WEBVTT 00:00:00.000 --> 00:00:05.600 Behind me is a PC with an all-in-one closed loop cooler to keep the CPU from overheating. 00:00:05.600 --> 00:00:09.760 And unless you're trying to perform an extreme overclock, it's a pretty good solution. 00:00:10.400 --> 00:00:15.680 But even with the best cooler on the market, your CPU will still be quite warm. 00:00:15.680 --> 00:00:20.960 You know what? Why do we need water blocks at all? I mean, look at this processor. 00:00:21.680 --> 00:00:28.160 All of this space inside that black square is dedicated to an integrated heat spreader. 00:00:28.160 --> 00:00:34.480 So would it be possible to instead build water cooling right into the dye itself? 00:00:34.480 --> 00:00:39.920 As it turns out, the answer is yes. And to find out more, we spoke to Ellison Matteoli, 00:00:39.920 --> 00:00:43.920 a professor of electrical engineering at the Swiss Federal Institute of Technology 00:00:43.920 --> 00:00:50.000 in Lausanne, where he and his team recently developed an all-new in-chip cooling solution. 00:00:50.000 --> 00:00:53.120 We'd like to give him a big thank you for taking the time to talk with us. 00:00:53.680 --> 00:00:58.640 So a big part of their approach was to look at the electronics and the thermal management 00:00:58.640 --> 00:01:02.960 as a single problem, rather than the paradigm that we're more used to where 00:01:02.960 --> 00:01:08.000 one company works on making chips and another one works on making heat sinks or water blocks. 00:01:08.560 --> 00:01:13.920 Instead, the idea was to design a chip in concert with an integrated liquid cooling solution 00:01:13.920 --> 00:01:17.280 so that you would already know where the hotspots were. 00:01:17.280 --> 00:01:22.480 To demonstrate this idea, they used a device that generates a lot of heat. In this case, 00:01:22.560 --> 00:01:28.400 a power converter, and then etched water microchannels into a layer directly below the 00:01:28.400 --> 00:01:34.560 gallium nitride chip that handled the power conversion, with small capillaries below hotspots 00:01:34.560 --> 00:01:37.920 and larger channels operating as inflow and outflow pipes. 00:01:38.560 --> 00:01:43.920 The small channels adjacent to the warmer areas of the chip provide lots of surface area for heat, 00:01:43.920 --> 00:01:48.640 making thermal dissipation very efficient compared to traditional solutions involving a 00:01:48.640 --> 00:01:54.400 separate heat spreader, heat sink, and thermal paste. Remember guys, you lose a lot of efficiency 00:01:54.400 --> 00:02:01.520 with each step. The chip only got up to 60 degrees Celsius compared to the blistering 250 degrees 00:02:01.520 --> 00:02:06.160 that we would have seen without cooling. In fact, their solution ended up being able to extract 00:02:06.160 --> 00:02:13.760 1.7 kilowatts per square centimeter using only half a watt of pumping power. Conveniently, 00:02:13.840 --> 00:02:18.960 a square centimeter is roughly the same size as a modern processor die, 00:02:18.960 --> 00:02:26.080 and even high-end processors have TDPs in the neighborhood of just 125 to 250 watts. 00:02:26.080 --> 00:02:32.000 So with a solution this efficient, you could cool a chip down to near room temperature, 00:02:32.000 --> 00:02:36.880 especially as the process would work just as well with more familiar silicon processors. 00:02:36.880 --> 00:02:41.520 I mean, the amount of heat that a typical chip puts out is a whole order of magnitude less 00:02:41.520 --> 00:02:47.520 than what in-chip cooling could handle. But hold on a second, is this some kind of pipe dream 00:02:47.520 --> 00:02:53.040 that's just never going to leave the laboratory like DNA storage, unless it does eventually? 00:02:53.040 --> 00:02:58.480 Or could we actually see this in our desktop rigs in the near future? Well, it turns out it would 00:02:58.480 --> 00:03:04.400 actually be fairly trivial for a big chip maker to replicate this microchannel process in a standard 00:03:04.400 --> 00:03:09.280 clean room, and the research team in Switzerland has already had a number of companies approach them 00:03:09.280 --> 00:03:14.240 about using this technology in the coming years. Now, the most obvious benefit will be 00:03:14.240 --> 00:03:19.920 for server farms and data centers, where huge amounts of money are already being spent to 00:03:19.920 --> 00:03:26.160 keep energy costs low and, well, processor temperatures lower. However, there's no reason 00:03:26.160 --> 00:03:30.480 at all that we couldn't see it in our home gaming computer at some point, which could allow for 00:03:30.480 --> 00:03:36.000 significant increases in overclocking for enthusiasts, or maybe even better gaming on 00:03:36.000 --> 00:03:42.000 mobile devices or small form factor PCs, since this in-chip solution doesn't need nearly as much 00:03:42.000 --> 00:03:47.120 liquid defunction correctly as a traditional water cooling loop. Understandably, the lab couldn't 00:03:47.120 --> 00:03:52.720 tell us much about exactly which companies were inquiring about in-chip cooling, but that shouldn't 00:03:52.720 --> 00:03:58.240 stop you or us from dreaming about the possibilities of having a water-cooled smartwatch one day. 00:03:58.880 --> 00:04:04.000 Why not, right? Thanks for watching, guys! Like, dislike, check out our other videos, 00:04:04.000 --> 00:04:08.560 leave a comment if you have a suggestion for a future Fast as Possible, and don't forget to 00:04:08.560 --> 00:04:12.880 subscribe, because if you do, we're gonna keep changing our sets every time! What do you guys 00:04:12.880 --> 00:04:17.360 think of this? Do you like it better than the green screen? We'd love to hear your feedback!