New Breakthrough Could Solve CPU Transistor Crowding

[barelysalted]

https://fossbytes.com/light-emitting-silicon-eliminates-need-cram-more-transistors-cpu/amp/

 

Seems like physicists have made a new breakthrough.

 

Super cool!

 

Quote

Research led by Erik Bakkers, a physicist at the Eindhoven University of Technology in the Netherlands has led to the invention of silicon alloy nanowires that can emit light. The breakthrough could allow the development of photon-based circuits instead of the currently used electron-transistor model.

 

In simple words, the newly invented silicon alloy nanowires could allow data to be transmitted through photos instead of electrons. This would eliminate prevalent issues like an electron traffic jam, overheating and slow transmission speed arising due to cramming of a large number of transistors in a chip. 

------
Since photons can transmit data at a much faster speed and across multiple channels as compared to electrons, we could see photonic circuits that can shuffle a large amount of data in a computer chip swiftly.

 

These photonic circuits could prove to be of great help in data-intensive applications where data needs to transferred around very quickly. For example, the photon-based circuits could find applications in self-driving cars for transmitting a large trove of data from a host of sensors in real-time.

 

Personally I am most excited for the possible thermal performance improvements. I wonder how long this will take to be adopted and if/when it is, will it be extremely expensive? Only for data center application rather than personal use? Where would the world go from here?  I definitely hope the manufacturing process isn't something that chip manufacturers deem to be inefficient. I'm sure there is more info in the research paper, but it is behind a pay wall. 

 

Link to the research paper; 

https://www.nature.com/articles/s41586-020-2150-y.epdf 

Edited April 14, 2020 by barelysalted
Updated to meet posting guidelines

[Pickles von Brine]

In 10 years maybe. Cool still. 

[TempestCatto]

Finally. Porn can come at me faster!

[barelysalted]

Just now, Pickles - Lord of the Jar said:

In 10 years maybe. Cool still. 

haha true. Def cool to think about tho. 

[SansVarnic]

@barelysalted Super cool BUT...

Your post does not meet the posting guidelines, please read the quote and link below and update your topic appropriately.

 

Quote

When creating a thread in the News subforum, please make sure your post meets the following criteria:

• Your thread must include some original input to tell the reader why it is relevant to them, and what your personal opinion on the topic is. This needs to be MORE than just a quick, single comment to meet the posting guidelines.
• Your thread must include a link to at least one reputable source. Most of the time, this should be a respected news site.
• Your thread should also include quotes from the cited source(s). While you shouldn't just copy the entire article, your quote should give the reader a summary of the article in a way that gives the key details, but also leaves room for them to read the full article on the linked website. Please use quote tags to show that you have copied this content from another site.
• The title of your thread must be relevant to the topic and should give a reader a good idea of the contents of the thread. Copying the title of the source is permitted but absolutely not required. It should be to the point and not be done in such a way as to mislead a reader, such as clickbait, etc.
• If your article is about a product or some form of media, images are always appreciated, although they are not required.

Failure to comply may result in your thread being locked or removed without warning.

 

[barelysalted]

4 minutes ago, SansVarnic said:

@barelysalted Super cool BUT...

Your post does not meet the posting guidelines, please read the quote and link below and update your topic appropriately.

 

 

will do. Apologies

[Dash Lambda]

That article doesn't say exactly what the alloy does. It says it's "light-emitting," but then it describes it as allowing 'free flow of photons,' which is transmissive. So does it produce photons or just let them through?

 

Either way, how is this a breakthrough? I thought the difficulty with optical computing had more to do with electron-photon-electron conversion efficiency and logic representation than miniaturization. Or is this purely for data transmission? If so, I guess it would be a breakthrough to have nano-scale light sources, but wouldn't converting to and from light add a ton of latency? We already use fiber-optics in large scale stuff where the conversion latency is negligible, but at the chip scale we're talking less than a mm. This is too small for RAM or PCIe, right?

 

If it is just for data transmission, how does that reduce the number of transistors needed? You're not replacing transistors with optical gates, you're just replacing buses with optical wires -if anything, that will mean more transistors for signal conversion.

 

I'm just so confused.

[leadeater]

Hope this doesn't take 20 years to make it in to products or be useful at all.

[rcmaehl]

So we're switching from Electron-ics to Protron-ics now?

[straight_stewie]

5 hours ago, barelysalted said:

I wonder how long this will take to be adopted and if/when it is

This was being talked about as being one potential solution to Moore's Law ending atleast as early as a 2008 talk titled "Indistinguishable From Magic".

I feel the need, largely because I said "solution to Moore's Law ending", to point out that there is no solution to the fundamental reasons that Moore's law is falling apart. Moore's law states "the number of transistors doubles every 18 months". The general case of Moore's Law would follow "the number of switching elements increases by some factor in a given amount of time". This would apply regardless of whether we are using transistors or optical methods, and the same result would happen: Things can neither be infinitesimal nor infinite in size, and therefore, at some point, there must be a limit to the number of switching elements in a computer. Ergo, Moore's law has been ending since before Gordon Moore even made the initial observation, and I, for one, am glad that people are no longer claiming that it's not ending.

