The Secret World of Ones and Zeroes: Moore's Law Explained

[@Thomas-ev4gm]

"Oh quantum mechanics you enchanting treacherous minx" 6:30 this is the best thing I've heard today

[@bbnbby2594]

I'm living in 2020 and we've achieved 7nm already. 5nm from TSMC is a work in progress, also Moore's Law is going to end , probably

[@nadirjofas3140]

In marketing terms we have.

[@jlp1528]

Came to the comments to say something like this. It used to be that X nm really did mean what you'd expect: that's the size of an individual transistor. Now the marketing games have gotten so crazy that Intel has abandoned units entirely and are just gonna call it the Intel 4 process...

[@abhijiths5237]

It still isnt completely dead

[@jonathanmichel7398]

​@@jlp1528aaaaaaaaaaaaaaaaaaa

[@ElijahDecker]

"Any sufficiently advanced technology is indistinguishable from magic." -Clarke's Third Law Therefore my sufficiently advanced glowing box of a monitor is magic. Huzzah!

[@paulmahoney7619]

Well, to someone 100 years ago, our computers would seem like magic, likely, technology from 100 years in the future would seem like magic to us.

[@ElijahDecker]

+Paul Mahoney Hence, our technology is sufficiently advanced compared to that of 100 years ago, though I would suspect it would depend on who you talked to. A scientist or engineer from the year 1916 would probably understand the basic operating principles of most of our technology if you explained it to them. However, a layperson from 1916 would probably consider it magic.

[@surajtiwari2614]

Nice quote.

[@ImDemonAlchemist]

No, just because two things are indistinguishable, that doesn't mean they are actually the same. Just that you can't tell whether it is one or the other.

[@Lexcepcion]

I fucking love computer science it's so intriguing

[@wick9427]

Not as much as your grammar though.

[@davidchall7684]

+ThaRealArabDude I believe you meant to say, "Please jump in front of a bus."

[@3800S1]

+ThaRealArabDude But AMD and Intel no longer use a front side bus anymore.

[@wick9427]

I love this comment thread!

[@Landrew0]

That sounds a bit extreme.

[@__-tz2yx]

I love this because after 3 years of programming classes and 6+ years of hanging out primarily with computer geeks, this is the first time anyone has ever explained transistors and how binary code works with hardware to me in a way that actually makes sense. Everyone before was always just like "oh it just sends ones and zeros" or "it turns on and off"" and considered that helpful when really it explains absolutely nothing. So thank you so much for explaining the actual "how" of it

[@NoblePineapples]

Good news everyone! We've hit 7 nm. IBM was the first to create a 7nm chip with 20 billion transistors.. I love technology.

[@ProfessionalGasLighting]

Hey, I’m here from the year 2021. Cerebras released a wafer scale engine chip with 2.6 trillion transistors.

[@peanuts2105]

just seen a KZbin advert for Intel. How fittingly appropriate

[@totinospizzarolls4737]

???

[@fallenup19]

+Colin "I have just seen.." is perfectly fine grammatically. He just cut the "i have" bit off because if you speak fast you wouldn't really use it

[@superperfectstranger815]

no because Intel had 22nm gaps in 2014, baytrail and ivybrindge and now has 15nm or so gaps so soon we get to much electronic noise and no more Intel crap😂

[@luke]

this is the scishow episode i have always been waiting for. thank you for finally making this :D

[@Kikilang60]

This is a science fan boy video. All this information would been informative thirty years ago. Why Moores law failed, is market forces are fucused elsewhere. Everyone is buying, and spending money on smartphone. Desk top computers are like fax machines, and pagers. No one wants them anymore. The big money is on making chips that don't use power. You might have desk top, but go to store, and tell me how many desk tops you see? Like all fan boys, he sounds like he's joking, but he really likes the idea of Lazer computers. The big money was on Quantum computers, not stupid light/photon computers. Light computer went out, with the turn of the millenium. Unless some crazy ass app come out of nowhere, desk tops are like fax machines, only people who need them, for specific needs will be buying them. This is the White Elephant in the room. Your personal electronic device is very personal. Your phone, or lap top is your personal property. It's like underwear, or a tooth brush, no one want to share their smartphone, it's just creepy, and violate your personal space. A desk top computer is like a dining room table. Everyone is using that thing, so don't do the nasty stuff on it. Eat, maybe homework, nothing gross. The money has left the cpu for computers.

[@thezuck1596]

So, when he got to the part where 32nm processors were small, I got confused until I looked at the year this was uploaded. Absolutely crazy how technology evolves, considering we are at 7nm processors as of now.

[@FloozieOne]

I am a dedicated SciShow follower so I don't know how I missed this one but I sure am glad I found it. As an X-ray tech I got kind of familiar with photons, (although I could never get one to slow down enough to actually pet it). I also studied computer science at Tufts University in Mass. for a year but my husband got lung cancer and I had to drop out to take care of him. So I have a little programming and a little physics but basically no hardware knowledge. Now I do thanks to you! The light of lasers has finally penetrated the "off" position in my brain switched it to "on".

[@asmi06]

As electronic chip designer by education, I gotta say it's a VERY simplified explanation. In reality open transistor (in so called "gate mode") would consume insane amount of power since it's resistance is very low, so engineers came up with the trick to put two transistors into each gate, such that at every moment of time one of them is always closed. Such structure is called "complementary pair" (which is where name CMOS - Complementary Metal-Oxide-Semiconductor - comes from), and is very energy-efficient, as it only consumes power during transitions from 0 to 1 and back. Each complementary pair performs logical function "NOT", and it's possible to implement any logical function using just that basic element.

[@cinder5lla]

I'm guessing it's simplified in order to appeal to a greater audience (like me), who pretty much switched off after your 4th line of comment. It's impressive and interesting for those who have an avid interest in such things but for everyone else who aren't in the profession of designing chips, we have Hank for the lowdown.

[@Falcrist]

Explaining the inner workings of a transistor in detail is beyond the scope of this video, and possibly beyond the interests of the target audience.

[@Brant92M]

AFAIK, transistor gate current is in the range of picoamps. The power dissipation comes from the internal resistance (usually in the milliohm range) when the current is actually passing through the transistor. I'm just an EE student, so I could be wrong *shrug*

[@asmi06]

Brant Martin I was talking about the current flowing through channel when transistor is open. In complementary pair channel current is only present during transitions from open state to closed and vice versa, since transistor have finite performance and can't open or close immediately (this is due to the very principle field-effect transistor - or FET - functions - in order for channel to open/close, conducting particles - either n or p - need to physically move, and as was pointed out in the video, they are pretty slow), so there is a (very short) period of time when both are partially open. That's where most of power consumption (and heat) comes from. Gate-to-channel current in FET is extremely small, although in modern ICs isolator layer between gate and channel is getting so thin that the current coming from tunnel effect becomes a problem. And this layer needs to be thinner to allow lower working voltage, which in turn would reduce heating.

[@Brant92M]

There's no current when the channel is closed? That doesn't sound right to me, there should still be some resistance there. It seems odd that most of the power dissipation would come from switching, but I guess when you're switching at 3GHz+ you're switching most of the time. That brings up some other questions; are we talking FETs or BJTs? I wonder how long the transistor stays closed (or open) per instruction. Hmmm...

[@istvankovasznai]

This is a great video, but Hank got some of his facts outdated or wrong: - Moore's law is just as true today as it was the day it was formulated. Intel's E3-12xx v3 and E5-16xx/26xx v3 Xeons based on the Haswell microarchitecture have 5.56 billion transistors. Some of those models were released one year before this video was created. - The "gap between transistor channels" he is referring to around 5:30 is actually called the "Feature size" and it's not a measure of distance between different elements of the same transistor, but instead refers to the average half-pitch (half the distance between identical semiconductor structures arranged in an array pattern, for instance two adjacent memory cells). The "gap between transistor channels", as he puts it, is about 0.5nm. - Most currently available consumer Intel CPU's are built with a feature size of 22nm since 2012, and the end of 2014 saw the launch of products built on 14nm tech. - Most of the heat generated by microprocessors is not due to the great number of transistors packed close together, but rather to a phenomenon known as "leakage current" - a very small amount of current is flowing through the transistors even when they are turned off. While this leakage is proportional to the number of transistors inside the chip, it is mainly a function of the manufacturing process. Newer manufacturing processes have greatly reduced the leakage current - that's why we can now have cellphones (that run off batteries) with CPUs that rival the processing power of desktops from less than a decade ago (that could be used as space heaters). - Multi-core consumer processing has been around since 2005. Nowadays most battery powered gadget has up to 8 CPU cores, and high-end server CPUs can have as many as 18 - that's 36 threads of code that can run in parallel with HyperThreading, and that's all in just one of potentially many CPUs inside a single system. - Software that makes use of multiple processors has been available for decades in the server space, but since the introduction of consumer multi-core processing, it has caught up nicely to use all the available resources. Most of the tasks that PCs, smartphones and tablets are used for actually lend themselves very well to parallel processing - video and audio decoding/encoding, multi-tab web browsing, anti-malware scanning etc. - sometimes all done at the same time :) (plus he gets bits and bytes mixed up) en.wikipedia.org/wiki/Haswell_(microarchitecture)#Server_processors en.wikipedia.org/wiki/22_nanometer

[@jamesm66]

Istvan Kovasznai While I agree with you better manufacturing processes cannot completely eliminate the heating, there is a fundamental physical limit to do with the irreversibility of the operations done by the computer so this is also a problem. While I know this is far less than the energy used by today's processors what you said seemed to imply that with perfect manufacturing processes there would be no heat produced. en.wikipedia.org/wiki/Landauer%27s_principle Interestingly this is also one of the advantages of quantum computers over classical computers. As the processes they rely on are reversible in nature they do not have this minimum heat production.

