https://www.linkedin.com/search/results/content/?fromMember=%5B"ACoAAACE44oBATlJnoYRAo3-jKfi-JfW2kTuRnA"%5D&heroEntityKey=urn%3Ali%3Afsd_profile%3AACoAAACE44oBATlJnoYRAo3-jKfi-JfW2kTuRnA&keywords=mark%20lutkowitz&position=1&searchId=89d98f44-c204-4bbf-ad64-b45cc89615c4&sid=xTu&update=urn%3Ali%3Afs_updateV2%3A(urn%3Ali%3Aactivity%3A7106618682453159936%2CBLENDED_SEARCH_FEED%2CEMPTY%2CDEFAULT%2Cfalse)
Mark LutkowitzMark Lutkowitz• Following• FollowingPrincipal at fibeReality, LLCPrincipal at fibeReality, LLC
2d • 2d •
Disappointingly, this past week, CIOE 2023 was full of linear-drive pluggable optics (please see: https://lnkd.in/evpGybMM) demonstrations, including for DR8 (please see: https://lnkd.in/e4DkNwnK) and SR4 modules. A main reason for the hype is the challenging conditions in the economy of China, while the data center/AI sector stands out in terms of growth. At present, many Chinese chip companies want to emulate MACOM in offering a linear driver and TIA chipset for LPO, seemingly to the exclusion of pursuing other IC business opportunities. (Of course, optics arguably has not been a top priority at MACOM, since the corporate leadership took power in 2019.) At least for the US hypersclalers, it is hard to imagine any of them giving up the assurance of link integrity in exchange for avoiding a cost premium for the DSP. The fanatical narrative behind LPO, which began in earnest at OFC 2023, is akin to proclaiming that a coherent transceiver must plug into a IMDD device, and each must continue operating, while complying with full specifications. On DR optics specifically – how does one make it practically work with LPO, given the bandwidth and the noise considerations?
While fibeReality understands that the results shown for VCSELs looked kind of good, in some previous crash programs (please see: https://lnkd.in/ev9RxNWS), VCSEL-based LPO could not meet any Ethernet specs. Yet, despite being less linear than silicon photonics (please see: https://lnkd.in/eV-Pb8HQ and https://lnkd.in/evQNmzU2), for non-Ethernet applications, it would still be fine for distances of 25 to 35 meters. At those lengths, we view VCSEL-based linear for active optical cables (the actual anticipated killer app for linear), as being the winner (please see: https://lnkd.in/e_9fywcq and https://lnkd.in/eA_hDczy). In fact, there is apparently a big expectation by optics suppliers at the show that Alibaba will be a primary purchaser of such AOCs, eventually culminating in a total market of tens of millions of LPO devices being shipped. Interestingly and unsurprisingly, there was not too much of an overt marketing push on linear AOCs at CIOE, other than to mention it as a future business. Suppliers did list LPO AOCs on their booth posters, such as ATOP Technologies, referring to 800G and 400G rates, using OSFP and QSFP112 form factors, respectively.
Activate to view larger image,
