Will optic ever replace copper interconnects? We ask a silicon photonics startup

https://www.theregister.com/2022/06/18/optical_interconnect_future/

by rntn • 4 years ago

55 16 4 years ago

imglorp• 4 years ago

Optical interconnect also solves a bunch of challenges offchip at human scale: RFI, impedance, crosstalk, power surges, but we keep shoring up cheap twisted pair copper because it kinda works. The cost per foot of the media is kinda similar but the transceiver gear is still expensive and needs commoditizing.

gizmo686• 4 years ago

Copper is also easier to work with. The wire is much more forgiving of bending, cutting and terminating it is easier, and is is more resilient to dirt on the endcap.

Also, the risk of power surges comes with the ability to transfer power, which is quite useful in many instances.

iLoveOncall• 4 years ago

And I'm sure if you ask tobacco companies they will tell you smoking is good for the health.

marcosdumay• 4 years ago

Yet they said you won't see any of it around any soon.

phoenix3200• 4 years ago

Photonics and copper both travel at about the speed of light. Photonics requires special circuitry to do the conversion to/from light. And don't get me started on how PCIe just isn't designed for optics, or really any kind of amplifier. Over a long distance, or due to signaling rate, it makes sense, but onboard it makes little sense unless we have optic switches which are faster than silicon.

dvh• 4 years ago

Wait till you learn about wavelength limit

tinco• 4 years ago

What's that?

gorkish• 4 years ago

Photonics and copper both travel at about the speed of light.

Just going to put this out there since it's totally non-obvious to a lot of armchair engineers, but this is not at all true. Velocity factor in most transmission lines (optical or electrical) is typically about 0.66c but varies depending on the material and the physical dimensions of the medium.

In very highspeed circuit design, (10s of ghz) both path length and velocity factor matching are considered when evaluating timing constraints for chipscale features.

rsynnott• 4 years ago

We ask a silicon photonics startup

Well, I mean, they're unlikely to say "no, we are simply wasting our time, for fun".

brk• 4 years ago

I guess based on Betteridge's law, we won't be expecting optic interconnects any time soon.

DesiLurker• 4 years ago

iirc the issue is that its very hard to do a optical transceiver on silicon in current node & cheaply. there were companies like luxtera trying to do it but I dont know what happened to them.

at one point I hoped that deep learning would bring about the change because of the large amount of data involved.

My naive guess is unless copper interconnects hit some kind of a wall and stop scaling we wont see widespread adoption.

jokabrink• 4 years ago

The article read a bit odd. You get to know a lot about said startup, who they are, what they do, how much funding they secured, etc.

For those who don't want to bother, the answer to the (IMHO clickbait) title is in the very last sentence:

However, he doesn’t expect this to happen anytime soon. “As you go into the millimeter scale, electrical I/O has, I think, a healthy roadmap in front of it,” he said. “Beyond 10-15 years, we might see… optical communication start to enter the millimeter scale regime.” ®

pvaldes2• 4 years ago

It does not matter who is better conductor or easier to work with, when the copper is being stolen each four months. This is the real factor in many countries, for a mile. Optic fiber is a game changer in that sense.

As long as optic fiber will be available doing a reasonable job at a reasonable price, many companies will jump over it like coyotes over a pile of chops.

otter-rock• 4 years ago

An "interconnect" here means the connections between components on a circuit board.

pvaldes2• 4 years ago

Oh, I see. I was thinking in railways. Said nothing then.

Scipio_Afri• 4 years ago

A lot of cables, other than ethernet, which are available in consumer products but are on the higher end of data transmission specs for 2021-2022... are able to meet these specs at lengths greater than about 6-12 feet (e.g hdmi 2.1+ or usb 3.1+) by very likely doing electrical to optical conversion, transmitting over an optical cable then converting back to electrical. This is especially true at the 15-20+ feet length.

If you've looked for the highest spec long length HDMI or USB cables you've probably seen the price jump somewhat significantly (from about $15 to $60-80+) when you hit a certain cable length. The ones that are long length but cheaper have many reviews stating they have failed to meet the data transmission rate, because they're not using the optical conversion technique and the ones that use it certainly cost more and are way more reliable (or simply aren't falsely advertised to be able to do something they can't, which is what the cheaper longer length purely copper cables tend to be).

With optical on the interconnects, you eliminate cross talk, parasitic capacitance, you reduce heat due to electrical resistance. It clearly does have its advantages. Just like chiplets and other design techniques that have become popular due physical constraints, optical will very likely one day too have the investment it needs to replace copper interconnects due to being worth the cost of investment by providing an advantage over them.

Will optic ever replace copper interconnects? We ask a silicon photonics startup

Related Stories