Researchers Create Photonic Microprocessor

Status
Not open for further replies.

turkey3_scratch

Polypheme
Ambassador
This is incredible, and definitely a start for what will have to come to the computing world. Once silicon limits are finally hit (5-10 years) quantum computing or now photonic computing are the next big things.

I would like more information from them on what tests have been done with this CPU and the reliability.
 

techguy911

Distinguished
Jun 8, 2007
1,075
0
19,460
58
This is incredible, and definitely a start for what will have to come to the computing world. Once silicon limits are finally hit (5-10 years) quantum computing or now photonic computing are the next big things.

I would like more information from them on what tests have been done with this CPU and the reliability.
Quantum computers are the next big thing Google already has one according to test is 100 million times faster than a single core cpu.
Next year they plan to buy a much larger one with NASA 1000 qubit model.
Also in Australia a group of scientist are working on a quantum computer printed on silicon.
D-Wave is the leading quantum computer mfg in the world.
I would say in about 10 years we will have hybrid computers with a quantum co-processor as they are for math and physics calculations only.
 
We also don't know about the cost of production yet. It might end up being that quantum computers are faster but far too expensive for every day devices like smartphones and PCs. I have no idea, as neither are in production now, but the photonic processors seem to be targeted at lower-end devices.
 

alextheblue

Distinguished
Apr 3, 2001
3,078
106
20,970
2
Quantum computers are the next big thing Google already has one.
1) Not for consumers. They require extremely low temps near absolute zero.
2) Extremely expensive.
3) Extremely limited use - they're actually next to useless right now and yes that includes the ones Google has purchased to tinker with.
4) Unknown reliability/accuracy.

D-Wave and MS are both working on some new designs, but anything remotely useful for real work are a ways off. But even then the other issues will remain extremely difficult to solve, in particular the operating temperature range.
 

anbello262

Distinguished
Sep 27, 2013
1,171
0
19,660
117
I don't think quantum PCs are going to be a consumer reality anytime soon (20+ years), because of the type of calculation they are built for. But, having photonic PCs in 15 years will really be a big step up for consumers. I sure hope this kind of technology is used for GPUs shortly after.
 

Darkbreeze

Retired Mod
I'd be surprised if it didn't find it's way into some kind of APU type configuration that handled both CPU and graphics functions. Virtual reality and holographic potential from this kind of tech is probably phenomenal. If you think those holograms they are using on stage now are something, think ACTUAL holosuites.
 

Joe Black

Honorable
Jul 3, 2013
88
0
10,640
1
Quantum computers are the next big thing Google already has one according to test is 100 million times faster than a single core cpu.
Next year they plan to buy a much larger one with NASA 1000 qubit model.
Also in Australia a group of scientist are working on a quantum computer printed on silicon.
D-Wave is the leading quantum computer mfg in the world.
I would say in about 10 years we will have hybrid computers with a quantum co-processor as they are for math and physics calculations only.
1000 qubits will be enough to process more "bits" of data than there are atoms in the observable universe if some docs I watched were telling the truth...

That said - I am not certain that q-computers will be ideal for all purposes. Precision is a big thing the way I understand it. They excel at certain types of computational tasks that conventional computers are abysmally bad at in comparison - Problems where brute iteration from step 0 - n to find an optimal solution would take too long. Problems that involve probability algorithms. Just like graphics processors can execute massively parallel tasks.

I am no expert on the matter, but it may very well be that the ideal chip of the future will have photonic GPU, CPU as well as QCPU cores all in one package. And it could be that QCPUs excel at graphics computations as well - I think its very possible.
 

aldaia

Distinguished
Oct 22, 2010
517
6
18,995
1
If you read the original paper published in Nature you will realize that this breakthrough is NOT a photonic computer. Its just a traditional electronic microprocessor that has integrated optical communication with the external world. Although a very interesting development, it's very very far from photonic computing.
By the way, just the tittle of the paper is revealing: "Single-chip microprocessor that communicates directly using light".
 

