Gaaah, please stop advertising optical computers as the technology that will overcome Moore's law. It makes no effing sense.
Wavelength of the light emitted by these devices: ~4000nm
Latest generation commodity CPU transistor structure size: 7nm
Add to that that photons really don't like being trapped; you essentially need a delay line and optical amplifier to hold them indefinitely (that's essentially the core technology my whole PhD thesis centers around), it makes them a really impractical thing to store bits with. Things with a rest mass can be stored easily, though. Things like, say, electrons!
It's definitely not a continuation of Moore's law as it has nothing to do with transistor density, but it may mean that the performance people expect from computers - which is why people are usually talking about Moore's law - may continue increasing.
I don't see how the wavelength is comparable to transistor size because as you switch to the optical realm, the benefit of information propagation at speeds near c (or c, if you're pulling a vacuum) means physical size doesn't matter as much. At 4Ghz you can move information 7.5cm in one cycle, and that's a pretty large distance compared to any integrated circuit I've ever seen.
Why is storage necessary? If you can move bits to optical gates and get a result back it seems to me like you can work around the fact that, in an electrical system, capacitance and heat (due to density achieved in the quest for minimizing capacitance) start to limit the computation you can do.
Only if you consider 70% or so to be close. There's some room for improvement over copper wires. Now, if there are any physicists here who want to jump in, I have a question about that. I heard waveguides are dispersive, would sending pulses of light through tiny channels slow it down as well?
Some people working in optics say it works "at the speed of light." That's true of course ... but the speed is no different from using copper.
Unfortunately something like 0.7c is about the fastest speed of EM wave propagation in an optical waveguide or along a copper waveguide. Another comment here gives a slightly faster example with n=1.3, which is maybe achievable in some kind of polymer. Or in highly purified water, for what it's worth.
You can get a mild speedup, 40% or something, by moving to free space. But that is an unbelievable can of worms, taking all the signals out of the waveguides and somehow still getting 1B signals going to the right place. The 40% speedup doesn't remotely pay for giving up solid state waveguides.
"Dispersive" fortunately doesn't mean a meaningful slowdown. It just means that a transmitted bit will travel at a range of slightly different speeds. If it goes very far, the shape of the pulse will get messed up. But that's a problem people are already pretty good at solving.
Nonetheless it feels like at least every few months or even weeks a new announcement appears in pop-science sites like eurekalert and phys.org. Feels a little bit like the always around the corner next big battery tech.
Most fascinating thing i've read years ago they'd be the prime candidate for manufacturing in space, because real vacuum.
Have you ever tried wiring any non-trivial logic without flip-flops? Say, a simple signal routing layer. Even the most basic bits of logic becomes much less efficient to downright impossible without storage.
You can use hybrid systems where, say, memory is conventional RAM but computation (maybe full cpu or submodule like apu) is done with photons. You can probably perform large numbers of concurrent operations by taking advantage of the wavelike properties of photons.
You can wave pipeline electrons too. It just very rapidly becomes an impossible design problem as the complexity of your design increases (and as the variation grows in significance with node shrinks)
In the article, they do say that the use will be in transmitting data from a processor to different components farther away from it, like RAM or sensors on a car.
The article specifically proposes using light for transmission between components (not storage), with the efficiency benefit of multiplexing (albeit not mentioned by name). As I read it they're talking nanoscale fiber optics, not optical transistors and memory. This sounds pretty reasonable to me, and your comment seems to not address it at all.
How do electronics compare to photonics thermal radiation wise?
Though as always with electrical based electronics - superconductors are room temperature are always heralded to be the big jump in many things. As always, soon, much like photonics or let alone the ability to easily design and implement asynchronous circuits, let alone CPU's.
Though I do wonder what other industries have the equivalent to moore's law driving them in both advancements and marketing?
I'm kinda drawing a blank of anything that has any progress metric defined. Though hopefully somebody else knows of something comparable in another form of production/business.
High-bandwidth plasmon resonator waveguides. This would allow multiple datapaths on a single 'wire'. Fermions are great for logic and storage, but not for comms. Currently we use ~90% of chip power moving around data. We need to use bosons for this. We need to make them in silicon, and reduce the waveguide dimension. That's where this is going.
I will be pedantic but there is no light at 4000nm. Light is by definition the radiation that is considered from the point of view of its ability to excite the human visual system (HVS). The HVS sensitivity, as given in ASTM E308-15 practise, is in range [360, 780]nm.
Well, the semantics of what constitutes light and what not are a bit murky. CO2 lasers are still considered LIGHT amplification by stimulated emission of radiation. CO2 lasers operate at ~10000nm.
In the optics community we usually consider everything we can manipulate with refractive optics as "light" – and yes, I am fully aware that this goes down well into what's considered microwave radio.
My personal cutoff for where optics begins is, where I no longer can use an antenna that is part of a resonant _circuit_ to emit / receive the radiation, and have to resort to quantum mechanical state transitions.
Not optical, but piezo electrical, usually with a crystal or air as the medium instead of mercury. Optical is much the same principle, a feedback loop incorporating the delay line, so the same bits get re-injected over and over again and can only be read out at specific points in time.
