records, they are pretty much all achieved in silicon nitride. 00:23:11.180 --> 00:23:15.080 align:middle line:84% And it also relates to the low loss of the optical materials. 00:23:15.080 --> 00:23:17.900 align:middle line:84% So amorphous materials, they tend 00:23:17.900 --> 00:23:19.797 align:middle line:84% to have very low intrinsic loss resulting 00:23:19.797 --> 00:23:22.380 align:middle line:84% from the fact that despite you have Rayleigh scattering that's 00:23:22.380 --> 00:23:22.840 align:middle line:90% present. 00:23:22.840 --> 00:23:24.257 align:middle line:84% But Rayleigh scattering comes from 00:23:24.257 --> 00:23:26.220 align:middle line:90% atomic-scale inhomogeneities. 00:23:26.220 --> 00:23:29.500 align:middle line:84% So because Rayleigh scattering scales wavelengths to the-- 00:23:29.500 --> 00:23:31.580 align:middle line:84% scale is essentially structure size 00:23:31.580 --> 00:23:33.460 align:middle line:90% to the negative fourth power. 00:23:33.460 --> 00:23:36.420 align:middle line:84% So much smaller length scale of inhomogeneity 00:23:36.420 --> 00:23:39.180 align:middle line:84% also corresponds to a much reduced Rayleigh scattering 00:23:39.180 --> 00:23:39.860 align:middle line:90% loss. 00:23:39.860 --> 00:23:42.300 align:middle line:84% And that's why you can achieve very low-loss waveguides 00:23:42.300 --> 00:23:45.420 align:middle line:90% in these amorphous materials. 00:23:45.420 --> 00:23:46.020 align:middle line:90% All right. 00:23:46.020 --> 00:23:48.780 align:middle line:84% So basically, for lithography, this 00:23:48.780 --> 00:23:52.020 align:middle line:84% is probably the most commonly used technique for fabrication. 00:23:52.020 --> 00:23:53.700 align:middle line:84% But there are also some other techniques 00:23:53.700 --> 00:23:56.460 align:middle line:84% that are unique only to amorphous materials. 00:23:56.460 --> 00:23:59.765 align:middle line:84% So one such technique is called ion exchange. 00:23:59.765 --> 00:24:01.140 align:middle line:84% We have talked about ion exchange 00:24:01.140 --> 00:24:04.120 align:middle line:84% in the case of strengthening for glass waveguides. 00:24:04.120 --> 00:24:07.700 align:middle line:84% In that case, we are trying to [? indiffuse, ?] 00:24:07.700 --> 00:24:10.400 align:middle line:84% say, potassium ion into a sodium-containing glass. 00:24:10.400 --> 00:24:14.860 align:middle line:84% Unfortunately, potassium and sodium, all those alkaline ions, 00:24:14.860 --> 00:24:19.140 align:middle line:84% they have very similar refractive indices. 00:24:19.140 --> 00:24:21.600 align:middle line:84% So when it comes to fabricating waveguides, 00:24:21.600 --> 00:24:24.060 align:middle line:84% people usually use different kind of ion combinations. 00:24:24.060 --> 00:24:27.660 align:middle line:84% So one common example is to actually use silver. 00:24:27.660 --> 00:24:31.180 align:middle line:84% By [? indiffusing ?] silver into a sodium glass, 00:24:31.180 --> 00:24:33.340 align:middle line:84% you can create essentially an optical waveguide 00:24:33.340 --> 00:24:36.660 align:middle line:84% in where you have essentially silver substitution. 00:24:36.660 --> 00:24:38.800 align:middle line:84% So silver is preferred for a couple of reasons. 00:24:38.800 --> 00:24:40.820 align:middle line:84% First thing, we have mentioned that silver 00:24:40.820 --> 00:24:44.340 align:middle line:84% has the electron cloud that's very easily deformable. 00:24:44.340 --> 00:24:47.260 align:middle line:84% So a deformable electron cloud optically essentially 00:24:47.260 --> 00:24:51.220 align:middle line:84% means that it's very polarizable. 00:24:51.220 --> 00:24:55.540 align:middle line:84% So it easily responds to an incident electric field 00:24:55.540 --> 00:24:56.620 align:middle line:90% of the light. 00:24:56.620 --> 00:25:01.997 align:middle line:84% So it corresponds to essentially a large refractive index. 00:25:01.997 --> 00:25:04.580 align:middle line:84% So you can create essentially waveguides with sufficient index 00:25:04.580 --> 00:25:05.820 align:middle line:84% difference between the core and the cladding 00:25:05.820 --> 00:25:07.780 align:middle line:84% when you substitute sodium with silver. 00:25:07.780 --> 00:25:09.580 align:middle line:90% So that's one. 00:25:09.580 --> 00:25:12.580 align:middle line:84% Second thing is that because it's such a deformable electron 00:25:12.580 --> 00:25:15.100 align:middle line:84% cloud, it means that silver ions, 00:25:15.100 --> 00:25:17.532 align:middle line:84% they tend to have high diffusivity. 00:25:17.532 --> 00:25:18.990 align:middle line:84% And that's why, if you recall, when 00:25:18.990 --> 00:25:20.830 align:middle line:84% we're talking about the ionic conduction, 00:25:20.830 --> 00:25:22.950 align:middle line:84% we said a lot of the superionic conductors 00:25:22.950 --> 00:25:27.712 align:middle line:84% or fast ionic conductors, they contain silver or copper ions. 00:25:27.712 --> 00:25:30.170 align:middle line:84% Again, because it's very easy to deform the electron cloud, 00:25:30.170 --> 00:25:31.870 align:middle line:84% you can easily squeeze these ions 00:25:31.870 --> 00:25:34.370 align:middle line:90% in between the solid structure. 00:25:34.370 --> 00:25:36.670 align:middle line:84% And therefore, they tend to have high diffusivity. 00:25:36.670 --> 00:25:39.190 align:middle line:84% So they also benefit essentially the throughput 00:25:39.190 --> 00:25:40.430 align:middle line:90% of the fabrication. 00:25:40.430 --> 00:25:44.470 align:middle line:84% So that's why silver is one of the preferred elements 00:25:44.470 --> 00:25:46.110 align:middle line:90% for this fabrication process. 00:25:46.110 --> 00:25:49.102 align:middle line:84% Now specifically, if you look at fabrication scheme here, 00:25:49.102 --> 00:25:51.310 align:middle line:84% you start out with essentially just a glass substrate 00:25:51.310 --> 00:25:52.550 align:middle line:90% containing sodium. 00:25:52.550 --> 00:25:55.550 align:middle line:84% You put a metal mask on top, you do lithography 00:25:55.550 --> 00:25:57.430 align:middle line:84% to pattern the metal mask, and then 00:25:57.430 --> 00:26:02.630 align:middle line:84% you use the metal mask essentially as a diffusion 00:26:02.630 --> 00:26:07.390 align:middle line:84% barrier to allow only ion exchange to take place where you 00:26:07.390 --> 00:26:09.110 align:middle line:90% have an opening on the mask. 00:26:09.110 --> 00:26:12.530 align:middle line:84% You then immerse it into a molten salt bath. 00:26:12.530 --> 00:26:14.870 align:middle line:84% You can apply electric field to essentially accelerate 00:26:14.870 --> 00:26:17.110 align:middle line:90% this kind of a diffusion here. 00:26:17.110 --> 00:26:21.230 align:middle line:84% And then, in the end, you can then remove this kind of mask 00:26:21.230 --> 00:26:22.570 align:middle line:90% to create a waveguide here. 00:26:22.570 --> 00:26:26.670 align:middle line:84% So this process usually, up to this step, 00:26:26.670 --> 00:26:29.830 align:middle line:84% essentially creates a waveguide that's actually on the surface, 00:26:29.830 --> 00:26:33.430 align:middle line:84% because that's where ions come in. 00:26:33.430 --> 00:26:35.870 align:middle line:84% For many applications, you want the waveguide core 00:26:35.870 --> 00:26:38.270 align:middle line:84% to be isolated from the environment 00:26:38.270 --> 00:26:40.590 align:middle line:84% to essentially increase its robustness 00:26:40.590 --> 00:26:41.690 align:middle line:90% against contamination. 00:26:41.690 --> 00:26:43.190 align:middle line:84% In that case, what you can do is you 00:26:43.190 --> 00:26:45.590 align:middle line:84% can continue to do a so-called field-assisted burial 00:26:45.590 --> 00:26:52.090 align:middle line:84% step that drives the silver ion further into the surface. 00:26:52.090 --> 00:26:54.150 align:middle line:84% So then, essentially, you now create 00:26:54.150 --> 00:26:56.990 align:middle line:84% this kind of buried or embedded optical waveguide here. 00:26:56.990 --> 00:26:59.950 align:middle line:84% So this is quite common fabrication process 00:26:59.950 --> 00:27:01.390 align:middle line:90% for glass waveguides. 00:27:01.390 --> 00:27:02.730 align:middle line:90% It's very scalable. 00:27:02.730 --> 00:27:04.470 align:middle line:84% It's a simple fabrication process. 00:27:04.470 --> 00:27:05.870 align:middle line:84% It also has the advantage that it 00:27:05.870 --> 00:27:08.750 align:middle line:84% tends to have a quite low optical loss. 00:27:08.750 --> 00:27:11.550 align:middle line:84% Its low optical loss come from two parts here. 00:27:11.550 --> 00:27:15.200 align:middle line:84% One is that its index contrast is not 00:27:15.200 --> 00:27:18.160 align:middle line:84% as large compared to, for example, silicon dioxide. 00:27:18.160 --> 00:27:21.240 align:middle line:84% Two is that this diffusion process, 00:27:21.240 --> 00:27:25.520 align:middle line:84% it creates this kind of a diffused interface. 00:27:25.520 --> 00:27:28.760 align:middle line:84% It does not create an abrupt index difference 00:27:28.760 --> 00:27:30.400 align:middle line:90% between the core and cladding. 00:27:30.400 --> 00:27:32.320 align:middle line:84% This diffuse interface also helps 00:27:32.320 --> 00:27:36.320 align:middle line:84% you to alleviate essentially the effective roughness you have 00:27:36.320 --> 00:27:38.840 align:middle line:84% and therefore lower the scattering loss. 00:27:38.840 --> 00:27:41.400 align:middle line:84% So that's why this is a quite commonly used waveguide 00:27:41.400 --> 00:27:42.840 align:middle line:90% fabrication technique. 00:27:42.840 --> 00:27:49.360 align:middle line:84% And a lot of the optical waveguide technologies, 00:27:49.360 --> 00:27:53.800 align:middle line:84% they rely on this process to create the devices. 00:27:53.800 --> 00:27:54.373 align:middle line:90% All right. 00:27:54.373 --> 00:27:56.040 align:middle line:84% Now, next technique that people are also 00:27:56.040 --> 00:27:58.240 align:middle line:84% using to create glass waveguides is essentially 00:27:58.240 --> 00:28:00.100 align:middle line:90% a nanoimprint process. 00:28:00.100 --> 00:28:02.320 align:middle line:84% Some people also call it embossing. 00:28:02.320 --> 00:28:06.360 align:middle line:84% But effectively what you would do, for standard nanoimprint 00:28:06.360 --> 00:28:10.540 align:middle line:84% process, you basically start out with substrate. 00:28:10.540 --> 00:28:12.220 align:middle line:90% You deposit your glass layer. 00:28:12.220 --> 00:28:14.100 align:middle line:84% You then deposit a layer of polymer. 00:28:14.100 --> 00:28:16.760 align:middle line:84% That's used essentially as a resist for imprint. 00:28:16.760 --> 00:28:20.040 align:middle line:84% You then press a mold with a pattern inverse 00:28:20.040 --> 00:28:24.120 align:middle line:84% what you want to get in the end against the surface. 00:28:24.120 --> 00:28:27.680 align:middle line:84% You then either apply a UV light, 00:28:27.680 --> 00:28:31.120 align:middle line:84% or you apply the heat to introduce 00:28:31.120 --> 00:28:33.140 align:middle line:90% reflow of the polymer resist. 