You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
The "copper cliff" is the biggest bottleneck in the data center today, but a new startup with a weird name thinks they’ve found a way to bridge the gap between heavy copper and expensive optics. In this video, we’re looking at Point2 Technologies and their E-Tube system—a hybrid interconnect that uses plastic waveguides to send millimeter-wave RF signals.
This was a very interesting listen. It sounds like it won't scale as high bandwidth-wise as fiber long term, but it finds a potential sweet spot in cost vs. distance and bandwidth inbetween copper and fiber.
The problem isn't so much long-term, it's real soon now...
Each "e-tube" carries two RF carriers at around 80GHz and 170GHz, each PAM4 modulated with 112Gbps, so 224Gbps per "e-tube" -- meaning, 8 "e-tubes" for 1.6T. There's no obvious way to scale this, doubling the RF frequencies is very difficult, higher modulation than PAM4 will run into SNR/BER problems.
That could be used to transport 8x224G down 8 "e-tubes" with a simple 1:2 reverse MUX at the input and MUX at the output. It's then competing against a 8x226G optical transport, which can use 8 fibres -- or 1 fiber with CWDM, which it can't do. If the 1.6T is 4x448G then you need 4:1 muxes but 8 "e-tubes". For 32T (8x448G) you need 16 "e-tubes". The problem rapidly becomes one of multiway connectors, as well as running out of bandwidth...
If copper still works, that's will carry on being the solution -- until it isn't. At that point you need so switch to something which is scaleable with increasing data rates without running out of connector poles in a few years time, which means optical and then WDM, and also means going CPO to avoid having to get all that data from the switch chip/GPU/AI chip to pluggable modules.
In theory they could take the same approach and move inside the package, but they'll very quickly hit the problem of connecting up large and increasing numbers of "e-tubes" because each one can only carry about the same amount of data as one optical carrier, but they can't be multiplexed together onto one fiber using WDM.
In theory they could take the same approach and move inside the package, but they'll very quickly hit the problem of connecting up large and increasing numbers of "e-tubes" because each one can only carry about the same amount of data as one optical carrier, but they can't be multiplexed together onto one fiber using WDM.
I tend to believe that CPO guys like LightMatter and Ayar Labs will provide better solutions - they seem to have figured out how to deal with the "limited shoreline" problem, at least for the next level of point to point complexity.
This was a very interesting listen. It sounds like it won't scale as high bandwidth-wise as fiber long term, but it finds a potential sweet spot in cost vs. distance and bandwidth inbetween copper and fiber.
Look at multimode fibre – another creature of marketing driven tech companies. They calculated that there must be a niche in the market in-between "expensive fibre" and "cheap copper". It faded. I never seen multimode fibre deployment anywhere outside of USA.
Look at multimode fibre – another creature of marketing driven tech companies. They calculated that there must be a niche in the market in-between "expensive fibre" and "cheap copper". It faded. I never seen multimode fibre deployment anywhere outside of USA.
Indeed a number of startups are attempting to find such sweet spot outside the mainstream SiPho/CPO, which will still need quite some time to get eco-system to HVM. Attotude is another one besides Point2, going up frequency in RF spectrum. A few others not very public yet are looking at plastic etc ...
One other problem with CPO is the SiPho tie at the moment, which is limiting the further performance and bitrate scaling at 400G/lane and beyond.
Indeed a number of startups are attempting to find such sweet spot outside the mainstream SiPho/CPO, which will still need quite some time to get eco-system to HVM. Attotude is another one besides Point2, going up frequency in RF spectrum. A few others not very public yet are looking at plastic etc ...
One other problem with CPO is the SiPho tie at the moment, which is limiting the further performance and bitrate scaling at 400G/lane and beyond.
If I continue the line about what has happened 15 years ago in the fibre market as an analogy to today: 112G over single fibre will become a reality soon, and a mainstream after.
112G over a single fibre will be enough for the industry for quite long while.
People with good memory will recall that there used to be an argument SMF can never be cheaper than the analogous copper solution, where copper's physical limits still allow, and so short range 10G will always stay with copper, or MMF. And so, what happened? SMF just got cheap enough as it had far bigger adoption already outside of the LAN niche. Massive FTTH adoption in Asian countries has driven the prices of the whole ecosystem down.