[@JF32304]

So what you're saying is if chips had no leakage current they would run cool?

[@sugarfree4073]

Show off

[@istvankovasznai]

@@sugarfree4073 Haha, by the time you posted this reply, half of the stuff I wrote in there is really outdated anyway :)

[@Neceros]

Ahh, quantum mechanics. You make life interesting.

[@dvoicer6785]

and makes physicists sh*t themselves in anger.

[@Mobius14]

dvoicer6785 and they don't shit themselves in anger. ..physics joke btw.

[@danny1111105]

Nobody that is one of the first physics jokes I actually laughed at. better than heisenberg getting lost by being pulled over for speeding.

[@dvoicer6785]

what is the joke? if he said strange that would make sense...

[@SlideRulePirate]

Nobody But they don't know until they check. 'ewww' joke btw.

[@anotherPhilosoraptor]

"Oh, quantum mechanics, you enchanting, treacherous minx." I love this guy. :D

[@sime3250]

Finally ... Someone sayed how that shit works ...

[@sime3250]

32nm a bit outdated but ok

[@Ricardordz11]

+SIME_LP Maybe because this video is 2 years old?...

[@hydrochloricacid2146]

+Ricardordz11 22nm/14nm was already a thing back in 2014

[@3800S1]

+Phoenix Wright yes, but not in the mainstream. Confined to testing and refinement back then. Those are current now in the mainstream. 10 and 7 nm are in development stages in the lab now and some test devices no doubt exist but we won''t see them available to purchase for quite a few years yet.

[@hydrochloricacid2146]

vkorinfsky well ivy bridge and haswell were out ( 22nm ) , and broadwell came out at the very end of the year (14nm )

[@DontMockMySmock]

For starters, transistors do not hold ones and zeros. One and zero are just two different voltage levels; wires hold them. Transistors just allow or deny a one or zero to be transferred from one wire to another - and the cool thing is that they allow or deny based on the voltage in a third wire. So you can have ones and zeros moving around wires based on other ones and zeros in other wires. Your characterizing transistors' on/off states as one/zero is wrong - on means the two wires are connected (both one or both zero), and off means the two wires are disconnected (and might be different). The bit about a byte being in eight transistors is wrong - a byte is in eight wires, which may be connected to transistors. And as for the storage of bits in a computer, well, storing a bit takes (at least) eight transistors wired together in a configuration generally known as a "flip flop" (there are several kinds), so the bit about a byte being in eight transistors is still wrong. For as well as you covered the operation of semiconductor devices, you might as well have said "they're magic." All you did was explain how insulating silicon can become partially conductive if you dope it. Well, we already have conductive things, we call them wires. Why are these particular types of silicon useful? What's the point of talking about doping and stuff without even mentioning p- and n-type semiconductors, or the way they are mashed together to make transistors? Honestly, it's fine if you don't want to explain that stuff - it's complicated and difficult to explain to laypeople who don't have any quantum mechanics and electronics background. But then why talk about doping? Why not simply talk about what transistors do (control the connection between two wires based on the voltage in a third wire) without trying (and failing) to get into the technical details? Also, talking about all those materials, but no love for gallium arsenide? The bit about optical computing made me cringe - the advantage has *nothing* to do with the speed of the constituent particles, it has to do with bandwidth. The speed that electrons drift in a wire is irrelevant to the speed that electrical signals flow, in the same way that the time that it takes the water to get from the water tower to your faucet is irrelevant to how fast the water starts flowing when you turn the knob. There was already water there; a wave of water pressure is pushing it out the faucet. There were already electrons there; a wave of electric field is pushing it. And because it depends on electric fields, electricity travels at the speed of light. Remember that light actually *is* electric fields, so there's really not a lot of conceptual difference between an optical computer and an electronic computer except the frequencies involved are much higher in an optical computer (and the devices and materials you use are different to deal with this). And that's the advantage, too; higher frequency means more bandwidth. Fortunately you did not talk about quantum computing in enough detail for me to get mad at you.

[@allanrempel437]

I appreciate your pedantry. I was just about to write a similar rant, until I saw that you had already covered everything here. I'm not sure if any laypeople will understand or care about these corrections, but at least technical people like us will be satisfied.

[@seriousbees]

yeah the bit about doping was kind of useless. I almost thought he was going to go into structure for a bit there, but nope. otherwise I thought his explanations were ok (excluding the drift velocity thing). He got people thinking of the transistor as a kind of switch, which is basically the core idea of digital hardware, and he sketched out some ways you can use binary data. Good enough for a non-technical approach.

[@miroslavstejskal8123]

Allan Rempel I am a layman in this field and I think I understand most of these remars and I surely do appreciate it.

[@blacktimhoward4322]

Never once has this man felt a vagina

[@DontMockMySmock]

​@@blacktimhoward4322 That's a lot of assumptions to pack into one comment. Here, I can do the same: You must be a really insecure, poorly-educated, brain-dead Neanderthal to associate knowledge of electrical engineering with inability to have sex, and even dumber to assume that a person with knowledge of electrical engineering must be straight and a man. Maybe if you spent some time acting like a mature adult instead of dropping lame junior high school insults on five year old youtube comments, you might eventually learn some stuff too.

[@jameswray50]

I have newfound respect for those who have to program computer parts to read all combination of 1's and 0's.

[@TVTacon]

I.e. People that write compilers in machine code.

[@bombmask]

Nah machine code is Hexadecimal. the CPU's have instructions, It was only in the very very beginning that we ever wrote On off explicitly.

[@lordofudead]

In all honestly not many people can actually read binary. and mind you, the first guy to make a computer did not just go, o ill make this thing do a thing. The current technology we are using is built upon years and decades of other peoples work. eg the first computer screen, the first sound card, the first operating system, the first game engine, eg...out computer runs the work of I dare say thousands or hundreds of thousands of peoples work. Our computers are their legacy....which is kind a cool way of looking at it.

[@mrsecify]

bombmask Nope, machine code is written in binary. Hexadecimal is simply a 16-digit number system, nothing else. So, technically, you could write could write code in base-64 as long as you can later translate it (i.e. compile it) to what a computer can understand.

[@Schindlabua]

Nezara I can teach you how to count in binary in 5 seconds - just count like you would normally and skip any number that contains the digits 2-9. 0,1,10,11,100,101,110,111,1000,1001,.. Likewise, reading binary is not hard either. bombmask Hexadecimal is merely there for convenience. 4 binary bits exactly fit into one Hex digit (2^4=16), so it's a way to make large chunks of binary data easier to work with. Like Hank said in the video, the only thing a computer understands is current/no current, one and zero, which makes machine code inherently binary. There have been experiments with ternary computers though - check out Setun!

[@merubindono]

Solution: make atoms smaller, obviously!

[@moro3797]

6 years after apple os about to blow your mind with a14 proccesor which has 12billion of transistors and preparing for 28 billion.

[@baruchben-david4196]

So simple when you spell it out like that....

[@professionalrose7889]

thats not how that works

[@rikaeriksharethisvid6911]

@@professionalrose7889 It’s a joke

[@surajtiwari2614]

Nice, detailed video. Crisp & clear!

[@randomjasmicisrandom]

As a Computing teacher in an English Secondary school I have just been made very happy! This is the basis for at least one lesson in September!

[@CommentCritic]

Topic about transistors and microchips. *looks at motherboard through case window* "Hmm...."

[@mihalis1010]

Just saying, the Core i7-5960X has 2.6 billion transistors... while that's not 5 billion, we should be seeing that with the release of the Broadwell architecture.

