WEBVTT 00:00:00.770 --> 00:00:03.790 And now we've reached the end of 6.004. 00:00:03.790 --> 00:00:08.200 Looking back, there are two ways of thinking about the material we've discussed, the skills 00:00:08.200 --> 00:00:12.379 we've practiced, and the designs we've completed. 00:00:12.379 --> 00:00:16.680 Starting at devices, we've worked our way up the design hierarchy, each level serving 00:00:16.680 --> 00:00:19.190 as building blocks for the next. 00:00:19.190 --> 00:00:23.380 Along the way we thought about design tradeoffs, choosing the alternatives that would make 00:00:23.380 --> 00:00:30.370 our systems reliable, efficient and easy to understand and hence easy to maintain. 00:00:30.370 --> 00:00:36.640 In the other view of 6.004, we created and then used a hierarchy of engineering abstractions 00:00:36.640 --> 00:00:41.330 that are reasonably independent of the technologies they encapsulate. 00:00:41.330 --> 00:00:46.100 Even though technologies change at a rapid pace, these abstractions embody principles 00:00:46.100 --> 00:00:47.730 that are timeless. 00:00:47.730 --> 00:00:53.780 For example, the symbolic logic described by George Boole in 1847 is still used to reason 00:00:53.780 --> 00:00:58.989 about the operation of digital circuits you and I design today. 00:00:58.989 --> 00:01:03.490 The power of engineering abstractions is that they allow us to reason about the behavior 00:01:03.490 --> 00:01:08.930 of a system based on the behavior of the components without having to understand the implementation 00:01:08.930 --> 00:01:11.590 details of each component. 00:01:11.590 --> 00:01:16.329 The advantage of viewing components as "black boxes" implementing some specified function 00:01:16.329 --> 00:01:23.170 is that the implementation can change as long as the same specification is satisfied. 00:01:23.170 --> 00:01:28.450 In my lifetime, the size of a 2-input NAND gate has shrunk by 10 orders of magnitude, 00:01:28.450 --> 00:01:33.439 yet a 50-year-old logic design would still work as intended if implemented in today's 00:01:33.439 --> 00:01:35.689 technologies. 00:01:35.689 --> 00:01:40.369 Imagine trying to build a circuit that added two binary numbers if you had to reason about 00:01:40.369 --> 00:01:44.819 the electrical properties of doped silicon and conducting metals. 00:01:44.819 --> 00:01:49.509 Using abstractions lets us limit the design complexity at each level, shortening design 00:01:49.509 --> 00:01:54.600 time and making it easier to verify that the specifications have been met. 00:01:54.600 --> 00:01:58.719 And once we've created a useful repertoire of building blocks, we can use them again 00:01:58.719 --> 00:02:02.099 and again to assemble many different systems. 00:02:02.099 --> 00:02:07.590 Our goal in 6.004 is to demystify how computers work, starting with MOSFETs and working our 00:02:07.590 --> 00:02:09.919 way up to operating systems. 00:02:09.919 --> 00:02:14.290 We hope you've understood the engineering abstractions we've introduced and had a chance 00:02:14.290 --> 00:02:19.280 to practice using them when completing the design problems offered in the labs. 00:02:19.280 --> 00:02:24.200 We also hope that you'll also understand their limitations and have the confidence to create 00:02:24.200 --> 00:02:28.250 new abstractions when tackling new engineering tasks. 00:02:28.250 --> 00:02:37.160 Good engineers use abstractions, but great engineers create them. 00:02:37.160 --> 00:02:40.640 principles used at each level of the design hierarchy. 00:02:40.640 --> 00:02:45.150 If a particular topic struck you as especially interesting, we hope you'll seek out a more 00:02:45.150 --> 00:02:49.900 advanced course that will let you dig deeper into that engineering discipline. 00:02:49.900 --> 00:02:54.120 Hundreds of thousands of engineers have worked to create the digital systems that are the 00:02:54.120 --> 00:02:56.320 engines of today's information society. 