Imagine if instead of 4ghz we had processos at around 100 ghz.
AMD currently offers 12 cores at 4.6 ghz.
12 * 4.6 ghz, this gives us around 55,2 ghz of clock.
If AMD made a single core processor at 100 ghz, it would be twice as fast as those 12 cores at 4.6 ghz.
Couldn't they just increase the size of the transistors? So it will handle better higher clocks and temperatures. Or even utilize a material stronger than silicon.
The same way to make car engines more powerful to handle higher RPMs.
Other urls found in this thread:
en.m.wikipedia.org
ieeexplore.ieee.org
ieeexplore.ieee.org
semiengineering.com
legitreviews.com
youtube.com
twitter.com
why do you even bother posting on a technology board when your understanding of technology is on a caveman's level
>shrinking transistors
>posts BJT
what's the n-type of a human?
Why you don't just do it then?
I'll bet those companies would pay you a shitload of shekels if you could increase processing power that easily.
No. Just no. There is no "quantum" anything, this isn't poorly understood near magic effects of some mythical theoretical particle. This is simply electrons being so small they can move through any material at the path of least resistance, because nothing can exert 100% perfect electrical control over them. It is current leakage. It is nothing but current leakage. It is current leakage in short channel devices, and it happens at literally every feature size, it is not exclusive to small FinFET devices like upcoming 5nm EUV FinFETs. Even planar devices have extremely high degrees of leakage through their channels, directly under the gates, electrons still leak out. Yet despite this the transistors still function.
Quantum tunneling is a meme regurgitated by people who know nothing about the field of FETs.
lol just increase the speed lol not hard.
freq++ so izy
This is a new fronteer in retardation
So that they don't have to start worrying about the CPU melting if the cooling system (which would have to be LN2 if you bumped the speed any more than it is right now) failed
>why don't we increase the speed of light instead of reducing the distance?
>posts a fucking BJT
The absolute state of this board.
Maybe because they're dumber than you and haven't thought of that.
oh my god
this is too long of a post to be bait and it looks like OP genuinely believes in what he wrote
holy shit
this is definitely one of the most retarded posts i've seen on Jow Forums in the last 10 years
bogan dumbass
If the transistors are big, you'll get the reverse of Dennard scaling, which will limit speed. If they're small, you'll be fucked over by subthreshold leakage and contact resistance, which will increase heat and again limit speed.
Yes zoomer,
They did that at the early 2000's but they realized that more clock speed = more heat and less reliable data transmission due to thermal noise.
So they had to go multicore, expand the cpu pipeline, other tricks like speculative execution etc, and keep the speeds at a managable rate.
low-frequency limit
back to school, op
Because Jow Forums is caveman level.
90% are neets and wanna be programmer
op probably has no idea what gate capacitance is, yet alone just capacitance
Couldn't a bigger transistor tolerate higher temps?
OP doesn't seem to have a firm grasp on what a transistor is
I only buy intel because despite the objectively better multi-core performance of AMD's multi-die process, all that truly matters is single thread performance, because all I ever use a CPU for is emulation and making sure my GPU has no bottleneck as I am nothing more than a mere consumer, with zero application for heavily multi-threaded task. Nothing 7nm that AMD has produced has even come close massively overclocked on LN2 to simply liquid cooling an Intel processor.
Prove me wrong Jow Forums?
Yes, and if magic were real lots of stuff would happen too!
>Couldn't they just increase the size of the transistors?
Think. Just for one second THINK.
7/10 pasta, could be better.
Why do ships have engines the size of houses instead of using a 4 cylinder engine that spins at 100k rpm?
Research, "nanoscale vacuum-channel transistor."
Electrons in CPUs don't move at C, per se.
Don't that move at the speed of C as far as Silicon allows?
>as far as the medium allows
Of course. That is why these can be faster at the same size.