Darkbreeze

Retired Mod
Actually, the article does not say that. It says:

Here we report an electronic–photonic system on a single chip integrating over 70 million transistors and 850 photonic components that work together to provide logic, memory, and interconnect functions
Which is as you say, far different from just interconnect functions which is what you are inferring when you say it only has integrated optical communication with the external world.
 

aldaia

Distinguished
Oct 22, 2010
517
6
18,995
1
Actually, the article does not say that. It says:

Here we report an electronic–photonic system on a single chip integrating over 70 million transistors and 850 photonic components that work together to provide logic, memory, and interconnect functions
Which is as you say, far different from just interconnect functions which is what you are inferring when you say it only has integrated optical communication with the external world.
Just keep reading, next sentence clarifies: "This system is a realization of a microprocessor that uses on-chip photonic devices to directly communicate with other chips using light.". Anyway this is just the abstract,the full paper and the drawings are self explanatory. I say it again, this is not a photonic computing device, its a traditional microprocessor that has an optical IO interface and has been connected to optical memory DIMMS (traditional DRAM memory with optical interface). It still is a breakthrough, the optical interface has been implemented using the same process as the rest of the processor (here is where the novelty resides). It's just not the kind of breakthrough that most people assume.
 

CaedenV

Splendid
having high bandwidth on a chip is all well and good, but it does not replace the fact that the CPU itself and the RAM is still traditional silicone. It is like giving a Ford pinto full access to a 100 lane highway... kinda cool, but mostly useless.

There are still 2 things that will be really neat to come from this tech:
1) There is still active development for actual photonic CPUs (not just CPUs with photonic buses) which will be really cool in 40-50 years.
2) Much sooner than that, imagine this kind of tech being used as a buss system on a motherboard to replace something like PCI Express. This would allow nearly infinite bandwidth to things like GPU, RAM, and storage with far fewer pins, and lower power consumption (and presumably heat).

For more info see IEEE's report on the chip here: http://spectrum.ieee.org/tech-talk/semiconductors/optoelectronics/processor-with-photonic-interconnects-built
 

Haravikk

Honorable
Sep 14, 2013
317
0
10,790
1
I wonder if these could appear sooner in applications like fibre-based networking? i.e- the computation would be comparatively simple (a network switch) but theoretically reducing the hardware complexity by pretty much doing the whole thing on the processor, albeit probably with some kind of amplification hardware for the external signals.

Getting these in computers too could be sweet, but a long ways off I expect; quantum processors have a lot of potential in certain applications due to the practically instantaneous calculation of complex problems, but configuring those problems is a very complex process, so I'm not convinced they'll be all that useful in general purpose systems like web-servers and gaming rigs. These photonic processors however sound very promising as a replacement to traditional silicon chips.
 

techguy911

Distinguished
Jun 8, 2007
1,075
0
19,460
58
Quantum computers are the next big thing Google already has one according to test is 100 million times faster than a single core cpu.
Next year they plan to buy a much larger one with NASA 1000 qubit model.
Also in Australia a group of scientist are working on a quantum computer printed on silicon.
D-Wave is the leading quantum computer mfg in the world.
I would say in about 10 years we will have hybrid computers with a quantum co-processor as they are for math and physics calculations only.
1000 qubits will be enough to process more "bits" of data than there are atoms in the observable universe if some docs I watched were telling the truth...

That said - I am not certain that q-computers will be ideal for all purposes. Precision is a big thing the way I understand it. They excel at certain types of computational tasks that conventional computers are abysmally bad at in comparison - Problems where brute iteration from step 0 - n to find an optimal solution would take too long. Problems that involve probability algorithms. Just like graphics processors can execute massively parallel tasks.

