Production of Photon Processors Expected in 2006 217
ThinSkin writes "Photon processors that transmit data via light, not electrons, are slated to enter production in mid-2006, ExtremeTech reports. Headed by a UCLA professor and a Nobel Prize winner, startup Luxtera claims that its optical modulator clocks in at 10-GHz, tens times that of Intel's optical modulator researchers talked about last year. Since the optical module exists as its own entity, it will require a standard CMOS processes to integrate the optical waveguides. Luxtera has worked closely with Freescale Semiconductor to develop this technology."
Re:Error In The Article (Score:5, Informative)
Article Text (Score:5, Informative)
minus the omniture spyware tracking and massive banners
________
Startup Luxtera has announced its plans to enter the CMOS photonics market, anticipating the day when microprocessors will transmit information via light, not electrons.
The company claims that its optical modulator for transforming electrons into photons runs at 10-GHz, ten times the speed of an optical modulator Intel Corp. researchers began talking about last year. Beginning in mid-2006, Luxtera hopes to enter production of photonic devices using standard CMOS manufacturing processes. ADVERTISEMENT
Although the majority of chip-to-chip communications are conducted using copper-based interconnects, researchers are already looking toward the day when the balance shifts toward optical transmissions, initially for chip-to-chip interfaces between microprocessors, or between a microprocessor and memory device. Fibre optics are a standard component of modern telecommunication infrastructures, and interfaces such as Fibre Channel also use optical fibre interconnects to link up devices.
Although light slows down by some degree when transmitted through an optical medium, shifting to optical-based components is still too expensive than relying solely on copper, even when factoring in the additional power, heat, and crosstalk issues.
"The problem here that we can solve is a matter of bandwidth," said Gabriele Sartori, Luxtera's vice president of marketing and a former advocate for the HyperTransport protocol developed by Advanced Micro Devices.
Part of the relatively high cost of photonics comes from the fact that converting electrons to photons requires an intermediary device, such as the modulator Luxtera is designing. Today, that device exists as a separate module. Intel, Luxtera, and others are trying to integrate the optical waveguides within standard CMOS processes, that can be controlled by the standard voltage swings of a microprocessor.
However, doing so requires that the optical vendor have close ties to a microprocessor manufacturer. At Intel, that's no problem. Luxtera, on the other hand, has worked closely with Freescale Semiconductor to develop the technology. Finding a partner like Freescale is "necessary," Sartori said. "You must walk before you can run."
Freescale has taped out several engineering samples of the optical technology, including a chip, one side of which includes the optical interface built in. The sample chip use a 130-nm SOI process, the same technology used to fabricate the G4 microprocessor. Part of Luxtera's job has been to develop silicon libraries, the files used to design the photonic chips in the same way other libraries serve as the blueprint for making more conventional semiconductors.
The 32-employee startup originally received $7 million funding from Sevin Rosen Funds and August Capital in 2001, followed by an additional $15 million by New Enterprise Associates in 2003. Eli Yablonovitch, a professor at UCLA who developed photoelectronic crystals, sits on the company's board, while Arno Penzias, who won the 1978 Nobel Prize for his work on the Big Bang theory, serves in an advisory role. Other board members include Andy Rappaport of August Capital, which funded Transmeta, among others.
Re:Error In The Article (Score:1, Informative)
Re:10Ghz? (Score:2, Informative)
The performance gain is up to the chip designers, who will design a chip as fast as they can. That wasn't the problem they were trying to solve.
Rather, the problem this addresses is off-chip interconnect. Today chips communicate with the rest of the system via solder joints; this provides for very limited bandwidths, far far less than 10 Giga items per second. This is mostly because process improvements that have allowed us to shrink our chips have not allowed us to shrink our solder joints. So off-chip bandwidth has not been scaling well over the years and is a significant bottleneck.
Re:Uh, okay (Score:5, Informative)
Whoever can afford them.
Are these like only special coprocessors for million-dollar supercomputers?