>increase freq
>power consumption grows exponentially
>try to decrease power consumption by lowering circuit capacitance
>howtodothis?.html
>decrease resistor size
>cpu still running at 15kw
fugg. realize that muh silicone can only handle temps lower than ~100C. when we get transistors made of carbon nanotubes your microchips will be able to runt at 1500C instead and you can have you 100GHZ (no idea why you would need this). until that the best ways of increasing performance is more cores, better pipelines and larger l1 caches
those companies will steal your idea before you even know it
At such frequencies even parts of the chip itself have trouble synchronizing with each other, communicate with other stuff like RAM is guaranteed to be slower than chip's frequency by even by the laws of physics alone. Only massive amounts of vertical on-chip cache could keep such frequencies from waiting for RAM.
>nanoscale vacuum-channel transistor
this only increases fmax though; silicon fmax is like 300-500GHz at channel lengths of 10nm, so this is not the big problem, there are power dissipation nuances though.
you can't dope graphene nanotubes easily, there's small molecular specimens such as F4TCNQ (polymer) and polymers/ionic fluids, CNT quality is also not easy to ensure and is very prone to defects. rGO is also a close band gap material so it's not going to behave like SiO2. MoS2 is a better novel material in nearly every single aspect for semiconductors. CNTs can be used for GAAFETs though.
RC term can also be improved by increasing dielectric constant of oxide, e.g. HfO (what Intel currently uses) and Al2O3. III-V and II-VI will never be used commercially though, it's not cost effective enough to justify switching and you can't build oxides predictably. There's some interesting ways to manufacture certain nitrides though it will not be cost effective which may be suitable for high frequency switching electronics.
retard
Silicone and electricity will be replaced with crystal matrix and magnetic monopoles like muh superman fortress.
I read it in muh news feed.
How are processors clocked to 8 GHz+ then?
Just the clock generator, the performance barely increases because everything is so much out of sync.
Switching away from III-V semiconductors to a substrate that allows faster electron flow is too scary because you essentially have to build ask new fabs. That's also why we are still eking gains out of extreme UV lithography instead of something else.
en.m.wikipedia.org
Even if heat wasn't an issue, there's this too
power consumption scale linearly with frequency because if you need x power for switching a transistor in 1 second if you do it twice it'll just take 2x. The problem is always voltage as power consumptions grow quadratically hence the run to lower working voltages.
No electron ever moves at the speed of light. In fact they move around human walking speed.
Electric fields propagate at the speed of light. That is what matters.
UHYHUYHYHYYHYUYHU WHY INSTED UF VEYMG A NIGGERFAGGOT WE INSTED B FAGGOTNIGGER UHGHUGHGUHGHGHUHYUHTHUHYYHU
Why don't they just decouple the frequency of different paths instead of trying to synchronize the entire thing. Imagine trying to get an major metropolis to synchronize seamlessly so everything runs on time. Oh sure it may be possible for God, but the lardass in the motorchair will slow everyone down
shit b8 m8
Why do you think they made microprocessors in the first place? Why not just one massive processor the size of a room?
Because the rate that electric energy can travel is limited, so the only way to make it "faster" is to reduce the distances it has to travel.
shrinking transistors does increase speed ya tard.
If you shrink a wire and send a electronic signal over it, it'll travel faster because less resistance.
same concept.
When the wire and its contacts are small enough, resistance actually increases
ieeexplore.ieee.org
ieeexplore.ieee.org
This is a problem with current nodes
smaller = faster
bigger = slower
>smaller wire has less resistance
based and redpilled.
based and physics pilled
wouldnt optical computers solve the heat issue of high frequency?
For a smaller processor, the signals can get around faster since there's less distance to travel.
HURR WHY PHYSICS DO THING
Ask yourself why car engines don't run at 100k rpm
OP, don't let these retards get you down. They're just jealous you struck a genius idea. I solder my own CPUs and even got the privilege of meeting President Barack Obama for it. Keep it up, you will change the world!
There are transistors that operate up to several hundred gigahertz.
Heterojunction bipolar transistors (HBT).
>A pseudomorphic heterojunction bipolar transistor developed at the University of Illinois at Urbana-Champaign, built from indium phosphide and indium gallium arsenide and designed with compositionally graded collector, base and emitter, was demonstrated to cut off at a speed of 710 GHz.