Like most things, the concept is much simpler than making one work — and I was hoping I could find some papers on the applied side of photon delay lines. (Since OP commented it was related to his/her PhD.)
I thought the mercury delay lines were the craziest thing I’d heard about in computer evolution until I learned about using a cathode ray tube as memory.
In a way that is a delay line too, the phosphor decay time allows you to read out the bits a bit later than you put them in. The big advantage is that it is theoretically random access.
I graduated from TU/e and in my experience their applied physics department, specifically nanomaterials like this are very well funded and attract a lot of international talent.
>> cubic crystal lattice that allows electrons to move within the lattice under certain voltage conditions. But it doesn’t allow similar movement for photons, and that’s why light can’t move through silicon easily.
Uhhh.. not really. I’ll try to explain (forgive my ad-lib MatSci from 20 years ago). Efficient light generation is a matter of direct or indirect bandgap. A direct transition is one where the electron wave number is unchanged in dropping from the high to low energy state, so it can be completed with a single photon (light). An indirect transition fails conservation of energy and momentum with one photon, so it requires phonon (heat) interactions. Semiconductors have an energy gap between the highest few occupied state and the lowest few unoccupied states, and these are the only states that can exchange energy. Direct transitions generate mostly photons, so even if it gets absorbed, it will get re-emitted intact until it leaves the material. Indirect transitions means that phonons remove energy each time, so it all becomes heat. In normal conditions, Indirect materials are more transparent, although direct materials can become transparent by population inversion, which is when there are more electrons in the high-energy states then the low-energy states for the bandwidth of the photons being generated. Then any photon generated is more likely to generate more photons on its way out (stimulated emission) than to be absorbed. This is what you want. Okay I’ll stop now, but there are tricks that you can use to get this behavior in silicon, an indirect-bandgap material, which is the topic of the article.
Super cool -- one challenge will be that in order to live on the hot processors of today, these emitters will need to still work at around 400K -- the good news is that these appear to still work well at around 300K -- but it sounds like these go to nearly no bandgap at higher temperatures and shift into the infrafred. But yea, pretty awesome stuff..
MOVPE is more of a self-assembly process than 3D printing, and it has been around for decades.
BTW It is worth noting from your link that the silicon is being grown on a GaAs substrate, so to be useful they would have to figure out how to grow the silicon wire on a silicon substrate. (GaAs already has many options for optical devices.)
Really impressed they stuck at it, by the sounds of it there's still a lot of work to do. Hopefully it won't be another 50 years before we see it adopted at scale.
Is the breakthrough here about emission or transmission? Or is the physics for these two connected? It's not clear to me, between the title and the article.
Secondly, could using a "photonic" memory bus bring RAM access speeds close to cache speeds, or is the transmission distance/time not the main issue there?
They were able to measure a strong photoluminescence signal from the silicon germanium nanowires. Photoluminescence is a good proxy for how efficiently it will light up - carriers are generated optically with a light source more energetic than the bandgap of the material. This is much easier than fabricating a full device with electrical contacts.
Silicon is typically a really lousy photon emitter because it’s an indirect bandgap material. Turning an electron/hole pair into a photon requires an interaction with a phonon. It seems by getting the silicon to grow in a hexagonal orientation, it becomes a direct bandgap material leading to much higher emission efficiency.
> Secondly, could using a "photonic" memory bus bring RAM access speeds close to cache speeds [...]
No, it really can't. The distance between CPU core and DRAM chips is approximately 10 cm, so at a typical electrical propagation speed of around 2/3 c, the round-trip time is 1 ns. A full DRAM access, however, is on the order of 100 ns. So physical transmission speed only accounts for about 1% of DRAM access times.
The speed is not limited by propagation delay, but by signal integrity. It is non trivial to have an external parallel bus operating at low Ghz speed. On chip signals can be much faster because they don’t have the capacitive load.
On top of that is power requirements which are again orders of magnitude higher than on chip signals.
Optical has a chance to fix that for the same reasons it works so well for longer distance networking.
What you're saying applies to throughput, not latency.
> On chip signals can be much faster because they don’t have the capacitive load.
You can routinely achieve > 30 Gb/s off-chip in copper cables over distances > 1 m using differential signaling [1]. Capacitive load is only a limiting factor if you directly drive the gate of a transistor.
[1] For example, high-end Xilinx FPGAs provide several of those transceivers. They really operate at > 30 GHz.
True it won't affect latency. Yet caches are pretty good at ensuring most DRAM requests take advantage of the hardware parallelization available.
That’s why despite DRAM having pretty much constant latency for the last 20 years bus speeds and bank counts have been consistently increasing. Optical interconnects will help immensely.
We may also see things like off chip SRAM come back into vogue once its feasible to take advantage of their performance.
Transmission time isn’t really the main issue, it’s more about the work required to get a memory request through the levels of the hierarchy to DRAM and back. Probing each level of cache, propagating through the miss queues, translation (maybe with TLB miss), waiting for the DRAM controller, etc.