00:28:33.140 --> 00:28:36.040 align:middle line:90% 00:28:36.040 --> 00:28:38.000 align:middle line:84% Once you have the polymer resist defined, 00:28:38.000 --> 00:28:39.680 align:middle line:84% you then do an etch pattern transfer 00:28:39.680 --> 00:28:44.360 align:middle line:84% to then transfer into the final waveguide material. 00:28:44.360 --> 00:28:46.160 align:middle line:84% This is one way to make waveguides. 00:28:46.160 --> 00:28:48.800 align:middle line:84% But for glass materials, there's actually a simpler way 00:28:48.800 --> 00:28:50.320 align:middle line:90% to actually create waveguides. 00:28:50.320 --> 00:28:53.400 align:middle line:84% Because glass themselves, if you heat them up, 00:28:53.400 --> 00:28:55.600 align:middle line:84% they will essentially turn to basically a viscous 00:28:55.600 --> 00:28:58.520 align:middle line:90% liquid or viscoelastic solid. 00:28:58.520 --> 00:29:00.960 align:middle line:84% So then you can just directly press 00:29:00.960 --> 00:29:02.960 align:middle line:90% a mold against glass material. 00:29:02.960 --> 00:29:05.840 align:middle line:84% And heating it up to well above the glass transition 00:29:05.840 --> 00:29:07.680 align:middle line:84% temperature, you can then introduce 00:29:07.680 --> 00:29:10.930 align:middle line:84% essentially a thermal reflow of the glass materials or viscous 00:29:10.930 --> 00:29:13.730 align:middle line:84% flow, which then directly allows you to structure 00:29:13.730 --> 00:29:15.930 align:middle line:90% the glasses into a waveguide. 00:29:15.930 --> 00:29:18.570 align:middle line:84% So the unique properties of glass 00:29:18.570 --> 00:29:21.010 align:middle line:84% enables this much simpler fabrication process 00:29:21.010 --> 00:29:23.290 align:middle line:84% using direct nanoimprint without having 00:29:23.290 --> 00:29:26.610 align:middle line:84% to use an intermediate resist-based pattern transfer 00:29:26.610 --> 00:29:27.770 align:middle line:90% step. 00:29:27.770 --> 00:29:30.462 align:middle line:84% So on the right side, this is just one example. 00:29:30.462 --> 00:29:32.170 align:middle line:84% You can see this is actually a waveguide. 00:29:32.170 --> 00:29:34.795 align:middle line:84% That's actually created, in this case in arsenic sulfide glass, 00:29:34.795 --> 00:29:37.570 align:middle line:84% a pattern using 3D printing technique. 00:29:37.570 --> 00:29:39.170 align:middle line:84% Also, one of the advantages here is 00:29:39.170 --> 00:29:42.610 align:middle line:84% that the roughness of the waveguide 00:29:42.610 --> 00:29:46.290 align:middle line:84% essentially is a direct replica from the mold. 00:29:46.290 --> 00:29:48.090 align:middle line:84% So if you can create a high-quality mold 00:29:48.090 --> 00:29:50.690 align:middle line:84% with very low roughness, then the resulting waveguide 00:29:50.690 --> 00:29:53.755 align:middle line:84% also inherits the property having a smooth surface. 00:29:53.755 --> 00:29:55.130 align:middle line:84% And that's beneficial to creating 00:29:55.130 --> 00:29:58.210 align:middle line:84% essentially a low-loss fabrication-- 00:29:58.210 --> 00:29:59.650 align:middle line:90% for low-loss propagation. 00:29:59.650 --> 00:30:02.330 align:middle line:90% 00:30:02.330 --> 00:30:04.410 align:middle line:90% All right. 00:30:04.410 --> 00:30:06.510 align:middle line:84% Now, last technique I'm going to talk about here, 00:30:06.510 --> 00:30:10.410 align:middle line:90% this is using laser writing. 00:30:10.410 --> 00:30:14.050 align:middle line:84% So for what's called direct laser writing or laser 00:30:14.050 --> 00:30:16.930 align:middle line:84% direct writing, you can start with a bulk piece 00:30:16.930 --> 00:30:18.650 align:middle line:90% of glass materials. 00:30:18.650 --> 00:30:23.210 align:middle line:84% You then focus the laser light tightly through basically a lens 00:30:23.210 --> 00:30:24.670 align:middle line:90% or microscope objective. 00:30:24.670 --> 00:30:27.810 align:middle line:90% 00:30:27.810 --> 00:30:31.610 align:middle line:84% And then, at the focal spot, you can 00:30:31.610 --> 00:30:34.290 align:middle line:84% induce essentially some refractive index change 00:30:34.290 --> 00:30:37.010 align:middle line:84% through a process of so-called two-photon polymerization. 00:30:37.010 --> 00:30:39.010 align:middle line:84% Sorry, Two-Photon Absorption, TPA. 00:30:39.010 --> 00:30:42.570 align:middle line:84% So two-photon absorption basically 00:30:42.570 --> 00:30:46.230 align:middle line:84% means that if you think about, this is the energy band. 00:30:46.230 --> 00:30:47.890 align:middle line:90% This is energy. 00:30:47.890 --> 00:30:49.330 align:middle line:90% This is Ec. 00:30:49.330 --> 00:30:51.610 align:middle line:90% This is Ev. 00:30:51.610 --> 00:30:54.410 align:middle line:84% Normally, in order for a photon to get absorbed, 00:30:54.410 --> 00:30:56.730 align:middle line:84% its energy has to be larger than the bandgap. 00:30:56.730 --> 00:31:00.350 align:middle line:84% However, if the light intensity is strong enough, 00:31:00.350 --> 00:31:04.010 align:middle line:84% you can introduce so-called two-photon absorption, 00:31:04.010 --> 00:31:07.470 align:middle line:84% where instead of having one photon, you use two photons. 00:31:07.470 --> 00:31:12.010 align:middle line:84% You combine the energy together to then promote one electron 00:31:12.010 --> 00:31:14.030 align:middle line:84% from the valence band to the conduction band. 00:31:14.030 --> 00:31:16.970 align:middle line:90% 00:31:16.970 --> 00:31:21.690 align:middle line:84% So in this case, you only need to have 00:31:21.690 --> 00:31:24.370 align:middle line:84% the photon energy to be larger than half of the bandgap. 00:31:24.370 --> 00:31:33.210 align:middle line:90% 00:31:33.210 --> 00:31:37.250 align:middle line:84% Now the unique behavior of this kind of absorption process 00:31:37.250 --> 00:31:39.170 align:middle line:90% is that it's nonlinear. 00:31:39.170 --> 00:31:44.850 align:middle line:84% So if you think about the absorption coefficient alpha, 00:31:44.850 --> 00:31:47.462 align:middle line:84% for typical linear absorption process, which 00:31:47.462 --> 00:31:49.170 align:middle line:84% is the regular absorption process that we 00:31:49.170 --> 00:31:52.970 align:middle line:84% see in materials, alpha for a given material 00:31:52.970 --> 00:31:55.550 align:middle line:84% at a particular wavelength is a material constant. 00:31:55.550 --> 00:31:59.287 align:middle line:84% It does not depend on light intensity. 00:31:59.287 --> 00:32:01.370 align:middle line:84% So essentially it means that the amount of photons 00:32:01.370 --> 00:32:04.880 align:middle line:84% absorbed scales linearly with the amount of photons 00:32:04.880 --> 00:32:06.660 align:middle line:90% you send into a system. 00:32:06.660 --> 00:32:10.020 align:middle line:84% That's why it's called the linear absorption process. 00:32:10.020 --> 00:32:15.180 align:middle line:84% For nonlinear absorption process for two-photon absorption, 00:32:15.180 --> 00:32:16.900 align:middle line:84% the absorption coefficient actually 00:32:16.900 --> 00:32:18.520 align:middle line:90% scales with light intensity. 00:32:18.520 --> 00:32:21.140 align:middle line:90% 00:32:21.140 --> 00:32:25.740 align:middle line:84% So in other words, the amount of light that's absorbed 00:32:25.740 --> 00:32:28.840 align:middle line:84% will scale quadratically with the light intensity. 00:32:28.840 --> 00:32:31.840 align:middle line:90% 00:32:31.840 --> 00:32:33.840 align:middle line:84% Now, intuitively, that's also easy to understand 00:32:33.840 --> 00:32:35.540 align:middle line:84% because this is basically a two-particle process. 00:32:35.540 --> 00:32:37.980 align:middle line:84% You need two photons in order to trigger absorption. 00:32:37.980 --> 00:32:40.500 align:middle line:84% So that's why its absorption coefficient should 00:32:40.500 --> 00:32:41.800 align:middle line:90% scale linearly with density. 00:32:41.800 --> 00:32:43.258 align:middle line:84% And the amount of absorption scales 00:32:43.258 --> 00:32:44.740 align:middle line:84% quadratically with light intensity. 00:32:44.740 --> 00:32:49.240 align:middle line:90% 00:32:49.240 --> 00:32:54.860 align:middle line:84% Now, why I have to mention this kind of TPA process here? 00:32:54.860 --> 00:32:58.260 align:middle line:84% So when it comes to waveguide fabrication, 00:32:58.260 --> 00:33:00.940 align:middle line:84% if you're trying to write a waveguide in a bulk glass-- 00:33:00.940 --> 00:33:05.060 align:middle line:90% 00:33:05.060 --> 00:33:08.820 align:middle line:84% so say this is the bulk piece of glass material, 00:33:08.820 --> 00:33:10.980 align:middle line:84% and you are trying to focus light down 00:33:10.980 --> 00:33:12.600 align:middle line:90% to this very small spot. 00:33:12.600 --> 00:33:16.940 align:middle line:84% And the idea is that this very small spot, this absorption 00:33:16.940 --> 00:33:18.480 align:middle line:90% induces some material change. 00:33:18.480 --> 00:33:21.180 align:middle line:84% You can either increase the index or decrease the index. 00:33:21.180 --> 00:33:25.520 align:middle line:84% Then, by translating this focal spot through all the material, 00:33:25.520 --> 00:33:30.020 align:middle line:84% you can essentially directly write a waveguide. 00:33:30.020 --> 00:33:33.380 align:middle line:84% Now if you leverage not the two-photon absorption process, 00:33:33.380 --> 00:33:36.500 align:middle line:84% by using linear absorption, then it 00:33:36.500 --> 00:33:40.220 align:middle line:84% means that, in addition to absorption 00:33:40.220 --> 00:33:43.480 align:middle line:84% at the focal spot, the light, as it passes through the material, 00:33:43.480 --> 00:33:46.480 align:middle line:84% in this shaded region, you also get absorption. 00:33:46.480 --> 00:33:50.100 align:middle line:90% 00:33:50.100 --> 00:33:53.700 align:middle line:84% So that means that you cannot spatially confine your material 00:33:53.700 --> 00:33:58.220 align:middle line:84% modification to only the focal spot. 00:33:58.220 --> 00:34:01.310 align:middle line:84% On the other hand, if it's two-photon absorption, then 00:34:01.310 --> 00:34:03.630 align:middle line:84% because of this kind of quadratic dependence 00:34:03.630 --> 00:34:06.270 align:middle line:84% of the absorption with intensity, 00:34:06.270 --> 00:34:08.632 align:middle line:84% it means that this absorption is really only significant 00:34:08.632 --> 00:34:10.590 align:middle line:84% where you have very high light intensity, which 00:34:10.590 --> 00:34:12.389 align:middle line:90% is the focal spot. 00:34:12.389 --> 00:34:14.389 align:middle line:84% And essentially, all the absorption 00:34:14.389 --> 00:34:17.110 align:middle line:84% in this kind of a material above the focal spot 00:34:17.110 --> 00:34:18.170 align:middle line:90% becomes negligible. 00:34:18.170 --> 00:34:23.464 align:middle line:90% 00:34:23.464 --> 00:34:27.670 align:middle line:84% So that's why you can actually use this kind of TPA process 00:34:27.670 --> 00:34:29.670 align:middle line:90% to create buried waveguides. 00:34:29.670 --> 00:34:33.290 align:middle line:84% Also with much higher resolution, again, 00:34:33.290 --> 00:34:35.750 align:middle line:84% because you can essentially confine the light interaction 00:34:35.750 --> 00:34:38.190 align:middle line:84% volume only to the focal spot very precisely. 