Metal–semiconductor field-effect transistor (MESFET)
>MESFETs are usually constructed in compound semiconductor technologies lacking high quality surface passivation such as gallium arsenide, indium phosphide, or Silicon carbide, and are faster but more expensive than silicon-based JFETs or MOSFETs. Production MESFETs are operated up to approximately 45 GHz, and are commonly used for microwave frequency communications and radar. The first MESFETs were developed in 1966, and a year later their extremely high frequency RF microwave performance was demonstrated.
High-electron-mobility transistor (HEMT)
>Applications (eg for AlGaAs on GaAs) are similar to those of MESFETs – microwave and millimeter wave communications, imaging, radar, and radio astronomy – any application where high gain and low noise at high frequencies are required. HEMTs have shown current gain to frequencies greater than 600 GHz and power gain to frequencies greater than 1 THz. (Heterojunction bipolar transistors were demonstrated at current gain frequencies over 600 GHz in April 2005.) Numerous companies worldwide develop and manufacture HEMT-based devices. These can be discrete transistors but are more usually in the form of a 'monolithic microwave integrated circuit' (MMIC). HEMTs are found in many types of equipment ranging from cellphones and DBS receivers to electronic warfare systems such as radar and for radio astronomy.
Just because you don't understand the technology doesn't mean it does not exist.
OP was talking about CPU speed, you doofus.
Well OP's a retard. Making transistors bigger or out of "strong material" won't help them work faster. The silicon is breaking down because of the high frequency. It's shit like parasitic capacitence and inductance creating distributed element low pass filters that put practical limits on your switching speed. Making components bigger is probably the worst possible thing you could do as you are increasing both capacitence and inducatance and thus lowering your maximum operational frequency. Anybody else saying smaller is bad because of resistance is retarded. More resistance is more bad true, but all the inputs in a CPU are high impedance. A little trace resistance means jack shit when that trace terminates into a buffer with a 1GΩ+ input impedance.
>the silicon **isn't breaking down because of high frequency
Fixed my typo
it's called "overclocking"
16nm was the golden size? Isn't the problem basically silicon at this point? If only IBM's Graphene research turned up
>make transistors out of graphene
>everything is a short
>No. Just no.
reddit pasta
>suggestion that 10nm and below might not be such a good idea after all
big process bad
>REEEE DELETE THIS AMD UNITE
retard
a thread died for this
So let me get this straight....
Can't make transistors smaller due to leakage and heat
Can't make the transistor clock speed higher due to power consumption and heat
Why don't we just make the multicores function as a single core? 24 cores @ 4.1666ghz would be the 100ghz single core we are after.
>12 * 4.6 ghz, this gives us around 55,2 ghz of clock.
Stop reading here.
>24 cores @ 4.1666ghz would be the 100ghz single core we are after.
What the fuck am I reading
Hahaguy.png
>processor
>at 100Ghz
>What are physical limitations
It's not that simple. You create more bear because you're pulsing the electricity that much more.
It's better to try to handle more instructions per cycle. I also think it'd be nice to have a balanced ternary processor. Maybe we can get more states with optronics by exploiting the properties of light.
Tl;Dr: Moar GHz is retarded.
24*4.16666 is approximately 100 so would be 100ghz. Now these processors use out of order execution BUT if you made them use in order execution you could use the cores as 1 core because tasks would be processed in order.
>your microchips will be able to run at 1500C instead
You do know that you can already have this, right?
newtype?
because web sites are developed by pajeet inc who can't grasp multithreading
Even fucking Youtube doesn't have multithreading support
>make them bigger so they can go faster
oh yea, haha, why didn't they think of that?
youtube.com
EE schizoposter detected
>Here's a microsecond. 984 feet. I sometimes think we ought to hang one over every programmers' desk... or around their neck... so they know what they're throwing away when they throw away microseconds.
brutal
You should have read some books about MicroElectronics...
It doesn't work like you think.
that was the plan, but there were unexpected problems with that idea
don't worry, the industry is well aware of the advantages of fast single cores, but it's not as easy as you think to keep making a single thing go faster
I don't think that what you said is actually accurate. I think its retarded
a useful comment from that page
notice;
"moores law will hold until about 2004 or 2005", the main article puts 4-8GHz in this range
"five silicon atoms", this is 5.55nm wide