00:02:56.320 --> 00:03:02.760 As you can imagine, there's no end of interesting engineering to explore and master - so roll 00:03:02.760 --> 00:03:06.090 up your sleeves and come join in the fun! 00:03:06.090 --> 00:03:09.090 What will be the engineering challenges of tomorrow? 00:03:09.090 --> 00:03:12.990 Here are a few thoughts about how the future of computing may be very different than the 00:03:12.990 --> 00:03:13.990 present. 00:03:13.990 --> 00:03:17.260 The systems we build today have a well-defined notion of state: 00:03:17.260 --> 00:03:22.200 the exact digital values stored in their memories, produced by their logic components, and traveling 00:03:22.200 --> 00:03:24.390 along their interconnect. 00:03:24.390 --> 00:03:29.090 But computation based on the principles of quantum mechanics may allow us to solve what 00:03:29.090 --> 00:03:34.990 are now intractable problems using states described not as collections of 1's and 0's, 00:03:34.990 --> 00:03:41.440 but as interrelated probabilities that describe the superposition of many states. 00:03:41.440 --> 00:03:45.650 We've built our systems using voltages to encode information and voltage-controlled 00:03:45.650 --> 00:03:51.100 switches to perform computation, using silicon-based electrical devices. 00:03:51.100 --> 00:03:55.350 But the chemistry of life has been carrying out detailed manufacturing operations for 00:03:55.350 --> 00:04:01.220 millennia using information encoded as sequences of amino acids. 00:04:01.220 --> 00:04:06.000 Some of the information encoded in our DNA has been around for millions of years, a truly 00:04:06.000 --> 00:04:08.900 long-lived information system! 00:04:08.900 --> 00:04:13.360 Today biologists are starting to build computational components from biological materials. 00:04:13.360 --> 00:04:18.529 Maybe in 50 years instead of plugging in your laptop, you'll have to feed it :) 00:04:18.529 --> 00:04:23.740 Instead of using truth tables and logic functions, some computations are best performed neural 00:04:23.740 --> 00:04:28.380 networks that operate by forming appropriately weighted combinations of analog inputs, 00:04:28.380 --> 00:04:32.400 where the weights are learned by the system as it is trained using example inputs that 00:04:32.400 --> 00:04:35.229 should produce known outputs. 00:04:35.229 --> 00:04:39.499 Artificial neural nets are thought to model the operation of the synapses and neurons 00:04:39.499 --> 00:04:41.110 in our brains. 00:04:41.110 --> 00:04:46.010 As we learn more about how the brain operates, we may get many new insights into how to implement 00:04:46.010 --> 00:04:51.159 systems that are good at recognition and reasoning. 00:04:51.159 --> 00:04:56.919 Again using living organisms as useful models, programming may be replaced by learning, where 00:04:56.919 --> 00:05:00.629 stimulus and feedback are used to evolve system behavior. 00:05:00.629 --> 00:05:05.999 In other words, systems will use adaptation mechanisms to evolve the desired functionality 00:05:05.999 --> 00:05:09.270 rather than have it explicitly programmed. 00:05:09.270 --> 00:05:13.860 This all seems the stuff of science fiction, but I suspect our parents feel the same way 00:05:13.860 --> 00:05:18.250 about having conversations with Siri about tomorrow's weather. 00:05:18.250 --> 00:05:21.050 Thanks for joining us here in 6.004. 00:05:21.050 --> 00:05:25.009 We've enjoyed presenting the material and challenging you with design tasks to exercise 00:05:25.009 --> 00:05:26.789 your new skills and understanding. 00:05:26.789 --> 00:05:31.539 There are interesting times ahead in the world of digital systems and we can certainly use 00:05:31.539 --> 00:05:34.719 your help in inventing the future! 00:05:34.719 --> 00:05:38.629 We'd welcome any feedback you have about the course so please feel free leave comments 00:05:38.629 --> 00:05:39.729 in the forum. 00:05:39.729 --> 00:05:43.069 Good bye for now... and good luck in your future studies.