00:34:38.190 --> 00:34:41.790 align:middle line:90% 00:34:41.790 --> 00:34:44.310 align:middle line:90% OK? 00:34:44.310 --> 00:34:44.810 align:middle line:90% OK. 00:34:44.810 --> 00:34:46.590 align:middle line:84% So then, depending on the glass material, 00:34:46.590 --> 00:34:49.969 align:middle line:84% most glass material exhibits so-called photosensitivity. 00:34:49.969 --> 00:34:51.469 align:middle line:84% So if you shine light on it, as long 00:34:51.469 --> 00:34:55.150 align:middle line:84% as you put enough energy into it, it's going to absorb light. 00:34:55.150 --> 00:34:57.350 align:middle line:84% As it absorbs light, local lattice structure 00:34:57.350 --> 00:34:59.950 align:middle line:90% actually gets slightly changed. 00:34:59.950 --> 00:35:01.850 align:middle line:84% So due to the local lattice change, 00:35:01.850 --> 00:35:04.190 align:middle line:84% it will also change the refractive index slightly. 00:35:04.190 --> 00:35:06.830 align:middle line:84% Some glass materials, their refractive index 00:35:06.830 --> 00:35:08.925 align:middle line:84% actually increases upon irradiation. 00:35:08.925 --> 00:35:10.550 align:middle line:84% In this case, it's very straightforward 00:35:10.550 --> 00:35:11.508 align:middle line:90% to write the waveguide. 00:35:11.508 --> 00:35:13.190 align:middle line:90% You just translate it. 00:35:13.190 --> 00:35:16.410 align:middle line:84% Essentially, the focal spot trace becomes a waveguide. 00:35:16.410 --> 00:35:18.190 align:middle line:84% Some glass materials, the index actually 00:35:18.190 --> 00:35:21.770 align:middle line:84% decreases as you expose it to these laser pulses. 00:35:21.770 --> 00:35:23.870 align:middle line:84% In that case, what you do is you go 00:35:23.870 --> 00:35:27.723 align:middle line:84% to a so-called depressed cladding design. 00:35:27.723 --> 00:35:29.890 align:middle line:84% You essentially write the cladding of the waveguide. 00:35:29.890 --> 00:35:32.670 align:middle line:84% You essentially write sort of like a tube. 00:35:32.670 --> 00:35:35.610 align:middle line:84% And in the center of the tube where you don't have a light, 00:35:35.610 --> 00:35:37.612 align:middle line:84% that's actually where you have the waveguide. 00:35:37.612 --> 00:35:39.570 align:middle line:84% You essentially write in this kind of patterns. 00:35:39.570 --> 00:35:42.270 align:middle line:90% 00:35:42.270 --> 00:35:46.205 align:middle line:84% So besides fabricating waveguide using this technique here, 00:35:46.205 --> 00:35:47.830 align:middle line:84% this is actually a technique now that's 00:35:47.830 --> 00:35:49.130 align:middle line:90% getting increasing interest. 00:35:49.130 --> 00:35:51.310 align:middle line:84% People like it because it's very versatile. 00:35:51.310 --> 00:35:55.690 align:middle line:84% And it's increasingly getting used for many applications, 00:35:55.690 --> 00:35:58.060 align:middle line:84% especially to create 3D waveguide structures, 00:35:58.060 --> 00:36:00.310 align:middle line:84% which is quite a unique capability that this technique 00:36:00.310 --> 00:36:02.030 align:middle line:90% can have. 00:36:02.030 --> 00:36:06.270 align:middle line:84% So besides for waveguide fabrication, 00:36:06.270 --> 00:36:09.830 align:middle line:84% I also want to mention that this is also a technique that people 00:36:09.830 --> 00:36:13.550 align:middle line:84% actually can use to actually store data in glass materials. 00:36:13.550 --> 00:36:17.070 align:middle line:84% So people are essentially also leveraging 00:36:17.070 --> 00:36:19.950 align:middle line:84% this photoinduced refractive index change 00:36:19.950 --> 00:36:22.330 align:middle line:90% to store digital data. 00:36:22.330 --> 00:36:25.402 align:middle line:90% [VIDEO PLAYBACK] 00:36:25.402 --> 00:36:42.190 align:middle line:90% 00:36:42.190 --> 00:36:44.942 align:middle line:84% JUEJUN HU: So this is a permanent storage. 00:36:44.942 --> 00:36:46.650 align:middle line:84% Once you actually write the index change, 00:36:46.650 --> 00:36:49.025 align:middle line:84% it's going to stay permanent there in the glass material. 00:36:49.025 --> 00:36:54.310 align:middle line:90% 00:36:54.310 --> 00:36:56.328 align:middle line:84% They use this kind of femtosecond laser pulse 00:36:56.328 --> 00:36:58.120 align:middle line:84% because you can achieve very high intensity 00:36:58.120 --> 00:36:59.940 align:middle line:90% in these very short pulses. 00:36:59.940 --> 00:37:03.160 align:middle line:84% And that would allow you to actually confine the interaction 00:37:03.160 --> 00:37:05.960 align:middle line:84% volume to a very small spot, leveraging a nonlinear 00:37:05.960 --> 00:37:06.620 align:middle line:90% absorption. 00:37:06.620 --> 00:37:17.360 align:middle line:90% 00:37:17.360 --> 00:37:19.505 align:middle line:84% So now the advantage of using two-photon absorption 00:37:19.505 --> 00:37:20.880 align:middle line:84% is that you can write essentially 00:37:20.880 --> 00:37:24.422 align:middle line:84% a multi-layer storage, because you 00:37:24.422 --> 00:37:25.880 align:middle line:84% don't have any kind of interference 00:37:25.880 --> 00:37:28.600 align:middle line:84% coming from light absorption above the focal spot. 00:37:28.600 --> 00:37:31.280 align:middle line:84% So you can essentially write multiple layers 00:37:31.280 --> 00:37:34.920 align:middle line:84% of this kind of voxels to increase the storage density. 00:37:34.920 --> 00:37:37.860 align:middle line:84% It's a 3D storage method rather than 2D storage method. 00:37:37.860 --> 00:38:00.302 align:middle line:90% 00:38:00.302 --> 00:38:01.760 align:middle line:84% The motivation there is that people 00:38:01.760 --> 00:38:03.960 align:middle line:84% are looking for essentially permanent storage 00:38:03.960 --> 00:38:06.200 align:middle line:84% media for the huge amount of data, 00:38:06.200 --> 00:38:09.380 align:middle line:84% and this is considered one of the most promising technologies. 00:38:09.380 --> 00:39:08.330 align:middle line:90% 00:39:08.330 --> 00:39:11.140 align:middle line:84% So that's about the difference of fabrication techniques. 00:39:11.140 --> 00:39:12.890 align:middle line:84% Any questions about that before we move on 00:39:12.890 --> 00:39:17.070 align:middle line:84% to talk about applications of fibers and waveguides? 00:39:17.070 --> 00:39:20.730 align:middle line:90% 00:39:20.730 --> 00:39:22.633 align:middle line:90% All right. 00:39:22.633 --> 00:39:25.050 align:middle line:84% There are lots of applications for this kind of integrated 00:39:25.050 --> 00:39:27.450 align:middle line:90% photonics and optics. 00:39:27.450 --> 00:39:28.970 align:middle line:84% I will talk about communications, 00:39:28.970 --> 00:39:30.503 align:middle line:90% augmented realities. 00:39:30.503 --> 00:39:32.170 align:middle line:84% But some other applications that I'm not 00:39:32.170 --> 00:39:35.050 align:middle line:84% going to mention in this class in detail. 00:39:35.050 --> 00:39:38.890 align:middle line:84% One of them is LiDAR, which refers to a light detection 00:39:38.890 --> 00:39:40.410 align:middle line:90% and ranging. 00:39:40.410 --> 00:39:44.130 align:middle line:84% It's similar to radar, which radar uses radio wave. 00:39:44.130 --> 00:39:46.930 align:middle line:90% LiDAR actually uses laser. 00:39:46.930 --> 00:39:49.370 align:middle line:84% So what it does essentially is send a laser beam or laser 00:39:49.370 --> 00:39:52.170 align:middle line:84% pulse, and then the laser pulse is 00:39:52.170 --> 00:39:54.690 align:middle line:84% going to get reflected if you see the object. 00:39:54.690 --> 00:39:57.790 align:middle line:84% And then, by collecting the reflected pulse, 00:39:57.790 --> 00:40:00.930 align:middle line:84% you can then basically calculate what 00:40:00.930 --> 00:40:03.370 align:middle line:90% is the distance of the object. 00:40:03.370 --> 00:40:07.130 align:middle line:84% So it gives you a way to perform 3D mapping of the surrounding 00:40:07.130 --> 00:40:10.170 align:middle line:84% environment with very high accuracy by scanning a laser 00:40:10.170 --> 00:40:11.210 align:middle line:90% beam. 00:40:11.210 --> 00:40:13.810 align:middle line:84% So LiDAR has attracted increasing attention 00:40:13.810 --> 00:40:18.250 align:middle line:84% in recent years because of the autonomous vehicles. 00:40:18.250 --> 00:40:21.130 align:middle line:84% It's considered as a way to actually get 00:40:21.130 --> 00:40:23.970 align:middle line:84% very precise three-dimensional data of the surrounding 00:40:23.970 --> 00:40:27.170 align:middle line:84% environments, with a lot higher resolution, much higher 00:40:27.170 --> 00:40:29.970 align:middle line:90% resolution compared to radar. 00:40:29.970 --> 00:40:35.370 align:middle line:84% And also a lot more precise and with much less 00:40:35.370 --> 00:40:38.850 align:middle line:84% ambiguity compared to direct imaging or cameras. 00:40:38.850 --> 00:40:42.310 align:middle line:84% So that's why there's a lot of attention to LiDAR there. 00:40:42.310 --> 00:40:44.630 align:middle line:84% But they are used not just for autonomous vehicles. 00:40:44.630 --> 00:40:47.210 align:middle line:84% They're used in robotics, used for UAVs. 00:40:47.210 --> 00:40:50.300 align:middle line:84% They're using for remote sensing. 00:40:50.300 --> 00:40:52.100 align:middle line:84% People can also perform chemical analysis 00:40:52.100 --> 00:40:54.240 align:middle line:84% using essentially laser spectroscopy techniques, 00:40:54.240 --> 00:40:55.500 align:middle line:90% basically based on LiDAR. 00:40:55.500 --> 00:40:58.180 align:middle line:84% So it's really attracting lots of attention in recent years. 00:40:58.180 --> 00:41:03.100 align:middle line:84% And the motivation there is that conventional LiDAR essentially 00:41:03.100 --> 00:41:05.620 align:middle line:84% uses mechanical components to rotate the laser 00:41:05.620 --> 00:41:07.560 align:middle line:90% beam to scan the environment. 00:41:07.560 --> 00:41:10.380 align:middle line:84% That's why, if you look at in the early days, when Google 00:41:10.380 --> 00:41:13.540 align:middle line:84% has their autonomous vehicles, before Waymo, the older 00:41:13.540 --> 00:41:15.620 align:middle line:84% vehicles, on the top of vehicle, they 00:41:15.620 --> 00:41:17.300 align:middle line:90% have little rotating things. 00:41:17.300 --> 00:41:18.280 align:middle line:90% That's actually LiDAR. 00:41:18.280 --> 00:41:21.740 align:middle line:84% It's actually scanning the laser beam by mechanical rotation. 00:41:21.740 --> 00:41:22.940 align:middle line:90% That's very expensive. 00:41:22.940 --> 00:41:25.700 align:middle line:84% It cost tens of thousands of US dollars. 00:41:25.700 --> 00:41:28.560 align:middle line:84% And that's why this adoption has been severely hampered. 00:41:28.560 --> 00:41:30.660 align:middle line:84% But in recent years, people have actually 00:41:30.660 --> 00:41:34.380 align:middle line:84% developed ways to replace that whole big mechanical rotating 00:41:34.380 --> 00:41:36.400 align:middle line:84% LiDAR with just a small photonic chip, 00:41:36.400 --> 00:41:38.140 align:middle line:84% essentially performing the same function. 00:41:38.140 --> 00:41:41.000 align:middle line:84% And that's why it's getting lots of interest 00:41:41.000 --> 00:41:43.580 align:middle line:84% now for autonomous driving industry. 