[S w a t s o n]

15 hours ago, rcmaehl said:

So we're switching from Electron-ics to Protron-ics now?

Photon-ics

https://en.wikipedia.org/wiki/Silicon_photonics

[Doobeedoo]

Amazing, new chip materials and this will be a complete new breakthrough in the future.

[Sauron]

18 hours ago, leadeater said:

Hope this doesn't take 20 years to make it in to products or be useful at all.

As far as I can tell they don't actually know how to make a transistor with this, so... it may be in 20 years or never. Unless it has some utility for inter-transistor connections or something.

On 4/14/2020 at 6:42 PM, barelysalted said:

For example, the photon-based circuits could find applications in self-driving cars for transmitting a large trove of data from a host of sensors in real-time.

Yeah, that doesn't actually mean anything for cpu transistor crowding... it's just a fast device connector.

[find_better_name_later]

3 hours ago, VegetableStu said:

Protonic Computer running Cinebench R20

The explosion would be three times as big if it was running Crysis.

[SpaceGhostC2C]

21 hours ago, Dash Lambda said:

 

Either way, how is this a breakthrough?

I'd would need to know far more physics to answer that (and the reader to understand the answer ), but this is research on materials, and the breakthrough is in terms of material properties.

 

The article's abstract (bold mine):

 

Quote

Silicon crystallized in the usual cubic (diamond) lattice structure has dominated the electronics industry for more than half a century. However, cubic silicon (Si), germanium (Ge) and SiGe alloys are all indirect-bandgap semiconductors that cannot emit light efficiently. The goal of achieving efficient light emission from group-IV materials in silicon technology has been elusive for decades. Here we demonstrate efficient light emission from direct-bandgap hexagonal Ge and SiGe alloys. We measure a sub-nanosecond, temperature-insensitive radiative recombination lifetime and observe an emission yield similar to that of direct-bandgap group-III–V semiconductors. Moreover, we demonstrate that, by controlling the composition of the hexagonal SiGe alloy, the emission wavelength can be continuously tuned over a broad range, while preserving the direct bandgap. Our experimental findings are in excellent quantitative agreement with ab initio theory. Hexagonal SiGe embodies an ideal material system in which to combine electronic and optoelectronic functionalities on a single chip, opening the way towards integrated device concepts and information-processing technologies.

 

So, taken at face value, the contribution is legitimate and important. The translation to "your computer will be photon-powered and cram moar corez by the end of the week" is more another case of the science news cycle:

Spoiler

 

[rcmaehl]

3 hours ago, VegetableStu said:

Protonic Computer running Cinebench R20

 

More like:

[Dash Lambda]

28 minutes ago, SpaceGhostC2C said:

-snip-

This makes more sense, though I do wonder what benefit hybrid electrical/optical chips will have. Signal propagation is one of the most important factors at play in chip design nowadays, but adding extra steps to sending a signal is usually more detrimental to latency than whatever your transmission medium is. Maybe this'll eliminate a couple of those steps? I guess we'll see, some day.

[cj09beira]

3 hours ago, SpaceGhostC2C said:

I'd would need to know far more physics to answer that (and the reader to understand the answer ), but this is research on materials, and the breakthrough is in terms of material properties.

 

The article's abstract (bold mine):

 

 

So, taken at face value, the contribution is legitimate and important. The translation to "your computer will be photon-powered and cram moar corez by the end of the week" is more another case of the science news cycle:

  Reveal hidden contents

 

pretty sure their main problem was that making light from electrons was taking too much space, now that they can make it smaller it starts to be more feasible to make photonic circuits 

[straight_stewie]

12 hours ago, Dash Lambda said:

adding extra steps to sending a signal is usually more detrimental to latency than whatever your transmission medium is

9 hours ago, cj09beira said:

their main problem was that making light from electrons was taking too much space

My understanding is that this is a material which just emits light when it sees a voltage differential, for whatever reason (I am not a material scientist).

If that understanding is correct (please fact check that), then there should be no latency, setup, or extra space required anymore than there would be for a copper wire.

But this material is only half of the story really. Sure this emits light when it sees electricity, but we don't have much use for that until we find a material, useful at similar scales, that emits electricity when it sees light. We can turn electrical signals into light signals until we're blue in the face, but it won't do any good unless we have a way to turn light signals back into electrical signals.

And that still ignores the problem of actually transmitting the light. Light transmission is very different than electron transmission. Is light transmission a ULSI scale thing yet?

My guess is that we will start seeing this material in LEDs and optocouplers first.

[RejZoR]

On 4/14/2020 at 11:51 PM, leadeater said:

Hope this doesn't take 20 years to make it in to products or be useful at all.