[@istvankovasznai]

The Haswell E3-12xx v3 and E5-16xx/26xx v3 Xeons have 5.56 billion transistors. Some of those models were released one year before this video was created. Most of the information in this video is either hopelessly outdated or simply wrong.

[@Willgtl]

Here's a more correct way of saying what he was trying to say. Transistor tech hasn't surpassed 14nm yet and it should've already (Most mainstream processors are actually 22-20nm). You can cram as many transistors on a chip that you want, but the more you have, the bigger the die will need to be. Take the Ivy Bridge HE-4, it has 8.75m transistors/mm2 on a 160mm die (1.4b transistor count). The Haswell-E5 HCC processor is only 8.6m transistors/mm2 on a massive 662mm2 die (5.69b transistor count). It's speculated we'll get down to 5nm process in 2020 (Simply the so-called "road map" of CMOS technology) but going by the past trend, we should be past that by 2020. Even getting to 10nm process isn't easy. We still don't have the lithography techniques for fabricating such tiny transistors. Unfortunately we don't even know if conventional CMOS technology can go that small. And as said in this video, 10nm is getting to the point where we have to consider quantum mechanics. It's just a big ass chip. Anyone can make a bigger chip, but the bigger the chip, the greater the chance of manufacturing defects and the greater the heat dissipation. The hotter the chip, the greater the gate delay. So bigger chip does not mean faster chip. It means hotter chip that's harder to keep cool and hence actually slower (Relatively). On top of that, because it's so large and harder to cool, it'll have a shorter life. Sun announced their new SPARC M7 processor will have 14b transistors. It'll be on the 20nm process, but that's still gonna mean a big die. However, last year Intel also announced the new Core M processors that will actually use the 14nm process. Better processors are still coming, but in only a couple years we're gonna hit a roadblock where we can't go smaller without moving onto a new semiconductor. My hope is we'll have quantum computers by 2025 (More than just D-Wave's quantum annealing processors).

[@TheEpicMusic161]

willgtl there are actually mobile chips that are 14 nm that released very recently they're in the latest macbook (the super thin one with only one port) and the ASUS UX305 and i think a few others

[@danielmastia87]

World Known Yep, that's right, and there are chips with even more than 5 billion transistors today. I think the "problem" is that it's come up to a point were consumer chips just don't need to follow Moore's Law, CPUs are plenty powerful for most people, so there isn't such a demand for faster ones. Or maybe I'm completely wrong, who knows.

[@TheEpicMusic161]

Daniel Mastía i think that's pretty much correct, i mean what would your average person want with a xeon or a haswell-E processor?

[@xelgringoloco2]

I'm kinda hoping that one day Hank will tell us that it is magic that makes things work...

[@OriginalRAB]

Quantum mechanics is the closest you'll get. Shit may as well be magic.

[@add111497]

OriginalRAB Agreed. That stuff is W-I-E-R-D!

[@AlexanderPavel]

Addison Waege Quantum tunneling scares me... It's like teleportation of particles into oblivion...

[@Bandit5317]

Alexander Pavel Electrical Engineering major here. Quantum tunneling, just like all of quantum mechanics, doesn't make sense when working with traditional physics. Intuitively, it shouldn't happen, but it does. We can calculate the probability of it happening under certain conditions, which is good enough for most engineering purposes, but, as my Semiconductor Devices professor stated, we don't actually know how it works or why it happens. Engineers and Scientists deal with a lot of complex problems, but that was the first and only time a professor told me that we just didn't know how or why something happens. This has actually been a problem in microprocessor design for several years now, since we hit 45nm gates. Since then, it has just been about using the right materials in order to lower the probability that quantum tunneling will occur. The patch has been to use "high-k" insulators.

[@petertang9611]

Bandit5317 Tunneling happens to all waves (ie. sound), basically stating that there is always some transmitted no matter how much the barrier resists it (and the intensity can be calculated). In quantum tunneling the electrons behave like waves (wave-particle duality), with the "wave intensity" being the probability that it is ends up in each specific location. Essentially when nothing is observing them, the electrons become waves, and some of the wave (some electrons) will transmit through the barrier.

[@TasseTina]

hello, i'm training at a giant company and they showed us this video as part of the training and it's the one that has made the most sense to me, thank you for making it!

[@adamgtrap]

This was one of my favorite episodes.

[@Fragmonkei]

We need to adopt the yes/no/maybe system instead

[@Eldorado1239]

There's still another way. We could fold the spacetime around the computer in the opposite direction gravity folds it - the computer could then do a year worth work in minutes. Or rather - we would get the result of the computer working for a year in minutes.

[@ZeNashB]

ElDoRado1239 I like the way you think. Its not hard to imagine a future where we can do manipulations at the planck length with strings and control the quantum gravity and warp spacetime around the transistors so as to achieve the slowed down time and thus, in a way, hack the system to get faster processing in computers. Just wow.

[@iliketrains0pwned]

ElDoRado1239 So, hack space and time so we can hack a computer. That's actually pretty genius.

[@eitkoml]

ElDoRado1239 Good luck building that. If you succeed you'll be one of the richest people on earth, just be sure to have a lawyer teach you how to not get ripped off. If that happened it would be one of the biggest rip offs in the world's history.

[@truthisrelative4815]

eitkoml awesome, just go write a freaking novel on that and in 20 years people will be like hey, that sci-fi shit might actually work. Just like everything else we invent

[@ComputersAreRealCool]

ElDoRado1239 Could you clarify what you mean? Are you suggesting leaving the computer running for years, but bending spacetime to grab the computer years ahead once the calculations start? that's pretty clever, but I get the feeling that it would cause some kind if paradox...

[@thesecondislander]

But the information in a circuit travels at 2/3 the speed of light, right? It is the individual electrons themselves that have a translational velocity of mm/cm per hour

[@LSC132]

drift velocity. yes

[@mathews1515]

Finally, to anyone who cares, the reason different components can interpret data in a different manner (although in a theoretically similar state) is because of 1) the Operating System and 2) the device controller, not to be confused with device driver. If anyone is interested, although it is a technical read, I suggest you look into Hardware Abstraction Layers and Operating System Kernels Sincerely, a friendly Software Engineer:)

[@darlingtonakogo6245]

This' is the Best video I've seen explaining Transistors, Moore's Law and the current limitations of it and processors with diff. possible options, awesome!!

[@maxscholz911]

Yeah boys we are at 7 NM rn

[@isky.h]

Moore, and Moore, and MOoooRe!

[@3800S1]

0:44 Old school 80-90s UV erasable EPROMs, Front one is an AMD brand 27C256 to be exact.

[@lollipopskulltagger]

Good job spotting that, I was wondering why there was a window on it, as I have only read about them and not seen one. Thank you for pointing it out. :)

[@jakehaas4919]

I have never read such a nerdy comment before...

[@Jessie_Helms]

I prefer long videos like this, I can do something while listening& they can actually fully explain things

[@mathews1515]

Hank, recently researchers in Silicon Valley have unveiled the Skyscraper theory (or Skyscraper processor) which uses Carbon Nano Tubes in a series of layers to which looks like... Well... A skyscraper. The skyscraper theory has been around of ages now but the largest hurdle was maintaining a consist temperature, as which you noted, could very easily melt through other components, however Carbon Nano Tubing allows the device to stay much cooler as long as it is in collaboration with other cooling technologies. The device implements a component known as an "Elevator" which allows different floors (or process layers) to communicate data with one one another. This is a significant find as one thing you had not explicitly mentioned was that, indeed several processors can crunch data synchronously or asynchronously a lot of limitations is head in the physical communicate line between the processors (Data Bus).

[@MRSketch09]

So basically were at a point with computers, almost.. we probably have 5 to 10 years to go, but wherever it goes next, its going to be flipping awesome next isn't it?

[@bakintoast]

Yeah, probably quite a bit of transition time, though. Optical computing looks like it's still being researched, so I don't see that coming into play for another 30 years or so. Quantum computers are apparently a thing, they just don't help with the normal computing we use everyday anyway, we would just be using a quantum computer for binary computing. I imagine we'll just be getting better at multiprocessing for a while, and then switch to optical as it becomes viable. Hopefully. Optical computers would be the shit.

[@1213JayStar]

I doubt it. Maybe but not likely. We start using multiple processors in computers. That would mean more efficient programs that utilize these processors and hardware that supports using them in parallel. It's already been done, the biggest problems are price and software using them. This at least saves the whole tedious part of creating an entirely new system of computing data. For now, this is probably the route we'll see taken.

[@theBabyDead]

***** Actually, the real problem is that with multiple cores, they need to communicate. This communication (basically deciding which core does what) slows the whole bunch down. In the beginning we lost 20% efficiency per core. Now we're at about 7% tho. But yeah, that's the actual problem.