00:41:43.580 --> 00:41:46.260 align:middle line:84% Sensing, you can use light to actually detect 00:41:46.260 --> 00:41:48.220 align:middle line:84% the presence of different kinds of molecules, 00:41:48.220 --> 00:41:50.330 align:middle line:90% by doing spectroscopy analysis. 00:41:50.330 --> 00:41:52.580 align:middle line:84% So that's also one big area that's actually attracting 00:41:52.580 --> 00:41:54.260 align:middle line:90% lots of attention recently. 00:41:54.260 --> 00:41:56.740 align:middle line:84% Now what I want to do, I want to focus on communications 00:41:56.740 --> 00:41:58.000 align:middle line:90% and augmented reality. 00:41:58.000 --> 00:42:01.620 align:middle line:84% So starting off by talking about optical communication systems. 00:42:01.620 --> 00:42:03.140 align:middle line:84% So I'm showing you this Great Wall 00:42:03.140 --> 00:42:04.660 align:middle line:84% here because this is what I consider 00:42:04.660 --> 00:42:06.820 align:middle line:84% as one of the oldest optical communication systems 00:42:06.820 --> 00:42:08.500 align:middle line:90% in the world. 00:42:08.500 --> 00:42:11.740 align:middle line:84% Its operation principle is very simple. 00:42:11.740 --> 00:42:15.680 align:middle line:84% So if there's no fire, that means that it's peace. 00:42:15.680 --> 00:42:16.740 align:middle line:90% Nothing happens. 00:42:16.740 --> 00:42:20.500 align:middle line:84% If there's actually invasion, people will light it up. 00:42:20.500 --> 00:42:22.320 align:middle line:90% And essentially you can see it. 00:42:22.320 --> 00:42:24.880 align:middle line:84% So now, essentially, it means that something happens. 00:42:24.880 --> 00:42:29.500 align:middle line:84% So this is basically a 1-bit or binary optical communication 00:42:29.500 --> 00:42:31.220 align:middle line:90% system. 00:42:31.220 --> 00:42:32.740 align:middle line:90% All right. 00:42:32.740 --> 00:42:36.500 align:middle line:84% So nowadays, state-of-the-art communication systems still 00:42:36.500 --> 00:42:38.880 align:middle line:84% operates on more or less the same principle, 00:42:38.880 --> 00:42:41.560 align:middle line:84% except that people do it a lot faster and more efficient. 00:42:41.560 --> 00:42:44.660 align:middle line:84% You can still encode digital signals, 1's and 0's, by turning 00:42:44.660 --> 00:42:46.460 align:middle line:90% light on and off, on and off. 00:42:46.460 --> 00:42:47.500 align:middle line:90% If it's on, it's 1. 00:42:47.500 --> 00:42:48.460 align:middle line:90% Off, it's 0. 00:42:48.460 --> 00:42:50.740 align:middle line:90% This is called on-off keying. 00:42:50.740 --> 00:42:56.080 align:middle line:84% It's the most simple form of doing a digital encoding. 00:42:56.080 --> 00:42:57.660 align:middle line:90% On-Off Keying, OOK. 00:42:57.660 --> 00:43:00.420 align:middle line:90% 00:43:00.420 --> 00:43:02.740 align:middle line:84% I mean, people have developed more advanced ways 00:43:02.740 --> 00:43:05.140 align:middle line:84% to encode information on optical chip, not just 00:43:05.140 --> 00:43:06.320 align:middle line:90% only the on-and-off light. 00:43:06.320 --> 00:43:07.840 align:middle line:84% You can essentially, for example, 00:43:07.840 --> 00:43:11.600 align:middle line:84% divide up light intensity into different levels 00:43:11.600 --> 00:43:14.112 align:middle line:84% so that when you send one optical pulse, 00:43:14.112 --> 00:43:16.320 align:middle line:84% you can actually encode multiple bits of information, 00:43:16.320 --> 00:43:18.920 align:middle line:90% not just like one bit, 1 and 0. 00:43:18.920 --> 00:43:21.540 align:middle line:84% You can also change the phase of the light 00:43:21.540 --> 00:43:23.660 align:middle line:84% or change the polarization of the light. 00:43:23.660 --> 00:43:25.500 align:middle line:84% All of those parameters of photons 00:43:25.500 --> 00:43:27.320 align:middle line:84% can be used to encode digital information. 00:43:27.320 --> 00:43:28.700 align:middle line:90% So it can be a lot more complex. 00:43:28.700 --> 00:43:29.800 align:middle line:90% But the idea is the same. 00:43:29.800 --> 00:43:32.820 align:middle line:84% You essentially turn light or switch light essentially 00:43:32.820 --> 00:43:36.820 align:middle line:84% very fast and use that to actually convey information. 00:43:36.820 --> 00:43:40.020 align:middle line:84% So now, if you look into optical communication system, 00:43:40.020 --> 00:43:42.360 align:middle line:84% you need to have two basic components. 00:43:42.360 --> 00:43:45.470 align:middle line:84% You need to have a transmitter which 00:43:45.470 --> 00:43:49.150 align:middle line:84% receives electronic signal, the electronic data, 1's and 0's, 00:43:49.150 --> 00:43:52.370 align:middle line:84% and converts it into some kind of optical property 00:43:52.370 --> 00:43:56.190 align:middle line:84% change of the photons, some property change of the photons. 00:43:56.190 --> 00:43:58.310 align:middle line:90% So it can be on or off, simply. 00:43:58.310 --> 00:44:01.870 align:middle line:84% So this kind of encoding unit is what's called transmitter. 00:44:01.870 --> 00:44:04.430 align:middle line:84% It allows you to transfer signal from electronic domain 00:44:04.430 --> 00:44:06.270 align:middle line:90% to the optical domain. 00:44:06.270 --> 00:44:09.070 align:middle line:90% This signal is then sent down-- 00:44:09.070 --> 00:44:11.230 align:middle line:84% this transmitter then sends the signal down 00:44:11.230 --> 00:44:13.870 align:middle line:90% through a fiber or waveguide. 00:44:13.870 --> 00:44:17.150 align:middle line:84% And then, at the other end you have a receiver, which basically 00:44:17.150 --> 00:44:19.470 align:middle line:90% is a high-speed photodetector. 00:44:19.470 --> 00:44:21.710 align:middle line:84% It detects a signal and then converts it back 00:44:21.710 --> 00:44:23.430 align:middle line:90% into a basically voltage. 00:44:23.430 --> 00:44:24.230 align:middle line:90% High-low voltage. 00:44:24.230 --> 00:44:26.270 align:middle line:90% Digital signals. 00:44:26.270 --> 00:44:29.950 align:middle line:84% So if you want to do a bidirectional transmission, 00:44:29.950 --> 00:44:32.510 align:middle line:84% you want to send signals in both directions, 00:44:32.510 --> 00:44:35.030 align:middle line:84% you then need to combine transmitter and receiver 00:44:35.030 --> 00:44:38.230 align:middle line:84% in one single unit so you can both send and receive signal. 00:44:38.230 --> 00:44:42.490 align:middle line:84% And that's what's called a transceiver. 00:44:42.490 --> 00:44:46.410 align:middle line:84% A transceiver is basically a transmitter plus a receiver. 00:44:46.410 --> 00:44:52.110 align:middle line:90% 00:44:52.110 --> 00:44:53.250 align:middle line:90% Transceiver. 00:44:53.250 --> 00:44:56.870 align:middle line:90% 00:44:56.870 --> 00:44:57.450 align:middle line:90% All right. 00:44:57.450 --> 00:45:00.530 align:middle line:84% So these transceivers, what I'm showing you here, 00:45:00.530 --> 00:45:03.190 align:middle line:84% this is a little picture of one example of optical transceiver. 00:45:03.190 --> 00:45:05.430 align:middle line:84% It can all be made on a small chip essentially using 00:45:05.430 --> 00:45:07.470 align:middle line:84% this integrated circuit technologies. 00:45:07.470 --> 00:45:11.670 align:middle line:90% 00:45:11.670 --> 00:45:18.130 align:middle line:90% Now, why is that important? 00:45:18.130 --> 00:45:21.550 align:middle line:84% Why is this kind of optical communication of light 00:45:21.550 --> 00:45:24.150 align:middle line:84% to actually send information, why is it important? 00:45:24.150 --> 00:45:26.230 align:middle line:84% So if you go back a few decades, people 00:45:26.230 --> 00:45:28.110 align:middle line:84% send information pretty much exclusively 00:45:28.110 --> 00:45:30.790 align:middle line:90% through copper cables. 00:45:30.790 --> 00:45:33.470 align:middle line:84% By essentially directly sending voltage 00:45:33.470 --> 00:45:35.350 align:middle line:90% signals across copper cables. 00:45:35.350 --> 00:45:39.470 align:middle line:84% What we are seeing today is that these copper cables 00:45:39.470 --> 00:45:42.540 align:middle line:84% are getting increasingly replaced by optical fibers 00:45:42.540 --> 00:45:43.400 align:middle line:90% and waveguides. 00:45:43.400 --> 00:45:44.340 align:middle line:90% So why is that? 00:45:44.340 --> 00:45:48.520 align:middle line:84% I'm showing you this slide basically from Intel. 00:45:48.520 --> 00:45:53.240 align:middle line:84% So vertical axis is basically the different hierarchies 00:45:53.240 --> 00:45:54.540 align:middle line:90% of the optical links. 00:45:54.540 --> 00:45:56.040 align:middle line:84% I mean, for hierarchy, I essentially 00:45:56.040 --> 00:45:58.040 align:middle line:84% mean more or less they [? call it a ?] distance. 00:45:58.040 --> 00:46:01.080 align:middle line:84% So for this one, this is basically 00:46:01.080 --> 00:46:02.940 align:middle line:84% long-distance optical communications. 00:46:02.940 --> 00:46:05.400 align:middle line:84% You are sending, for example, signals through a fiber 00:46:05.400 --> 00:46:06.600 align:middle line:90% to your home. 00:46:06.600 --> 00:46:08.430 align:middle line:84% So this is a communication distance 00:46:08.430 --> 00:46:10.680 align:middle line:84% of the order of hundreds of meters or kilometers, tens 00:46:10.680 --> 00:46:11.555 align:middle line:90% of kilometers meters. 00:46:11.555 --> 00:46:14.320 align:middle line:84% Now if you go to shorter distances, 00:46:14.320 --> 00:46:15.780 align:middle line:90% this could be rack to rack. 00:46:15.780 --> 00:46:17.840 align:middle line:84% This could be from one rack of servers 00:46:17.840 --> 00:46:20.700 align:middle line:84% to another rack of servers, for example, inside a data center. 00:46:20.700 --> 00:46:22.880 align:middle line:90% So this is a shorter distance. 00:46:22.880 --> 00:46:24.460 align:middle line:90% You can go to the extreme. 00:46:24.460 --> 00:46:26.240 align:middle line:84% The very extreme is you are trying 00:46:26.240 --> 00:46:29.503 align:middle line:84% to send signals from one chip to another chip on the same board. 00:46:29.503 --> 00:46:30.920 align:middle line:84% This could be, for example, you're 00:46:30.920 --> 00:46:32.920 align:middle line:84% trying to send signal from essentially 00:46:32.920 --> 00:46:37.040 align:middle line:84% your CPU to the memory, which are only a few centimeters apart 00:46:37.040 --> 00:46:38.480 align:middle line:90% on the board. 00:46:38.480 --> 00:46:43.958 align:middle line:84% So what we also see on the horizontal axis 00:46:43.958 --> 00:46:45.500 align:middle line:84% here-- the vertical axis is distance, 00:46:45.500 --> 00:46:51.060 align:middle line:84% horizontal axis is basically time as well as the bandwidth. 00:46:51.060 --> 00:46:52.720 align:middle line:90% How fast you send your signal. 00:46:52.720 --> 00:46:56.460 align:middle line:84% So this is quite an outdated chart, as I mentioned. 00:46:56.460 --> 00:46:59.720 align:middle line:84% If you look back 20 years, you are talking about 3 gigabit 00:46:59.720 --> 00:47:03.040 align:middle line:90% per second transmission speed. 