Next gen chip-to-chip optical interconnect

[Keith Lofstrom]

I've written before that the best first use of space solar
power is computation in orbit, beaming results to Earth
rather than the megawatts/gigawatts...exawatts that power
the computation, dissipating "cold" waste heat directly
into the 2.7K void, rather than my favorite biosphere.

I can yammer on about the intrinsic rad-hardness of very
small, deep submicron, redundant transistors and gates
on thin unshielded chips, but when the circuits get
very small, the signal wires connecting them might
comprise most of the mass of the system.

However:

The NEXT BIG THING in electronics, both terrestrial and
(someday) off planet, is between-chip optical interconnect
with lasers, etalons, and wavelength division optical
multiplexing on chiplets bonded to the top of the silicon.
Perhaps 200 THz per fiber, approaching petabits per second,
packable to kilofiber per square millimeter density.
An unclad optical fiber in vacuum is cheap, thin, and very
very low mass.

Upcoming online IEEE Solid State Circuits Society
presentation about the chiptech on August 12,
link here: https://sscs.ieee.org/events/

I'm currently paying a mere $60 per month for 200/200
megabit-per-second internet fiber; my provider
(northwest company Ziply fiber) gives me 350/350 for
the same price because the hardware is so damned
cheap. A little extra Just In Case of minor defects.
My fiber connection is combined with dozens of
neighbors using a wee bit of etched glass called an
etalon, which acts as an optical wavelength-division
multiplexer. The fiber bundle from their switching
center 10 miles south to eastern Beaverton (tens of
thousands of customers like me) is pencil diameter,
including armor.

In the near future, when cheap chips route petabits,
data bandwidth may be "too cheap to meter". The optical
fibre itself is nearly weightless, and absurdly rad-hard
(after error-correction for the occasional scintillation
photon).

Such technology will obsolete ancient satellites that are
big shielded boxes of circuit boards and wire bundles.
We will still need to move DC power watts from the P.V.
to the ULSI digital chips - but those power watts will
enable vastly more computation.

Information physics tells us that a bit can
(theoretically) be represented with ln2 kT joules;
If we can radiate at 200 Kelvins into 2.7 Kelvin deep
space, Shannon limit ln2 kT approaches 12 milli-electron
volts: 5e20 bits per watt-second. I have no clue how to
accomplish this with 2024 technology ... but the sky is
no longer the limit.

I've written more about this; you can find some of it
at http://server-sky.com. I need to upgrade that server
to Debian 12 (help!?), but when that is done I will add
more information about the best ways to do VAST amounts
of computation, near the Earth-Sun L1 and L2 points.

10 seconds round trip speed of light delay; but if the
computation is kilo-Earth-power-scale AI, I want enough
buffer time to dynamite the terrestrial transceivers,
when AI starts going off the rails.

Thats a lot to grok. My wife informs me that I am
going for a walk now. Later!

Keith L.

--
Keith Lofstrom kei...@keithl.com