[@christiantaylor12]

iVulgarThrust yes but the problem with quantum computers is that they run really deep underground in really cold areas and they are really big they are based on the idea of quantum entanglement and you can only hack them in ONE situation... You break the laws of physics so not able for consumer use but progress is being made

[@TonyStark-dw6xp]

we go where no man has gone before(porn sites with automated masturbation techniques)

[@frostyw]

"Oh, quantum mechanics, you charming, treacherous minx." XD

[@LUXAETERNA6603]

explain quantum computing more please

[@aoifeanonymous]

Quantum computing is based on the principle of superposition - basically a single particle (though not really a particle because those have mass; photon) can exist in multiple states at one given time. For problem solving, this would mean the computer could test all possible scenarios at the same time, or at least test multiple scenarios at the same time. Basically superposition reduces runtime.

[@lordofudead]

Ill give you a run down by what I understand of it (this may be pretty simplified,maybe even wrong at times, i mean, its quantum mechanics, its hard to explain) Ever heard of Schroedinger Cat? This is actually a thought experiment that explains more or less how quantum mechanics work. because things are so small, when we try and scan them we are more or less changing the state by which the exist with whatever we are scanning them with (even light). (think of how our computer searches for data it needs, memory). For example, so we scan a atom once to see the location of its electron, we might see that its above the atom, scan it again and its below, scan it again and its beside it. The data is more or less indicating that electron is everywhere....so while its not being seen, that electron exists all around that atom at exactly the same time. quantum computers more or less use this principle of objects existing in different states by scanning them in different ways. (I think) This pretty much makes any data you create with a quantum computer almost indecipherable and impossible to hack or crack the code of by anyone who did not create that piece of data. ergo, quantum computers are generally used to store important information.... They would be incredibly fast due to the multitude of different states a single electron can exist in, instead of 1 or 0 you could have....well..near infinite I think. 1, 0, 244, 64, 1000, anything. There is also this quantum tunnelling effect hank mentioned. This more or less means distance is not an issue and things literally jump (wormhole style) distances. This would indicate that distance/speed is never an issue when dealing with quantum states or computers, because everything can be everywhere, at once. (trippy as hell I know) This also kinda touches on the issue that Quantum Physics and well...Physics dont get along so well, they literally play by different rules.

[@EmanuelMay]

Nezara Well.... Kinda. QC is way more complicated and at the same time way simpler than you describing it to be. Please let me correct you on a few things: Quantum Mechanics is based on the principle, that a state of a particle is absolutely not known, until you observe it. Somehow they have all possible states at the same time, until you check which exactly it is. This is called Superposition. After you checked them, they keep their state pretty reliable. So your example of checking the same electron again and again is not true, but the general assumption was right. Next to QCs: You are mixing a whole bunch of different stuff with "quantum" in front, like quantum computers, quantum algorithms, quantum encryption, and so on. First of all: The general problem with QCs is that they don't exactly store anything. They work in the principle of Superposition and very complex stochastic calculations. You feed them information and an algorithm to work with and then you let them do their work. The problem is: You don't know when they are finished. They can work on a problem very fast, but they can also take hours or days to find a solution. And because they use particles in Superposition to calculate, the moment you check if they are finished, you force said particles into some state and thus stopping the calculation. If they were finished: Great. But if the calculations were not finished, you get a false result. What makes QC interesting is, that they don't actually "calculate" anything. They simply "try" every possible solution until you check. The algorithms are merely there to try to disregard any known false solutions or force the QC to try more possible solutions. So QCs can be pretty fast, but they are absolutely not deterministic. That means: They can give you the right solution to your problem in a split second, but they can also give you a false solution anytime. The currently best QC algorithm has about a 90% chance of giving the correct answer. So you see QCs are definitely not replacing regular computers anytime soon (or ever). But they are pretty good in solving very specialised problems, are great for encryption and pretty much the best way to research specific quantum mechanical phenomena. But don't expect them to dramatically change the way our home computers work ;) Oh and Quantum Tunneling has absolutely nothing to do with Quantum Computers. It would even be a problem, because you don't want to have your carefully trapped and freezed particles be moved at all.

[@lordofudead]

Emanuel May Thank you so much for cleaning that up for me. It being well, quantum mechanics/computer and not something one has apple opportunity to look into most of my knowledge was from scraps of information I have heard here and there. cheers for correcting me, it can be a hard thing to explain without a deeper understanding.

[@EmanuelMay]

Nezara I know that too well ^^ I learned most of the stuff by just going to our local university and attend a physics lecture about QCs. I am not studying physics, though, just went in with the other students and listened to the prof.

[@MyrmidonsProductions]

As a computer Science/ I.T ,CNET Major this video is exellent. We are actually working on alternatives to how chips are made, biggest contender being gemstones, Specifically Emeralds. That's why we have multicored CPU's to adjust for the more chips issue. Software actually has not been able to keep up with hardware advances.We actaully recently started bringing back multi soketed motherboards again and more monsterous CPU's like Threadripper

[@mkostrub]

Do a video about quantum computing! I've seen videos about it, but SciShow has a knack for making it easier to understand! Keep up the good work!

[@EdEditz]

Hmmm quantum computers. That would be cool. The whole bandwidth and clockspeed issues will be obsolete :)

[@TheDesius]

nope quantum computers are good at solving problems but just as slow as normal computers when youbrun programs or games on them.

[@EmanueleZanetti]

EdEditz actually if a program isn't optimized for quantum computation it will run slower on a quantum PC than on a "normal" PC. _if_ someone developed software specifically thought for execution on a quantum computer then it _might_ have some benefits (though it would imply more time to be developed and new hardware, so that brings downsides too), but as far as I know quantum computing is useful just for resolving _really_ hard problems that normal PCs handle poorly, due to the fact that they only use two states, faster.

[@EdEditz]

TheDesius Oh right. I didn't know that. I thought it would be the answer to all speed problems. Alas. Thanks for the info though :)

[@EdEditz]

***** Ah okay, thanks. I thought they would be much faster but I see I was wrong :)

[@amihartz]

Quantum computing actually won't ever likely be a replacement to classical computers, at least not any time soon. They don't necessarily process things any faster. They're only better for specific algorithms, like Shor's algorithm.

[@ChristieNel]

My computer runs on faith, not ones and zeros.

[@NahinAnnan]

lol

[@FarmYardGaming]

Your computer runs?!

[@Nohandle4u]

Well, bless you.

[@jacobb1442]

This pleases the great machine spirit.

[@anubis2814]

Moore's law is actually 18 months not 2 years.

[@EmanuelMay]

Moore's Law started with "Every year", later Moore himself changed it to "Two Years". After that, people tried to correct the Law a few times, so it still makes sense for our declining transistor rate ;)

[@superdogmeatmeat]

It started out with a year, in '75 he changed it to two years. The law itself largely doesn't mean anything / it's a self-fulfilling prophecy.

[@anubis2814]

Yeah, it also depends on how much government spending we feel the research deserves. Every part of the cell phone was invented by government research and then private sector engineers combined the parts and brought them into our lives.

[@moncayoda]

Hank, this is the most entertaining one you have done so far. Keep up the humor.

[@MAGonzzManifesto]

HOLY CRAP HANK THIS HAS BEEN THE BEST EPISODE OF SCISHOW IN HISTORY!!!....Really though, this is my favorite show so far. Please keep all the tech computer stuff coming. I feel like I got more out of this than I did from one semester of a Programming Fundamentals class at my local community college 1+ years ago. Where can I go to learn more about this stuff?

[@EcceJack]

"Because I know you like it!" Really, Hank? *Really??!* xD

[@reckarthack3018]

AMD watching this & seeing 7:56 : *insert gif of Spongebob saying "write that down"

[@TheJaredtheJaredlong]

Faster computers would be amazing for games. If a computer ever got to the point where it could render scenes faster than the brain could process the scene, then graphics would literally be indecipherable from reality.

[@bombmask]

we already render faster then you can process. thats a matter of refresh rate. And its not really about faster now its about making efficient multitasking code and more multitasking on the hardware side. 4GHZ we barely ever use a full 3

[@JamesRJKR]

"If a computer ever got to the point where it could render scenes faster than the brain could process the scene" Shit, reminds me of the We are from the future video, you should check it out if you haven't.

[@comicbstudios]

Not really, that would just be really high FPS, you can run pacman at 300 FPS, wont make it look real.

[@jamez6398]

That's not likely to happen.

[@onwul]

Not really. We already can run some games at like 200+ FPS, it doesn't make them real. To make them real, we firstly need some kind of 3D vision, like oculus rift, we then need high quality textures, anti-aliasing, shadows, lighting, and so on. As for FPS, i think 95+% will absolutely feel it is real under 50FPS.