00:47:03.040 --> 00:47:08.240 align:middle line:84% This, of course, increases exponentially over time. 00:47:08.240 --> 00:47:15.535 align:middle line:84% And what we are seeing here is that the optics would gradually 00:47:15.535 --> 00:47:20.920 align:middle line:84% penetrate into the domain where, traditionally, people use copper 00:47:20.920 --> 00:47:22.800 align:middle line:90% cables to do communication. 00:47:22.800 --> 00:47:27.040 align:middle line:84% So rule of thumb is that for longer distances, 00:47:27.040 --> 00:47:29.240 align:middle line:90% optics is better. 00:47:29.240 --> 00:47:30.080 align:middle line:90% Why? 00:47:30.080 --> 00:47:31.712 align:middle line:90% Because you have long distances. 00:47:31.712 --> 00:47:33.920 align:middle line:84% In order to send signals through a copper cable, what 00:47:33.920 --> 00:47:35.640 align:middle line:84% you have to do, you essentially have 00:47:35.640 --> 00:47:37.570 align:middle line:90% to charge up the optical cable. 00:47:37.570 --> 00:47:40.330 align:middle line:84% That's how you send voltage signals across the copper cable. 00:47:40.330 --> 00:47:44.130 align:middle line:84% Now, the longer the cable is, you need a higher voltage 00:47:44.130 --> 00:47:45.230 align:middle line:90% to properly charge it up. 00:47:45.230 --> 00:47:47.030 align:middle line:84% You need to put more charges into it. 00:47:47.030 --> 00:47:48.610 align:middle line:84% The capacitance actually increases. 00:47:48.610 --> 00:47:50.770 align:middle line:84% So you actually need to put more charges in order 00:47:50.770 --> 00:47:53.410 align:middle line:90% to efficiently charge it up. 00:47:53.410 --> 00:47:56.650 align:middle line:84% Also, if the cable gets longer, you 00:47:56.650 --> 00:47:59.890 align:middle line:90% have more electrical resistance. 00:47:59.890 --> 00:48:04.160 align:middle line:84% So you also have to waste more energy there. 00:48:04.160 --> 00:48:10.950 align:middle line:84% So bottom line-- the longer the cable, the less energy efficient 00:48:10.950 --> 00:48:13.090 align:middle line:84% it's going to be for electrical communication. 00:48:13.090 --> 00:48:17.690 align:middle line:84% So that's why you see that several decades back, people 00:48:17.690 --> 00:48:20.570 align:middle line:84% have already replaced copper cables 00:48:20.570 --> 00:48:22.310 align:middle line:84% for long-haul telecommunications. 00:48:22.310 --> 00:48:25.490 align:middle line:84% You are trying to send signals over kilometers. 00:48:25.490 --> 00:48:28.650 align:middle line:84% Decades ago, people are already using light. 00:48:28.650 --> 00:48:32.750 align:middle line:84% But also, if you go to higher and higher bandwidth, 00:48:32.750 --> 00:48:35.490 align:middle line:84% essentially faster and faster signal transmission speed, 00:48:35.490 --> 00:48:37.530 align:middle line:90% optics also has an advantage. 00:48:37.530 --> 00:48:39.550 align:middle line:84% This also comes from the same argument here. 00:48:39.550 --> 00:48:41.210 align:middle line:84% When you are trying to do communication 00:48:41.210 --> 00:48:44.370 align:middle line:84% through copper cables, you need to charge it up. 00:48:44.370 --> 00:48:47.750 align:middle line:84% But if you are trying to go to very high-speed communications, 00:48:47.750 --> 00:48:50.530 align:middle line:84% you need to essentially charge and discharge the copper cable 00:48:50.530 --> 00:48:53.010 align:middle line:90% at very high speed. 00:48:53.010 --> 00:48:54.690 align:middle line:84% This becomes quite energy efficient 00:48:54.690 --> 00:48:57.810 align:middle line:84% because any cables always has an inherent capacitance associated 00:48:57.810 --> 00:49:00.170 align:middle line:90% with it. 00:49:00.170 --> 00:49:02.185 align:middle line:84% So if you are trying to charge and discharge 00:49:02.185 --> 00:49:03.810 align:middle line:84% at a very high speed, it means that you 00:49:03.810 --> 00:49:06.210 align:middle line:84% need to spend lots of energy just trying to charge 00:49:06.210 --> 00:49:08.530 align:middle line:90% and discharge this cable there. 00:49:08.530 --> 00:49:11.090 align:middle line:84% That becomes very, very inefficient 00:49:11.090 --> 00:49:15.330 align:middle line:84% when you need to do this repeatedly at very high speed. 00:49:15.330 --> 00:49:17.750 align:middle line:84% Well, for optics, you don't have that limitation here. 00:49:17.750 --> 00:49:19.850 align:middle line:84% You can send the pulse through long distances 00:49:19.850 --> 00:49:22.450 align:middle line:84% as long as the waveguide or fiber is low loss. 00:49:22.450 --> 00:49:25.850 align:middle line:84% Your energy consumption barely depends on the distance 00:49:25.850 --> 00:49:27.170 align:middle line:90% you send the signal through. 00:49:27.170 --> 00:49:29.970 align:middle line:84% And it also barely depends on how fast 00:49:29.970 --> 00:49:31.290 align:middle line:90% you modulate the signal. 00:49:31.290 --> 00:49:34.210 align:middle line:90% You can always use the-- 00:49:34.210 --> 00:49:37.030 align:middle line:84% in fact, you can use less amount of energy 00:49:37.030 --> 00:49:39.750 align:middle line:84% if you go to higher speed in some cases. 00:49:39.750 --> 00:49:43.470 align:middle line:84% So as a result, what you're seeing here, 00:49:43.470 --> 00:49:47.010 align:middle line:84% the general trend is that as you go to higher and higher speed 00:49:47.010 --> 00:49:50.490 align:middle line:84% communication, optics start to penetrate 00:49:50.490 --> 00:49:54.010 align:middle line:84% into a shorter and shorter distance communication links. 00:49:54.010 --> 00:49:58.170 align:middle line:90% 00:49:58.170 --> 00:50:01.030 align:middle line:84% I mean, despite that this graph is somewhat outdated, 00:50:01.030 --> 00:50:04.170 align:middle line:84% this general trend is what I'm trying to show here. 00:50:04.170 --> 00:50:05.930 align:middle line:84% Optics is becoming more important 00:50:05.930 --> 00:50:08.370 align:middle line:84% as you go to higher high-speed communication. 00:50:08.370 --> 00:50:11.230 align:middle line:84% So now we are not talking about a few gigabit per second. 00:50:11.230 --> 00:50:16.970 align:middle line:84% We are talking about, in fact, in this case, 00:50:16.970 --> 00:50:21.290 align:middle line:84% terabit per second per communication channel nowadays. 00:50:21.290 --> 00:50:23.890 align:middle line:84% And the capacity for optical communication 00:50:23.890 --> 00:50:25.907 align:middle line:84% can also be further boosted by using 00:50:25.907 --> 00:50:27.490 align:middle line:84% a technique called wavelength-division 00:50:27.490 --> 00:50:30.716 align:middle line:90% multiplexing. 00:50:30.716 --> 00:50:34.160 align:middle line:84% So it's not just one single channel 00:50:34.160 --> 00:50:35.520 align:middle line:90% the data speed is increasing. 00:50:35.520 --> 00:50:39.820 align:middle line:84% You can also essentially throw in tens or hundreds of channels 00:50:39.820 --> 00:50:41.460 align:middle line:90% into one single optical fiber. 00:50:41.460 --> 00:50:44.750 align:middle line:84% And the reason this wavelength-division multiplexing 00:50:44.750 --> 00:50:46.500 align:middle line:84% works is essentially different wavelengths 00:50:46.500 --> 00:50:48.083 align:middle line:84% of light or different colors of light, 00:50:48.083 --> 00:50:53.020 align:middle line:84% they propagate independently in a fiber. 00:50:53.020 --> 00:50:56.060 align:middle line:84% This goes back to essentially so-called linearity 00:50:56.060 --> 00:50:58.012 align:middle line:90% of the system. 00:50:58.012 --> 00:50:59.720 align:middle line:84% When the light intensity is not too high, 00:50:59.720 --> 00:51:01.595 align:middle line:84% you don't have any kind of nonlinear effects, 00:51:01.595 --> 00:51:02.980 align:middle line:90% like two-photon absorption. 00:51:02.980 --> 00:51:05.060 align:middle line:84% Then if it's in the linear domain, then 00:51:05.060 --> 00:51:07.980 align:middle line:84% all these different wavelengths, there was very little crosstalk 00:51:07.980 --> 00:51:09.225 align:middle line:90% in between them. 00:51:09.225 --> 00:51:11.100 align:middle line:84% They will essentially propagate independently 00:51:11.100 --> 00:51:13.700 align:middle line:84% as they travel down the fiber or waveguide. 00:51:13.700 --> 00:51:16.860 align:middle line:84% So therefore, what you can do is that you can now 00:51:16.860 --> 00:51:22.300 align:middle line:84% use a large number of closely packed wavelength 00:51:22.300 --> 00:51:26.020 align:middle line:84% where each of these wavelengths you can independently encode 00:51:26.020 --> 00:51:28.500 align:middle line:90% one channel of information. 00:51:28.500 --> 00:51:31.340 align:middle line:84% So by doing so, you can use the same fiber 00:51:31.340 --> 00:51:35.580 align:middle line:84% to achieve much higher capacitance or communication 00:51:35.580 --> 00:51:36.200 align:middle line:90% capacity. 00:51:36.200 --> 00:51:38.822 align:middle line:84% So this is one example of how people 00:51:38.822 --> 00:51:40.780 align:middle line:84% do essentially wavelength-division multiplexing 00:51:40.780 --> 00:51:42.020 align:middle line:90% of optical fibers. 00:51:42.020 --> 00:51:45.220 align:middle line:84% You can pack this large number of channels near 00:51:45.220 --> 00:51:47.120 align:middle line:90% the 1,550-nanometer wavelength. 00:51:47.120 --> 00:51:49.402 align:middle line:84% This is where you have a low-loss fiber channel. 00:51:49.402 --> 00:51:50.860 align:middle line:84% So you can essentially squeeze lots 00:51:50.860 --> 00:51:54.500 align:middle line:84% of channels around this wavelength range here. 00:51:54.500 --> 00:51:56.160 align:middle line:84% I'm just showing you one example here. 00:51:56.160 --> 00:51:59.242 align:middle line:84% This is actually what people have done a few years back, 00:51:59.242 --> 00:52:00.700 align:middle line:84% where they were able to demonstrate 00:52:00.700 --> 00:52:05.180 align:middle line:84% that one single optical fiber was able to transmit over 00:52:05.180 --> 00:52:08.140 align:middle line:90% a petabyte signal per second. 00:52:08.140 --> 00:52:11.460 align:middle line:84% A petabyte is 10 to the 15th bits per second-- 00:52:11.460 --> 00:52:17.320 align:middle line:84% 10 to 15 bits, which means that it's over a million gigabits. 00:52:17.320 --> 00:52:20.340 align:middle line:84% So you are sending over a million gigabits signal 00:52:20.340 --> 00:52:22.760 align:middle line:84% through one single optical fiber per second. 00:52:22.760 --> 00:52:27.180 align:middle line:90% 00:52:27.180 --> 00:52:31.430 align:middle line:84% So the way they can achieve this remarkably high information 00:52:31.430 --> 00:52:36.990 align:middle line:84% capacity is that they actually use WDM techniques. 00:52:36.990 --> 00:52:41.590 align:middle line:84% They use a total of 382 wavelength 00:52:41.590 --> 00:52:44.110 align:middle line:84% as a separate channel to transmit information. 00:52:44.110 --> 00:52:45.890 align:middle line:84% In fact, they also use modes as well. 00:52:45.890 --> 00:52:48.150 align:middle line:84% They actually use a multi-mode fiber. 00:52:48.150 --> 00:52:51.830 align:middle line:84% They were able to selectively excite each mode here 00:52:51.830 --> 00:52:54.270 align:middle line:84% so that each mode can be used as an independent channel 00:52:54.270 --> 00:52:56.070 align:middle line:90% to actually send information. 