IIRC, Intel uses tape in for internal manufacturing and tape out for external manufacturing in the same way. I'm less than convinced by attempts to draw meaningful conclusions from their use of that term. More to the point, perhaps:
In an earnings call, Intel CEO Bob Swan said the company identified a 'defect mode' in the 7nm process that resulted in 'yield degradation' when manufacturing the chips.
www.pcmag.com
If not Meteor Lake, then what would that be? Makes far more sense than a stopgap product pushing things out by a whole year.
No first product on a node tapes out on a 1.0 PDK. But that distinction doesn't matter in this context.
Well, even Gelsinger tripped up and used tape-in and tape-out in the same sentence, and
the corresponding tweet from Gregory Bryant is notable for using the progressive form of the verb, i.e. "taping in". I believe the consensus was that "taped-in" indicates that design work has been completed, however the final mask images have not yet been generated (traditional tape-out), nor do they have a complete set of photomasks (which is a prerequisite for wafer production). So the start of Meteor Lake risk production will likely be 1 to 2 quarters out from May 24, 2021, which would be sometime in the second half of this year.
Your other quote is from
Bob Swan during the Q2 2020 earnings call on July 23, 2020, nearly a year ago:
We now expect to see initial production shipments of our first Intel-based seven-nanometer product, a client CPU, in late '22 or early '23.
We are also focused on maintaining an annual cadence of significant product improvements independent of our process road map, including the holiday refresh window of 2022. In addition, we expect to see initial production shipments of our first Intel-based seven-nanometer data center CPU design in the first half of '23.
Ignoring the fact that Intel has updated both their CEO and roadmap since then, you still need to parse that statement correctly. The term "initial production shipments" when addressing shareholders means the same thing as "shipping for revenue", which hinges entirely on the Product Release Qualification (PRQ) engineering milestone. What he's actually saying is, "We hope to achieve Meteor Lake lead-die PRQ during Q1 2023." (I don't care how much of an eternal optimist you are, if you expect Intel to ever hit the earlier side of a stated production window, then you're only setting yourself up for disappointment.) RTS will trail that by 1-2 quarters as Intel undertakes the volume ramp and channel fill, which puts you right at an August 2023 launch date. Consider Intel's press embargo dates over the past eight years:
Jun 02, 2013: Haswell U/H/S
Sep 01, 2013: Haswell Y/U
Sep 05, 2014: Broadwell-Y
Aug 05, 2015: Skylake-S (Core 17-6700K and Core i5-6600K only)
Sep 01, 2015: Skylake Y/U/H/S
Aug 30, 2016: Kaby Lake-Y/U
Aug 21, 2017: Kaby Lake Refresh-U
Aug 28, 2018: Amber Lake-Y / Whiskey Lake-U
Aug 01, 2019: Ice Lake-Y/U
Sep 17, 2020: Tiger Lake-UP4/UP3
The small-die Y/U platforms have consistently launched within a 7 week window in Q3 each year, and have represented the lead die for every generation except for Haswell and Skylake. Which is also why I still think we might see an Alder Lake-P announcement in August this year.
Intel 7nm risk production has been underway for quite some time at this point with Ponte Vecchio. And I'll allow my unrelenting optimism show through for a moment and say that I think they'll be able to ramp just fine with much shorter cycle times and better defect densities and overall yields than 10nm. I think part of the reason Intel failed so hard on 10nm was because they were trying to achieve densities that were nearly impossible without EUV. However, the term "compute tile" being used in conjunction with Meteor Lake also indicates advanced packaging techniques will be used. So even if cycle times are improved on 7nm, test, sort and packaging will almost certainly take longer. So I'm expecting a similar amount of time between PRQ and RTS as Ice Lake and Tiger Lake.
ark.intel.com
Facebook
Twitter