[@neverchk]

I think that may be the longest introduction segment in any SciShow video I've ever seen. ...Love this guy, BTW

[@shapiroezra]

Really nicely done! This is the first time I think I have managed a rudimentary understanding of the topics in the video.

[@LurkTheDarK]

I certainly hope this takes a bit longer, or by the time i finish my computer engineering major it'll all be irrelevant.

[@cbrazy2263]

which year are you? :)

[@dotnet97]

more like we'll get to be the ones to solve these problems and move things forward :P

[@outlaws9295]

Luckily, I'm going for computer science and not computer engineering. My main selling point is writing code, and we're always going to need programs.

[@dotnet97]

Only, computer engineers are capable of both writing software and designing hardware for the software, a field that broad won't ever become irrelevant either. At least that's what made me pick computer engineering over computer science.

[@outlaws9295]

Himanshu Goel The main skill of a computer engineer is hardware design. You're certainly able to do both as a computer engineer, but from what I've gathered, you'll rarely touch software development. At my university, for example, I'm required to take two computer engineering classes. I'm allowed and highly encouraged to take more, but two is the minimum. Computer engineers, on the other hand, are required to take physics, electrical engineering classes, specialized computer engineering classes, and so on. Typically, they never end up in a C# or Java programming class with the CS students.

[@4w7thncwt48oi]

Well like 99% of your phone is the battery so we need to fix the battery before we do anything else...

[@MrOreoMuncher]

*Draws Rainbow in mid air* "Graphene"

[@4w7thncwt48oi]

OreoMuncher what about graphene?

[@MrOreoMuncher]

caleb trollston You haven't heard of Graphene super capacitors? www.technologyreview.com/view/521651/graphene-supercapacitors-ready-for-electric-vehicle-energy-storage-say-korean-engineers/ There ya go.

[@The_Reductionist]

OreoMuncher why is this stuff not in my phone yet?

[@MrOreoMuncher]

Boby Gandhi Money, the only reason. It's not economically viable to implement this stuff on a mass scale.Yet.

[@DFPercush]

If current gaps are ~30nm, having a photon computer would greatly reduce the number of "transistors" or whatever the fundamental component of a light computer would be, because visible light is in the 100-1000 nm range. But perhaps more could be accomplished with a smaller number of components. I don't quite understand how the speed of the actual electrons matters here, because the electric field propagates at light speed. Any change in electrical current and voltage will be "felt" down the line however fast one electron's motion can affect adjacent electrons, and then how fast those electrons affect the ones behind them, etc. I must be missing something, because some very smart people are convinced that this would be better. Best of luck to them. Let me know when there's a compiler for it.

[@pg2826]

The current processors in the market are 22nm. That is the channel length. It is not related to the wavelength of light. The channel length is in simple terms is the distance between where electrons are and where they want to go to change a 1 to a 0. This requires a minimum amount of voltage to trigger the transfer. The electricity in transistors does not flow in a "line". It is more complex than that which is why you can't directly apply the electric field propagation theory to it.

[@DFPercush]

Right, current solid state electronics are not restricted by EM wavelengths until you start talking about interference, but I was speaking of a hypothetical light-based computer. Light can't pass through something smaller than its wavelength, can it?

[@jarjar561]

DFPercush But why does it need to be visible light?

[@bryanleebmy]

DFPercush Remember that light acts as a wave AND a particle, so not necessarily. Also, the problem with electron's movement speed is that it is not the current we are measuring but rather the literal and physical time it takes for the electron to move into the gap. Try not to think of the electrons as a current but as an entity that has to move into and out of a gap :)

[@DFPercush]

Bryan Lee That's why I don't have a degree in computer engineering right now. Can you recommend any online resources that would be good to learn about solid state physics? Listening to a professor who could barely speak english didn't do it for me.

[@amberlee4536]

Thank you, Lord Helix, for the close captioning. I am terrible at listening to words and retaining them. I need to read them for anything to make it into my head properly.

[@johnbell7802]

Fantastic summary. A couple of details I'd like to clarify though: 1. Transistors don't just have two modes (on/off), they can actually have infinitely many modes since the amount of current it passes when it is "on" can be controlled rather precisely. This is how things like audio amplifiers work. However, early on we developed a way of using transistors called "digital logic" that uses only two possible states for a number of reasons. For example, it has the potential to use less power to perform a given calculation and it can also be much more robustness in tough environments. 2. Although the speed of electrons themselves are, in fact, quite slow through normal materials (this is called its "drift velocity"), the speed of the information through electrical systems does pass at near the speed of light. This is because the information we are interested in travels as an electromagnetic field. Electrons are, of course, affected by the field and it is the electrons themselves that make the switching action of a transistor work. However, a single electron is never used to pass information from one place to another by itself. Instead, think of moving a column of water through a hose (to use a cliched example - better suggestions are appreciated!): a section of water 100m from the end will take time to exit, but the water begins to flow almost instantly (assuming the hose is already full) because of the pressure that is formed in the hose. In the case of electronics, it is the pressure (em field, voltage) and the flow (current) that we are generally interested in, rather than the exact location of any single electron. I hope that makes sense.

[@johnbell7802]

tl;dr - 1. transistors are way more complicated than just being on or off, but we often choose to use them this way because it helps us do certain things very efficiently. 2. Electrons do travel slowly, but the information transferred in an electrical system does travel at the speed of light (via an electromagnetic field).

[@cheatingthesystem21]

Temporary solution; make the processors bigger in size. Sure we'll need to develop new motherboards that support them but so what? Processors are some of the smallest components in a pc- but why

[@Zerepzerreitug]

For what I understand the more travel time between the transistors, the chip would get slower instead of faster. Not to mention the processor is expensive. It may be small, but it is one of the most expensive parts of the motherboard

[@BloodEyePact]

They get incredebly hot as is, making them even bigger would mean hotter. Also, the more transistors, the more expensive. So you'd have a really expensive box that sets fire to your desk.

[@mr.barnacle]

As said by Arturo above, the size of the processor inhibits the speed of the processor. Larger scales means larger distances for information to travel, thus wouldn't work.

[@matthehat]

What it boils down to is the practical application of these chips. There is already research going on into stacking dies (the actual part of the chip package that contains the silicon). The problem we face is that as you add complexity to the die (more transistors/layers) the thrmal output increases. The current mid to high end offerings from Intel and AMD run hotter than intended even when cooled by air, so if the complexity of the chip is doubled then the cooling requirements would increase significantly. That's not a huge problem for desktops because you can use massive heat sinks or water cooling systems, but laptops, tablets and smartphones are limited by physical space.

[@yumri4]

cheatingthesystem21 yes ... kinda as atm about only half of the current gen CPUs is actually a CPU and the rest a GPU on the same die. The reason why AMD has a more powerful GPU is because of the larger die size yes but even if Intel and/or AMD were to make the die size larger there is a point where as it is so large that it isn't practical anymore to do either due to heat and/or expense. Though as AMD chips are usually bigger than Intel chips and the manufacturing process is very close in transistor size AMD will continue to have the stronger GPU due to just having more space to fit it. Stacking chips is one thing which nVidia has looked at and is doing with the memory units on the GeForce 800 series cards so the technology is there it is mainly just the heat which it would generate which is the problem. Water Cooling is one cooling solution though as we almost all have gotten use to air cooling by using heatsinks and/or heatsink + fan for like 2 decades so far going back to water cooling would be a hard sell for Intel and/or AMD to make for a better chip. Water cooling also is not possible on a laptop nor a smart phone nor a tablet. What direction i see the CPU companies going in is actually for lower power consumption and better integrated GPU on die also with AMD better Crossfire support between integrated GPU and discrete Graphics card.

[@JukesMcGee]

I dont normally comment on clothing, being a dude and all, but that shirt he's wearing is FLY.

[@SK8BANDANA]

eh, looks like minecraft water.

[@wick9427]

+SK8BANDANA best. Comparison. Ever.

[@UmVtCg]

my gtx 980ti has 8 billion transistors, the chip is bigger in size as a cpu though as it is created on 28nm process

[@Neocasko]

+Rem ko Well, a 980 Ti is a graphics card. The heart of it being a GPU. A processor that consists of FAR more cores than a CPU. :p

[@Grasshoppa065]

Awesome video. Just thought I'd point something out incase anyone is curious. This video implied that a single transistor represents a bit of information. This actually isn't the case, as transistors are not as simple as being either on or off (it would be awesome if they were). This is an extreme oversimplification. Memory elements, such as bits, are actually composed of a circuit of transistors called flip flops. The amount of transistors in a flip flop varies depending on how it is implemented and what type of memory you want (volatile or non-volatile) but they typically cost around 8 transistors. This isn't even including the high enable needed on each flip flop in order for the computer to address the memory.