No. These are not "processors" of any sort. It is a new way to modulate signal between CMOS and optical at high frequency and small scale. It may provide faster bus speeds, assuming the reality matches the funding hype.
Are they going to be x86-compatible? MIPS compatible? What?
It will be "compatible" with any CMOS device that needs a bus to communicate with some other device. Since that includes all useful CMOS devices, it will be compatible with everything!
Re:Uh, okay (Score:2, Informative)
--
Want a free iPod? [freeipods.com]
Or try a free Nintendo DS, GC, PS2, Xbox. [freegamingsystems.com] (you only need 4 referrals)
Wired article as proof [wired.com]
More detailed article at Forbes (Score:4, Informative)
Interestingly, the 10Ghz figure comes from a measurement made a researcher at Sun Labs, who have been working with Luxtera for more than a year now. The article also talks about what other companies such as Intel and IBM are up to.
FYI Freescale=Motorola (Score:5, Informative)
Just thought I'd clear up that potential confusion...
solution for wiring problem? (Score:5, Informative)
A major problem as CMOS processes get smaller and smaller is wires and wiring. Its really bad at 90nm and it looks like its going to be way worse at 65nm.
Even if optical interconnects can just be used for long intra-unit busses (think L1 cache to fetch/decode unit, and there to integer unit and float unit, etc) we could see great performance gains.
Something like when the upper metal layers in CMOS went to copper a few years ago.
Optical interconnects (Score:5, Informative)
Other groups working on optical interconnects: (incomplete list)
Heriot Watt [hw.ac.uk]
Cornell University [cornell.edu]
IBM Zurich [ibm.com]
Delft [tudelft.nl]
UIUC [uiuc.edu]
Intel [intel.com]
Stanford [stanford.edu]
New meaning to deeply pipelined architectures (Score:4, Informative)
Its always interesting to see what happens when the relative speeds of processor, memory, and interconnects change.
Re:Error In The Article (Score:2, Informative)
Photons are light. Protons are not.
A Proton is a neutron with a positron.
Electrons, positrons, protons, and neutrons are particles with mass and they are not light.
When electrons or positrons move (current) they produce electro magnectic waves which are light.
What they develop is the ability of devices like CPU and memory to comunicate using light and thus giving them more bandwidth.
It's a small step toward faster computing and eventualy quantum computing...
Re:Error In The Article (Score:3, Informative)
So they are what, heavy? It's a joke. Perhaps a little too subtle, but a joke none the less. Laugh.
Re:maybe someone smarter than I can.... (Score:4, Informative)
That's not an issue here, from what I can tell. The 10 GHz number is modulating light to electrical signals. All the actual storage and processing will be done just as before; you still have your Flip Flops and storing the bits. The only difference here is that instead of copper interconnects, we use light pulses. The benefit of this new technology is that it can be done with normal CMOS fabrication techniques.
Anyone with more experience with this stuff is free to correct/clarify.
Slashdot mislead (Score:5, Informative)
If you read the article carefully (which is laced with marketing hype and was obviously written by someone only passingly familiar with the technologies involved), you will see that nobody's promising optical cpu's in 2006. In anticipation of future optical chips and other technologies, Intel has begun developing one of the stepping stones toward this technological era, which is an optical/electrical gateway of sorts which can be built on a standard electrical chip to allow it to interface with optical components. Think a modern cpu, with some low level optical/eletrical interface on the edge of it so that a row of optical "pins" can stick out one side in addition to the normal electrical pins on the bottom.
This little startup company is working on the same thing, and hopes to have it out soon. Their marketing article is trying to build hype so they can get more cash. Nobody will be selling anyone an all-optical cpu in 2006 (or 2007, or 2008, etc).
Latency != Frequency (Score:5, Informative)
It may take a few nanoseconds for the light to bounce around, but that light can be modulated at extremely high rates (that electrical wires cannot). Managing latency is a well understood problem, generally solved by using speculation, buffering, etc..