I am no expert on the matter, but it may very well be that the ideal chip of the future will have photonic GPU, CPU as well as QCPU cores all in one package. And it could be that QCPUs excel at graphics computations as well - I think its very possible.
You are right quantum computers are for only crunching math calculations they are math co-processors on steroids they are totally useless at graphics that is what Google is trying to tell every one maybe in 10 years or so quantum math co-processors + regular cpu technology will be on same motherboard.
Quantum computers would only help gaming computers by taking the load off the cpu for physics calculations that is where they excel.
http://www.theregister.co.uk/2015/10/06/quantum_pentium_appears_on_the_horizon/
 

Rhinofart

Distinguished
Jan 30, 2006
977
0
19,360
106
I for one don't want any part of that! Especially if they get things so fundamentally wrong such as the directions on a compass. When did East and West switch?
 

Darkbreeze

Retired Mod


No, it's really not like that. It's more like giving whatever car you're driving a straight shot on a desert highway with a guarantee of no tickets. However fast your mechanics enable you to go is your limit. Not the road itself.
 

Darkbreeze

Retired Mod


I still don't understand what any of that has to do with logic functions. Logic is a feature of the actual processing, being mathematical in nature, and must have something to do with the actual internal functions of the processor. I think there is more to this or something missing, or, the original article is wrong in saying that logic is relevant to the photonic nature of the device unless they are considering the mathematical nature of the data pipelines which is still intentionally misleading.
 

bit_user

Splendid
Ambassador
I think that must be the bottom of the chip. It was probably labeled to remind people working on it of that fact.
 

bit_user

Splendid
Ambassador
Um, all computation reduces to math, including graphics. It's right there in the word!

What current quantum computers excel at is energy minimization problems. And yes, due to the overheads involved, they aren't useful for graphics.
 
having high bandwidth on a chip is all well and good, but it does not replace the fact that the CPU itself and the RAM is still traditional silicone. It is like giving a Ford pinto full access to a 100 lane highway... kinda cool, but mostly useless.

There are still 2 things that will be really neat to come from this tech:
1) There is still active development for actual photonic CPUs (not just CPUs with photonic buses) which will be really cool in 40-50 years.
2) Much sooner than that, imagine this kind of tech being used as a buss system on a motherboard to replace something like PCI Express. This would allow nearly infinite bandwidth to things like GPU, RAM, and storage with far fewer pins, and lower power consumption (and presumably heat).

For more info see IEEE's report on the chip here: http://spectrum.ieee.org/tech-talk/semiconductors/optoelectronics/processor-with-photonic-interconnects-built
It reduces the power consumption significantly and is probably very good for signal integrity (hugely important for high capacity and/or high performance memory). This is in no way useless.

It's silicon, not silicone. One letter can make a huge difference :p

Darkbreeze is also correct in that your analogy is flawed. Regardless of that, it isn't just more bandwidth between parts of the CPU (which is much more useful than you realize), it's more bandwidth per square mm and that's hugely important. That alone saves power, reduces latency, improves yields, and reduces costs. Furthermore, simple improved bandwidth between parts of the CPU improves performance, especially multicore performance. I imagine signal integrity can also be improved as the light wouldn't have to deal with electrical resistance, potentially increasing maximum stable clock frequency.

I doubt fully photonic CPUs are even a realistic goal. Light-based memory is nowhere near as effective as electrical/electrically controlled memories, let alone any difficulties in making a practical CPU core that processes with light.

I can see your 2) suggestion being practical. Light-based data transmission has been shown time and time again to be superior in most ways to electrical data transmission and that's one of the next logical steps. I don't think storage pin count can be reduced much (SATA already only has four data pins), but PCIe data pin count certainly could be reduced.
 

bit_user

Splendid
Ambassador
Intel announced integration of a Omni-Path interconnect in its upcoming Knights Landing processor and Purley enterprise CPUs. I thought it would be optical, but now I'm not sure.

I wonder if we're not headed for a world where memory controllers move back off-chip. CPUs could have 8-16 GB of HBM stacked next to the cores (accessed by an integrated memory controller), and that would be all that desktop users get. But for servers & workstations, you'd have an optical interconnect for multi-CPU communication, interfacing with a set of shared memory banks, and of course the I/O backplane.
 
Status
Not open for further replies.

ASK THE COMMUNITY

TRENDING THREADS