[@20hkn23]

the best video i have ever watched about transistors. Thank you!

[@PhynixPc]

Just a little clarification. Intel's14nm process is ready and they have said commercial shipments of processors based on 14nm will ship by the end of the year. Samsung and Glo-Fo will be ready to start using their own 14nm fab this year as well and TSMC is working on a 12nm fab process. Also D-Wave has quantum computers out in the world. NASA and Google each have one and Lockheed Martin just bought one. These are not gpcpus but specialized for specific purposes however they are still quantum computers and they are on the market for a ton of cash.

[@EmanuelMay]

Well, I haven't seen a single proof that D-Wave's are actually real QCs. Google and others use them mainly for research. As far as I know they are regular computers specialised to simulate QCs as close to original QCs as possible, so researchers use them to test their algorithms, because real QCs are very unstable.

[@loshan1212]

Yes, what Emanuel has said is true... There are only simulators available for programming if you purchase something of D-Wave. Also TSMC is currently having problems with its 14nm process, I doubt that they will put a lot of work into it now, before fixing their problems with the 14nm process.

[@PhynixPc]

"Ever since D-Wave arrived on the scene with a type of quantum computer capable of performing a problem-solving process called annealing, questions have flown thick and fast over whether or not the system really functioned - and, if it did function, whether it was actually performing quantum computing. A new paper by researchers who have spent time with the D-Wave system appears to virtually settle this question - the D-Wave system appears to actually perform quantum annealing. It would therefore be the first real quantum computer." www.extremetech.com/computing/184242-d-wave-confirmed-as-the-first-real-quantum-computer-by-new-research

[@PhynixPc]

And yes TSMC is having issues as they almost always do with a new fab especially since this is their first 3d transistor process. That doesnt change the time table they recently laid out by more then a couple months. It also doesnt change the info I put here because Hank said that MAYBE we can go to 22nm or lower with current materials, when in fact 22nm process microprocessors have been available to the public since 2011 and microprocessors based on even smaller processes are coming out soon.

[@loshan1212]

Apple supposedly ordered 14nm A9 chips from TSMC.

[@Elfos64]

It seems like there are multiple ways each problem can be addressed, each having varying degrees of practicality. For example, generating too much heat. Some might say the obvious solution is to make it generate less heat, some might say the solution is to remove the heat it generates fast enough. Is there a way we can USE the heat productively rather than just thinking of it as an unwanted by-product to be rid of?

[@DJRAMO4LYF]

That may be the best idea I've ever heard...

[@ImTabe]

Currently? Too expensive I would say, the quality of the energy is super important, that's why oil for example is so popular, because it's easy to extract the energy from it.

[@JamesBrown059]

Can't exactly put it to any use. The heat produced by chips is very, VERY insignificant in the real world. Remember, the chips auto shut off at 100C. That's a hardware limitation because of the materials we use for production. Most computers run at much lower temperatures, thereby, the potential energy gain from somehow harvesting that heat is literally zero. We're trying to keep the size of the heat-sink small; otherwise you would simply see bigger heat-sinks and the heat issue would never rise again.

[@Elfos64]

Xthreeo In other words: no, there is no way to use the heat. Is there any way at all for it to be something other than an unwanted by-product people wish they could be rid of, or is it simply impossible to be anything else?

[@AZURA888]

The solution is a new architecture with no clocks, a signal enters, a signal go out in cascade, a network not a circuit, check out the new IBM chip Truenorth, with 5.4 BILLIONS transistors consumes only 70mWatts/h of energy not needing heatsinks or fans, so for first time it will be possible to put a lot of processors working together.

[@le4kyf4ucet54]

Or make the chips bigger

[@ChristopherWeaver1]

Yeah but that would mean the cooling system needs to be bigger which effectively makes the computer bigger.

[@fareesha2303]

Electrons take time to Travel . So increasing the size of the chip will make it slower. Increase size of the chip will require more positive energy for the electrons to flow . So , a larger processor will take more power to run. Silicon has defects. Eg, If you take 1 layer of silicon and make 100 small chips , 99 of them will be good and 1 will be bad. This will give you 99% yield. But, If you take 1 layer of silicon and make 1 big chip , it will have a defect which will give you 0% yield .

[@rakibhassan5946]

Fareesha Zaman you just cleared all my confusions, Thank you

[@kiddybopp7110]

Tj Moran they should hire Lays maybe that will teach them both

[@willyou2199]

Modern day CPUs run at 2-5GHz, 2-5 billion cycles per second. 1 cycle is 0.2-0.5 nano seconds. This is just enoguh time for a electrical signal, travelling at the speed of light, just enough for a singal to move just mere inches. Your CPU cannot be bigger than that.

[@CCrazyA]

I participated in an internship at the University of Arizona's Physics department this summer working on Magnetic Tunnel Junctions (MTJs) as replacements for transistors. They can be much much smaller than transistors in that they inherently rely on the electrons tunneling through the separative insulating dielectric. Instead of relying on the electron's charge, they rely on the electrons spin (spin up or spin down), and the magnetic orientation of two ferromagnetic conductive layers on either side of the dielectric determines the resistance of the MTJ (High resistance equals 1, low resistance equals 0, similar to a transistor). On top of their size, MTJs are also resistance to power loss or power surges due to their magnetic nature, they are resistant to radiation (so applications in space could be on the horizon), and their heat density is practically nothing due to the lack of electrons being bled into the surroundings of the microchip. What do you guys at SciShow think of the idea of MTJs? They have been around since 1975, and modern samples have been able to get resistance changes of ~400% at room temperature, so they could be ready for modern applications very soon!

[@romanovergerio1174]

Need more hank on SciShow

[@3800S1]

we are currently at the 16-22nm fabrication node.

[@KaymationNation]

14 nm on the highest consumer grade broadwell proccessors

[@TipCounter4You]

Actually all of Intel's current CPU lineup is using a 14nm process (skylake)

[@FlyingGospel]

Electrons are not slow. Depending on the medium they are traveling in, thry go between 97 and 98% of the speed of light. The 8.5 cm/hour was poorly explained and misguided lots of people. It's the flow of AC that is travelling at 8.5cm/hour, not the electrons themselves.

[@mr.barnacle]

Here is the explanation to why you are wrong (with math). TLDR: Electrons don't need to move that fast because there are just so many of them in wire. What moves so fast is the propagation of the electromagnetic fields since one electron moving causes a ripple effect. "Electric current is essentially a measure of how many charge carriers you can move through a given cross-section of conductor in a given amount of time. This will depend on the size of the cross section, the number of charge carriers, and their velocity. A current of 1 A corresponds to a transfer of 1 Coulomb of charge per second. An electron carries 1.6*10-19C so you need to move 6.3*10^18 electrons/sec. Divide by the density of electrons in a copper wire (about 8.45*10^22 electrons/cm^3) and the cross section of the wire (for AWG 18 this is pi*(1.02mm/2)^2 or 0.008 cm^2) and you get 0.0093 cm/s."

[@michaelkenner3289]

You actually have that backwards. The flow of electricity travels at close to the speed of light, individual electrons flow at 8.5cm/hour. Unless we're talking particle accelerators and similar high-energy astronomical phenomena, individual electrons never move near the speed of light.

[@FlyingGospel]

Michael Kenner When lightning strikes, do you see it go down at 8cm/hour? Electrons have almost no mass. Therefore the slightest input of momentum and they're almost travelling at C.

[@Vulcapyro]

Yan Leduc-Chun That isn't the case. Electrons go "pretty fast" as lightning because they're being accelerated in a direction by the rather strong electric field present. But they still don't go anywhere near the speed of light _in the given direction_, and it's simply wrong that "the slightest input of momentum" will accelerate electrons at such speeds as they would achieve significant relativistic mass. It isn't as if electrons are somehow normally super slow and then applying some few volts accelerates them to near light speed, either. The free electrons in the medium do bounce around _much faster_, at the Fermi velocity (still not close to the speed of light at all; on the order of 1%). The electrons in a conductor are accelerated _in a direction_ when DC voltage is applied, again proportional to the strength of the electric field. The average velocity of the electrons in the direction applied by the field is called the drift velocity and this is what we're talking about when describing the velocity of electrons when an electric field is applied. And with AC the electrons might as well not be going much of anywhere because the applied current alternates and they barely have any net movement.

[@mcool13thebass]

It's actually exactly the opposite of what you said. Electric current (or more properly, the *electromagnetic wave*) travels at nearly the speed of light down a wire, but the electrons themselves move slower than a snail.