The fact is, if these parts are running at 10ghz, you will have 10ghz connections between connected parts (with a few nanoseconds of latency, which is mostly irrelevant).
Bandwidth is a measure of frequency and number of communication channels. This advancement does indeed provide more bandwidth, mostly because it can be clocked higher. All computer configurations could see substantial benefits because current electrical designs have highly limited bus speeds (it is not signal propagation that matters, but signal modulation speed "frequency").
Again, signal propagation speed is mostly irrelevant. Signal modulation speed is what is important. Latency != Frequency.
Re:Error In The Article (Score:3, Informative)
Re:Error In The Article (Score:4, Informative)
A Proton is a neutron with a positron.
No, it's not. A proton is three quarks. From Wikipedia [wikipedia.org]:
Protons are classified as baryons and are composed of two Up quarks and one Down quark, which are also held together by the strong nuclear force, mediated by gluons. The proton's antimatter equivalent is the antiproton, which has the same magnitude charge as the proton but the opposite sign.
A neutron may decay into a proton+electron pair, but a proton is most definitely not composed of neutron+something else. If nothing else, this should be the proof: neutron is heavier than proton---by conservation of mass and energy, neutron cannot be a component of proton.
When electrons or positrons move (current) they produce electro magnectic waves which are light.
No, it's not when they move that they produce EM waves. It's when they accelerate that it does (if you had been a physicist, this difference would have been carved into your very being). Moving charge only creates a magnetic field, which doesn't necessarily propagate as an oscillating field in space (i.e. EM wave). What you need is not a current but an alternating (as one example) current.
It's a small step toward faster computing and eventualy quantum computing...
Er... I know that you don't know what you are talking about, but this has nothing to do with quantum computing. (O.K. I haven't RTFM (nor do I have interest or time to do so), so I may be wrong on this, but...) This development is analogous to moving to fiber optics from copper cables---it does use a less "lossy" and perhaps faster medium, but it is in no way related to quantum computing.
Re:Uh, okay (Score:3, Informative)
Really? I wasn't aware that Freescale made ARM processors, too. After all, when it comes to microprocessors, they're primarily known for 68k and PowerPC.
Re:Error In The Article (Score:3, Informative)
The nucleus of every atom (except for hydrogen, obviously) is smaller than the sum of the masses of its nucleons. If your proof were valid, then they couldn't consist of the nucleons they consist of. The point is that the binding energy also contributes to the mass, and since the binding energy is negative (the bound state has less energy than the unbound state), this means it reduces the mass.
No, the ultimate proof that a neutron is not part of the proton are accelerator experiments which agree with the established theories where the proton is not a neutron with a positron.
Re:Latency != Frequency (Score:2, Informative)
The extra bandwidth does indeed allow more in-flight memory accesses, but there are many problems involved with this.
First of all, there are implicit problems in the memory-level parallelism in applications. How many memory accesses are independent of each other? For example, code that manipulates hash tables or linked lists does not profit from additional bandwidth because the next memory access depends on the current one. Such code normally does things like:
LD R1, 0(R1)
do something on R1
LD R1, 0(R1)
etc
See the dependency on R1
Second, there are problems with the microarchitecture. Microprocessors contain two structures necessary to handle off-chip memory accesses: The Load/Store Queues (LSQ) and a special table to track these external accesses (called something like the miss address table, where miss refers to the corresponding cache miss). The LSQs track all in-flight loads and stores and they are used to check dependencies between loads and stores. The largest implementation I remember can track a total of 48 loads (is it p4?). The Miss Address Table contains references to all off-chip accesses. This table is usually smaller. Thus, even if all the memory accesses are independent and can be issued in parallel you cannot really take profit of all that bandwidth. Theoretically you can issue hundreds of memory accesses during the time an off-chip access is in progress. In reality you will end up with 20 or 30 (that does not include prefetches and alike).
sorry for not including references