00:52:56.070 --> 00:52:58.470 align:middle line:84% We mentioned that multi-mode fiber usually 00:52:58.470 --> 00:53:00.845 align:middle line:84% is not preferred for long-distance communication. 00:53:00.845 --> 00:53:03.470 align:middle line:84% That's in the case where all the modes are essentially randomly 00:53:03.470 --> 00:53:04.485 align:middle line:90% excited in the fiber. 00:53:04.485 --> 00:53:06.110 align:middle line:84% In that case, all the modes essentially 00:53:06.110 --> 00:53:07.235 align:middle line:90% travel at different speeds. 00:53:07.235 --> 00:53:10.310 align:middle line:84% Then you cannot have a very good signal integrity. 00:53:10.310 --> 00:53:12.950 align:middle line:84% However, in this case, they developed a way 00:53:12.950 --> 00:53:16.150 align:middle line:84% to very selectively excite each order mode 00:53:16.150 --> 00:53:18.950 align:middle line:84% so that each mode corresponds to one specific information 00:53:18.950 --> 00:53:20.005 align:middle line:90% channel. 00:53:20.005 --> 00:53:21.630 align:middle line:84% So that's why they were able to achieve 00:53:21.630 --> 00:53:23.830 align:middle line:84% this very impressive communication capacity 00:53:23.830 --> 00:53:27.250 align:middle line:84% number using essentially one single optical fiber. 00:53:27.250 --> 00:53:29.830 align:middle line:90% 00:53:29.830 --> 00:53:30.330 align:middle line:90% All right. 00:53:30.330 --> 00:53:32.690 align:middle line:84% So bottom line is that using light, 00:53:32.690 --> 00:53:34.950 align:middle line:84% you can achieve a much higher information transmission 00:53:34.950 --> 00:53:37.110 align:middle line:84% capacity compared to copper cables. 00:53:37.110 --> 00:53:41.570 align:middle line:84% Now, in terms of the examples of this kind of communication, 00:53:41.570 --> 00:53:44.550 align:middle line:84% this is one example which has been very widely available. 00:53:44.550 --> 00:53:46.712 align:middle line:84% You probably heard about Verizon fiber to the home. 00:53:46.712 --> 00:53:48.670 align:middle line:84% This is a technique that, in fact, in my house, 00:53:48.670 --> 00:53:50.503 align:middle line:84% we actually use a Verizon fiber to the home. 00:53:50.503 --> 00:53:55.230 align:middle line:84% So they have the so-called Tri Play service. 00:53:55.230 --> 00:53:56.790 align:middle line:84% They use one single optical fiber 00:53:56.790 --> 00:53:58.730 align:middle line:84% sent to your house, connected to your house. 00:53:58.730 --> 00:54:00.810 align:middle line:84% You can have essentially high-definition TV. 00:54:00.810 --> 00:54:02.542 align:middle line:90% You can have your phone service. 00:54:02.542 --> 00:54:04.750 align:middle line:84% You also have your internet data all through that one 00:54:04.750 --> 00:54:06.750 align:middle line:90% single optical fiber. 00:54:06.750 --> 00:54:10.550 align:middle line:84% So to see how that works, I'm going to actually show you 00:54:10.550 --> 00:54:12.510 align:middle line:90% a little quick video here. 00:54:12.510 --> 00:54:13.170 align:middle line:90% Let me see. 00:54:13.170 --> 00:54:17.850 align:middle line:90% 00:54:17.850 --> 00:54:18.350 align:middle line:90% Oh. 00:54:18.350 --> 00:54:21.750 align:middle line:90% 00:54:21.750 --> 00:54:22.870 align:middle line:90% Hold on. 00:54:22.870 --> 00:54:24.578 align:middle line:90% [VIDEO PLAYBACK] 00:54:24.578 --> 00:54:26.370 align:middle line:84% JUEJUN HU: Now when you look at this video, 00:54:26.370 --> 00:54:29.370 align:middle line:84% try to catch what is the wavelength. 00:54:29.370 --> 00:54:31.502 align:middle line:84% - You'll get your guaranteed maximum refund. 00:54:31.502 --> 00:54:33.710 align:middle line:84% JUEJUN HU: Try to catch what is the wavelength that's 00:54:33.710 --> 00:54:35.990 align:middle line:90% actually referred to here. 00:54:35.990 --> 00:54:38.650 align:middle line:84% - Actually, it's a variety Fios tech. 00:54:38.650 --> 00:54:41.110 align:middle line:84% It brings fiber optics right to your door 00:54:41.110 --> 00:54:43.370 align:middle line:84% on three different spectrums of light. 00:54:43.370 --> 00:54:49.630 align:middle line:84% You got your 1,310 your 1,490, and the 1,550. 00:54:49.630 --> 00:54:52.290 align:middle line:84% - The light is so clean it's plus 20 dB hot. 00:54:52.290 --> 00:54:55.070 align:middle line:90% It's true quantum. 00:54:55.070 --> 00:54:57.563 align:middle line:90% - You should see his truck. 00:54:57.563 --> 00:54:58.350 align:middle line:90% [END PLAYBACK] 00:54:58.350 --> 00:54:59.010 align:middle line:90% JUEJUN HU: OK. 00:54:59.010 --> 00:55:01.005 align:middle line:84% AUDIENCE: What a knowledgeable kid. 00:55:01.005 --> 00:55:01.630 align:middle line:90% JUEJUN HU: Yes. 00:55:01.630 --> 00:55:04.870 align:middle line:84% Well, he knows the wavelength, obviously. 00:55:04.870 --> 00:55:07.490 align:middle line:84% So how many wavelength number have you heard about? 00:55:07.490 --> 00:55:08.070 align:middle line:90% Three. 00:55:08.070 --> 00:55:10.372 align:middle line:90% 1,310, 1,490, 1,550. 00:55:10.372 --> 00:55:12.830 align:middle line:84% So if you look at how they actually divide up the different 00:55:12.830 --> 00:55:16.790 align:middle line:84% wavelength band here, you see that for 1,310-- 00:55:16.790 --> 00:55:19.830 align:middle line:84% so if you look at how they divide up the band here, 00:55:19.830 --> 00:55:23.500 align:middle line:84% for 1,310, they use it essentially for upstream, 00:55:23.500 --> 00:55:26.560 align:middle line:84% essentially uploading of voice and data. 00:55:26.560 --> 00:55:28.560 align:middle line:84% So this includes your phone service as well as 00:55:28.560 --> 00:55:29.960 align:middle line:90% internet data upload. 00:55:29.960 --> 00:55:38.280 align:middle line:84% And 1,490, this is the waveband used for voice and data 00:55:38.280 --> 00:55:39.480 align:middle line:90% download. 00:55:39.480 --> 00:55:42.440 align:middle line:84% So this is, again, for your phone and your internet service. 00:55:42.440 --> 00:55:47.280 align:middle line:84% At 1,550, they use it for high-definition TV. 00:55:47.280 --> 00:55:50.318 align:middle line:84% Now for TV, you only need to download information. 00:55:50.318 --> 00:55:51.860 align:middle line:84% You don't need to upload information. 00:55:51.860 --> 00:55:55.640 align:middle line:84% That's why there's only one single channel there. 00:55:55.640 --> 00:55:58.480 align:middle line:84% Now if you look at the three wavebands, 00:55:58.480 --> 00:56:00.780 align:middle line:84% 1,550 is right at the low-loss band. 00:56:00.780 --> 00:56:03.400 align:middle line:84% 1,490 is also pretty close to the low-loss band here. 00:56:03.400 --> 00:56:04.340 align:middle line:90% This is 1,550. 00:56:04.340 --> 00:56:05.800 align:middle line:90% This is 1,490. 00:56:05.800 --> 00:56:09.080 align:middle line:84% 1,310 is used because this is also a low-loss transmission 00:56:09.080 --> 00:56:10.680 align:middle line:90% window in the fiber. 00:56:10.680 --> 00:56:13.780 align:middle line:84% Turns out, this is also the so-called minimum dispersion 00:56:13.780 --> 00:56:14.280 align:middle line:90% window. 00:56:14.280 --> 00:56:17.200 align:middle line:84% It means that, essentially, the chromatic dispersion 00:56:17.200 --> 00:56:19.520 align:middle line:84% is actually minimum at this particular window here. 00:56:19.520 --> 00:56:22.640 align:middle line:84% So that's why 1,310 is also a pretty commonly used wavelength 00:56:22.640 --> 00:56:24.480 align:middle line:84% for these kind of communications. 00:56:24.480 --> 00:56:26.440 align:middle line:84% But what you are seeing here is essentially 00:56:26.440 --> 00:56:28.920 align:middle line:84% one example where people can use different wavelengths 00:56:28.920 --> 00:56:33.080 align:middle line:84% to transmit different channels of data independently. 00:56:33.080 --> 00:56:33.580 align:middle line:90% All right. 00:56:33.580 --> 00:56:36.640 align:middle line:84% So that's for long-distance fiber-optic communication. 00:56:36.640 --> 00:56:39.240 align:middle line:84% As I mentioned, over time, optic communication 00:56:39.240 --> 00:56:42.320 align:middle line:84% is increasingly penetrating into shorter-reach communication 00:56:42.320 --> 00:56:43.240 align:middle line:90% links. 00:56:43.240 --> 00:56:45.743 align:middle line:84% So one such example is actually data centers. 00:56:45.743 --> 00:56:47.160 align:middle line:84% So what I'm showing you here, this 00:56:47.160 --> 00:56:49.302 align:middle line:84% is actually a photo of a Google data center. 00:56:49.302 --> 00:56:51.760 align:middle line:84% So if you go on to Google, you try to search for something, 00:56:51.760 --> 00:56:54.360 align:middle line:84% like search for cat photo, then what's 00:56:54.360 --> 00:56:56.240 align:middle line:84% going to happen there is then your query 00:56:56.240 --> 00:56:58.600 align:middle line:84% is going to send one to the data centers. 00:56:58.600 --> 00:57:00.280 align:middle line:84% The server is then going to process 00:57:00.280 --> 00:57:02.720 align:middle line:84% that, now try to fetch information essentially 00:57:02.720 --> 00:57:05.560 align:middle line:84% from either this data center or other data centers 00:57:05.560 --> 00:57:06.600 align:middle line:90% around the globe. 00:57:06.600 --> 00:57:08.937 align:middle line:84% So it's a pretty complex operation 00:57:08.937 --> 00:57:11.020 align:middle line:84% for every search you're actually performing there. 00:57:11.020 --> 00:57:12.800 align:middle line:84% There's lots of internet data traffic 00:57:12.800 --> 00:57:14.080 align:middle line:90% that's actually involved. 00:57:14.080 --> 00:57:16.640 align:middle line:84% So to efficiently transmit information 00:57:16.640 --> 00:57:18.282 align:middle line:84% between the different servers here, 00:57:18.282 --> 00:57:20.490 align:middle line:84% that's where actually optical communication comes in. 00:57:20.490 --> 00:57:24.370 align:middle line:84% So nowadays, for communicating information between one 00:57:24.370 --> 00:57:28.368 align:middle line:84% of these big server racks to another server rack, 00:57:28.368 --> 00:57:29.910 align:middle line:84% people used to be using copper cable. 00:57:29.910 --> 00:57:32.650 align:middle line:84% But nowadays, exclusively, pretty much, people only 00:57:32.650 --> 00:57:35.650 align:middle line:84% use optical fibers to do this kind of communication. 00:57:35.650 --> 00:57:40.210 align:middle line:84% And people are also-- one example 00:57:40.210 --> 00:57:44.570 align:middle line:84% for doing this kind of rack-to-rack communication here, 00:57:44.570 --> 00:57:47.370 align:middle line:84% people use a technology called active optical cables. 00:57:47.370 --> 00:57:50.930 align:middle line:90% 00:57:50.930 --> 00:57:53.630 align:middle line:84% So an active optical cable, from the outside, 00:57:53.630 --> 00:57:56.130 align:middle line:84% it looks exactly like a copper cable. 00:57:56.130 --> 00:58:01.010 align:middle line:84% Its connector would be exactly the same as a copper cable. 00:58:01.010 --> 00:58:03.250 align:middle line:84% So from the outside, it can be used essentially 00:58:03.250 --> 00:58:05.150 align:middle line:90% in a technology-agnostic way. 00:58:05.150 --> 00:58:07.610 align:middle line:84% You can plug it, use it to connect different server 00:58:07.610 --> 00:58:10.050 align:middle line:84% racks without having to worry about whether it's actually 00:58:10.050 --> 00:58:12.890 align:middle line:84% a light inside or it's actually electrons running inside. 