The technology Rolland is referring to is a "miniaturized on-die version" of Intel's silicon photonics technology that's "used as a super high-speed optical interconnect between a Xeon server CPU and an Altera FPGA."
Rolland says this technology will allow Intel to build "best-in-class CPU/FPGA combinations proprietary to Intel."
Here's why I'm not as enthusiastic as Rolland is about that technology being used for this application.
Intel's Altera end game
Back in March, Intel said it would begin offering samples of chips with Xeon processors and FPGA chips from Altera, now Intel Programmable Solutions Group, integrated onto the same chip package later in the year.
Putting the two chips onto the same package does deliver some efficiency improvements over connecting two discrete components -- say, an FPGA via a PCI Express link -- on the order of two times.
However, Intel has indicated that it can do better.
Intel's end-game vis-a-vis Altera's FPGA technology is to take Altera's core FPGA technology and build it into the same piece of silicon as the main processor. By integrating the technologies into a single silicon die, rather than as two distinct dies on the same package, Intel expects to be able to realize further performance and power efficiency benefits compared with what a co-packaged part can deliver.
It's hard to see how a technology that allows two discrete chips to communicate data more quickly -- when eventually the FPGA and the CPU will both be in the same piece of silicon -- will eliminate the need for a "super high-speed optical interconnect between a Xeon server CPU and an Altera FPGA."
Putting Intel's silicon photonics efforts into perspective
Even if Intel's silicon photonics technology isn't a big deal for the company's combination CPU and FPGA parts, the technology seems to represent a significant opportunity for Intel. At Intel's late 2015 investor meeting, Intel data-center group chief Diane Bryant said that by 2020, the total addressable market for the company's silicon photonics technology will be approximately $5 billion.
It's not clear how much of that revenue opportunity Intel will be able to capture by that time. However, even 20% of that would be about $1 billion in incremental revenue for the company's data-center group. Considering that Intel's data-center business generated about $16 billion in revenue during 2015, an extra $1 billion in revenue is by no means chump change.
We should get an update on this tech in February
Intel is planning to host an investor meeting in early February of next year. At this event, the company is likely to delve deeply into what kind of growth rate it expects in its data-center group over the next several years. It'll be interesting to learn management's current sales and market-share expectations around silicon photonics for both the near term and the longer term, as Intel reportedly began volume shipments of silicon photonics products earlier this year.