Report: Intel 4 Process is On Track For Production in 2022
Intel has been hard at work over the last few years trying to recover from its years-long stumble on 10nm. To catch up with its competitors, most notably TSMC, Chipzilla has launched an audacious plan. Catching up and then overtaking TSMC requires Intel to advance five nodes in four years, which is an unprecedented pace of development. Given the company’s recent track record you would be in good company if you were skeptical of its ability to adhere to such a schedule. Intel, however, seems hellbent on following through with it.
To these ends, a new report from Digitimes claims sources say Intel is right on schedule for its next major node shrink. It will be transitioning from Intel 7 (10nm) to Intel 4 (7nm) later this year. It’s the first major node shrink for Intel since it moved from 14nm to 10nm in 2018.
Digitimes cites sources within Intel confirming Intel’s right on target for Meteor Lake according to Tom’s Hardware. It’s expected to enter high volume production in the second half of 2022. This will be a breakthrough product for Intel. It’s not only its first 7nm CPU, but its first tile-based chip as well. It also marks the beginning of Intel’s use of EUV lithography, making it the last of the big foundries to adopt this technology. Intel has lofty goals for its 7nm process. It expects to offer a 20 percent performance boost in the same power envelope as Intel 7. It can also offer a 40 percent reduction in power at the same clocks. Intel says this node will be optimized for high-performance silicon. Thus, it will likely be used for its own processors as opposed to products for foundry customers.
Although Intel’s 12th and 13th generation CPUs use hybrid cores, they still use a monolithic design. That ends with Meteor Lake, as the company switches to a tile-based layout. This will mark the first time the company combines tiles from different nodes on one package. It will create the I/O and CPU tiles on Intel 4. The other two will be made by TSMC. The GPU tile will be a 3nm part, while the SoC will be N4/N5. If all that wasn’t enough to chew on, it’s also using Foveros to connect the dies laterally and vertically.
As you can see, Intel 4 is a lot of “firsts” for the company. Several novel technologies have to come together perfectly, making it quite the engineering feat. Now, you could say AMD has been doing tiles for years now, and Intel is late to the party. That’s also true, but as you can see the implementations are radically different. Either way, it’s a good sign that Intel is on track to deliver on its roadmap. If it’s able to successfully execute this node shrink, it should help silence the naysayers. If that’s not enough, Pat Gelisinger recently brandished an 18A wafer (top of article). He bragged that the company’s most advance node is already six months ahead of schedule. It is due to arrive in 2024.