[@JKCWvids]

i want biocomputers. a system that includes living organisms like in Hitch Hikers guide.

[@ltericdavis2237]

That's called a brain...

[@BragoTHEgraviyKING]

Factoid Mother brain* Cough cough AURO UNIT ... HACK... phew practically wheezed out a lung

[@ltericdavis2237]

BragoTHEgraviyKING Meh, killed it enough times in metroid.

[@BrunoPontoTxT]

welcome to matrix

[@BrunoPontoTxT]

***** xD

[@johenrique21]

I don't know you guys, but I think this was his best show. I laugh so much! Thank you man that talks fast.....I mean..... real fast!

[@davidblanchard8052]

My mind was blown indeed. I didn't expect electrons to be so slow, yet electrical current holds so much power.

[@Andytlp]

Graphene where art thou. Personally i'm waiting for graphene, cause they're actively researching the matter and it's feasible unlike lasers or quantum computing. They're too advanced. graphene is here and now.

[@IceMetalPunk]

Quantum computers are here and now. They already exist and are being used. they're just not the miracle system the media has touted them as. But they do work well for certain types of problems (combinatorics and probability theory problems, specifically).

[@matthelton6637]

Plus, I'm pretty sure graphene holds the charge a lot longer so using it for batteries would increase the capacity of batteries exponentially.

[@shawniscoolerthanyou]

Waiting for it? Research it! Develop it! Experiment with it!

[@Andytlp]

IceMetalPunk Yeah they're slower than a pocket calculator and cost 10 million usd for the cheapest one.

[@yourface747]

IceMetalPunk I cant give a link to the vid where i saw it, but i watched a video where this guy was explaining quantum computing. He said that it would not replace ordinary computers. Although it could be better than the ones we have now at doing some of the things you mentioned. Like probability theory etc. But it would not be better than the computers we have now for the everyday use. especially considering the cost.

[@ananyakalia5]

Now we have 14nm FinFET by AMD.

[@BlackoulisP]

you lost me at "it's not magic"

[@ciCCapROSTi]

I'm so happy that finally a video for the layman states that the smallest unit is the byte, not the bit. The computer doesn't think in binary, it thinks in base 256.

[@lesageethan2380]

The bit is the smallest difference that makes a difference. So, technically, it is the smallest data storage

[@RED89P13]

Wow those first microchip pictures were of EEPROM chips! My dad used to work for general dynamics in San Diego and he was telling me that he saw these one day at work and was blown away by how small these were back in his day. He hardly can grasp that we do have machines with lasers to make the new chips we have today. He thinks they are technology from some other place not of earth hahaha. I love this show it has so much information to offer and I love crash course too! You made me pass my psychology final with flying colors by reviewing with your videos rather than the book and in all honesty you did a better lecture than the professor!

[@Tuttomenui]

Disappointed that you didn't cover the option of making chips layered.

[@EmanuelMay]

Because it's not very good. Heat problems increase tremendously and the connections between the layers are too long to be practical in a single chip. Some multicore chips are layered, though, but mostly low-voltage microchips, not high-power chips like the CPU or GPU in your computer.

[@Tuttomenui]

Yeah I thought about all of that when I wrote my comment. I figured by not addressing it It could spark conversation and I wouldn't have to get all long winded. Layering is something best suited for different materials that can handle the heat or an architecture that doesn't produce as much. The pathways are no different if you don't limit yourself to connecting the layers at the edges but throughout the chip like you do with PCBs.

[@kurtilein3]

Tuttomenui thermal load is the problem. basically processors need to be small, so that they can be fast, and they need to be flat, so that they can be cooled. if you could stack hundreds of them into little cubes where the paths are still short, that would be insanely powerful, but it would burn out. here the solution could be superconducting computers. basically they work like regular ICs, its just that the heat problems go away and now you can stack them and do really compact designs.

[@frollard]

Tuttomenui to be fair, it was just announced that commercially viable 3d storage (nand flash and ram) solutions were coming about...so that will be cool. As for CPU technology as it was said, the heat would be a nightmare.

[@kurtilein3]

frollard flash storage could really work in 3d or layered structures, you just need to slow down the data flow so that no area gets too hot because too much is happening at the same time.

[@TheSenileOldMan]

... why don't they just make the cpu's bigger, i mean obviously not in like phones or tablets or any small "handheld" device, but i mean what does it matter if you make the chip 2x or even 4x bigger in a desktop/ mid-large laptop? i mean the laptop will still be just as thin and who moves their desktop regularly. even if you made a chip as big as a modern mobo, you would just need to make a new form for mobo's and cases. if it meant 10x faster processing then sure why not? why try getting smaller when you could go big?

[@nickmoody4493]

The most expensive component of your computer just quadrupling in size? Just saying, that'd start getting expensive fast. It's not a long-term solution if we want computers to keep evolving.

[@bombmask]

Keep in mind, that its how close things are that make things fast, if you move things farther apart things have to travel farther and it all adds up in latency

[@Falcrist]

Assuming all other things are equal, 4 times the size means 4 times the cost (actually it would be more like 10x), 4 times the power consumption (thereby requiring larger power supplies), 4 times the heat dissipation (requiring much larger heatsinks), 4 times as much infrastructure, less space for other components, etc. Not to mention the additional latency that the greater distances will cause. That isn't a viable strategy for consumer components.

[@dvoicer6785]

The heat produced by such a chip would exponentially increase. You heard Hank talk about chips melting through the casing? That kind of heat.

[@JohnKha]

Falcrist Not only the heat issue, but in reality, the wafers that the silicon chips are made on only get so big, and the cost and reliability of the whole chip goes down as the size gets up. Some of the newer processors, like the Intel Core i7s actually have 8 processing cores on them, but only 4 or 6 are running, which lets them sell the dies that still have issues as a different (cheaper) product.

[@sehrgut42]

Transistors are not binary: they're analog, with a response curve based on their input. A lot of engineering has gone into discovering how to design circuits that can use these decidedly NOT binary elements to implement binary functions.

[@K1LLW1LL056]

Cool channel...I found out about you by sitting on the L train in NYC and the whole cart was wrapped with your ad. I'm talking about the most psychedelic trippy train I've ever seen...your ad is like tripping acid! Good Job

[@MariusSchar]

I found this episode to be exeptionally great! I have no idea why, but so it is.

[@jeffthompson4276]

Great episode, but how do you get 2,292 thumbs up with just 362 views, without magic...or quantum computing?

[@dvoicer6785]

Views in multiples states, OMG.

[@ConiferGrower]

Views are tallied slower than likes, it will update with time.

[@jjbonedi1]

dvoicer6785 Im in Maine :)

[@magzire]

How do people still not now this, been explained for the past 10 years.

[@isoen4557]

How don't you now you've got to now

[@XatolosWired]

Great video, poor mastering. You can hear someone tapping the mic at 2:22, and more background and other sounds in the video at different places

[@theuglyzone]

Wow, you totally ruined this video for me.

[@Alakabram]

Make your own perfect videos so we can judge them

[@ParanoidPixel]

Its the mic on his shirt.... doy

[@OriginalRAB]

It's his tie mic. I'm constantly hearing the rustling from the mic against his shirt.. This should have been picked up during filming let alone post processing. Was the sound guy sick and they just went 'Fuck it'?

[@UltimateVegetto]

Trrrraaaaaaaagedyyyyyyy!!!!

[@scottycatman]

Neat stuff except I don't know where you were going with the 6.5cm/hr electrons to a lamp analogy. we're talking data transfer, not raw power. Data represented by electrons travels down copper at something like .6 * c (speed o' light). can't look up exact figure... it's like two in the morning and I'm nodding off.

[@scottycatman]

I re-did my research and your phrasing is a bit ambiguous. Wave propagation (the important speed) is what I was referring to moving at speeds of .6*c. I believe the video was correct about 6.5cm/hr speed of a single electron along a wire, but it's a pointless figure.

[@scottycatman]

Except that's wrong. The structure of copper allows for an input voltage on one end to be detectable on the other end at a rate of roughly .65*c. This is taking the non-linear movement into account. Individual electrons may be moving slowly, but it's a mass movement; huge magnitudes of electrons are making jumps to other copper atoms. So wave propagation is the important thing; it's the speed at which an input can cause an output on a length of copper or graphene or what have you. The 6.5cm/s might only jump up to 10cm/s with graphene, I don't know the numbers there. The revolutionary thing about graphene isn't so much its speed but rather its power efficiency due to extremely low resistance. Accurate interpretation of an input on graphene can be done with smaller input on graphene than copper.

