Earlier this year, chip giant Intel (INTC -0.38%) launched a new data center platform known as Purley and, at the heart of the platform, was a new family of processors based on the Skylake Server architecture.
The processors are built using Intel's 14nm+ technology, a second-generation version of the 14nm technology used to build the company's prior generation Broadwell-based data center chips.
Image source: Intel.
Next year, Intel is expected to release a revised version of the Skylake Server processors, known as Cascade Lake. The company hasn't disclosed much in the way of details about these chips, but Intel has indicated that they will support Intel's 3D XPoint memory modules, a platform feature that was originally targeted for the original Skylake Server chips but was pushed out.
Additionally, it stands to reason that the updated chips will take advantage of Intel's 14nm++ manufacturing technology to help deliver performance and power efficiency improvements.
Per a tweet from David Schor, an industry analyst who publishes on WikiChip.com, Intel is planning to launch the successor to Cascade Lake, known as Ice Lake Server, in May of 2019.
Can Intel hit this schedule?
Ice Lake Server is likely to be manufactured using Intel's second-generation 10nm technology, marketed as 10nm+. Intel's first, and likely only, product to be built using its first-generation 10nm technology is expected to go into production later this year, with shipment volumes ramping up during the first half of 2018.
If Intel is planning to launch this processor in May 2019, then it'll need to begin mass producing it in late 2018 or early 2019, depending on several factors.
Generally speaking, Intel's data-center specific chips are very large and are therefore quite difficult to manufacture. That inherent difficulty is compounded by the fact that Intel's 10nm+ technology won't have been in mass production for very long before these chips are released.
Large, complex chips on a relatively immature manufacturing technology is risky.
One way that Intel could possibly get around this is to do exactly what the company said it planned to do at its investor meeting back in February: use EMIB.
Former Intel Data Center Group (DCG) chief Diane Bryant said then that one of the obstacles to having its data center chips adopt new manufacturing technologies first was the fact that data center chips tend to be quite large.
Intel's plan to get around that is to partition its future chips into smaller pieces, build the elements that can benefit from the latest manufacturing technology using that technology, build the parts of the chip that can be kept on older technologies without compromising performance on those older technologies, and connect them using Intel's Embedded Multi-Die Interconnect Bridge technology, or EMIB.
Intel didn't say when it planned to utilize its EMIB technology in its mainstream PC processor or data center processor products (the technology is being used to build Intel's Stratix 10 FPGA), but using it to build the Ice Lake Server parts could make it easier for Intel to hit its designed product launch schedule.
Not only would it be easier to build a chip out of smaller building blocks (smaller chips are easier to build than bigger ones, all else equal), but it could cut down on the time it'd take Intel to develop and validate the design.
So, I'd say this: Intel using EMIB to build Ice Lake Server wouldn't be a guarantee that the chip-maker could hit its schedule, but it's almost certainly make it easier for Intel to hit it.