Well, we're reaching physical limits of silicon with methods that we know currently. 7nm was already a significant challenge, 5nm will be even bigger and from there on, we don't even know for sure if 3nm and 1nm are even possible. And what then as we switch from nanometers to picometers (though we're probably remain in nanometer realm with decimals like 0.25nm for a while instead of defining it in massive picometer numbers).

[CarlBar]

Hmmm i wonder is this might have some application in room temperature quantum computing. From what i understand a q-bit just requires that it output one of 3 state's state 1, state 2 or state 1 and 2 at the same time. If the light frequency output is subject to variation base don the input and if that variation is controllable and it';s capable of covering a broad enough rnage at one time it could potentially work by emitting light of one frequency for state 11 and light of a second frequency for state 2 or a mixture of both frequency for the combination.

 

I could be completely off base ofc. I really don't understand how the idea of those 3 states translates to actual on die logic. transistors are easy as each one is basically an IF gate and everything else builds off that so it's easy to understand how the concept, (2 output states depending on a single input), translates to work, understanding how a Q-biot translates is not somthing i've ever read an explanation i can wrap my head around so maybe i'm wildly off targets with this...

[greenhorn]

From what I understand, if the emission wavelength can be continuously tuned over a broad range, you could encode information in the emitted wavelength/frequency. For conventional semicoductors, the size of the bandgap (which corresponds to an energy difference) is fixed and corresponds to a certain wavelength (not necessarily in the visible light range). That's why different LED types have different colors. If you need another color, you can use an LED and apply a coating that absorbs the LED's light an emits light of another color/filters the original light (that's how the phosphorous coating in LED filaments works).
Those tunable new semiconductors could also be used for LED strips with more interesting color options and perhaps for much smaller RGB LEDs. For the use in microchips, you have a continuous spectrum of light that you can emit so in theory you could convey infinite amounts of information in just 1 pulse, but in practice you would need to define intervals to get all the advantages of digital systems.
WIth that same technology, you can then also build a light detector sensistive to a part of the spectrum to detect the transmitted symbols (which consist of a number of bits depending on the symbol alphabet that you manage to define).
These things are useful for data transmission. If you need something like transistors, there are phototransistors that open depending on the light received at the base and a multitude of other optoelectronic devices.

[cj09beira]

5 hours ago, straight_stewie said:

My understanding is that this is a material which just emits light when it sees a voltage differential, for whatever reason (I am not a material scientist).

If that understanding is correct (please fact check that), then there should be no latency, setup, or extra space required anymore than there would be for a copper wire.

But this material is only half of the story really. Sure this emits light when it sees electricity, but we don't have much use for that until we find a material, useful at similar scales, that emits electricity when it sees light. We can turn electrical signals into light signals until we're blue in the face, but it won't do any good unless we have a way to turn light signals back into electrical signals.

And that still ignores the problem of actually transmitting the light. Light transmission is very different than electron transmission. Is light transmission a ULSI scale thing yet?

My guess is that we will start seeing this material in LEDs and optocouplers first.

exiting something with light shouldn't be an issue 

[Dash Lambda]

4 hours ago, CarlBar said:

Hmmm, I wonder is this might have some application in room temperature quantum computing. From what I understand a qubit just requires that it output one of 3 states; state 1, state 2 or state 1 and 2 at the same time. If the light frequency output is subject to variation based on the input and if that variation is controllable and it's capable of covering a broad enough range at one time it could potentially work by emitting light of one frequency for state 11 and light of a second frequency for state 2 or a mixture of both frequency for the combination.

A qubit can exist in a superposition of states, which means that it has a certain probability of being in either state when measured. When it is measured, though, it will be in one state or the other.

 

The problem with room-temperature supercomputing is suspending and controlling the particles. The higher the temperature, the more noise you get and the harder it is to carefully isolate and manage individual electrons (or sometimes photons). Superconductors are also commonly used in a couple critical components (I'm not entirely certain for what role), which thus far require cryogenic temperatures.

[CarlBar]

4 hours ago, Dash Lambda said:

A qubit can exist in a superposition of states, which means that it has a certain probability of being in either state when measured. When it is measured, though, it will be in one state or the other.

 

Ok this is fairly different from the regular explanation you hear but makes waaaaay more sense why it's so difficult.

 

If i'm understanding it right, (i'm asking to check because i could be waaaay off base again), a q-bit is still like a transistor in that it;'s an IF gate, but the output is subject to probabilistic variation depending on some factor, (i assume an input?). With each measurable output probabilistic distribution being a final output state. In layman's terms assuming you can measure it measuring an average of 80% state 1 and 20% state 2 isOutput A, but 79%state 1 and 21% state 2 is a different output, ditto for 78% state 1 and 22% state 2 and so on and so forth? That would allow a single Q-Bit to produce a truly enormous number of potential final outputs many, many orders of magnitude more than a single transistor could.