[@MrBranboom]

***** +Scotty Catman Come on guys, get back at it. You gotta find out who's righter.

[@bryanleebmy]

Scotty Catman Correct me if I'm mistaken, but this isn't a matter of current but rather we are dealing with how fast the actual electron can move into and out of the silicon gap. Try not to think of the electron as part of a current but instead a physical entity that has to move 8.5cm/h into a gap :)

[@scottycatman]

Bryan Lee You don't treat it like individual electrons though. It's treated as a chain of electrons... you move one end of the chain, the other moves as well. Can you imagine how slow a computer would be if we were relying on a speed of 8.5cm/h? Additionally, he's relating copper to graphene, not silicon to graphene. Silicon is a semiconductor, graphene is a highly-efficient conductor.

[@benefit14snaake]

Great topic. Was very excited to watch it. Keep up the good work.

[@Eliminator5555]

Really great video. Helps one understand the potential future of computing.

[@7777TheShade]

Why don't we just make the chips slightly bigger?

[@tarnvedra9952]

Because bigger the chip gets, you more likely mess it up during manufacture and you increase chance of total failure by combination of little ones. When you look at chips from same architecture and core count they are are manufactured on same assembly line, and they tested and priced accordingly to how much broken they are. Those part are just too small to be 100 % perfectly manufactured, and at some point its just not profitable anymore.

[@magicstix0r]

Propagation delay. The delay across a circuit gets longer the bigger or longer it is. This is one of the main things that sets clock speeds. Each piece in the circuit has to reach steady state to be usable, and the amount of time it takes for that steady state to be reached across the circuit controls how fast you can clock it.

[@schwarzerritter5724]

Cherno Alpha The smallest impurity can ruin the entire chip. Making the chips bigger will greatly increase the chance of failed chips. Imagine you can cut 100 chips out a silicone wafer. Impurities might ruin 50 of them. I am pulling these numbers out of my ass, but they really do have a lot of defective chips. Now imagine you make the chips bigger that you get only get 50 out of a wafer.

[@overwrite_oversweet]

Cherno Alpha Costs more.

[@jessicadineen4755]

Cherno Alpha To add to Schwarzer's comment, some flawed chips are sold. A good example are AMD tri core processors, which are actually quad cores with a disabled defective core (or at least this was the case for specific models). They are sold simply because the rest of the chip works fine, and it cuts down on costs by recouping some of what would otherwise simply be waste.

[@Morpheuss]

Isn't it possible to just make a bigger processor? With a custom motherboard and...

[@cwispynoodles7672]

He just said the more there are the more heat it produces

[@Morpheuss]

I know but I mean like multiple processors on top of each other with airways in between them but still like 1 processor

[@Morpheuss]

If that makes sense :/

[@stickmenproducts21]

morpheus970 heat would still be an issue, small airways aren't enough, most computer cpus require fans just to keep heat down, often multiple fans

[@BragoTHEgraviyKING]

Zope50 or for the extreme computer modders, water cooling tanks...

[@abhiram7379]

fast forward to 2019 and we have processor with 7nm technology at a very reasonable price..

[@MattGreenisawesome]

This is SO COOL! One step closer to the singularity.

[@zinebjedraoui7987]

omg, this is like the best channel ever, EVER!!

[@Mazaroth]

Excuse me, i identify myself as a transistor, and this is really turning me off.

[@XgamerdaveX]

👏🏻

[@knightwing5169]

Very funny, man.

[@Mazaroth]

The Illusionist May i inquire, how was my comment racist?

[@knightwing5169]

The Illusionist Are you a troll or are you serious?

[@Mazaroth]

The Illusionist Ok, now i just feel sorry for you.

[@yourfullofsheite]

How about magic? With all of the advances in creation science ,I'm sure they have a computer that runs off faith by now.

[@Jazma_edits]

YOU JUST HAVE TO BELEIVE

[@crunchycheeto99]

Why not make the chip bigger..

[@crunchycheeto99]

I watched the whole video... I must have zoned while out watching it xD

[@Epicshadow123456789]

Because then the heat it generates will still cause a problem.

[@W3NNIS]

Well then the motherboard would need to be bigger, then the cases, the the computer needs more power, it would cost more money to make, and yeah...

[@nussiskate3]

TheSharp0ne they didn't say anything about bigger chips, only about using multiple chips. so why not just use bigger ones?

[@stickmenproducts21]

Kevin Nusshold it would be more costly

[@sumitchakraborty9988]

This was like CS101 how computers work. Absolutely fantastic!

[@everestfalls]

I started watching your videos last night.....it's morning now.....I can't stop lol

[@XxXVideoVeiwerXxX]

So who made the binary dictionary??? How do we know what all these different binary sequences mean?

[@Fortstorm]

We invented them.

[@XxXVideoVeiwerXxX]

so is there like a list or compilation that says 10100 means this for that

[@Fortstorm]

XxXVideoVeiwerXxX I'm sure there's a list somewhere.

[@daanwilmer]

XxXVideoVeiwerXxX There are many standards detailing which binary sequence means what, depending on the purpose. In a processor, it's the processor manufacturer that decides which binary sequence corresponds to which operation, which is called the instruction set. Years ago, Intel invented the x86 instruction set, and it has become the standard in desktop computing. Your phone, on the other hand, probably has a chip designed by ARM, which uses a different instruction set (actually several, slightly different instruction sets, but that's not the point). For text, for example, there is unicode, managed by the unicode consortium, which is a dictionary between binary sequences and characters.

[@Darkfeyttt]

Basically programmers and hardware manufacturers decide which binary value corresponds to which 'logic value'. For example, a simple character convention is ASCII (look for it on internet), the character 'a' in ASCII is represented by the binary value 01100001. But what makes it 'a' (why 01100001 represent 'a'), it's because there's a small piece of software/hardware that when it sees 01100001 it says "turn on these pixels" and you see a 'a' on your screen. (it's actually a bit more complex in true life, but that's the basic of why 01100001 can represent 'a') How do we (programmers and/or hardware manufacturers) decide which value represent what ? We decide it using our need. For example, let's say I need to represent a forecast state, I'm lazy so forecast is represented by only four state : not sunny and not raining not sunny and raining sunny and not raining sunny and raining There's four state I can decide to represent it on 2 bits, with one bit representing the "sunny" information, and another bit representing the raining information, that would make the list look like that : 00 not sunny and not raining 01 not sunny and raining 10 sunny and not raining 11 sunny and raining The first bit representing the sunny information, the second one the raining information. So my program displaying the weather information would look like that : If the first bit is activated Then print "sunny" on screen Else print "not sunny" on screen print " and " on screen If the second bit is activated Then print "raining" on the screen Else print "not raining" on the screen So if we input '01', the program will print "not sunny and raining" on the screen. We created a new convention of forecast encoding ! If you're interested in learning computer stuff, the computerphile channel might interest you.

[@blakkwaltz]

graphics cards actually have 7 billion transistors so moore's law only really stopped for cpus

[@algotq]

Arent graphics cards just "bigger". If not care to explain?

[@RowanIngram]

Except what's really in question is transistors per unit space. Graphics cards just have more space, but the same density.

[@GaryBusey-sLaserdiscCollection]

That's because the die size of a GPU is far larger than that of a CPU.

[@johnneijzen]

True but gpu have way bigger to area to build unlike.

[@blakkwaltz]

It is true that gpus have more space to work with. I just meant that even though cpus were slowing down there were still chips out there that are improving still. If know how gpus work you'll realize they still can push moore's law forward. well for next 20 years but yeah we've basically hit the end of the road.

[@42Fossy]

If electrons are slow, how come they're able to buzz around atoms so fast?

[@Neocasko]

+Reba Fossy Atoms are TINY!!! XD

[@sapiranimations]

Actually they don't buzz around atoms, this is actually mind blowing. The electron is not in one specific location, it just has a higher probability of being in a specific location at the specific time. an electron is sort of everywhere around the atom until you observe it. It's amazing

[@Neocasko]

sapiranimations It sure is. How matter exists in that scale doesn't really match how we experience it.

[@sapiranimations]

+Neocasko Experience is a big word for the illusion of matter our brains create XD, We base our representation of the universe just by the light we observe, and people seem to forget that light only interacts with matter. I just wonder how it would be evolving life on the more energetic perception of the universe, such energy creatures could exist? idk it's really interesting tho thinking about our perception and how much we miss and can't see

[@Neocasko]

sapiranimations Indeed. Our perception is quite limited. We only see things with a TINY section of the electromagnetic spectrum. Imagine what we miss. I do imagine how life can exist in forms other that how we react. An interesting thought.

[@99thTuesday]

Thanks for explaining this SciShow.

[@EdwardDrachenbergJr]

Where was this video when I gave a report on Moore's Law like 10+ years ago...? well done!