What? That doesn't make sense. If cache probing would be the cause for DRAM accesses being slow, we wouldn't need caches. We would just access DRAM directly!
It's the other way around: DRAM accesses are slow, that's why we need caches.
> translation (maybe with TLB miss)
In most architectures, the caches are physically addressed, so TLB lookups occur before even L1 cache access. Successful TLB lookups are extremely fast! And you can't skip the TLB, even if you don't have any data caches.
I wasn’t suggesting probing caches is the main cost, I only wanted to describe that there is a long journey to DRAM in current architectures of which signal propagation is such a small part.
You are totally right that if you can make the resultant communication speed faster you could theoretically do away with caches. However this approach wouldn’t solve that problem on its own. Also forget not that cache is expensive and DRAM is cheap!
Yes I’m aware that caches can be physically addressed and you could reorder the sequence I described. No you can’t skip the TLB, but a hit will be faster since you don’t have to perform translation.
> In most architectures, the caches are physically addressed, so TLB lookups occur before even L1 cache access.
So to see if a memory location is contained in a cache line, a TLB lookup is needed to first get the physical address? I wouldn't have expected this, can you expand on why this is the case?
Two reasons: 1) a virtual address might refer to different physical addresses (see pwildani's comment), and 2) a physical address can be mapped to different virtual addresses – a virtually addresses cache has to keep track of that somehow, otherwise the cache will become incoherent.
Interesting - if that's the case I would imagine it make sense for some systems to feature an architecture which skips the idea of multilevel cache entirely and has only RAM connected over a photonic bus. No probing, no cache misses.
> Modern transistors, which function as a computer’s brain cells, are only a few atoms long. If they are packed too tightly, that can cause all sorts of problems: electron traffic jams, overheating, and strange quantum effects. One solution is to replace some electronic circuits with optical connections that use photons instead of electrons to carry data around a chip.
Journalists need to be educated: Transmission lines are photonic, so silicon already has connections carrying data around using photons. As you would expect, those photons are traveling at the speed of light in the material.
If I were king, I would demand that every optical-silicon publication explicitly describe why their optical photons are more desirable than microwave photons that are already in widespread use.
"Integrating photonic circuits on conventional electronic chips would enable faster data transfer and lower energy consumption without raising the chip’s temperature, which could make it particularly useful for data-intensive applications like machine learning."
I don't think the win here will be speed, particularly chip level as much as enabling lower power and higher throughout architectures at the system level. As in this is mote likely to be used for medium range links where infiniband/pcie/nvlink are used today.
Really not sure this is the correct way to think about it, but if the silicon is now transparent and parts of it are emitting light, how is the focus limited / controlled ie. how does a receiver know from emitter it should be receiving from?
All you need to do is shoot it with a photon bean, and the when material can no longer absorb the photons you send at it, it will begin releasing them as reflection. But the one you are shooting in aren't the same ones that are coming out.
> Modern transistors, which function as a computer’s brain cells
Who's the target audience for this analogy? If you understand what a brain cell is then you probably know what a transistor is too. I would bet that more people know what a transistor is than what a brain cell is.
You don't need to know the details of a brain cell to get the analogy. If I say "transistor" to my mom, she wouldn't be able to tell me what it is. If I say "It's like the brain cells for the computer", she would probably understand that "Ah, so a computer has many transistors that helps it think", which seems good enough for an article with a broad audience.
> I would bet that more people know what a transistor is than what a brain cell is
I'm happy to take you up on that bet, but it depends on what bubble you ask. In San Francisco/Silicon Valley, that's probably true, but outside any high-tech bubbles, more people know that we have brain cells in our heads, than we have transistors in our computers, I'm fairly sure.
I would take that bet, too. If you say “transistor” to non-technical people over 50 or so, they more likely would think of small portable radios (https://en.m.wikipedia.org/wiki/Transistor_radio). To them, and many others, computers don’t have transistors, they have chips.
They _might_ know the transistor replaced vacuum tubes, but I doubt many would be able to tell what function either had, or be able to point out the transistors inside such a radio.
Is that a USA/North America thing, based on calling radio sets "transistor radios"?
Here in the UK we had "the wireless", and I'm confident that my parents - late 70s - who were the generation of first domestic computer ownership in the UK would associate "transistor" primarily with computers.
I would bet that more people know what a transistor is than what a brain cell is.
No chance. Everyone knows what a brain cell is, to some extent. Even people who have too few to rub together don't really feel like they know what a transistor is after reading the Wikipedia page for them, twice...
I used to know what a transistor was. And then we shrunk them to the point where they are no longer the same thing and yet they are still a thing, so I don't know what they are anymore.
Wavelength of the light emitted by these devices: ~4000nm
Latest generation commodity CPU transistor structure size: 7nm
Add to that that photons really don't like being trapped; you essentially need a delay line and optical amplifier to hold them indefinitely (that's essentially the core technology my whole PhD thesis centers around), it makes them a really impractical thing to store bits with. Things with a rest mass can be stored easily, though. Things like, say, electrons!