00:58:12.890 --> 00:58:15.130 align:middle line:84% But if you look inside this connector 00:58:15.130 --> 00:58:18.872 align:middle line:84% here, what it contains here, this is not a copper cable. 00:58:18.872 --> 00:58:21.330 align:middle line:84% It has actually several optical fibers running in the cable 00:58:21.330 --> 00:58:21.968 align:middle line:90% here. 00:58:21.968 --> 00:58:23.510 align:middle line:84% And also, in one of these connectors, 00:58:23.510 --> 00:58:27.170 align:middle line:84% it actually contains essentially a photonic transceiver. 00:58:27.170 --> 00:58:30.310 align:middle line:84% It converts the electrical signal you get from the socket 00:58:30.310 --> 00:58:31.930 align:middle line:84% and converts it to optical signal. 00:58:31.930 --> 00:58:34.703 align:middle line:84% You then send it through optical fibers. 00:58:34.703 --> 00:58:36.370 align:middle line:84% In this case, in this particular example 00:58:36.370 --> 00:58:39.930 align:middle line:84% here, there are actually four independent fiber channels. 00:58:39.930 --> 00:58:43.570 align:middle line:84% And there are actually four transceivers. 00:58:43.570 --> 00:58:46.090 align:middle line:84% So this allows you to essentially 00:58:46.090 --> 00:58:48.930 align:middle line:84% increase the communication bandwidth by four-fold 00:58:48.930 --> 00:58:53.130 align:middle line:84% because there are four fibers actually involved. 00:58:53.130 --> 00:58:53.670 align:middle line:90% OK. 00:58:53.670 --> 00:58:55.770 align:middle line:84% So this is one technology that people commonly 00:58:55.770 --> 00:58:58.570 align:middle line:84% use in data centers to connect different racks using 00:58:58.570 --> 00:59:03.205 align:middle line:84% essentially what they call active optical cables. 00:59:03.205 --> 00:59:04.830 align:middle line:84% This technology has been pretty mature. 00:59:04.830 --> 00:59:08.210 align:middle line:84% It has been around for past decade. 00:59:08.210 --> 00:59:13.850 align:middle line:84% In more recent years, due to the huge surge of data 00:59:13.850 --> 00:59:18.540 align:middle line:84% requirements, especially related to the emergence of AI, 00:59:18.540 --> 00:59:21.900 align:middle line:84% people are now trying to push this interconnect 00:59:21.900 --> 00:59:24.100 align:middle line:90% to even shorter length scales. 00:59:24.100 --> 00:59:25.680 align:middle line:90% So it's not just rack to rack. 00:59:25.680 --> 00:59:28.420 align:middle line:84% People are now talking to essentially communications 00:59:28.420 --> 00:59:31.220 align:middle line:84% between the different chips, different GPUs, for example. 00:59:31.220 --> 00:59:34.500 align:middle line:84% So in this slide here I'm showing you 00:59:34.500 --> 00:59:38.060 align:middle line:84% the architectural tradition for essentially a communication 00:59:38.060 --> 00:59:40.700 align:middle line:84% between a large AI computing node. 00:59:40.700 --> 00:59:45.140 align:middle line:84% So the key takeaway message here is that AI is actually 00:59:45.140 --> 00:59:46.980 align:middle line:90% very communication-intensive. 00:59:46.980 --> 00:59:52.100 align:middle line:84% So for many AI operations, you want to run the task in parallel 00:59:52.100 --> 00:59:53.485 align:middle line:90% across many GPUs. 00:59:53.485 --> 00:59:55.860 align:middle line:84% But in order to be able to do that, you want to make sure 00:59:55.860 --> 00:59:57.443 align:middle line:84% that all the GPUs, they are constantly 00:59:57.443 --> 00:59:59.300 align:middle line:90% communicating with each other. 00:59:59.300 --> 01:00:04.700 align:middle line:84% So in lots of cases, essentially the efficiency 01:00:04.700 --> 01:00:08.340 align:middle line:84% and the speed of computation is not limited by the GPU itself. 01:00:08.340 --> 01:00:10.860 align:middle line:84% The bottleneck is actually in the communication-- 01:00:10.860 --> 01:00:13.060 align:middle line:84% how well the GPUs can talk to each other. 01:00:13.060 --> 01:00:16.187 align:middle line:84% So conventionally, this is the conventional architecture 01:00:16.187 --> 01:00:18.020 align:middle line:84% about how people actually interlink together 01:00:18.020 --> 01:00:20.380 align:middle line:90% different GPUs. 01:00:20.380 --> 01:00:22.540 align:middle line:84% You essentially have copper cables 01:00:22.540 --> 01:00:26.260 align:middle line:84% trying to link all the GPUs into one network switch. 01:00:26.260 --> 01:00:28.260 align:middle line:84% The switch essentially is a device 01:00:28.260 --> 01:00:30.220 align:middle line:84% that allows you to create different connections 01:00:30.220 --> 01:00:32.880 align:middle line:84% and reconfigure the connections between the different units. 01:00:32.880 --> 01:00:36.860 align:middle line:84% And then this switch, then you have optical fibers 01:00:36.860 --> 01:00:40.060 align:middle line:84% to use it to connect to essentially other racks. 01:00:40.060 --> 01:00:46.100 align:middle line:84% So this is not very efficient because as you can tell, 01:00:46.100 --> 01:00:47.660 align:middle line:84% you have this kind of middleman here. 01:00:47.660 --> 01:00:49.962 align:middle line:84% In order to communicate from one GPU to another, 01:00:49.962 --> 01:00:52.420 align:middle line:84% you always have to go through this kind of a network switch 01:00:52.420 --> 01:00:54.700 align:middle line:84% here, which is done electrically. 01:00:54.700 --> 01:00:57.200 align:middle line:84% So as you go to higher data speed, 01:00:57.200 --> 01:01:00.540 align:middle line:84% as you go to longer distances, it becomes very inefficient. 01:01:00.540 --> 01:01:03.180 align:middle line:84% So what people are trying to push for nowadays 01:01:03.180 --> 01:01:06.540 align:middle line:84% is they want all the GPUs to be able to directly talk 01:01:06.540 --> 01:01:10.300 align:middle line:84% with each other through light, rather than through copper 01:01:10.300 --> 01:01:13.260 align:middle line:90% traces or copper cables. 01:01:13.260 --> 01:01:19.180 align:middle line:84% So that's why now people are now investigating 01:01:19.180 --> 01:01:22.980 align:middle line:84% a technology of so-called co-packaged optics. 01:01:22.980 --> 01:01:25.780 align:middle line:84% So without going into too much details, what 01:01:25.780 --> 01:01:27.860 align:middle line:84% I'm showing you here is essentially 01:01:27.860 --> 01:01:31.700 align:middle line:84% people are trying to push optics closer and closer 01:01:31.700 --> 01:01:32.960 align:middle line:90% to the electronic chip. 01:01:32.960 --> 01:01:37.100 align:middle line:90% 01:01:37.100 --> 01:01:40.460 align:middle line:84% So the earlier generation of optical communication 01:01:40.460 --> 01:01:43.200 align:middle line:84% technologies, you essentially will have your ASIC here. 01:01:43.200 --> 01:01:45.340 align:middle line:84% Your ASIC is your Application-Specific Integrated 01:01:45.340 --> 01:01:45.940 align:middle line:90% Circuit. 01:01:45.940 --> 01:01:47.440 align:middle line:84% This could be your GPU, for example. 01:01:47.440 --> 01:01:48.648 align:middle line:90% It could be your memory chip. 01:01:48.648 --> 01:01:51.220 align:middle line:84% So this GPU or memory chip, they will 01:01:51.220 --> 01:01:53.300 align:middle line:84% be mounted on essentially a circuit board, 01:01:53.300 --> 01:01:54.580 align:middle line:90% printed circuit board. 01:01:54.580 --> 01:01:59.020 align:middle line:84% You have copper traces that link these electrical circuits now 01:01:59.020 --> 01:02:01.400 align:middle line:90% to the optical transceiver. 01:02:01.400 --> 01:02:03.860 align:middle line:84% You have essentially a plug here where 01:02:03.860 --> 01:02:06.780 align:middle line:84% you can plug in, for example, your active optical cable. 01:02:06.780 --> 01:02:09.020 align:middle line:84% And then you use the active optical cable 01:02:09.020 --> 01:02:11.230 align:middle line:84% to then communicate to another board. 01:02:11.230 --> 01:02:13.710 align:middle line:84% So the downside about this configuration 01:02:13.710 --> 01:02:16.430 align:middle line:84% here is that the electrical signal still 01:02:16.430 --> 01:02:18.630 align:middle line:84% has to travel through these copper traces 01:02:18.630 --> 01:02:21.610 align:middle line:84% on the printed circuit board for pretty long distances, 01:02:21.610 --> 01:02:26.950 align:middle line:84% like tens of centimeters, or at least a few centimeters. 01:02:26.950 --> 01:02:29.870 align:middle line:84% So as you go to higher and higher data rate, 01:02:29.870 --> 01:02:33.270 align:middle line:84% people are talking about, for example, 100, 400, 800 gigabit. 01:02:33.270 --> 01:02:35.750 align:middle line:84% So with higher data rate, this kind 01:02:35.750 --> 01:02:39.190 align:middle line:84% of information transmission through the copper trace 01:02:39.190 --> 01:02:41.670 align:middle line:90% becomes very energy inefficient. 01:02:41.670 --> 01:02:44.790 align:middle line:84% So what people want to do, as I mentioned, 01:02:44.790 --> 01:02:47.970 align:middle line:84% for electrical communication, you want to make it shorter 01:02:47.970 --> 01:02:51.170 align:middle line:84% so you don't have to charge up very long cables. 01:02:51.170 --> 01:02:53.950 align:middle line:84% So then what people want to do is 01:02:53.950 --> 01:02:57.870 align:middle line:84% try to push these optics closer to the electronic chip 01:02:57.870 --> 01:03:01.990 align:middle line:84% to essentially minimize the length of the copper trace. 01:03:01.990 --> 01:03:04.470 align:middle line:84% And that's why people are now evolving towards what's 01:03:04.470 --> 01:03:07.190 align:middle line:84% called the co-packaged optics, where essentially 01:03:07.190 --> 01:03:11.310 align:middle line:84% the optical transceiver is packaged essentially 01:03:11.310 --> 01:03:15.670 align:middle line:84% in the same package, along with your GPU or your computer chips. 01:03:15.670 --> 01:03:17.370 align:middle line:84% They are all inside one single package. 01:03:17.370 --> 01:03:22.490 align:middle line:90% 01:03:22.490 --> 01:03:26.250 align:middle line:84% So I just want to show you this video here. 01:03:26.250 --> 01:03:27.850 align:middle line:84% This is actually a very recent video. 01:03:27.850 --> 01:03:31.790 align:middle line:84% This is something unveiled by NVIDIA just last month. 01:03:31.790 --> 01:03:34.110 align:middle line:84% They have demonstrated that they can 01:03:34.110 --> 01:03:36.310 align:middle line:84% use this kind of co-packaged optics technologies 01:03:36.310 --> 01:03:38.550 align:middle line:84% to enable some of the most powerful AI engines 01:03:38.550 --> 01:03:39.710 align:middle line:90% in the world. 01:03:39.710 --> 01:03:41.010 align:middle line:90% So let me see. 01:03:41.010 --> 01:03:43.830 align:middle line:90% 01:03:43.830 --> 01:03:46.150 align:middle line:84% I'm hoping the internet connection works. 01:03:46.150 --> 01:03:46.810 align:middle line:90% Come on. 01:03:46.810 --> 01:03:54.268 align:middle line:90% 01:03:54.268 --> 01:03:56.043 align:middle line:90% [VIDEO PLAYBACK] 01:03:56.043 --> 01:03:57.210 align:middle line:90% JUEJUN HU: There's no sound. 01:03:57.210 --> 01:04:28.600 align:middle line:90% 01:04:28.600 --> 01:04:29.700 align:middle line:90% So you see the interface. 01:04:29.700 --> 01:04:33.360 align:middle line:84% They have optical fiber bundles going directly into the package, 01:04:33.360 --> 01:04:35.947 align:middle line:90% into the ASIC. 01:04:35.947 --> 01:04:36.780 align:middle line:90% Going into the ASIC. 01:04:36.780 --> 01:04:48.440 align:middle line:90% 01:04:48.440 --> 01:04:49.540 align:middle line:90% Now this is outside. 01:04:49.540 --> 01:05:08.360 align:middle line:90% 01:05:08.360 --> 01:05:09.540 align:middle line:90% This is outside the server. 01:05:09.540 --> 01:05:12.100 align:middle line:84% You see the large number of fiber connections coming out. 01:05:12.100 --> 01:05:19.440 align:middle line:90% 01:05:19.440 --> 01:05:20.600 align:middle line:90% OK. 01:05:20.600 --> 01:05:23.880 align:middle line:84% So this is what's actually driving today's state-of-the-art 01:05:23.880 --> 01:05:25.200 align:middle line:90% AI computing there. 01:05:25.200 --> 01:05:28.880 align:middle line:84% Essentially, there's lot of optics communication channels 01:05:28.880 --> 01:05:29.880 align:middle line:90% that's involved. 01:05:29.880 --> 01:05:33.600 align:middle line:90% 01:05:33.600 --> 01:05:34.360 align:middle line:90% OK. 01:05:34.360 --> 01:05:37.160 align:middle line:84% So now next application I'm going to mention here 01:05:37.160 --> 01:05:39.240 align:middle line:90% is about augmented reality. 01:05:39.240 --> 01:05:42.680 align:middle line:84% I have this little short video here. 01:05:42.680 --> 01:05:46.666 align:middle line:84% I don't know why the sound is not working here. 01:05:46.666 --> 01:05:50.140 align:middle line:90% [VIDEO PLAYBACK] 01:05:50.140 --> 01:05:51.625 align:middle line:90% 01:05:51.625 --> 01:05:54.000 align:middle line:84% JUEJUN HU: For those of you who are not familiar with AR, 01:05:54.000 --> 01:05:55.640 align:middle line:84% or Augmented Reality, I just have 01:05:55.640 --> 01:06:00.440 align:middle line:84% a very short video that shows some envisioned applications 01:06:00.440 --> 01:06:01.100 align:middle line:90% of AR. 01:06:01.100 --> 01:06:26.000 align:middle line:90% 01:06:26.000 --> 01:06:26.800 align:middle line:90% All right. 01:06:26.800 --> 01:06:31.560 align:middle line:84% So what AR is basically is you have 01:06:31.560 --> 01:06:36.000 align:middle line:84% this kind of see-through glasses where they can be operated just 01:06:36.000 --> 01:06:36.940 align:middle line:90% like normal glasses. 01:06:36.940 --> 01:06:38.380 align:middle line:84% You can see the things around you. 01:06:38.380 --> 01:06:41.280 align:middle line:84% But at the same time, you have essentially this optical gadget 01:06:41.280 --> 01:06:42.660 align:middle line:90% that's actually embedded. 01:06:42.660 --> 01:06:45.440 align:middle line:84% So they can actually project images into your eye, 01:06:45.440 --> 01:06:47.440 align:middle line:84% and it's actually superimposed onto whatever you 01:06:47.440 --> 01:06:48.880 align:middle line:90% are seeing in the surroundings. 01:06:48.880 --> 01:06:52.060 align:middle line:84% So if you look closely into how this augmented reality, 01:06:52.060 --> 01:06:54.400 align:middle line:84% how it works, essentially you always 01:06:54.400 --> 01:06:56.000 align:middle line:90% have a little projector here. 01:06:56.000 --> 01:06:59.420 align:middle line:84% These are microdisplay modules with a very high resolution, 01:06:59.420 --> 01:07:03.730 align:middle line:84% very small display panels placed on the side. 01:07:03.730 --> 01:07:07.090 align:middle line:84% Essentially you have three colors, red, green, and blue. 01:07:07.090 --> 01:07:08.890 align:middle line:84% These three colors display and project 01:07:08.890 --> 01:07:11.610 align:middle line:84% light into these glass waveguides. 01:07:11.610 --> 01:07:13.890 align:middle line:84% You have these sheets of glasses that essentially 01:07:13.890 --> 01:07:16.410 align:middle line:90% serve as waveguides. 01:07:16.410 --> 01:07:18.870 align:middle line:84% To couple the incident light into the waveguides, 01:07:18.870 --> 01:07:21.430 align:middle line:84% you have essentially some coupling elements here. 01:07:21.430 --> 01:07:23.810 align:middle line:84% These are typically diffractive gratings 01:07:23.810 --> 01:07:26.450 align:middle line:84% that allow you to bend the propagation direction of light 01:07:26.450 --> 01:07:30.210 align:middle line:84% and then allow them to couple into these waveguides here. 01:07:30.210 --> 01:07:32.850 align:middle line:84% You have three sheets of waveguides to deliver 01:07:32.850 --> 01:07:34.130 align:middle line:90% three different colors. 01:07:34.130 --> 01:07:37.330 align:middle line:84% And then at the output end, you have some outcoupling elements 01:07:37.330 --> 01:07:37.830 align:middle line:90% here. 01:07:37.830 --> 01:07:40.570 align:middle line:84% Typically these are so-called holographic gratings 01:07:40.570 --> 01:07:44.090 align:middle line:84% which, again, redirect the light from being trapped 01:07:44.090 --> 01:07:46.740 align:middle line:84% inside the waveguide essentially into free space 01:07:46.740 --> 01:07:47.990 align:middle line:90% and then couple into your eye. 01:07:47.990 --> 01:07:50.770 align:middle line:84% So that's why you can actually see the projected images here. 01:07:50.770 --> 01:07:54.570 align:middle line:84% At the same time, any kind of a light coming 01:07:54.570 --> 01:07:56.030 align:middle line:84% from the surrounding environment, 01:07:56.030 --> 01:07:58.690 align:middle line:84% they can also just transmit through this glass deck 01:07:58.690 --> 01:08:00.430 align:middle line:84% with minimal amount of attenuation. 01:08:00.430 --> 01:08:02.690 align:middle line:84% So that's why you can see superimposed images 01:08:02.690 --> 01:08:08.410 align:middle line:84% of this kind of images from this display with the surrounding 01:08:08.410 --> 01:08:09.190 align:middle line:90% environment. 01:08:09.190 --> 01:08:12.730 align:middle line:90% 01:08:12.730 --> 01:08:16.649 align:middle line:84% Now, about why glass material is important for augmented reality 01:08:16.649 --> 01:08:18.210 align:middle line:90% applications. 01:08:18.210 --> 01:08:20.550 align:middle line:90% I hope this YouTube video plays. 01:08:20.550 --> 01:08:22.842 align:middle line:90% [VIDEO PLAYBACK] 01:08:22.842 --> 01:08:23.970 align:middle line:90% 01:08:23.970 --> 01:08:26.649 align:middle line:84% - Experience augmented reality like never 01:08:26.649 --> 01:08:29.689 align:middle line:90% before with SCHOTT RealView. 01:08:29.689 --> 01:08:32.010 align:middle line:84% Augmented reality is merging our view 01:08:32.010 --> 01:08:35.210 align:middle line:84% of the real world with virtual experiences, 01:08:35.210 --> 01:08:39.370 align:middle line:84% and it will be a vital part of our life for daily usage 01:08:39.370 --> 01:08:42.890 align:middle line:90% in business and private life. 01:08:42.890 --> 01:08:47.130 align:middle line:84% The applications include manufacturing through education 01:08:47.130 --> 01:08:48.649 align:middle line:90% to entertainment. 01:08:48.649 --> 01:08:51.490 align:middle line:84% But what role does SCHOTT play in this progress? 01:08:51.490 --> 01:08:54.170 align:middle line:84% SCHOTT RealView empowers the optical engine 01:08:54.170 --> 01:08:57.410 align:middle line:90% of augmented reality devices. 01:08:57.410 --> 01:09:01.100 align:middle line:84% Our glass wafers or substrates for optical waveguides 01:09:01.100 --> 01:09:04.540 align:middle line:84% are produced in an innovative fabrication process. 01:09:04.540 --> 01:09:07.060 align:middle line:84% Layered over one another, ultra-thin glass 01:09:07.060 --> 01:09:09.140 align:middle line:84% slices collect the light waves of the image 01:09:09.140 --> 01:09:11.180 align:middle line:84% from the input structure and guide them 01:09:11.180 --> 01:09:15.020 align:middle line:84% to a similar output structure, projecting the image to the eye. 01:09:15.020 --> 01:09:18.680 align:middle line:84% Our glass waveguides directly impact the picture quality, 01:09:18.680 --> 01:09:20.220 align:middle line:90% simulating reality. 01:09:20.220 --> 01:09:22.340 align:middle line:84% The exceptionally high refractive index 01:09:22.340 --> 01:09:24.359 align:middle line:84% enlarges the total reflection angle, 01:09:24.359 --> 01:09:26.300 align:middle line:90% expanding the field of view. 01:09:26.300 --> 01:09:29.460 align:middle line:84% SCHOTT RealView is robust and durable, but also 01:09:29.460 --> 01:09:32.960 align:middle line:84% thin and lightweight, suited for everyday use. 01:09:32.960 --> 01:09:35.899 align:middle line:84% The excellent glass surface enables high picture quality 01:09:35.899 --> 01:09:38.819 align:middle line:84% and contrast, and the added optical coatings 01:09:38.819 --> 01:09:41.460 align:middle line:90% create excellent transparency. 01:09:41.460 --> 01:09:42.340 align:middle line:90% The result? 01:09:42.340 --> 01:09:45.740 align:middle line:84% The image merges perfectly with the world's real view. 01:09:45.740 --> 01:09:48.660 align:middle line:84% Our waveguide guarantees brilliant image quality 01:09:48.660 --> 01:09:52.540 align:middle line:84% for full, immersive mixed-reality experiences. 01:09:52.540 --> 01:09:55.300 align:middle line:90% [MUSIC PLAYING] 01:09:55.300 --> 01:09:56.220 align:middle line:90% 01:09:56.220 --> 01:09:58.380 align:middle line:84% All these are outstanding qualities 01:09:58.380 --> 01:10:01.100 align:middle line:84% are the result of a specialized know-how 01:10:01.100 --> 01:10:04.780 align:middle line:84% and innovative fabrication processes of the global SCHOTT 01:10:04.780 --> 01:10:05.480 align:middle line:90% network. 01:10:05.480 --> 01:10:08.740 align:middle line:90% 01:10:08.740 --> 01:10:11.420 align:middle line:84% SCHOTT RealView provides manufacturers 01:10:11.420 --> 01:10:15.100 align:middle line:84% with the ability to develop augmented reality devices that 01:10:15.100 --> 01:10:19.690 align:middle line:84% will redefine our view of the world. 01:10:19.690 --> 01:10:21.115 align:middle line:90% [END PLAYBACK] 01:10:21.115 --> 01:10:22.540 align:middle line:90% 01:10:22.540 --> 01:10:23.620 align:middle line:90% JUEJUN HU: Oh. 01:10:23.620 --> 01:10:24.120 align:middle line:90% All right. 01:10:24.120 --> 01:10:27.000 align:middle line:84% I think we actually are running out of time here. 01:10:27.000 --> 01:10:29.780 align:middle line:84% So I will wrap up about the glass application 01:10:29.780 --> 01:10:32.480 align:middle line:84% for augmented reality in the next lecture. 01:10:32.480 --> 01:10:32.980 align:middle line:90% Yeah. 01:10:32.980 --> 01:10:35.680 align:middle line:84% Sorry for the poor image quality here. 01:10:35.680 --> 01:10:38.300 align:middle line:84% That's why we need better optical connections, better 01:10:38.300 --> 01:10:39.080 align:middle line:90% interconnects. 01:10:39.080 --> 01:10:39.580 align:middle line:90% Yeah. 01:10:39.580 --> 01:10:40.660 align:middle line:90% That's why. 01:10:40.660 --> 01:10:41.160 align:middle line:90% OK. 01:10:41.160 --> 01:10:43.530 align:middle line:90% Anyway, thank you. 01:10:43.530 --> 01:11:55.000 align:middle line:90%